Datasets:

eswardivi

/

books

like 0

e up-conversion/ down-conversion of the band-pass signal transmission. Ideally, the spectrum of the oscillator is expected to have an impulse at the frequency of oscillation with no frequency compo- nents elsewhere. However, the spectrum of a practical oscillator's output does have random variation around the oscillation frequency due to phase noise. The impact of local oscillator phase noise on the performance of an OFDM system has been extensively studied in the lit- erature [30]. It has been shown that phase noise may have significant effects on OFDM sig- nals with small subcarrier spacing (i.e., large OFDM symbol duration in time). Long symbol duration is required for implementing a long guard interval that can mitigate long multipath delay in single-frequency operation without excessive reduction of data through- put. The studies suggest that phase noise in OFDM systems can result in two effects: a common subcarrier phase rotation on all the subcarriers and a thermal-noise-like subcarrier de-orthogonality. The common phase error, that is, constellation rotation, on all the demodulated subcarriers, is caused by the phase noise spectrum from DC (zero frequency) up to the frequency of sub- carrier spacing. This low-pass effect is due to the long integration time of the OFDM sym- bol duration. This phase error can in principle be corrected by using pilots within the same symbol (in-band pilots). The phase error causes subcarrier constellation blurring rather than rotation. It results from the phase noise spectrum contained within the system bandwidth. This part of the phase noise is more crucial, since it cannot be easily corrected. The SNR New Radio Access Physical Layer Aspects (Part 1) 329 degradation caused by the common phase error can be quantified as SNRphase-rotation = [I(a),BAf] where Af is the subcarrier spacing, B denotes the upper bound of the phase noise spectral mask, a is the ratio of the equivalent spectrum mask noise bandwidth and the subcarrier spacing, and I(a) = with I(0.5) = 0.774, I(1) = 0.

903, and I(00) = 0.774. It can be seen that, when a > 1, the com- mon phase error decreases as the subcarrier spacing decreases. To mathematically model the effect of the phase noise, let's consider the noisy output of an oscillator which contains phase noise 4(t) as follows v(t) = Let's further assume that the stochastic variation of the phase can be modeled as the output of a system with a step function impulse response and input perturbation n(t) as follows: Based on the above assumption, the single-sided power spectral density (PSD) of the phase can be written as Sq(f) = Sn(f)/(2r)2 in which Sn(f) denotes the noise PSD function. As an example, if Sn(f) is modeled as white noise, then Sq(f) 2 f-2 and if Sn(f) is modeled as flicker noise, then Sy(f)~f-3. Considering that the PSD of the phase is difficult to observe, one may alternatively look at the PSD of the oscillator's noisy output v(t). It can be shown that the PSD of v(t) can be calculated as follows [30]: where {ak} and {bk} denote the Fourier series coefficients of v(t) and B is a constant. Given that we are only interested in evaluating Sv(f) at fc, the above equation can be simplified as follows: The above function is a Lorentz distribution. We now define function If(f) as the ratio of noise power in 1 Hz bandwidth at offset f from center frequency to carrier power which is expressed in dBc/Hz. As theoretically expected, having a higher phase noise in the signal 11 The Cauchy distribution is a continuous probability distribution which is also known as Lorentz distribution or Cauchy-Lorentz distribution, or Lorentzian function. It describes the distribution of a random variable that is the ratio of two independent standard normal random variables, with the probability density function f(x; 0,1)= [n(1+x2)] 330 Chapter 3 near does fc not and increase the total power. A signal with higher phase noise will have smaller with lower will have a broader spectrum around the center frequency. Conversely, a power signal Therefore su(f) can be expressed as follows

efficient frequency-domain equalization at the receiver. The DFT-spread-OFDM generated in frequency domain, similar to OFDM as illustrated in Fig. 3.14, where the signal mon processing blocks in OFDM and DFT-S-OFDM are distinguished from those that com- are downlink specific to DFT-S-OFDM. This allows for a relatively high degree of commonality with the OFDM baseband processing using the same parameters, for example, clock fre- quency, subcarrier spacing, FFT/IFFT size. The use of DFT-S-OFDM in the LTE uplink Channel Processing block specific to DFT-S-OFDM Figure 3.14 Transmitter structure for DFT-S-OFDM with localized subcarrier mapping schemes (note that NDFT < NFFT) [47]. New Radio Access Physical Layer Aspects (Part 1) 331 was mainly due to relatively inferior PAPR properties of OFDM that resulted in worse uplink coverage compared to DFT-S-OFDM. The PAPR characteristics are important for cost-effective design of UE's PAs. The principles of DFT-S-OFDM signal processing can be explained as follows. The ith transmitted symbol in a DFT-S-OFDM system without CP in single transmit/receive antenna case can be expressed as a vector of length NFFT samples defined by = FODx where X = XM) T is an NDFT 1 vector with NDFT QAM-modulated symbols (the superscript "T" denotes matrix transpose operation), D is an NDFT X NDFT matrix which per- forms NDFT-point DFT operation, O is the NFFT X NDFT mapping matrix for subcarrier assignment, and F performs NFFT-point IFFT operation. After the propagation through the multipath fading channel and addition of the AWGN and removing the CP and going through the NFFT-point FFT module, the received signal vector in the frequency domain can be expressed as =F-1-HFODx + W where H is the diagonal matrix of channel response and W is the noise vector. Note that the maximum excess delay of the channel is assumed to be shorter than the CP; therefore, the ISI can be mitigated by the CP. The amplitude and phase distortion in the received signal due to the multipath channel is compensated by a frequ

ency-domain equalizer (FDE) and the signal at the FDE output can be described as V = Cz where CNFFT) is the diagonal matrix of FDE coefficients. The FDE complex coefficients can be derived using minimum mean square error (MMSE) criterion as Ck = 02/02) where k denotes the subcarrier index, o2 denotes the variance of the additive noise, and o2 is the variance of the transmitted pilot symbol. Following the subcarrier demapping function and IDFT despreading, an NDFT X 1 vector X containing NDFT QAM-modulated symbols as an estimate to the input vector X is obtained at the receiver. The IDFT despreading block in the receiver averages the noise over each subcarrier. A particular subcarrier may experience deep fading in a frequency-selective fad- ing channel. The IDFT despreading averages and spreads the fading effect, which results in a noise enhancement to all the QAM symbols. Therefore the IDFT despreading makes DFT-S-OFDM more sensitive to the noise. As shown in Fig. 3.14, the modulation symbols in blocks of NDFT symbols are processed through an NDFT-point DFT processor, where NDFT denotes the number of subcarriers assigned to the transmission of the data/control block. The rationale for the use of DFT pre- coding is to reduce the cubic metric of the transmitted signal. From an implementation point of view, the DFT size should ideally be a power of 2. However, such a constraint would limit the scheduler flexibility in terms of the amount of resources that can be assigned for an uplink transmission. In LTE, the DFT size and the size of the resource allocation is lim- ited to products of the integers 2, 3, and 5. For example, the DFT sizes of 60, 72, and 96 are allowed, but a DFT size of 42 is not allowed [30]. Therefore, the DFT can be implemen- ted as a combination of relatively low-complexity radix-2, radix-3, and radix-5 FFT proces- sing blocks. The subcarrier mapping in DFT-S-OFDM determines which part of the 332 Chapter 3 spectrum is used for transmission by inserting a number of zeros (i.e., null subcarriers in

serted between or around the data subcarriers) in the upper and/or lower end of the fre- quency region. The goal of equalization is to compensate the effects of channel distortion due to frequency selectivity and to restore the original signal. One approach to signal equalization is in the time domain using a linear equalizer, which consists of a linear filter with an impulse response w(t) operating on the received signal. By selecting different filter impulse responses, different receiver/equalizer strategies can be implemented. For example, the receiver filter can be selected to compensate the radio channel frequency selectivity. This can be achieved by configuring the receiver filter impulse response to satisfy w(t)*h(t) = 1 where the operator denotes linear convolution. This method of filtering is known as zero-forcing equalization, which compensates the channel frequency selectivity. However, the ZF equalization may lead to significant increase in the noise level after equalization, degrading the overall link performance. This will be the case especially when the channel has large variations in its frequency response. Another alternative is to select a filter which provides a tradeoff between signal distortion due to channel frequency selectivity and the corruption due to noise/interference, resulting in a filter impulse response that minimizes the mean squared error between the equalizer output and the transmitted signal. The linear equalizers are typically implemented as a discrete-time FIR-filter with certain number of taps. In general, the complexity of such a discrete-time equalizer increases with increasing bandwidth of the signal [30,46]. An alternative to time-domain equalization is frequency-domain equalization which can sig- nificantly reduce the complexity of linear equalization. In this method, the equalization is performed on a block of data. The received signal is transformed to frequency domain using a DFT operation. The equalization is done as a frequency-domain filtering operation, where the

frequency-domain filter W(k) is the DFT of the corresponding time-domain impulse response w(n). The equalized frequency-domain signal is then transformed to the time domain using an inverse-DFT operator. For processing of each signal block of size N = 2m. the frequency-domain equalization would include two N-point DFT/IDFT operations and N complex multiplications. With the introduction of a CP, the channel would appear as a circular convolution over a receiver processing block of size N. Therefore, there would be no need for overlap- and-discard in the receiver processing. Furthermore, the frequency-domain filter taps can now be calculated directly from an estimate of the sampled channel frequency response. Similar to the OFDM case, the drawback of using CP in conjunction with single-carrier transmission is the overhead in terms of extra power consumption and bandwidth. One method to reduce the relative CP overhead is to increase the block size N of the FDE. However, the accuracy of block equalization requires that the channel to be approximately constant over a period of time corresponding to the size of the proces- sing block. New Radio Access Physical Layer Aspects (Part 1) 333 Equalizer Detector Equalizer Detector DFT-S-OFDM Figure 3.15 Illustration of different equalization/detection aspects of DFT-S-OFDM and OFDM [30]. The detection procedure for a DFT-S-OFDM signal is illustrated in Fig. 3.15 and is com- pared with that of an OFDM waveform. The transmission through a time-dispersive or equivalently a frequency-selective channel will distort the DFT-S-OFDM signal and equalizer shown is needed to compensate for the effects of channel frequency selectivity. However, an as in Fig. 3.15, a simple one-tap equalizer can be applied to each subcarrier in OFDM, whereas in the case of DFT-S-OFDM, the frequency-domain equalization function prior to the IDFT operation. is applied to the complex-valued symbols at the output of the subcarrier demapping and It order must be noted that in OFDM downlink parameterization, t

he DC subcarrier is unused in is to support direct conversion receiver architectures. In contrast, nulling DC subcarrier of not the possible in DFT-S-OFDM since it affects the low cubic metric (CM)/PAPR property transmit signal. Direct conversion transmitters and receivers can introduce distortion tor at the carrier frequency (zero frequency or DC subcarrier in baseband) due to local oscilla- rier. leakage. In LTE downlink, this issue is avoided by inclusion of an unused DC subcar- However, for the uplink when using the DFT-S-OFDM waveform, the same solution the may adversely impact the low CM property of the transmitted signal. In order to minimize impact of such distortion on the packet error rate and the CM/PAPR, in LTE, the DC riers subcarrier of the DFT-S-OFDM signal is modulated in the same way as all the other subcar- ing in but the subcarriers are all frequency-shifted by half a subcarrier spacing Af/2, result- the an offset of 7.5 kHz relative to the DC subcarrier. Therefore two subcarriers straddle DC location; hence, the amount of distortion affecting any individual resource block is reduced by half. In LTE, the DC subcarrier was not used because it might be subject to 334 Chapter 3 disproportionally high interference due to local-oscillator leakage and intermodulation pro- ducts. In fact, all LTE devices could receive the full carrier bandwidth, which was centered around the carrier frequency. The NR devices, on the other hand, may not be centered around the carrier frequency and each NR device may have its own DC located at different locations in the carrier, unlike LTE, where all devices typically have their DC coinciding with the center of the carrier. Since special handling of the DC subcarrier would have been difficult in NR, it was decided to exploit the DC subcarrier for data transmission, under- standing that the quality of this subcarrier may be degraded in some conditions. In order to demonstrate the similarities and differences between OFDM and DFT-S-OFDM processing, let's assume that one w

ishes to transmit a sequence of eight QPSK symbols as shown in Fig. 3.16 [30]. In the OFDM case assuming NDFT = 4, four QPSK symbols would be processed in parallel, each of them modulating its own subcarrier at the appropriate QPSK phase. After one OFDM symbol period, a guard period or CP is inserted to mitigate the multi- path effects. For DFT-S-OFDM, each symbol is transmitted sequentially. With NDFT = 4, there are four data symbols transmitted in one DFT-S-OFDM symbol period. The higher rate data symbols require four times the bandwidth and SO each data symbol occupies NDFT X Af Hz of spectrum assuming a subcarrier spacing of Af Hz. After four data symbols, the CP is inserted. Note the OFDM and DFT-S-OFDM symbol periods are the same [30]. Input QPSK symbols in time (1,1) (-1,-1) (-1,1) (1,-1) (-1,-1) (1,1) (1,-1) (-1, 1) (-1,1) (1, 1) (-1,-1) (1,1) QPSK modulation Frequency NDFTX Af DFT-S-OFDM Data symbols occupy Af in frequency domain and Data symbols occupy NDFTX of in frequency domain and 1/NDFT one OFDM symbol period in time domain DFT-S-OFDM symbol period in time domain Figure 3.16 Comparison of OFDM and DFT-S-OFDM using QPSK modulation with NDFT = 4 [30]. New Radio Access Physical Layer Aspects (Part 1) 335 OFDM (16QAM) OFDM (QPSK) SC-FDMA (QPSK) SC-FDMA (16QAM) PAPR (dB) Figure 3.17 Comparison of OFDM and DFT-S-OFDM PAPRs (5 MHz bandwidth) (30,46]. As mentioned earlier, the PAPR of OFDM intrinsically is inferior to DFT-S-OFDM. Fig. 3.17 shows the comparison of complementary CDF (CCDF) of OFDM and DFT-S- OFDM PAPRs. It can be seen that the PAPR of DFT-S-OFDM is approximately 3 dB better than that of OFDM with probability of 0.99. In the case of 16QAM modulation, the PAPR of DFT-S-OFDM increases relative to that of DFT-S-OFDM with QPSK modulation, whereas in the case of OFDM, the PAPR distribution is independent of the modulation scheme because the OFDM signal is the sum of a large number of independently modulated subcarriers; thus the instantaneous power has an approximately exponential distribution, re

gardless of the modulation scheme applied to different subcarriers [30,46]. 3.2.3 Other Waveform Candidates In the study phase of 3GPP NR, a number of waveforms promising to improve upon OFDM waveform and to overcome the limitation of the latter were proposed and evaluated. However, when the practical aspects of implementation complexity and analog RF proces- sing were considered, many of those candidate waveforms fail to provide any significant improvement over the status-quo, and thus 3GPP agreed to specify the OFDM as the base- line waveform for the new radio. In the following sections, we briefly describe those candi- date waveforms and their respective advantages and disadvantages over OFDM. 336 Chapter 3 3.2.3.1 Filtered-OFDM To mitigate the limitations of OFDM waveform, filtered-OFDM (F-OFDM) waveform was proposed wherein subband-based splitting and filtering were used to allow independent OFDM systems operate in the assigned bandwidth. In this way, F-OFDM can overcome the drawbacks of OFDM while retaining the advantages of it. With subband-based filtering, the requirement on system-wide synchronization is relaxed and inter-subband asynchronous transmission can be supported. Furthermore, with suitably designed filters to suppress the OOB emissions, the guard band size can be reduced to a minimum. Within each subband, optimized numerology can be applied to suit the needs of certain type of services. Fig. 3.18 shows the block diagram of a frequency-localized OFDM-based waveform. As shown in the figure, the baseband OFDM signal of each subband with its specific numerol- ogy is independently generated by processing through a spectrum shaping filter. The main purpose of this filtering is to avoid interference to the neighboring subbands. There are vari- ous approaches to the design of the spectrum shaping filter. In subcarrier filtering, the sinc(.) pulse shape of each individual subcarrier within the subband is filtered to make it more localized in frequency. An example of this method is the windowed OFDM wher

e the subcarrier filtering is performed in the time domain by modifying the rectangular pulse shape of CP-OFDM to have smoother transitions in time at both ends. In an alternative approach known as subband filtering, the PSD of the entire subband is made well-localized without changing the CP-OFDM symbol's rectangular pulse. For this purpose, the subband CP-OFDM signal is passed through a frequency-localized filter whose bandwidth is close to the size of the subband. As a result, only a few subcarriers close to edges of the subband in frequency domain are affected by the filter, as the filter suppresses their out-of-subband side-lobes. This leads to F-OFDM signal generation. A key property of this approach is that the filter length can exceed the CP length, which allows better frequency localization than the subcarrier-based approach without causing any ISI. Although the subcarrier-based approach provides a lower complexity, it cannot achieve the frequency localization Subband 1 IFFT with Spectrum input subcarrier CP 1 insertion shaping filter symbols spacing 1 h1(n) hi(n) Subband 2 IFFT 2 with Spectrum input subcarrier CP 2 insertion shaping filter Spectrum FFT k with Subband k symbols spacing 2 h2(n) shaping filter CP k removal subcarrier hk(n) spacing k detection Subband N IFFT N with Spectrum input subcarrier insertion shaping filter symbols spacing N hn(n) Transmitter Receiver Figure 3.18 Illustration of transmit/receive processing for frequency-localized OFDM-based waveforms 49]. New Radio Access Physical Layer Aspects (Part 1) 337 performance of the subband-based approach, and further causes ISI in multipath channels with large delay spread. A combination of the two approaches can provide a tradeoff between complexity and frequency localization. In particular, in the latter approach, the CP- OFDM signal of each subband is first subcarrier-filtered with an excess window length smaller than that of the original subcarrier-based approach. Then, the windowed signal is subband filtered with a filter length smal

ler than that of the original subband-based approach. In the receiver side, in order to filter out the signals of the neighboring subbands, the received signal in baseband is first passed through the receiver spectrum shaping filter. Subcarrier-based, subband-based, or a composite approach can be employed by the receiver, independent of the approach employed by the transmitter. After the subband spectrum shaping, the resulting signal is processed by the regular OFDM processing within that subband. The filters used in F-OFDM processing must satisfy a number of criteria. The passband of the filter should be as flat as possible over the subcarriers contained in the subband. This ensures that the distortion caused by the filter in the data subcarriers, especially the subband edge subcarriers, is minimal. The frequency roll-off of filter should start from the edges of the passband and the transition band of the filter should be sufficiently steep. This ensures that the system bandwidth is utilized as efficient as possible and the guard band overhead is minimized. Also, the neighboring subband signals with different numerologies can be placed next to each other in frequency with minimal number of guard subcarriers. The filter should further have sufficient stop-band attenuation to ensure that the leakage into the neighboring subbands is negligible. The F-OFDM waveform processing introduces negligible delay at the receiver side. Signal processing delay of F-OFDM depends on the receiver processing capabilities. It should be noted that the only extra signal processing block in F-OFDM receiver compared to CP- OFDM is the receiver subband spectrum shaping filter. The delay due to spectrum shaping filter is implementation-specific and depends on the receiver processing capabilities. The rest of receiver processing blocks in F-OFDM, for example, FFT block size, channel estima- tion/equalization are the same as those in CP-OFDM [49]. The F-OFDM concept can be used in the asynchronous access of multiple UEs in the uplink as sho

wn in Fig. 3.19. 3.2.3.2 Filter Bank Multicarrier Filter bank multicarrier (FBMC) is an OFDM-based waveform wherein subcarriers are indi- vidually processed through filters that suppress their side-lobes, making them strictly band- limited. The transmitter and receiver may still be implemented through FFT/IFFT blocks or polyphase filter structures and band-limitedness may offer larger spectral efficiency than OFDM. During the study of waveforms, FBMC was found promising mainly due to signal band-limitedness in order to relax synchronization requirements in the uplink and/or in the 338 Chapter 3 IFFT/CP Spectrum f((n) Spectrum CP removal/ insertion shaping f. (n) Channel 1(n) shaping filter synchronization f1 (-n) IFFT/CP Spectrum Spectrum CP removal/ insertion shaping f(n(n) Channel hn(n) shaping filter synchronization fn (-n) gNB processing Figure 3.19 F-OFDM uplink asynchronous communication. downlink with coordinated transmission, its greater robustness to frequency mis-alignments among users when compared to OFDM, and its more flexible exploitation of frequency white spaces in cognitive radio networks. The rectangular impulse adopted in OFDM sys- tems is not well-localized in time and frequency, making it sensitive to timing and fre- quency offsets (e.g., introduced by channel, or local oscillator mismatch). As we discussed earlier, ideal time and frequency well-localized pulse does not exist in practice for the conventional OFDM according to Balian- - Low theorem. 12 However, if pulse amplitude modulation (PAM) symbols instead of QAM symbols are considered, time and frequency well-localized pulse can be achieved in a multicarrier system called FBMC. The transmit signal can be expressed as k=-00 where g(t) is a square-integrable function on real domain (Gabor set), which is manifested as the rectangular pulse in OFDM, and Sk,n denotes real-valued data symbols. In FBMC, the pulse g(t) can be designed to achieve better time and frequency localization properties using filter design methods. Usually, the prototyp

e filter g(t) spans an integer K (overlapping factor) multiple length of symbol period TF = KT. It must be noted that real mathematics, the Balian-Low theorem in Fourier analysis states that there is no well-localized window func- tion or Gabor function either in time or frequency domain for an exact Gabor frame. Let g denote a square- integrable function on the set of real numbers, and consider the so-called Gabor system gm,n(x) = g(x-na)e2njmbx for integers m and n, and a,b>0 satisfying ab = 1. The Balian-Low - theorem states that if {8m,n:m,neZ} is an orthonormal basis for the Hilbert space L2(R), then either = 00 [ Note g(o) is the Fourier transform ofg(x)]. The Balian-Low theorem has been extended to exact Gabor frames. New Radio Access Physical Layer Aspects (Part 1) 339 and imaginary data values alternate on subcarriers and symbols, which is called offset QAM (OQAM). Since PAM symbols convey only one half of information content compared to QAM symbols, a data rate loss factor 2 is implicit. Nevertheless, the symbol period in FBMC is also halved to T/2 in order to compensate for the efficiency loss of OQAM modu- lation. Furthermore, CP is not essential anymore in FBMC due to the well-localized pulse shape. Filtering can use different overlap factors (i.e., K factor) to provide varying levels of OOB rejection. As K factor is reduced, the OOB characteristics have a spectrum-rejection profile similar to that of OFDM. The critical step for FBMC design is to implement filters for each subcarrier and to align multiple filters into a filter bank. One way to implement the filter bank is to design a prototype filter. Once the prototype filter is designed, the next step is to make a copy of the prototype filter and shift it to neighboring subcarriers as illustrated in Fig. 3.20. The comparison of the PSDs of the OFDM and FBMC signals is shown in Fig. 3.21. The FBMC signal was processed with overlapping factors K = 2 and 3. It is shown that with the increase of the overlapping factor, the OOB emissions of the FBMC sig

nal significantly decreases; however, the signal processing complexity and latency prohibitively increases relative to that of the OFDM processing. go(t) exp [j2nt(f-mT)] g1(t) x(t)= S(N-2)m gn-2(t) exp [/27t(N-1)Af(t-mT)] XN-1 (t) = g(t-mT)exp[j2u(N-1)Af(t -mT)] S(N-1)m gn-1(t) Figure 3.20 Illustration of FBMC concept and transmitter/receiver architecture [47]. 340 Chapter 3 FBMC signal in the frequency domain (K = 3) FBMC signal in the frequency domain (K = 2) OFDM signal in the frequency domain FFT bins Figure 3.21 Comparison of OFDM and FBMC signals in the frequency domain [38]. 3.2.3.3 Universal Filtered Multicarrier Universal filtered multicarrier (UFMC) is a generalization of OFDM and FBMC. The ulti- mate goal of UFMC is to combine the advantages of OFDM and FBMC while avoiding their main drawbacks. By filtering groups of adjacent subcarriers, the side-lobe levels (com- pare to OFDM) and the prototype filter length (compare to FBMC) can be simultaneously and significantly reduced. The kth OFDM signal over the ith physical resource block (i.e., 12 adjacent subcarriers in NR) can be expressed as follows [38]: m = 0, , Nsub where Si is a set which contains consecutive subcarrier indices that are assigned to the ith physical resource block. This signal is then filtered by an FIR-filter fi(n) with the length of New Radio Access Physical Layer Aspects (Part 1) 341 LF. The UFMC scheme applies filtering on a per subband basis, reducing complexity of the baseband processing algorithms. Thus, the kth transmit symbol can be written as The FIR-filter can be differently designed for each physical resource block. Let's assume that we use an identical Chebyshev filter with variable side-lobe attenuation for all physical resource blocks and the filter is shifted to the center frequencies of the physical resource blocks. The filter ramp-up and -down regions at the beginning and the end of individual UFMC symbols provide somewhat ISI protection, in the presence of channel delay spreads and timing offsets. With very high del

ay spreads, sophisticated multi-tap equalizers must be applied. There is no time overlap between subsequent UFMC symbols. The symbol duration Nsub + LF - 1 with Nsub being the FFT size of the IFFT spreaders (the size of the subbands) and LF the length of the filter. Similar to FBMC, in UFMC typically the FFT window size is increased, resulting in a higher implementation complexity. Also in UFMC the insertion of a guard interval or CP is optional. Another feature of the unified frame structure is the usage of multiple signal layers. The users can be separated based on their interleavers as it is done in interleave division multiple access scheme. This will introduce an additional degree of freedom for the system, improve robustness against cross-talk, and help to exploit the capacity of the multiple access channel (uplink) [45]. The comparison of the PSDs of the OFDM and UFMC signals is shown in Fig. 3.22. In the processing of the UFDM signal, a Chebyshev filter with LF = 74 and side-lobe level attenu- ation of 40 dB has been used. Furthermore, we assume NFFT = 1024 and Ng = 0. It is shown that while the relative complexity of UFMC is more manageable than FBMC, the OOB components are significantly more suppressed compared to that of the OFDM signal. An alternative mathematical representation of the UFDM signal generation and processing can be given as follows: [(++ LF - 1),1] [N+LF - 1),1] [N,Nn] [N,1] where N, LF, NSB, and Nn denote the FFT size, the filter length, the number of subbands, and the number of complex QAM symbols, respectively; [n,k] represents the subband index and user number; Fn,k,Vn,k: and Sn,k denote a Toeplitz matrix comprising the filter impulse response, an IDFT matrix corresponding to the subband location, and a symbol matrix, respectively. The above mathematical model is illustrated in Fig. 3.23. Chapter 3 UFMC signal in the frequency domain OFDM signal in the frequency domain FFT bins Figure 3.22 Comparison of OFDM and UFMC signals in the frequency domain [38]. Transmit filter length IFFT

output symbol 1 IFFT output symbol 2 Transmit filtering x[k] = F. n, k Vn, n, S n. Nsb-1 Transmit waveform symbol 1 Transmit waveform symbol 2 S(Nsb-2)n UFMC processing at the transmitter FFT size S(Nsb-1)n Received waveform Zero padding Subband block filtering 2x size FFT input UFMC processing at the receiver Figure 3.23 UFDM signal processing [42,47]. New Radio Access Physical Layer Aspects (Part 1) 343 3.2.3.4 Generalized Frequency Division Multiplexing Generalized frequency division multiplexing (GFDM) is a flexible multicarrier modulation scheme. The process is performed block-by-block, where each GFDM block consists of a number of K subcarriers and M sub-symbols. By setting the number of subcarriers and the number of sub-symbols to one, GFDM reduces to single-carrier frequency domain equaliza- tion and CP-OFDM as its special cases. Furthermore, pulse shaping with a prototype filter go,((n) is another flexibility in GFDM to reduce OOB emissions. In contrast to linear convo- lution used in FBMC, GFDM uses circular convolution. Let gk,m(n) denote the pulse- shaping filter corresponding to the data symbol Sk,m that is transmitted at subcarrier m and time k. It can be shown that In the above equation N denotes the number of symbols within a GFDM block (GFDM block size). Thus, the time-domain signal x(n) of a GFDM block is expressed as n=0,1,...,N - A CP and a cyclic suffix can be optionally added in the GFDM data block. Furthermore, a raised-cosine filter with configurable roll-off factor B is used for filtering. The GFDM sig- nal processing stages are illustrated in Fig. 3.24. M samples across S0,0 *** SO,M- SK-1,0 ... SK-1,M-1 g(n) n =0,1 LEN- LF M S N (n) pj2tknIN OFDM Data Data Data Data Data (K-2)n (K-2)n W(K-1)n (K-1)n Figure 3.24 Illustration of GFDM signal processing [45,47]. Chapter 3 GFDM signal in the frequency domain OFDM signal in the frequency domain FFT bins Figure 3.25 Comparison of OFDM and GFDM signals in the frequency domain [38]. In one aspect, GFDM is similar to FBMC where a prototype filter

is used to suppress OOB emissions. However, for GFDM, multiple OFDM symbols are grouped into a block and a CP is added to the block. Within a block, the prototype filter is cyclic-shift in time, for dif- ferent OFDM symbols. Therefore, better OOB leakage suppression can be achieved relative to CP-OFDM. However, the approach results in a complicated receiver to handle the ISI and ICI. Furthermore, the prototype filter may require more complicated modulation, for example, OQAM as in FBMC, and more complex receiver architecture. Higher block pro- cessing latency is inevitable in GFDM given that there is no possibility for pipelining, and multiplexing with CP-OFDM requires a large guard band, which would add to the overhead. The comparison of the PSDs of the OFDM and GFDM signals is shown in Fig. 3.25. In the processing of the GFDM signal, a pulse-shaping filter with roll-off factor B = 0.1 and N = 10 symbols were used. Furthermore, we assume that NFFT = 1024 , OFDM Ng = 10 and GFDM Ng = 100 samples. It is shown that while the relative complexity of GFDM pro- cessing is more than that of OFDM, the OOB components are significantly more suppressed compared to that of the OFDM signal. 3.2.3.5 Faster Than Nyquist Signal Processing Faster than Nyquist (FTN) is a non-orthogonal transmission scheme which was one of the approaches initially considered for 5G systems that was expected to improve the spectrum New Radio Access Physical Layer Aspects (Part 1) 345 efficiency by increasing the data rate. It was observed a few decades ago that the binary sinc pulse can be transmitted faster than what the Nyquist theorem states without increasing the bit error rate and despite of ISI. The idea was extended to frequency domain to reduce subcarrier spacing. The transmit signal of FTN can be expressed as follows [4]: where AT < 1 is the time compression factor, which means that the pulses are transmitted faster a factor of 1/AT and AF < 1 is the frequency compression factor, which means the spectral efficiency is increased by a factor

of 1/\F. The FTN transmitter structure is depicted in Fig. 3.26. Since the time and frequency spacing varies for different FTN sys- items, direct implementation cannot provide sufficient implementation flexibility. An FTN mapper based on projection scheme has been designed and used in FTN signaling systems to provide flexibility. The FTN mapper is shown in Fig. 3.26. A cyclic extension is needed in the modulation block, with which the system can switch easily between FTN and Nyquist modes. As mentioned before, FTN signaling inevitably introduces ISI in the time domain and/or ICI in the frequency domain when its baud rate is over the Nyquist one (see Fig. 3.27). Therefore a very important issue is how to design a receiver with ISI and/or ICI suppression capability to recover the original transmitted data. 3.2.3.6 Comparison of the Candidate Waveforms A number of candidate waveforms were studied by 3GPP for NR, before CP-OFDM was selected as the default waveform for the downlink and uplink, which were based on FFT/IFFT processing, additional filtering, windowing, or precoding, in order to achieve higher time-frequency localization and lower OOB spectral leakage, and higher throughput. Filtering is a straightforward way to suppress OOB emissions by applying a digital filter with prespecified frequency response. Some waveforms like F-OFDM and UFDM belong to this category. However, the delay spread of the equivalent composite channel may exceed the CP size and guard period in TDD systems, which results in ISI and imposes restrictions on downlink-to-uplink switching time. Furthermore, the promised OOB emission performance may diminish significantly when PA nonlinearity and other non-ideal RF processing effects are taken into account. At the cost of increased PAPR, filtering techniques are generally known to be unfriendly to communication at high carrier frequencies. Windowing is used to prevent steep changes across two consecutive OFDM symbols in order to reduce the OOB emissions. Multiplying the time-domain samples l

ocated in the extended symbol edges by raised-cosine window coefficients is a widely used realization as chosen by windowed OFDM and weighted overlap-and-add OFDM waveforms. This 346 Chapter 3 Af = 1/T Af=alT (a<1) ATAF<1 F = alT ATAF = 1 x(t) = s(n)*h(t-nT) x(t) = s(n)*h(t-ndT,) Nyquist ISI criterion Faster-than-Nyquist ISI criterion Time-frequency sampling Faster-than-Nyquist sampling 1 W fs Frequency Nyquist sampling Frequency Channel Interleaver Modulator encoder mapper FTN transmitter Pulse sampler shaper Channel Equalizer demapper interleaver decoder Matched Whitening filter sampler filter FTN receiver Figure 3.26 Illustration of FTN concept and transmitter and receiver architecture [44]. technique generally has little or no effect on PAPR increase and has lower complexity com- pared to that of filtering techniques. Nevertheless, the detection performance might be degraded because of ISI caused by symbol extension. The linear processing of input data prior to IFFT is usually known as precoding, and may be helpful to improve OOB emissions and PAPR reduction. One example is DFT-S-OFDM New Radio Access Physical Layer Aspects (Part 1) 347 Windowed CP-OFDM Windowed GFDM Windowed DFT-S-OFDM Zero-tail DFT-S-OFDM QPSK single-carrier (SC) QPSK + hybrid PSK SC PAPR (dB) Figure 3.27 Comparison of PAPR performance of some prominent variants [47]. waveform that was adopted in LTE uplink transmission because of its low PAPR properties relative to conventional OFDM. Some variants of DFT-S-OFDM were proposed for NR such as zero-tail (ZT) DFT-S-OFDM by omitting the CP and letting the tail samples taper to zero. The DFT-S-OFDM-based waveforms, in contrast to filter-based waveforms, usu- ally make it easier to maintain PA linear operation with less deterioration from lowering OOB emissions. Moreover, an appropriate modification of modulation schemes, such as 2-BPSK can greatly assist such waveforms in achieving an extremely low PAPR. Note that in the absence of redundant intervals, ISI can still occur. Among DFT-based precod-

ing techniques, other types of precoding matrices often have undesirable complexity and compatibility issues. PAPR is one of the often-mentioned disadvantages of OFDM waveform. In practice, the crest factor reduction (CFR) techniques are applied to reduce the PAPR and digital predis- tortion algorithms will then correct for any distortion caused by the analog hardware used 348 Chapter 3 to amplify the signal. Both techniques will allow more efficient PA design and help mitigate major limitations of PAPR and spectral regrowth. Traditionally these techniques were only applied at the base station side, but currently, they are also used in mobile devices, mainly from the aspect of reducing power consumption. The use of envelope tracking 14 to reduce the static power consumption is an example of such techniques. Fig. 3.27 compares the PAPR performance of the prominent waveforms and shows that despite additional com- plexity and latency, the relative PAPR performance of multicarrier techniques remains about the same as OFDM and inferior to DFT-S-OFDM [47]. In the spectrum of rectangular pulse (i.e., a sinc function), besides the desired peak, there are some side-lobes that result in a theoretical infinite bandwidth of the rectangular pulse function, causing OOB emissions. Moreover, consecutive OFDM symbols are independent of each other; thus there is an inherent discontinuity in the time domain between them. In this way OFDM differs from single-carrier modulated signals after digital filtering. This dis- continuity translates into spectral spurs in the frequency domain. This typical characteristic can be improved by applying time-domain windowing that smooths out the transition from one symbol to another. However, this technique introduces an overlap between consecutive symbols that impacts signal quality and results in higher EVM. The transition time defines the duration of the overlap between two symbols. For a sampling rate of 30.72 MHz (20 MHz LTE signal), a transition time of 1 us translates into 30 samples overl

ap. Fig. 3.28 shows the ideal case by means of connecting the signal generator directly to a spec- trum analyzer. The idea is to demonstrate the impact of a nonlinear device on the highlighted advantage of 5G waveform candidates, where any nonlinearity will result in a spectral regrowth, and there is a risk that this spectral regrowth may undermine any optimization due to the waveform design. The improved spectral characteristics of the candidate waveforms are clearly visible in the top snapshot. In a second step, a nonlinear amplifier is introduced into the signal path. The top and bottom snapshots in Fig. 3.28 compare the LTE OFDM signal with FBMC, UFMC, and GFDM waveforms under two different conditions. A generic PA, which supports a frequency range of 50 MHz to 4 GHz, is used to demonstrate the effects of nonline- arity on the waveforms. The maximum input power for the PA is 0 dBm, and it has a typical Intermodulation products or spurs can develop within the analog and digital transmitters in combined systems using high-level injection. In some cases spurs can result in suboptimal signal quality or even cause stations to interfere with each other's signals. The term spectral regrowth is used to describe intermodulation products generated when a digital transmitter is added to an analog transmission system. Envelope tracking is an approach to RF amplifier design in which the power supply voltage applied to the RF power amplifier is continuously adjusted to ensure that the amplifier is operating at peak efficiency for power required at each instant of transmission. A conventional RF amplifier is designed with a fixed supply voltage and operates most efficiently only when operating in compression region. Amplifiers operating with a constant supply voltage become less efficient as the crest factor of the signal increases, because the ampli- fier spends more time operating below peak power and, therefore, spends more time operating below its maximum efficiency. New Radio Access Physical Layer Aspects (Part 1) 349

10 dB -10 dB 20 MHz -20 dB -30 dB -40 dB -50 dB 60 dB -70 dB -80 dB Spectral density of the candidate waveforms under ideal conditions 10 dB 110 BB 20 MHz 20-BB 30 dB 404 BB 50 BB 6-60 BB +70 dB -80 BB Spectral density of the candidate waveforms with nonlinear power amplification Figure 3.28 Comparison of a 20 MHz LTE downlink OFDM signal (yellow) with FBMC (blue), UFMC (green), and GFDM (orange) signals under ideal (top) and non-ideal conditions [45]. For color interpretation of this figure, please refer web version. gain of 20 dB. Maximum achievable output power for the PA is + 20 dBm. At 0 dBm input power, the PA starts to enter the saturation region. Higher input power would mean that the PA would be operating in the compression region. Fig. 3.28 further shows the result of a measure- ment with the same signal configuration for an input power of -2 - dBm to the amplifier. The spectral advantages of the candidate waveforms seem to have almost disappeared compared to 350 Chapter 3 a 20 MHz LTE downlink signal. When using typical input power of 0 dBm, the advantages of non-OFDM waveforms will completely vanish [45]. 3.3 Multiple-Access Schemes A cellular network consists of a number of fixed base stations distributed across a geographi- cal area. The coverage area is divided into cells and a mobile station communicates with one or more base stations in its proximity. There are two main issues in the physical and medium access layers of cellular communication schemes: multiple access and interference manage- ment. The first issue addresses how the overall radio resources of the system are shared by the users in the same cell (intra-cell) and the second issue addresses the interference caused by simultaneous signal transmissions in different cells (inter-cell). At the network layer, an important issue is to provide and maintain seamless connectivity to the users as they move from one cell to another and thus switching communication link from one base station to another through an operation known as handover. There

are various multiple-access schemes that have been studied and used in wireless sys- tems in the past decades, which allow the network to share the available radio resources (i.e., time, frequency, code, space, power) among a number of active users in the cell in the down- link and uplink. Fig. 3.29 illustrates the concept of resource sharing in some prominent multiple-access schemes. As mentioned earlier, orthogonal frequency division multiple access (OFDMA) has been a promising MA scheme that has been used in mobile broadband radio access technologies such as NR and LTE. The new radio uses a symmetric OFDMA scheme in the downlink and uplink, whereas LTE uses OFDMA and SC-FDMA as the MA schemes in the downlink and uplink, respectively. In addition, the non-orthogonal concept can be applied to MA scenarios. Sparse code multi- ple access (SCMA), non-orthogonal multiple access (NOMA), and mult-iuser shared access (MUSA) are examples of non-orthogonal multiple access schemes that were studied in 3GPP Rel-16 [23,50-52]. These techniques can superimpose signals from multiple users in the code domain or the power domain to enhance the system-access performance and poten- tially allow asynchronous access in the uplink. NOMA/SCMA/MUSA User K User K User K-1 Users multiplexed in time and User K-1 User K-1 User K frequency User 3 User3 User3 User 2 User1 User2 User 1 Figure 3.29 Illustration of various multiple access concepts [47]. New Radio Access Physical Layer Aspects (Part 1) 351 3.3.1 Orthogonal Frequency Division Multiple Access OFDMA is the multi-user variant of the OFDM scheme where multiple access is achieved by assigning subsets of time-frequency resources to different users, allowing simultaneous data transmission from several users. In OFDMA, the radio resources are 2D regions over time (an integer number of OFDM symbols) and frequency (a number of contiguous or non-contiguous subcarriers). Similar to OFDM, OFDMA employs multiple closely spaced subcarriers that are divided into groups of subcarriers where eac

h group is called a resource block. The grouping of subcarriers into groups of resource blocks is referred to as sub- channelization. The subcarriers that form a resource block do not need to be physically adjacent. In the downlink, a resource block may be allocated to different users. In the uplink, a user may be assigned to one or more resource blocks. Sub-channelization defines subchannels that can be allocated to mobile stations depending on their channel conditions and service requirements. Using sub-channelization, within the same time slot (i.e., an integer number of OFDM symbols) an OFDMA system can allocate more transmit power to user devices with lower SNR and less power to user devices with higher SNR. Sub- channelization also enables the base station to allocate higher power to sub-channels assigned to indoor mobile terminals resulting in better indoor coverage. In OFDMA, an OFDM symbol is constructed of subcarriers, the number of which is determined by the FFT size. There are several subcarrier types: (1) data subcarriers are used for data trans- mission, (2) pilot or reference-signal subcarriers are utilized for channel estimation and coherent detection, and (3) null subcarriers that are not used for pilot/data transmission. The null subcarriers including the DC subcarrier (if it exists) are used for guard bands. The number of used (or occupied) subcarriers is always less than the FFT size. The guard bands are used to allow spectrum sharing and to reduce the adjacent channel interference and OOB emissions. The sampling frequency is selected to be greater than or equal the channel bandwidth. The number of time samples in a radio frame is always an integer and to further simplify the design of analog transmit filter, the sampling frequency is scaled by a factor greater than one (e.g., in LTE, the sampling frequency for 20 MHz bandwidth is 30.72 MHz). In order to explain the signal processing concepts involved in an OFDMA transmission sys- tem, we use the generic transmitter model that is illustrated i

n Fig. 3.30, which shows the baseband structure of a general multicarrier transmitter that is applicable to a variety of multicarrier MA schemes such as OFDMA and SC-FDMA. Blocks of data represented by vector s of size M X 1 are precoded with an MXM precoding matrix P. The M X output vector is then mapped to M out of N inputs of the inverse-DFT block according to the subcarrier mapping N transform matrix L. To overcome the effects of frequency- selective channel fading, a CP of length NCP is appended to beginning of each NX1 block output by the inverse-DFT function. Transmission with different rates among users is available according to each user's requirement, as a different number of Chapter 3 Subcarrier Inverse Cyclic mapping Precoding Fourier (zero insertion) prefix transform insertion Multicarrier (Size N) (M<N) data symbols (N+Ncp)x1 Data block (Mx1Vector) Figure 3.30 General multicarrier transmission scheme [47]. subcarriers and a different modulation and coding schemes can be applied to each user. Let s(n) denote the information symbols which are parsed into data blocks of size M. The ith data block Si can be written as S = [s(iM), s(iM+M-1)]T. Let's further denote by X the Kronecker product, by O the all-zero matrix of size MXN and by I the MXM identity matrix. We assume that the size of the inverse DFT is a multiple of the block size N = MK. The special case of P = I results in OFDMA where the user-specific data blocks are mapped to a subset of M <N subcarriers, which are selected by the user-specific subcarrier mapping matrix L. The vector Ns is fed to the inverse-DFT function. The form of the matrix N might lead to either a localized or a distributed subcarrier mapping as follows: 01 SQNXM=IMXM x Onx1 By assigning different groups of subcarriers to different users, each user's transmit power can be concentrated in a restricted part of the channel bandwidth, resulting in significant coverage enhancement. Different user signals remain orthogonal only if time/frequency syn- chronization is maintained and

an appropriate CP is appended to compensate for timing misalignments at the receiver. In order to maintain good performance in frequency-selective fading channels, robust forward error correction schemes must be employed. Precoded OFDMA is a variant of OFDMA in which a precoding matrix P is used that spreads the energy of symbols over the subcarriers allocated to the user. Uniform energy distribution is desirable in practice. One of the most well-known precoding matrices is the Walsh-Hadamard matrix P = ( Po PM-1) where the row vectors Pi are orthog- onal Walsh-Hadamard sequences of length M [47]. New Radio Access Physical Layer Aspects (Part 1) 353 3.3.2 Single-Carrier Frequency Division Multiple Access SC-FDMA has been used as the uplink multiple-access scheme in LTE systems. The use of SC-FDMA was motivated by the fact that a single-carrier system with an OFDMA-type multiple-access would combine the advantages of the two techniques, that is, low PAPR and large coverage. The first SC-FDMA concept was interleaved FDMA (IFDMA), which was based on compression and block repetition of the modulated signal in the time domain. It can be theoretically shown that the spectrum of the compressed and K times repeated sig- nal has the same shape as the original signal, with the difference that it presents exactly K - 1 zeros between two data subcarriers. This feature enables us to easily interleave differ- ent users in the frequency domain by applying to each user a specific frequency shift, or equivalently, by multiplying the time-domain sequence by a specific phase ramp. In addi- tion, similar to OFDMA, robustness to inter-cell interference can be achieved by coordinat- ing resource allocation between adjacent cells. The same waveform can be obtained in the frequency domain, if discrete Fourier transform matrix P = [Pk,n]; Pk,n exp(j2nkn/M) used as the precoding matrix in Fig. 3.30, resulting in DFT-precoded OFDMA, which is mathematically identical to IFDMA in a distributed scenario. The precoding operation Ps is equival

ent to an M - point DFT operation. With a subcarrier mapping matrix N as given in Section 3.3.1, the spectrum of the distributed DFT-precoded OFDMA signal is identical to the IFDMA signal spectrum, thus it corresponds to the same waveform. This is also called DFT-spread OFDM. The two techniques are different implementations of SC-FDMA. The advantage of DFT-precoded OFDMA is in its more flexible structure. While IFDMA imposes a distributed signal structure, DFT-precoded OFDMA allows the use of an appro- priate subcarrier mapping matrix L. Localized variants of implementation or channel- dependent mappings are also possible. A pulse-shaping filter can be further applied in the frequency domain, with a lower complexity than the time-domain filtering. Note that in frequency-selective channel scenarios, interference may occur among the M elements of each data block. This degradation, which is more important in a distributed subcarrier map- ping, also impacts Walsh-Hadamard precoded OFDMA [47] 3.3.3 Non-orthogonal Multiple-Access Schemes Previous generations of cellular standards relied on orthogonal MA, where each time/fre- quency resource block was exclusively assigned to one of the active users to ensure no inter-user interference would occur. In 3GPP NR, synchronous/scheduling-based orthogonal MA continues to play an important role in uplink/downlink transmissions. Non-orthogonal multiple access transmission, which allows multiple users to share the same time/frequency resource, was recently proposed to enhance the system capacity and to accommodate mas- sive connectivity through asynchronous uplink access. Unlike orthogonal MA, multiple Chapter 3 non-orthogonal multiple access users' signals are multiplexed using different power alloca- tion coefficients or different signatures such as codebooks/codewords, sequences, interlea- vers, and preambles. The fundamental theory of non-orthogonal multiple access has been extensively studied in network information theory. The uplink and downlink non-orthogonal multiple acce

ss can be theoretically modeled as a multiple-access channel and a broadcast channel, respectively. The capacity region of the Gaussian broadcast channel can be achieved by power-domain superstition coding with a successive interference cancellation (SIC) receiver. Meanwhile, the capacity region of Gaussian multiple-access channel corre- sponds to CDMA, where different codes are used for the different transmitters, and the receiver decodes them in an SIC manner. In general, a user with poor-channel condition tends to allocate more transmission power, SO this user would decode its own messages by treating the co-scheduled user's signal as noise. On the other hand, a user with good chan- nel condition applies the SIC strategy by first decoding the information of the poor- channel user and then decoding its own, removing the other users' information. The results of studies in 3GPP suggest that using a non-SIC receiver results in negligible per- formance degradation in many cases [23]. Relaxing the need for an SIC receiver signifi- cantly would reduce the decoding complexity for the downlink case as the others' codebooks are no longer required. In addition to the orthogonal MA scheme, the 3GPP NR may support an uplink non- orthogonal transmission (Note: at the time of publication of this book, 3GPP has decided not to specify non-orthogonal multiple access in Rel-16 and instead only specify a two-step RACH) to provide the massive connectivity that is desperately required for applications in mMTC as well as other scenarios. 3GPP has further studied grant-free uplink multiple-access schemes for mMTC scenarios. Since there is no need for a dynamic and explicit scheduling grant from gNB, latency reduction and control signaling minimization could be expected. For uplink non-orthogonal multiple access, network information theory suggests that CDMA with a SIC receiver provides a capacity achieving scheme. However, securing uplink non- orthogonal multiple access gain requires further system design enhancement. As the number o

f co-scheduled users increases, the decoding complexity of the SIC receiver increases. The message passing algorithm (MPA), as a less-complex decoding algorithm, as well as other low-complexity receiver designs have recently drawn attention. Several code-spreading- based techniques, including SCMA, MUSA, and PDMA and several others were the candi- dates under consideration in 3GPP Rel-16 NOMA study item. It has been shown that one can potentially achieve higher spectral efficiency, larger connectivity, and better user fairness with non-orthogonal multiple access relative to orthogonal MA schemes [50-52]. While the interference-free condition between orthogonally multiplexed users might facili- tate multi-user detection at the receivers, it is widely known that orthogonal MA cannot achieve the sum capacity of a wireless system. The orthogonal MA also has limited granu- larity of resource scheduling, SO it struggles to handle a large number of active connections. Non-orthogonal multi-user transmission/access has been recently investigated in a New Radio Access Physical Layer Aspects (Part 1) 355 Bit-level operations Symbol-level operations Bit-level Forward error Modulated symbol Symbol to interleaver/ sequence resource element correction (FEC) scrambler generator mapping Figure 3.31 High-level block diagram for uplink non-orthogonal multiple-access schemes [22,23]. Input data Transport block Channel encoder UE-specific segmentation and Rate matching Bit scrambling Modulation CRC addition interleaving UE-specific Transform Subcarrier spreading/ Spatial precoding OFDM processing DFT-S-OFDM precoding mapping scrambling Subcarrier Spatial precoding OFDM processing CP-OFDM mapping Figure 3.32 General framework for non-orthogonal multiple access uplink transmission [23]. systematic manner to deal with the above problems. Interference can be controlled by non- orthogonal resource allocation at the cost of increased receiver complexity. The non-orthogonal multiple access schemes that were studied for the uplink transmissi

on in 3GPP have the following features in common: (1) they use MA signature(s) at the trans- mitter side and (2) they allow multi-user detection at the receiver side. The MA signatures are typically used to differentiate the users. Thus all proposed non-orthogonal MA schemes for the uplink transmission at a high level can be described with the basic diagram shown in Fig. 3.31. As we stated earlier, in non-orthogonal multiple access uplink transmission, multiple UEs share the same time/frequency resources via non-orthogonal resource allocation. There are various non-orthogonal multiple access schemes that can be derived from the general con- cept shown in Fig. 3.31. Fig. 3.32 shows a unified framework for non-orthogonal multiple access based on UE-specific spreading/scrambling/interleaving for the uplink data transmis- sion. For data transmission based on UE-specific spreading, the existing solutions can be classified into two categories: linear spreading and nonlinear spreading. The category of lin- ear spreading includes solutions such as resource spread multiple access (RSMA), MUSA, WSMA, and NCMA, while the category of nonlinear spreading includes SCMA. Linear spreading can be used in conjunction with DFT-S-OFDM and CP-OFDM waveforms. Although receiver implementation relatively less-complex and straightforward in orthogonal MA systems, the successful deployment of non-orthogonal multiple access depends on advanced receivers with inter-UE interference cancellation capabilities [23]. 356 Chapter 3 Input data Transport block segmentation and Channel encoder Rate matching Bit scrambling Modulation CRC addition UE-specific Transform Subcarrier spreading/ Spatial precoding OFDM processing DFT-S-OFDM precoding mapping scrambling Subcarrier Spatial precoding OFDM processing CP-OFDM mapping Figure 3.33 Linear hybrid spreading for non-orthogonal multiple access uplink transmission [23]. Fig. 3.33 shows non-orthogonal multiple access uplink transmission blocks based on linear hybrid spreading. In particular, the assignme

bination with spatial multiplexing schemes (in the presence of multiple transmit antennas). Compared to long spreading codes, short spreading codes need smaller spreading factor and higher spectral efficiency. The short spreading codes can be optimized to achieve the Welch bound1 on cross-correlation, which can be leveraged for multi-user detection and inter-UE In mathematics, Welch bounds are a family of inequalities corresponding to the problem of evenly spreading set of unit vectors in a vector space. If {X1 Xm} are unit vectors in Cn. We define (Xi,Xj) where the inner product is defined in Cn. Then, the following inequalities are given Ak = If m VI n, then the vectors {Xn} can orthonormal set in Cn. In this case, Cmax 0 and the bounds are void. Consequently, interpretation of the bounds is only meaningful if m > n. New Radio Access Physical Layer Aspects (Part 1) 357 interference cancellation for synchronized reception. Moreover, it can be easily combined with spatial precoding to further mitigate the cross-correlation and enhance the non- orthogonal multiple access capacity. A large number of the existing non-orthogonal multiple access schemes are based on linear spreading. For this type of spreading-based NOMA schemes, the same type of NOMA receivers can be utilized. The NOMA receiver consists of three parts. The first part is multi- user detector where the superposed received signal is processed jointly across the UEs to derive the LLR for each UE. The second part is the channel decoder which receives the LLR from multi-user detector and decodes the transmitted codeword. The output from the channel decoder can be either a decoded codeword when the decoding is successful or an intermediate LLR for each bit refined through the message passing decoding process. The third part is the iteration between the multi-user detector and the channel decoder, where they can exchange both soft-LLR information and hard decision information. We will focus on multi-user detector here and in particular, we will focus on elem

entary signal estimator (ESE) and linear MMSE (LMMSE) estimator. Without loss of generality, we consider sin- gle symbol processing. Let's assume that there are J users and K resources (spreading fac- tor). The received signal at resource k can be written as yk = ;=1 hkjXkj where hkj is the channel coefficient corresponding to resource k from user j, Xkj is the transmitted signal by user j on resource k, and n ~C(0,No). For linear spreading codes, each user is assigned a spreading code sequence. Let Ckj be the kth coefficient of the spreading code for user j. We assume that all users share the same modulation alphabet W = WM}. Then, = 1,2, ., K where Sj W is the transmitted symbol by user j. The received sig- nal can be written as y = Hs + n where s=[s]...sj]T, n=[n1...nj]T, and [hkilk, is a K X J matrix with entries hkj = hkjCkj. Let hj denote the jth column of matrix H. The multi-user detector estimates the LLRs for S based on y [23]. Matched filter and ESE, which is a generalization of the matched filter that can accommo- date soft interference cancellation, can be used as multi-user detectors. The ESE multi-user detector first compresses the received signals to scalar values for each UE by matched filter- ing. The output of the matched filter can be written as y = [ = H4y. In order to take advantage of soft information computed at the channel decoder, we can apply the ESE to y, which approximates signal and interference as Gaussian random variables as follows hswhere Ej is residual interference plus noise. The residual interference Ej is approximated as a Gaussian random variable which can be described by its mean and variance as follows: M(E))= 358 Chapter 3 where u(si) and o2(si) are the a priori mean and variance of the symbol transmitted from the ith user, which can be computed using a priori bit LLRs. The LMMSE estimation can be used to estimate Sj from . From the estimation, LLR for each bit can be derived from conventional marginalization. 16 It can be noted that the ESE multi-user detector without mat

ched filter and symbol spreading can also be used. For random symbol interleaver cases, the ESE multi-user detector can be applicable assuming that = The ESE multi-user detector is also applicable to bit-level interleaving scenarios. It can be shown that the computational complexity of ESE multi-user detector scales as O(K²) for J UEs and spreading factor K [23]. Unlike ESE multi-user detector, the LMMSE estimator treats the received signal as a vector and applies LMMSE estimation matrix to the transmitted signal estimation of each UE. Let ,ndvp vectors denote the priori mean and variance of each UE transmitted signal derived from LLRs. The receiver applies the following LMMSE filter to the received vector where the output of the LMMSE filter is the mean and variance vectors. mH(H)y-p As mentioned earlier, up is the a priori mean vector and Vp is a diagonal matrix whose diag- onal entries are a priori variance values for the corresponding transmitted symbols. A priori mean and variance values can be computed using the bit LLRs computed at the channel decoder. Based on the LMMSE output, the receiver generates extrinsic bit LLR values for channel decoder by marginalization. It can be shown that the computational complexity of LMMSE multi-user detector scales as O(K3 + K2.J + KJ2) for J UEs and spreading factor of K [23]. Table 3.6 summarizes the use cases and operation modes of Rel-16 NOMA candidates. The features in the characteristics column reflect the potential benefits of NOMA over Rel-15 NR MA scheme [6], which are considered in the design, evaluation and comparison of NOMA transmitter and receiver schemes. The studies conducted in 3GPP suggest that when In probability theory and statistics, the marginal distribution of a subset of a collection of random variables is the probability distribution of the variables contained in the subset. It gives the probabilities of various values of the variables in the subset without reference to the values of the other variables. This contrasts with a conditional distribut

ion, which gives the probabilities contingent upon the values of the other vari- ables. Marginal variables are those variables in the subset of variables being retained. The distribution of the marginal variables (the marginal distribution) is obtained by marginalizing; that is, focusing on the sums in the margin, over the distribution of the variables being discarded, and the discarded variables are said to have been marginalized. New Radio Access Physical Layer Aspects (Part 1) 359 Table 3.6: Rel-16 non-orthogonal multiple-access use cases and features supported by different operation modes [23]. Dynamic Operation Mode MCS Support Characteristics Use Case RRC_INACTIVE, grant-free with Reduction in system overhead, mMTC and contention, tracking area-free latency, and power consumption (asynchronized) RRC_CONNECTED Grant-free with Reduction in system overhead and mMTC, (Synchronized) contention latency URLLC, and Grant-based Limited downlink overhead with overloading reduction the network operates in grant-based mode, transmission schemes proposed for NOMA can be applied to MU-MIMO; however, the relative spectral efficiency advantage of NOMA over MU-MIMO is not clear under underloaded scenarios. When the network operates in grant-free mode and the uplink access is contention-free, the relative gain of NOMA over MU-MIMO in terms of spectral efficiency is not proven. The most significant gain of NOMA over MU-MIMO may be attributed to contention-based, grant-free transmission and small data transmission in RRC_INACTIVE state scenarios. 3.3.3.1 Sparse Code Multiple Access SCMA is a frequency-domain non-orthogonal multiple-access scheme, which can improve the spectral efficiency of wireless radio access. In SCMA, different incoming data streams are directly mapped to codewords of different multi-dimensional cookbooks, where each codeword represents a spread transmission layer. Each layer or user has its own dedicated codebook. Multiple SCMA layers share the same time-frequency resources of OFDMA. The sparsity of codew

ords makes the near-optimal detection feasible through iterative MPA. Such low complexity of multi-layer detection allows excessive codeword overload- ing in which the dimension of multiplexed layers exceeds the dimension of codewords. Belief propagation, also known as sum-product message passing, is a message-passing algorithm for perform- ing inference on graphical models such as Bayesian networks and Markov random fields. It calculates the marginal distribution for each unobserved node conditional on any observed nodes. Belief propagation is commonly used in artificial intelligence and information theory and has demonstrated empirical success in numerous applications including low-density parity-check codes, turbo codes, free energy approximation, and satisfiability. The algorithm was formulated as an exact inference algorithm on trees, which was later extended to polytrees. While it is not accurate for general graphs, it has been shown to be a useful approxi- mation algorithm. If X = {Xi} is a set of discrete random variables with a joint mass function p, the marginal distribution of a single Xi is the summation of p over all other variables Px,(Xj) = However, this would become computationally prohibitive, whereas by exploiting the polytree structure, belief propaga- tion would allow the marginals to be computed more efficiently [50-52]. 360 Chapter 3 Optimization of overloading factor along with modulation/coding levels of layers provides a more flexible and efficient link-adaptation mechanism. On the other hand, the signal spread- ing feature of SCMA can improve link-adaptation as a result of less colored interference. In SCMA, incoming bits are directly mapped to multi-dimensional complex codewords selected from predefined codebook sets. The co-transmitted spread data are carried over super-imposed layers. Since layers are not fully separated in a NOMA system, a nonlinear receiver is required to detect the intended layer of each user. Therefore, additional detection complexity is the cost of the nonorthogo

nal multiple-access especially when the system is heavily overloaded with a large number of multiplexed layers. Low-density spreading (LDS) is a special form of SCMA. In LDS, codewords are built by spreading of modulated symbols using LDS signatures with few non-zero elements within a large signature length. Despite the moderate complexity of detection, LDS suffers from poor performance espe- cially for large constellation sizes beyond QPSK. All CDMA schemes, and in particular LDS, can be considered as different types of repetition coding in which different variations of a QAM symbol are generated by a spreading signature. Repetition coding is not able to provide desirable spectral efficiency for a wide range of SNR. In SCMA the QAM mapper and linear operation of sparse spreading are merged to directly map incoming bits to a com- plex sparse vector called a codeword. Both LDS and SCMA are based on the idea that one- user information is spread over multiple subcarriers. However, the number of subcarriers assigned to each user is smaller than the total number of subcarriers, and this low spreading (sparse) feature ensures that the number of users utilizing the same subcarrier is not too large, such that the system complexity remains manageable [50-52]. In LDS, each user spreads its data on a small set of subcarriers. There is no exclusivity in the subcarrier allocation and more than one user can share each subcarrier. The interference pattern at the receiver will generate a low-density graph, and graph theory-based techniques can be utilized. The main features of the LDS scheme can be summarized as follows. At each subcarrier, a user will have relativity small number of interferers comparing to the total number of users. Consequently, the search space will be smaller and more complex multi- user detection techniques can be implemented. Higher SINR can be achieved at each sub- carrier, which results in reliable detection process. Each user will experience interference from different users at different subcarriers, w

hich results in interference diversity by avoid- ing strong interferers to destroy the signal of a user on all the subcarriers. Belief propaga- tion based multi-user detection can be implemented with linear complexity in the number of subcarriers [50-52]. The LDS and SCMA share the same concept which is to use a low-density or sparse non- zero element sequence to reduce the complexity of MPA processing at the receiver. However, in SCMA, bit streams are directly mapped to different sparse codewords. An exam- ple is illustrated in Fig. 3.34, where each user has a codebook and there are six users. All codewords in the same codebook contain zeros in the same two dimensions, and the positions New Radio Access Physical Layer Aspects (Part 1) 361 B11b12... SCMA modulation FEC encoder 1 codebook mapping SCMA codeword 1 SCMA block 1 SCMA modulation FEC encoder 2 codebook mapping B31b32--- SCMA modulation FEC encoder 3 codebook mapping B41b42.. SCMA modulation FEC encoder 4 codebook mapping B51b52.. SCMA modulation FEC encoder 5 codebook mapping Frequency B61b62... SCMA codeword 6 SCMA modulation FEC encoder 6 codebook mapping Figure 3.34 Illustration of SCMA concept [52]. of the zeros in the different codebooks are distinct to facilitate the collision avoidance of any two users. For each user, two bits are mapped to a complex codeword. Codewords for all users are multiplexed over four shared orthogonal resources. The key difference between LDS and SCMA is that a multi-dimensional constellation for SCMA is designed to generate codebooks, which provides a shaping gain that is not possible with LDS. In order to simplify the design of the multi-dimensional constellation, a baseline constellation can be generated by minimizing the average alphabet energy for a given minimum Euclidian distance between constellation points, and also taking into account the codebook-specific operations such as phase rotation, complex conjugate, and dimensional permutation [50-52]. We use the uplink multiple-access system shown in Fig. 3.35 to exp

lain the SCMA proces- sing stages. Let's assume that the system comprises K users whose information bits are spread over N resource elements. In an orthogonal scenario, K < N to ensure that each user is assigned to an orthogonal resource element, while in the non-orthogonal scenarios, K > N and the ratio K/N is defined as the overloading factor. The transceiver structure of SCMA can be mathematically modeled as follows. Let b=b = bk] denote the infor- mation bits transmitted by K uplink users and represent the transmitted symbols by the kth user. An SCMA encoder at the kth user is defined to be a one-to-one mapping fk:Bk Xk with E Bk and Xk E Xk, where the cardinality of Bk and Xk is 2NB where NB denotes the number of information bits in bk. Note that due to the sparse nature of SCMA scheme, Xk may contain zero symbols. The received signal at the base station y, after passing through a block fading multiple-access (uplink) channel, can be expressed as = where Hk = diag(hi h IN) represents the channel between the base station and the kth user, z=zz Z2 ZN] is the additive white Gaussian Chapter 3 Codebook 1 User 1 B11b12.. Baseband and FEC encoder 1 SCMA encoder MPA receiver/SCMA decoder RF processing User 2 Baseband and FEC encoder 2 SCMA encoder RF processing User 3 B. b32 Baseband and FEC encoder 3 SCMA encoder RF processing User 4 B41b42.. Baseband and FEC encoder 4 SCMA encoder RF processing User 5 B511552 Baseband and FEC encoder 5 SCMA encoder RF processing Function nodes User 6 Variable nodes Baseband and FEC encoder 6 SCMA encoder RF processing Codebook 6 Layer 1 Layer 2 Layer 6 (Codebook 1) (Codebook 2) (Codebook 6) (1,1) (1,0) Streams are mapped to (0, 1) codebooks 6 Sparse codebooks are MPA-based sent over 4 resources MUD receiver Figure 3.35 Example SCMA processing with K = 6, N = 4 and 150% overloading factor [52]. noise with zero mean and unity variance. Given the received signal y = y1 y2 and the channel knowledge H = {Hk|k = 1, 2, ., K}, the joint maximum-A-posteriori detec- tion18 18 of X = X1 X2

XK] can be written as X max p(X|y). eral, the solution to the above problem requires a global search over the joint space of K Maximum-a-posteriori (MAP) estimate of random variable X given that we have observed Y = y, is given by the value of x that maximizes fx|y(x|y) if X is a continuous random variable, and Px|y(x|y) if X is a discrete random variable. The MAP estimate is shown by MAP = arg maxxfxy(x)y). New Radio Access Physical Layer Aspects (Part 1) 363 uplink users X1 X X2 X ... X . Due to the sparse nature of SCMA transmission scheme, the MPA detector can be applied to reduce the decoding complexity, which iteratively updates the belief associated with the underlying factor graph. Once X has been estimated, we can use the inverse mapping function (fk)-1 to recover the original user information bits Bk [50-52]. 3.3.3.2 Power-Domain Non-orthogonal Multiple Access Power-domain NOMA can serve multiple users in the same time slot, OFDMA subcarrier, or spreading code, and multiple-access is realized by allocating different power levels to different users depending on their relative position to the base station. Fig. 3.36 illustrates the concept of power-domain NOMA in the downlink with two UEs that utilize SIC receiver. For simplicity, we assume in this section the case of single trans- mit/receive antennas. The overall system transmission bandwidth is assumed to be 1 Hz. The base station transmits signal Xi to the ith UE with transmit power P where E{Ixil}} = 1 assuming . In power-domain NOMA, X1 and X2 are superposed as X = + 2; thus the received signal at the ith UE can be written as yi = hix+Wi, in which hi is the complex-valued channel coefficient between the ith UE and the base sta- tion, and Wi denotes a zero-mean AWGN plus inter-cell interference. The PSD of Wi is Noi. In the downlink NOMA, the SIC receiver is implemented at the UE receiver. The optimal order for decoding is in the order of decreasing channel gain normalized by noise and P2 (UE) Total P1 (UE1) SIC of UE2 signal Frequency Decode UE1 sig

nal Decode UE2 signal Received SINR Figure 3.36 Illustration of the principle of downlink power-domain NOMA [52]. 364 Chapter 3 inter-cell interference power, that is, |hil2/Noi. Based on this order, we assume that any user can correctly decode the signals of other users whose decoding order comes before the cor- responding user. Therefore, the ith UE can remove the inter-user interference from the jth user whose channel gain is less than |hil2/Noi- In the example with two UEs, assuming that UE2 does not have to perform interference cancellation since it comes first in the decoding order. UE1, on the other hand, has to first decode X2 and subtract that component from received signal y1, then to decode X1 without interference from X2. Assuming successful decoding and no error propagation, the throughput of the ith UE, Ri, is given as follows [52]: It can be seen that power allocation for each UE greatly affects the user throughput and thereby the modulation and coding scheme used for data transmission of each UE. By adjusting the power allocation ratio P1/P2, the base station can effectively control the throughput of each UE. The overall cell throughput, cell-edge throughput, and user fairness are closely related to the adopted power allocation scheme [52]. In a system that uses orthogonal MA scheme and hypothetically serves two UEs, if normal- ized bandwidth 0<B<1 is assigned to the first UE, the remaining bandwidth 1 - B will be assigned to the second UE to maintain orthogonality between the users. The throughput of the ith UE Ri is given as follows: In power-domain NOMA, the performance gain relative to orthogonal MA increases when the difference in channel gains, for example, path loss between UEs, is large. The uplink capacity can be calculated in the similar manner as the downlink, although the formula is somewhat different. Defining Pr1 and P2 as the received powers at the base sta- tion from UE1 and UE2, respectively, the rate of each user in the case of non-orthogonal uplink access can be written as follo

ws: Multicarrier NOMA can be viewed as a variation of NOMA, where the users in a network are divided into multiple groups. The users in each group are served in the same orthogonal resource block following the NOMA principle, and different groups are allocated to New Radio Access Physical Layer Aspects (Part 1) 365 different orthogonal resource blocks. The motivation for employing hybrid NOMA is to reduce the system complexity. For example, assigning all the users in the network to a sin- gle group for the implementation of NOMA in one orthogonal resource block is problem- atic, since the user having the best channel conditions will have to decode all the other users' messages before decoding its own message, which results in high-complexity and high-decoding delay. Hybrid NOMA is an effective approach to make a tradeoff between system performance and complexity. Let's consider multicarrier NOMA as an example. The users in the cell are divided into multiple groups which are not necessarily mutually exclu- sive. The users within one group are assigned to the same subcarrier, and intra-group interference is mitigated using the NOMA principle. Different groups of users are assigned to different subcarriers, which effectively avoids inter-group interference. As a result, overloading the system, which is necessary in order to support more users than the number of available subcarriers and is required to enable massive connectivity, can be realized by the hybrid NOMA scheme. It is noted that, with hybrid NOMA, overloading is realized at reduced complexity since the number of users assigned to each subcarrier is limited [50-52]. The base station scheduler in power-domain NOMA searches and pairs multiple users for simultaneous transmission at each subband. To determine the set of paired users and the allocated power set at each subband, a multi-user proportional fairness (PF) scheduler may be used. The PF scheduling metric attempts to find the candidate user sets U and power sets Ps that maximize the following expression

over each sub- band S: QUAP) where(Umax,Psmax)=max Q(U,Ps) denotes the maximum argument of PF scheduling met- ric Q(U,Ps) for candidate user set U and allocated power set Ps over all users in the user set, n(k, U,Ps,t) is the instantaneous throughput of user k in subband S at time instance t (the time index of a subframe), whereas L(k,t) is the average throughput of user k. For power-domain NOMA, if we assume the possibility of dynamic switching between NOMA and orthogonal MA, then NOMA can be used only when it provides performance gains. Moreover, the number of users to be multiplexed over each subband is decided by searching all possible candidate user sets of different sizes up to m. The number of candidate user sets to be searched is given by [50-52]: In orthogonal MA schemes, the same MCS is selected over all subbands allocated to a sin- gle user. Therefore, the average signal-to-interference plus noise power ratio over all 366 Chapter 3 allocated subbands is used for MCS selection. However, when power-domain NOMA is uti- lized over each subband, user pairing and power allocation are performed over each sub- band. With such a mismatch between wideband MCS selection and subband power allocation granularities, the full-scale NOMA gains would not be realized. Furthermore, the higher the power allocation granularity, the more signaling overhead and thus performance degradation. Power control of uplink NOMA is different from that of downlink in two aspects. In the downlink direction, the transmission power is limited by the maximum transmission power of the base station; however, in the uplink, the transmission power optimization is con- strained by the maximum transmission power of individual UEs. In addition, there is a dif- ferent approach to transmit power control in the uplink. In the downlink, the superposed signal received at a UE experiences the same channel, that is, the signals of different UEs have the same channel gain at each UE receiver. Therefore, the design of downlink power control tries to crea

te sufficiently large difference among the signals of different UEs in the power domain in order to enable signal separation at the (SIC) receiver. For the uplink, due to different channels experienced by signals transmitted via different UEs, the received sig- nal powers of different UEs already have differences in the power domain. On the other hand, when NOMA is applied in the uplink, the ICI greatly increases since multiple UEs are allowed to simultaneously transmit, whereas in the downlink, ICI does not increase when NOMA is applied because generally the base station has fixed transmission power regardless of the number of multiplexed UEs [23]. 3.3.3.3 Scrambling-Based and Spreading-Based NOMA Schemes In addition to the NOMA schemes discussed in the previous sections, there are other schemes proposed as part of 3GPP Rel-16 study item on NOMA. Scrambling-based NOMA schemes use different scrambling signatures for each user and utilize a low-rate forward error correction code or code repetition for multi-user decoding. The scrambling operation is carried out after the modulation. MMSE with SIC (MMSE-SIC) and ESE are used for the multi-user detection. RSMA is one of the scrambling-based schemes under consideration which utilizes low cross-correlation properties of long pseudo-random scrambling codes. Long scrambling sequences are used in RSMA. However, a long user signature causes high- decoding complexity and latency. Following the descrambling step, the ratio of signal-to- interference power is directly proportional to the scrambling code length. It must be noted that each user can transmit signal at any time using the asynchronous RSMA. Depending on the application scenarios, single-carrier RSMA or multicarrier RSMA can be utilized. Single-carrier RSMA can be used in the uplink access to reduce the PAPR of the UE. Multicarrier RSMA can be utilized in the downlink access to simplify the receiver complex- ity in the frequency-selective wireless fading channels. RSMA can be extended to multi- ple layers. Treatin

g layers as virtual users, the data is split into multiple parallel layers New Radio Access Physical Layer Aspects (Part 1) 367 for each user. The complexity of multi-layer RSMA is higher than that of single-layer RSMA. The RSMA uses hybrid short-code spreading and long-code scrambling as the MA signatures. The generation of scrambling sequences can be UE-group and/or cell specific, wherein the sequence ID of scrambling code is a function of the cell ID and UE-group ID. One or multiple UE groups can be configured in a cell. The sequences used for scrambling code can be Gold sequences, Zadoff-Chu sequences, or a combina- tion of the two [12,23]. In Welch bound equality (WBE)-based spreading schemes such as RSMA, the design metric for the signature vectors is the total squared cross- correlation The lower bound on the total squared cross-correlation of any set of K vectors of length N is K2/N . The WBE sequences are designed to meet the bound on the total squared cross-correlations of the vector set with equality The spreading-based NOMA schemes use non-orthogonal short spreading sequences with relatively low cross-correlation, for distinguishing multiple users, and the spreading sequences are non-sparse. The spreading sequences and the decoding algorithm are dif- ferent for this category of NOMA schemes. In MUSA, modulation symbols of multiple users are spread by specially designed short sequences. All spreading symbols are trans- mitted over the same time-frequency resources. Multiple spreading sequences constitute a pool from which each user can randomly select. The spreading sequences of MUSA are complex-valued, in which the real part and imaginary part are both drawn from a real- valued multi-level set with uniform distribution. At the receiver, the codeword-level SIC detection is used to separate the target UE signal from the overlapped signals [23]. The average mutual information19 can be used as a performance metric to compare the spectral efficiencies of various NOMA schemes with OFDMA. This performance me

tric provides the maximum information rate that can be reliably transmitted for a given channel state information. In a single-user case, the average mutual information is calculated for the signal after constellation mapping and before the soft demapping. This analysis can be extended to the multiuser case for evaluating the achievable sum-rate of the NOMA schemes. Let X = XJ] denote the multi-user modulation symbol vector before the NOMA signature pattern, and y=[y1 y2 denote channel output bol vector at the receiver, in which J, K, M, andNrx represent the number of users, the NOMA signature length, the order of modulation, and the number of receive antennas, respectively. Assuming equi-probable input constellation points and the high-dimensional The average mutual information is defined as the weighted sum of the mutual information between each pair of the input and output events Xi and yj. The average mutual information is a measure of independence between the two random variables X and Y. In mathematical terms I(X; = E[I(x;yi)] = and I(x{;yi) = 368 Chapter 3 constellation set given by S = {W1,W2, 1}, the average mutual information in the multi-user case can be written A Monte Carlo simulation can be used to calculate the expectation function in the above equation. Assuming a six-user uplink multiple access channel, a tapped delay line TDL-A- 30 ns channel H and ideal channel estimation, multi-user average mutual information of the NOMA schemes under consideration in 3GPP for Rel-16 and OFDMA have been calculated and compared in Fig. 3.37. In this analysis, the overloading factor is 150%, spectral effi- ciency per user is 0.25 bps/Hz, and the sum-rate is 1.5 bps/Hz. The SNR is defined as the ratio of average total received multi-users' power to the noise power at each receive antenna for given bandwidth. Multiple UEs are assumed to share the same six physical resource blocks. The number of users, SE per user, and transmission bandwidth are identical for both NOMA and OFDMA schemes to ensure fairness. Fig. 3.3

7 further compares the BLER performance of the NOMA and OFDMA schemes for six and eight UEs, respectively. LTE turbo code is used for the channel coding and QPSK 1/2 for NOMA QPSK 3/4 is used for OFDMA, the overloading factor is 150%, spectral effi- ciency per user is 0.25 bps/Hz, and the sum-rate is 1.5 bps/Hz. In theory, the multi-user average mutual information analysis suggests that NOMA schemes provide higher capacity relative to OFDMA for given achievable sum-rate. In addition the, coding-based NOMA schemes (SCMA) have some performance advantage over other schemes. When the number of UEs increases the OFDMA system needs to use higher order modulation which would -1.48 OFMDA OFDMA SNR (dB) SNR (dB) Figure 3.37 Comparison of average mutual information and BLER of NOMA and OFDMA schemes [51]. New Radio Access Physical Layer Aspects (Part 1) 369 suffer from performance loss, while NOMA schemes with low order modulation can take advantage of superposition coding for higher overloading factor with slight performance degradation. The BLER performance advantage of NOMA schemes over OFDMA grows with the increase of the number of UEs [51]. 3.4 Duplex Schemes One of the key elements of any radio communication system is the way in which radio communications are maintained in the downlink and uplink. For cellular systems, it is nec- essary to enable simultaneous transmission of data in both directions, which creates a num- ber of constraints on the schemes that may be used to control over the air transmission. As a result, the choice of duplex scheme becomes the basic part of the overall specification for the cellular or any radio communications system. The term duplex refers to the bidirectional communications between two devices. When unpaired spectrum or alternatively the same RF carrier is used for downlink and uplink communications, the transmit/receive functions are time-multiplexed. When paired spectrum or alternatively two RF carriers are used for downlink and uplink communications, the transmit/receive function

s are frequency- multiplexed. 3.4.1 Frequency and Time Division Duplex Schemes The Frequency Division Duplex is a duplex scheme in which uplink and downlink transmis- sions occur simultaneously using different frequencies. The downlink and uplink frequen- cies are separated by sufficiently large frequency offset. For the FDD scheme to properly operate, it is necessary that the frequency separation, that is, channel separation between the transmission and reception frequencies, to be sufficient in order to prevent the receiver blocking due to high-power transmitter signal. The receiver blocking is an important issue in FDD schemes and often highly selective filters may be required. For cellular systems using FDD, filters are required in the base station and the user terminal to ensure suffi- cient isolation of the transmitter signal without desensitizing the receiver. While imple- mentation cost is not a significant constraint for the base stations, placing a filter in the user terminal involves higher complexity and cost. The use of an FDD system does enable simultaneous transmission and reception of signals. However, two RF channels are required, which in some cases may not be the efficient use of the available spectrum. The spectrum used for FDD systems is allocated by the regulatory bodies. Since there is a fre- quency separation between the uplink and downlink directions, it is not typically possible to reallocate spectrum to change the balance between the capacity of the uplink and down- link directions, if the capacity requirements for each direction vary over time. 370 Chapter 3 The Time Division Duplex is a duplex scheme where uplink and downlink transmissions occur at different times but may share the same frequency. In other words, the downlink and uplink transmissions are multiplexed in time and are not concurrent. While FDD transmissions require a large frequency separation between the transmitter and receiver frequencies, TDD schemes require a guard time or guard interval between transmission and rec

eption. This gap must be sufficient to allow the signals traveling from the remote transmitter to arrive before a transmission is started and the receiver is shut down. Although this delay is relatively short, switching between transmission and reception several times in a second, even a small guard time can reduce the spectral efficiency of the system since a percentage of time must be used for the guard interval. For small- sized cells, for example, up to one mile, the guard interval is typically small and accept- able. However, for large cell sizes, it may become an issue and may introduce signifi- cant overhead. FDD has been the dominating duplex scheme since the beginning of the mobile com- munication era. In the 5G era, FDD will remain the main duplex scheme for lower fre- quency bands; however, for higher frequency bands, especially above 10 GHz and targeting very dense deployments, TDD will play more important role. In very dense deployments with low-power nodes, the TDD-specific interference scenarios (direct base station-to-base-station and device-to-device interference) will be similar to the base-station-to-device and device-to-base-station interference that also occurs in FDD schemes. Furthermore, for the dynamic traffic variations expected in very dense deployments, the ability to dynamically assign transmission resources (e.g., time slots) to different transmission directions may allow more efficient utilization of the avail- able spectrum. To reach its full potential, 5G will allow for very flexible and dynamic assignment of TDD transmission resources. This contrasts with current TDD-based mobile technologies, including TD-LTE, for which there are restrictions on the down- link/uplink configurations; thus there typically exist assumptions about using the same configuration for neighboring cells and between neighboring operators. The guard interval required for TDD will comprise two main elements: (1) A time- allowance for the propagation delay for any transmission from a remote transmitter to arri

ve at the receiver. This will depend upon the distances involved (i.e., cell radius) and (2) a time-allowance for the transceiver to switch from receive-to-transmit mode. The switching times can vary considerably depending on the implementation but can take a few microse- conds. As a result, TDD is not normally suitable for use over very large cell sizes as the guard time increases and the spectral efficiency decreases. is often found that traffic in both directions is not balanced. Typically, there is more data transmitted in the downlink direction of a cellular system. This means that the capacity should be ideally greater in the downlink direction. Using a TDD system, it is possible to New Radio Access Physical Layer Aspects (Part 1) 371 change the capacity in either direction simply by changing the number of time slots allo- cated to each direction. This is often dynamically configurable SO it can be adapted to meet the demand. A further aspect to be noted with TDD transmission is the latency. Since data may not be able to be routed immediately to the transmission chain as a result of the time multiplexing between transmit and receive circuitry, there will be a small delay between the data being generated and being actually transmitted. Typically, this may be a few millise- conds depending on the frame timing. Both TDD and FDD have their advantages and each can be used in different deployment scenarios. Before deciding on a particular type of duplex scheme, it is necessary to analyze the advantages and disadvantages of each duplex mode. Table 3.7 summarizes and compares the relative attributes of TDD and FDD systems. The new radio can operate in paired and unpaired spectrum using a common frame structure unlike LTE where two different frame structures were used, which was later expanded to three for support of unlicensed operation introduced in 3GPP Rel-13. The basic NR frame structure is designed such that it can support both half-duplex and full-duplex operation. In half duplex the device cannot transmit an

d receive at the same time. A major issue in limiting the capacity of non-cooperative cellular massive MIMO networks operating in TDD mode is the pilot contamination. The rise of asymmetric uplink/downlink traffic patterns necessitates that the ratio between downlink/uplink traffic changes over the time; thus the static paired spectrum for downlink/uplink is not efficient for supporting such dynamic asymmetric traffic, particularly in UDNs. Flexible duplex can better adapt to dynamic asymmetric traffic. With flexible duplex, the uplink spectrum defined in FDD sys- tems can be reallocated for downlink transmission with high flexibility. Considering the potential cross-link interference, that is, downlink to uplink and uplink to downlink, the transmission power for downlink transmission on the uplink spectrum should be constrained to a relatively low. Flexible duplex can be applicable to small cells with low transmission power and relay base stations. 3.4.2 Half-Duplex and Flexible-Duplex Schemes The LTE and NR support TDD and FDD schemes with a great extent of commonality in the baseband processing. In order to reduce the implementation complexity and cost of FDD terminals and further to increase the reuse of baseband functional elements, a half- duplex FDD (H-FDD) operation is supported where the downlink and uplink transmissions are not simultaneous, but occur in two different frequencies. A classic H-FDD operation does not efficiently utilize the radio resources on the downlink and uplink RF carriers. The complementary grouping and scheduling of users would allow efficient use of downlink and uplink resources in an H-FDD operation. For H-FDD operation, a guard period is virtually 372 Chapter 3 Table 3.7: Comparison of time division duplex (TDD) and frequency division duplex (FDD) attributes. Attribute Paired Does not require paired spectrum as both Requires paired spectrum with sufficient spectrum transmit and receive occur on the same frequency separation to allow simultaneous channel transmission and receptio

n Hardware cost Lower cost as no diplexer is needed to Diplexer is needed and the implementation isolate the transmitter and receiver cost is higher Channel Channel propagation is the same in both Channel characteristics are different in both reciprocity directions which enables estimation of the directions as a result of the use of different downlink channel from the uplink channel frequencies UL/DL It is possible to dynamically change the UL UL/DL bandwidths are determined by asymmetry and DL ratio based on the traffic volume in frequency allocation designated by the each direction regulatory authorities. It is therefore not possible to dynamically adapt to the traffic volume Guard period/ Guard period required to ensure uplink and Frequency separation is required to provide frequency downlink transmissions do not interfere. sufficient isolation between uplink and separation Large guard period will limit the capacity. downlink. However, large frequency Larger guard period normally required if separation does not impact the capacity. distances are increased to accommodate Note that a guard band in frequency larger propagation delays. Note that a domain is required to suppress interference guard band in frequency domain is required to adjacent bands. to suppress interference to adjacent bands. Discontinuous Discontinuous transmission is required to Continuous transmission is possible transmission allow both uplink and downlink transmissions. This can degrade the performance of the RF power amplifier in the transmitter Switching point Base stations are required to be Not applicable synchronization synchronized with respect to the uplink and downlink transmission times. If neighboring base stations use different uplink and downlink assignments and share the same channel, then interference may occur between cells. aA diplexer is a passive device that implements frequency domain multiplexing. Two ports are multiplexed onto a third port. The signals on each input ports occupy nonoverlapping frequency bands. Consequent

ly, the signals on the input ports can coexist on the output port without interfering with each other. On the other hand, a duplexer is a device that allows bidi- rectional communication over a single channel. In radar and radio communications systems the duplexer isolates the receiver from the transmitter while permitting them to share a common antenna. Most radio repeater systems include a duplexer. Duplexers are designed for operation in the frequency band used by the receiver and transmitter and must be capable of handling the output power of the transmitter. They must provide sufficient isolation between transmitter and receiver to prevent receiver desensitization [30]. created by the UE by not receiving the last OFDM symbol(s) of a downlink subframe immediately preceding an uplink subframe in which the UE is active. The length of the guard period is the sum of the maximum round-trip propagation delay in the cell, transmit- to-receive and receive-to-transmit switching delay at the UE. New Radio Access Physical Layer Aspects (Part 1) 373 Cell A Classic Cell B Cell A Flexible duplexing Cell B Cell A Flexible duplexing with Cell B synchronization Interference Victim cell Figure 3.38 Examples of flexible duplexing schemes in TDD mode. The choice of duplex scheme is typically determined by the spectrum allocation. For lower frequency bands, allocations are often paired, implying FDD mode is dominant. However, in higher frequency bands, unpaired spectrum allocations are increasingly common; thus TDD mode is used. 3GPP NR supports both duplexing methods. However, unlike LTE where the TDD uplink-downlink allocation does not change over time, NR supports dynamic TDD as a key technology component. In dynamic TDD, parts of slot can be dynamically allocated to either uplink or downlink based on the scheduler decision. This enables network adaptation to traffic variation, which is particularly common in dense deployments with a relatively small number of users per base station. The device follows any scheduling decisions

, and it is up to the network scheduler, if necessary, to coordinate the scheduling decisions between neighboring sites to avoid inter-cell interference. There is also a possibility to semi-statically configure the transmission direction of some of the slots, a feature that can allow reduced device energy consumption as it is not necessary to monitor downlink control channels in slots that are a priori known to be reserved for uplink usage (Fig. 3.38). 3.4.3 Full-Duplex Schemes The term full-duplex was traditionally used to describe a simultaneous bidirectional commu- nication, in contrast to half-duplex, which assumed time division duplexing. However, in recent years, the term has carried a new concept and that is when a device can transmit and receive at the same time and over the same carrier frequency. Some authors refer to the lat- ter scheme as in-band full duplex in the literature to distinguish this new concept from its traditional usage. It is intuitive that enabling wireless devices to operate in full-duplex mode offers the potential to double the spectral efficiency, considering that traditional approaches for increasing spectral efficiency such as adaptive coding and modulation, MIMO and smart antennas have almost reached their maximum limits. In addition, full-duplex scheme 374 Chapter 3 improves the reliability and flexibility of dynamic spectrum allocation in cognitive radio networks and enables the small cells to reuse radio resources simultaneously for access and backhaul [29]. The main challenge for a full-duplex radio is self-interference and how to manage and sup- press it. Self-interference or transmitter leakage was studied earlier and refers to the signal that leaks from the device transmitter to its own receiver. In general, the transmitter signal is about 100 dB stronger than the reference sensitivity level of the receiver. A considerable part of this transmitted signal leaks into the receiver chain, causing serious issues in decod- ing the desired signal, which could be considered noisy,

with a dramatically affected SNR. To achieve the best performance of a full-duplex system, the self-interference signal must be suppressed to reach the receiver's noise floor. The self-interference may be originated from the linear components and the RF carrier itself, which is attenuated and reflected from the environment. The received distortion can be modeled as a linear combination of differ- ent delayed copies of the original carrier. It can be further generated by nonlinear compo- nents because the imperfect radio circuits typically create third and fifth intermodulation products of the transmit signal. These higher order intermodulation terms have significant frequency content at frequencies close to the transmitted frequencies, which directly corre- spond to other harmonics. The self-interference can also be caused by the transmitter noise, which appears as an increase of about 50 dB over receiver noise floor. Due to random nature of this component, it can only be canceled by subtracting the appropriately weighted transmitter signal sampled in the analog domain from the received signal. For narrowband systems, the self-interference channel can be modeled as gain and delay functions, whereas wideband systems require a more complex model, because the reflected-path self-interfer- ence channel is often frequency-selective as a result of multipath propagation. In general, one can model the process in the digital domain that is valid for both the nar- rowband and wideband scenarios as r(n) = ra(n) + wr(n) + rdsi(n) + rssi(n) where r(n) is total received complex-valued baseband samples, ra(n) is the desired signal from the remote node, rdsi(n) denotes the complex-valued samples caused by the direct self-interference component signal between the transmit and receive antennas in case of two antennas, or leaked signal in the circulator in case of one antenna, rssi(n) represents the complex-valued samples caused by scattered self-interference components, and wr(n) denotes additive white Gaussian noise. A circulato

r is a passive three-port device in which an RF signal entering any port is only transmitted to the next port in rotation. Both direct and scattered self- interference can be represented as a combination of linear and nonlinear components. The suppression in the propagation domain can mitigate both linear and nonlinear self- interference at the same time and with the same isolation value. Meanwhile, cancellation techniques in the analog and digital domains have different performances for the two com- ponents. The self-interference cancellation is implemented in three domains: propagation, analog, and digital. Since none of these domains can meet the required cancellation New Radio Access Physical Layer Aspects (Part 1) 375 requirements, hybrid solutions have been proposed in the literature [40,41]. The primary role of self-interference cancellation in the propagation and analog domains is to avoid the saturation of the receiver due to the high power of the self-interference signal as this power exceeds the ADC dynamic range and limits its precision after conversion because the desired signal is much weaker than the self-interference. Self-interference comprises several components with different characteristics depending on the specifics of the full-duplex system implementation, such as the number of antennas, the characteristics of the RF components and the environment. The constituents of the self- interference can be classified by linearity. Linear components involve multipath propagation between the transmit/receive antennas. For a single-antenna system, linear components include the leakage of the circulator or the reflections from impedance mismatch. Components of RF circuits, such as attenuators and delay lines, are also modeled as linear systems for analog self-interference cancelation. The linear components of self-interference can be removed by the existing channel estimation methods, as in most conventional wire- less communication systems. Nonlinear components are usually created by PAs in transmit- te

rs and low-noise amplifiers in receivers. The nonlinearity of the PA is generated because the power of the output signal is saturated for the high-power input signal, which worsens for modulation schemes with high PAPR such as OFDM and wideband CDMA. Intermodulation distortion, caused by the nonlinearity, interferes with the linear model of self-interference. The intermodulation can be theoretically calculated by a Volterra series or approximated by a Taylor series. Since the even-ordered terms are out-of-band, the Taylor series would include only odd-ordered terms. Other RF imperfections/impairments, for example, I/Q imbalance, phase noise and transmitter noise can occur in the transmitter side. The I/Q imbalance occurs when there is mismatch between the gain and phase of the two sinusoidal signals, which deteriorates the baseband transmitter signal. The imperfection of the local oscillator can also degrade the linearity of the transmitter signal. In general, most of the impacts of the oscillator impairment is noticeable in random deviations in the output frequency, which can be modeled as phase noise. Transmitter noise also includes Volterra series is a mathematical model for nonlinear behavior of systems in which the output of the nonlin- ear system always depends on the input. This provides the ability to capture the memory effect of devices. In mathematics, Volterra series denotes functional expansion of a dynamic, nonlinear, time-invariant function. The Volterra series, which is used to prove the Volterra theorem, is an infinite sum of multidimensional con- volutional integrals. A continuous time-invariant system with x(t) as input and y(t) as output can be expanded Volterra series as y(t) = ho + s sb hn Tn) Im=1x(t-tm)dTm where the constant term ho on the right-hand side is often set to zero by suitable choice of output level y. The function hn (T1, Tn) is called the nth order Volterra kernel. It can be regarded as a higher-order impulse response of the system. If N is finite, the series can be truncated.

If a, b, andN are finite, the series is called doubly finite. Since in any physically realizable system, the output can only depend on previous values of the input, the kernels tn) will be zero, if any of the variables t1, t2, tn are negative. The integrals may then be written over the half range from zero to infinity. Therefore, if the operator is causal a > 0. 376 Chapter 3 thermal noise which is typically generated by RF components. Using the estimated linear wireless channel, this method can mitigate transmitter impairments. Unlike other SIC meth- ods, the distortion of the transmitter signal by PA nonlinearity or by I/Q imbalance is obtained directly by the auxiliary receive chain. The specifications of a full-duplex system can be classified into three categories: main specifications, ADC specifications, and self-interference specifications. While the residual self-interference power PRSI is higher than the receiver noise floor level, the signal to self- interference plus noise ratio in a full-duplex system is lower than the SNR of a half-duplex system receiver. This means that the maximum efficiency of full-duplex cannot be achieved. In general, self-interference cancellation solutions are a combination of several techniques to help meet the system requirements. Fig. 3.39 provides an example, showing the average performance value achieved in each domain. In case of a shared transmit/receive antenna system, the suppression is performed using a three-port RF circulator. The ferrite within the device can be considered as a propagation domain. Achievable isolation by the circulator is between 15 and 30 dB, and in the case of wideband operation, the maximum value would decrease. In a separate-antenna system, sev- eral self-interference cancellation techniques can be used. The two transmit antennas can be placed at distances d and d + X/2 away from the receive antenna. Separating the two trans- mitters by half a wavelength causes their signals to cancel one another. For narrowband sig- nals, this technique is exp

erimentally proved to be sufficiently robust; however, the suppression drastically decreases in case of wideband signals. The antenna directionality isolates the receiving antenna from the interfering signals of the transmitting antenna; TX data Baseband processing module Path loss Analog SIC control Digital SIC Analog SIC Full scale Baseband processing RX data module SSINR Dynamic range Digital Digital SIC Analog SIC Passive suppression (20-35 dB) using separate (20-45 dB) before LNA to (20-30 dB) suppressing linear and nonlinear SIC avoid ADC saturation and linear and nonlinear SIC limiting precission Quantized noise floor Figure 3.39 Example Wi-Fi or LTE signals, which are typically transmitted at an average power of 23 dBm (200 mW). The thermal noise level at 20 MHz bandwidth is about - 101 dBm 174 dBm/ Hz + 73 dB) [40,41]. New Radio Access Physical Layer Aspects (Part 1) 377 however, such an approach would not work for point-to-point full-duplex scenarios. Electromagnetic shielding can enhance the isolation between transmit and receive anten- nas. Nevertheless, one disadvantage is that the shielding affects the far-field antenna pat- terns because it prevents the antenna from transmitting to/receiving from the shielding direction; thus it is only relevant to the case of directional antennas. Self-interference can be mitigated using orthogonal polarization between the transmit/receive antennas, achiev- ing about 10-20 dB isolation in an anechoic chamber and 6-9 dB in a reflective room at 2.4 GHz. The dual-polarized antenna can suppress self-interference by transmitting and receiving through orthogonal polarization. The isolation characteristic of a dual-polarized antenna can be expressed by its XPD factor. In transmit mode, XPD is the proportion of the signal that is transmitted in the orthogonal polarization to the desired direction. In receive mode, it is the antenna's ability to maintain the incident signal's polarization purity. For example, if a perfectly vertically polarized signal (containing no horizo

ntal component) was incident upon a single polarized receive antenna, the electrical and mechanical imperfections would introduce a small amount of ellipticity to the polarization of the signal. The signal can be thought of as having both vertical and horizontal compo- nents. The ratio of the resulting horizontal to vertical components is the defined as XPD. Active self-interference cancellation techniques in the analog domain generate a replica of the transmit signal and then adjust it to match the self-interference channel, making the replica similar to the self-interference signal in order to subtract it from the total received signal. This replica can be generated either in the analog domain or in the digi- tal domain before the DAC. The self-interference cancellation signal is performed in the same domain from which it was created; thus no additional ADC/DAC is required. Replication of the transmission signal in the analog domain can be achieved by tapping the transmitter chain, using a power splitter, or using a balanced-to-unbalanced circuit in the case of two separate antennas. Experiments show the practical benefits of the latter approach relative to phase shifter, notably the flatter response within a wide frequency band. After creating an exact negative replication of the RF reference signal, the replica is adjusted by delay and attenuation elements to match the self-interference [40,41] (Fig. 3.40). It can be observed in the experiments that the nonlinear components of self-interference can be 80 dB higher than the receiver noise floor. A large fraction of this component may be eliminated along with the linear self-interference by self-interference cancellation techni- ques in the analog and propagation domains, but the residual nonlinear self-interference in the digital domain, which amounts to 10-20 dB, needs to be canceled. In general, nonlinear self-interference cancellation methods are added to the linear methods to achieve optimal performance. A generic model for approximation of the nonlinear f

unction is based on a Taylor series, where the output transmitted signal is represented as y(n) = (n) where xm(n) is the ideal passband analog signal for the digital representation of x(t). It can Chapter 3 TX radio TX radio processing processing Self-interference signal AUX TX AUX RX radio Digital SIC radio processing processing Self-interference signal RX radio RX radio processing processing Self-interference cancelation signal Self-interference cancelation signal Figure 3.40 Auxiliary transmit/receive chain in full-duplex systems (40,41]. be shown that for practical wireless systems, only the odd orders of the polynomial contrib- ute to the in-band distortion. Furthermore, only a limited number of odd orders contribute to the distortion, and higher orders can be neglected. The nonlinearity is typically character- ized by the third-order intercept point, which is defined as the point at which the power of the third harmonic is equal to the power of the first harmonic. Another source of impairment is IQ imbalance which is caused by the gain and phase mismatches between I and Q branches of the transmitter and receiver chains. This imbalance results in the com- plex conjugate of the ideal signal to be superimposed with some level of attenuation. Thus the output of an imperfect IQ mixer is a transformation of an input signal x(t) where both direct and conjugated signals are filtered and then summed together. The IQ imbalance can be modeled and compensated using widely linear filters. The results of studies suggest that the oscillator phase noise is one of the main self- interference cancellation challenges that limit the performance of full-duplex systems. It was assumed that when transmitter and receiver use a common local oscillator, the level of phase noise would remain at a tolerable level. However, this consideration is not always valid, especially in the case of OFDM systems. The studies show that with a phase noise variance between 0.4 and 1.0 degrees, the reduction in self-interference cancellation perfor-

mance is about 20-25 dB for OFDM systems. This can be explained by the phase noise causing two effects: common phase error and inter-carrier interference. The former may have acceptable levels as previously assumed, but the latter stimulates an enhancement in Widely linear filters augment the data vector with its conjugate, thus providing the complete second-order statistical information when computed using the minimum mean square error cost function. Widely linear fil- ters have been proposed for applications such as interference cancelation, demodulation, and equalization for direct sequence code-division-multiple-access systems, and array receivers. A widely linear filter forms the estimate of a desired sequence d(n) through the inner product ywL(n) = wHx(n) where the weight vector W1 W2N-1] has double dimension compared to the linear filter and X(n) = [x(n)x*(n)] T with the definition x(n) = [x(n) x(n-1) x(n-N+1)]T. New Radio Access Physical Layer Aspects (Part 1) 379 self-interference cancellation performance, which is achieved by consecutively estimating and suppressing the ICI. The goal of digital self-interference cancellation is to remove the residual self-interference after analog self-interference cancellation especially that originated from NLoS reflections. Various signal processing and calculations performed after ADC for the purpose of self- interference cancellation are classified as digital self-interference cancellation. Since they operate in the digital domain, the baseband equivalent models of self-interference should be determined before calculation. The models include linear and nonlinear self-interference models. Once the self-interference signal is modeled as a function of the transmitted signal, it can be estimated from the transmitted and received signals. The residual self-interference is then reconstructed as the output of the function, and subtracted from the received signal [40,41]. 3.5 Operating Frequency Bands The new radio has been developed to operate in two distinct frequency re

Table 3.8: NR operating bands in frequency range (FR) 1 and FR2 [9]. Uplink Operating Bands Downlink Operating Bands Total Total Frequency Operating FULLOW-FUL_High Bandwidth FDL_Low-FDL_High Bandwidth Duplex Range (MHz) (MHz) (MHz) (MHz) 1920 - 1980 2110-2170 1850-1910 1930-1990 1710 -1785 1805-1880 824-849 869-894 2500-2570 2620-2690 880-915 925-960 832-862 791-821 703-748 758-803 2570-2620 2570-2620 2496-2690 2496-2690 1432-1517 1432-1517 1427-1432 1427-1432 1710-1780 2110-2200 1695-1710 1995-2020 663-698 617-652 1427-1470 1475-1518 1432-1517 1427-1432 3300-3800 3300-3800 3300-4200 3300-4200 4400-5000 4400-5000 1710-1785 880-915 832-862 703-748 1920-1980 26,500-29,500 26,500-29,500 24,250-27,500 24,250-27,500 37,000-40,000 37,000-40,000 consists of 12 subcarriers, the maximum number of resource blocks for each channel bandwidth depends on the subcarrier spacing. The spacing between carriers will depend on the deployment scenario, the size of the fre- quency block available and the channel bandwidths. The nominal channel spacing dfs between two adjacent NR carriers with 100 kHz channel raster is defined dfs = (BW ch1 + BW (ch2)/2 where BW ch1 and BWch2 denote the channel bandwidths of the two respective NR carriers. The channel spacing can be adjusted depending on the channel raster to optimize performance in a particular deployment scenario. For NR carriers in FR1 New Radio Access Physical Layer Aspects (Part 1) 381 Channel bandwidth (MHz) Guard band Transmission bandwidth configuration NRB [RB] Guard band Transmission bandwidth [RB] Spectrum Resource Active resource Note: Guard bands can be asymmetric emission mask block blocks Figure 3.41 Definition of channel and transmission bandwidth configuration for one NR channel [9]. Table 3.9: Maximum transmission bandwidth configuration in NRB for frequency ranges 1 and 2 [9]. Bandwidth (MHz) Bandwidth (MHz) Subcarrier Spacing (SCS) (kHz) Number resource blocks operating bands with 15 kHz channel raster, the nominal channel spacing is defined as dfs = (BW ch1 + BW /

2 + [5kHz, 0]. Furthermore, for NR carriers in FR2 operating bands with 60 kHz channel raster, the nominal channel spacing is defined dfs = (BW ch1 + BW ch2)/2 + [20kHz, 0] [9]. In the case that multiple numerologies are multiplexed over the same symbol, the mini- mum guard band on each side of the carrier is the guard band applied at the configured gNB channel bandwidth for the numerology that is transmitted/received immediately adja- cent to the guard band. Nevertheless, if multiple numerologies are multiplexed on the same symbol and the gNB channel bandwidth is wider than 50 MHz when operating in FR1, the guard band applied adjacent to 15 kHz SCS is the same as the guard band 382 Chapter 3 defined for 30 kHz SCS for the same channel bandwidth. If multiple numerologies are multiplexed on the same symbol and the gNB channel bandwidth is wider than 200 MHz when operating in FR2, the guard band applied adjacent to 60 kHz SCS is the same as the guard band defined for 120 kHz SCS for the same channel bandwidth. For each numerol- ogy, the starting point of its transmission bandwidth configuration on the common resource block (CRB) grid for a given channel bandwidth is indicated by an offset to ref- erence point A in the unit of the numerology while the indicated transmission bandwidth configuration must fulfill the minimum guard band requirements [9]. The channel raster defines a subset of RF reference frequencies that can be used to identify the uplink and downlink RF channel positions, where the RF reference frequency corre- sponding to an RF channel is mapped to a resource element on that carrier. A global fre- quency raster is further defined for all frequencies from 0 to 100 GHz and is used to define the set of RF reference frequencies FREF that are used for signaling the location of RF chan- nels and synchronization signal blocks. The granularity of the global frequency raster is defined as \F global. For each operating band, a subset of frequencies from the global fre- quency raster are applicable for that ban

d and form a channel raster for that band with granularity \Fraster \F global. The channel raster is mapped to physical resource block NPRB = NRB/2 with resource element index k = 0 or 6 depending on whether NRBMOD2 =0 or 1, respectively [9]. The RF reference frequency in the uplink and downlink is identified by the NR absolute radio frequency channel number (NR-ARFCN) in the range [0. 13279167] on the global frequency raster. The relationship between the NR-ARFCN and the RF reference frequency FREF in MHz for the downlink and uplink is given as FREF = FREF-offset + \Fraster (NREF - NREF-offset), where FREF-offset and NREF-offset are given in Table 3.10 and NREF represents the NR-ARFCN [9]. For the supplementary bands and bands (n1, n2, n3, n5, n7, n8, n20, n28, n66, n71) (see Table 3.8), the reference frequency is defined as FREF-shift = FREF shift where shift==0 or 7.5kHz is signaled by the network. A channel raster of 100 kHz is used for some NR operating bands, in which \Fraster = 20\Fglobal; thus every 20th NR-ARFCN within the operating band can be used for the channel raster with the step size of 20. For NR operating bands below 3 GHz with 15 kHz channel raster \Fraster NstepAF global In this case, every nstep E {3,6} NR-ARFCN within the operating band is a candidate channel raster. There are NR operating bands above 3 GHz where the channel raster is either 15 or 60 kHz. In that case, \Fraster = NstepAF global and every nstep E {1,2} NR-ARFCN within the operating band is a candidate channel raster. New Radio Access Physical Layer Aspects (Part 1) 383 Table 3.10: NR absolute radio frequency channel number parameters for the global frequency raster [9]. Frequency Range (MHz) Fglobal (kHz) FREF-offset (MHz) FREF-offset NREF Range 0-3000 0-599,999 3000-24,250 600,000 600,000-2,016,666 24,250-100,000 24,250 2,016,667 2,016,667-3,279,167 Table 3.11: Global synchronization channel number (GSCN) parameters for the global frequency raster [9]. Frequency (MHz) Synchronization Block Frequency Position SSREF Range of G

SCN 0-3000 1200 N kHz + 50 M kHz; N = 1:2499, ME{1,3,5} 3N (M - 3)/2 2-7498 3000-24,250 3000 MHz + 1.44 N MHz; N = 0:14,756 7499 + N 7499-22,255 24,250-100,000 24250.08 MHz + 17.28N MHz; N = 0:4383 22,256 + N 22,256-26,639 The synchronization raster signifies the frequency positions of the synchronization block that can be used by a UE for frequency acquisition when the UE has not received an explicit signaling indicating the synchronization block position. A global synchronization raster is defined for all frequencies. The frequency position of the synchronization block is defined by parameter SSREF with a corresponding global synchronization channel number (GSCN). The GSCNs are numbered in increasing frequency order as shown in Table 3.11. The physical resource block number is NPRB = 10 for the synchronization raster mapping to the synchronization block resource elements [9]. 3.6 Frame Structure and Numerology 3GPP NR is designed to operate from 450 MHz to 100 GHz with a wide range of deploy- ment scenarios, while supporting a variety of services. Given OFDMA was chosen as the multiple-access scheme for the downlink and uplink, it is not possible for a single numerol- ogy to satisfy the requirements of various use cases. Therefore, NR defines a family of OFDM numerologies for various frequency bands and deployment scenarios. The advantage of 3GPP NR relative to LTE is that it defines multiple numerologies which can be mixed and used simultaneously. A numerology is defined by a subcarrier spacing and a CP. The requirements for the OFDM subcarrier spacing is determined based on the carrier frequency, phase noise, delay spread, and Doppler spread. The use of smaller subcarrier spacing would result in either large EVM due to phase noise or more stringent requirements on the local oscillator. The small subcarrier spacing further leads to performance degradation in high Doppler scenarios. The required CP overhead and thus anticipated delay spread sets an 384 Chapter 3 Table 3.12: Supported OFDM parameters in NR [12].

Supported for Supported for Cyclic Subcarrier Cyclic Synchronization Useful Prefix Symbol Spacing Prefix (PDSCH, Blocks (PSS, SSS, Symbol Length Length symbol Af = 2H X 15 kHz PUSCH, etc.) PBCH) Length us Nslot subframe Normal 66.67 71.35 Normal 33.33 35.68 Normal/ 16.67 17.84 extended Normal Normal upper limit for the subcarrier spacing. A large subcarrier spacing would result in unwanted overhead due to CP. The maximum FFT size of the OFDM modulation along with subcar- rier spacing determines the channel bandwidth. Based on these observations, the subcarrier spacing should be as small as possible, while the system is still robust against phase noise and Doppler spread and supports the desired channel bandwidth. As shown in Table 3.12, in NR, the Primary and Secondary Synchronization Signals (PSS/SSS) and the Physical Broadcast Channel (PBCH), which are collectively known as SS block, will use 15/30 kHz SCS for sub-6 GHz and 120/240 kHz for above 6 GHz frequency bands. In the case of Physical Random-Access Channel (PRACH), the long preamble sequence utilizes 1.25/5 kHz SCS, in addition to the short preamble sequence using 15/30/60/120 kHz SCS. In other words, NR exploits a scalable OFDM subcarrier spacing (powers of 2) to support various frequency bands and deployment scenarios, where Af = 2H X 15kHz with u = {0, 1, 4} considered for PSS, SSS, and PBCH, and u = {0, 1, 2, 3} designated for other physical channels. The normal CP is supported for all subcarrier spacing values, whereas extended CP is only supported for u = 2. From the network perspective, multiplexing of dif- ferent numerologies over the same NR carrier bandwidth is possible in TDM and/or FDM manner in the downlink and uplink. From the UE perspective, multiplexing different numer- ologies is performed in TDM and/or FDM manner within or across a subframe. Regardless of the numerology used, the lengths of radio frame and subframe are always 10 and 1 ms, respectively. Different numerologies will then translate into the number of slots per sub- frame.

The higher the subcarrier spacing, the more slots that can be accommodated per sub- frame [12]. Fig. 3.42 illustrates the NR frame, subframe, slot and mini-slot structure. Different numerologies can be used in diverse deployment scenarios with their correspond- ing performance requirements. For example, the lower the subcarrier spacing, the larger the size, which will be suitable for the lower frequency deployments. At the same time, larger subcarrier spacing will allow shorter latency since the symbol duration will be shorter New Radio Access Physical Layer Aspects (Part 1) 385 10 ms Frame i Frame i + 1 Slot 0 SCS = 15 kHz Slot 0 Slot 1 SCS = 30 kHz Slot 0 Slot 1 Slot 2 Slot 3 SCS = 60 kHz Slot 0 Slot 1 Slot 2 Slot 3 Slot 4 Slot 5 Slot 6 Slot 7 SCS = 120 kHz OFDM symbols Figure 3.42 3GPP NR frame structure [12]. (see Fig. 3.42). Fig. 3.43 shows the relationship of the numerology, cell size, latency, and the carrier frequency [53]. The frame structure provides the basis for the timing of physical signal transmission. The timing scale is different for data, control, and synchronization physical channels. The sam- pling time in NR is defined as T = 1/(AfmaxNFFT) where fmax = 480kHz and NFFT = 4096. In order to support multiple OFDM numerologies, the parameter u and the corresponding CP for a BWP are signaled (configured) by the RRC parameters DL-BWP- mu and DL-BWP-cp for the downlink and UL-BWP-mu and UL-BWP-cp for the uplink [12,21]. The downlink and uplink transmissions are structured in the form of radio frames in the time domain with frame duration Tframe = (AfmaxNFFT/100)2 T = 10 ms, where each frame consists of ten subframes of Tsubframe = (AfmaxN//1000): duration. The num- consecutive OFDM symbols in each subframe is defined as symbols subframe subframe (u). In FDD mode, there is one set of frames in the uplink and one set of frames in the downlink on a carrier. As shown in Fig. 3.44, the ith uplink frame number starts at TTA = (NTA + NTA-offset) T before the start of the corresponding downlink frame at the

UE in order to ensure uplink frame synchronization where NTA-offset depends on the frequency band of operation [18]. It can be seen that for the 15-kHz subcarrier spacing, Chapter 3 Cell size Latency 15 kHz 15 kHz 30 kHz 30 kHz 15 kHz 30 kHz 120/240 30 kHz 60 kHz 60 kHz 60 kHz Medium Very high Frequency Figure 3.43 Relationship of numerology, carrier frequency, latency, and cell size [53]. Downlink frame Uplink frame TTA = (N TA + N TA-offset 3GPP 38.133 Table 7.1.2-2 Frequency range of uplink transmission NTA-offset (XT) Timing advance from FR1 FDD band without LTE-NR coexistence case or FR 1 25,600 MAC CE or RAR TDD band without LTE-NR coexistence case message FR1 FDD band with LTE-NR coexistence case FR1 TDD band with LTE-NR coexistence case 39,936 or 25,600 13,792 NTA 16.64T /2" TA=0, 1,..., 3846 (From RAR message) T = 509 ns NTA-new = NTA-old + (T -31)16.64/2 T=0, 1,..., 63 (From MAC CE) Figure 3.44 Illustration of uplink timing calculation in NR [ [12,39]. New Radio Access Physical Layer Aspects (Part 1) 387 an NR slot has the same structure as the LTE subframe, which is important for supporting LTE/NR coexistence scenarios. In the case of co-located deployment, slot and frame structures may be aligned to simplify cell search and inter-frequency measurements [9,16,17,19]. Coordination of control signals and channels in time domain will also be possible to avoid inter- ference between LTE and NR. Given that a slot is defined as a fixed number of OFDM sym- bols, a larger subcarrier spacing results in a shorter slot duration, which can be used to support low-latency applications. 3GPP NR further supports a more efficient approach to low-latency transmissions by allow- ing scheduling shorter slot sizes known as mini-slots (see Fig. 3.42). The mini-slot-based transmissions can also preempt an already ongoing slot-based transmission to another device, allowing immediate transmission of application data requiring very low latency. Mini-slots can be used for low-latency applications such as URLLC and operation in u

nlicensed bands, for example, to start transmission directly after a successful listen-before-talk procedure without waiting for the slot boundary. Mini-slots can consist of two, four, or seven OFDM symbols, where the first symbol includes (uplink or downlink) control information (see Fig. 3.42). For low-latency applications, the HARQ protocol can be configured either on a slot or a mini-slot basis. For the regular frame structure used by delay-tolerant applications, slot bundling as in LTE is also possible. Mini-slots may also be used for fast flexible scheduling of services (pre- emption of URLLC over eMBB). However, mini-slots are likely to be supported by few UE categories. A major difference between LTE and NR in terms of scheduling granularity is that LTE transmission time interval is fixed at 1 ms whereas NR transmission interval is a slot or a fraction of slot whose length is a function of the subcarrier spacing. For given subcarrier spacing parameter 11, the slots are numbered as ascending order within a subframe. There are N symbol slot conse- cutive OFDM symbols in a slot where N slot symbol depends on the CP. The start of slot nslot in a subframe is aligned in time with the start of OFDM symbol NslotNsymbol in the same subframe [12]. In the TDD mode, the OFDM symbols in a slot can be classified as downlink, flexible, or uplink. Table 3.13 shows possible slot formats. In a downlink frame slot, the UE assumes that downlink transmissions only occur in downlink or flexible symbols, whereas in an uplink frame slot, the UE only transmits in uplink or flexible symbols. In LTE TDD, there are a number of predefined patterns for uplink/downlink OFDM symbol allocation in a radio frame, while NR does not define any preset uplink/downlink patterns (see Fig. 3.45). A slot format indication (SFI) parameter informs the UE whether an OFDM symbol is downlink, uplink or flexible. The SFI can indicate link direction over one or many slots when configured through RRC. The SFI carries an index to Table 3.13 (pre-con- figur

ed UE-specific slot configuration table) configured through RRC. The SFI can be either dynamically configured through a DCI or statically or semi-statically configured through RRC. The UE assumes there is no conflict between dynamic SFI and downlink control 388 Chapter 3 Table 3.13: Slot formats for normal cyclic prefix ("D/U" denotes flexible downlink/uplink symbols) [12]. Symbol Number in a Slot Slot Format (Continued) New Radio Access Physical Layer Aspects (Part 1) Table 3.13: (Continued) Symbol Number in a Slot Slot Format 56-255 Reserved information (DCI) DL/UL assignments, thus when operating in NR TDD mode, one has to clearly define how available time slots are allocated to downlink and uplink transmissions. The NR defines those patterns in more flexible manner using the following parameters (see Fig. 3.46) [21]: dl-UL-TransmissionPeriodicity: Periodicity of the DL-UL pattern nrofDownlinkSlots: Number of consecutive full DL slots at the beginning of each DL-UL pattern 390 Chapter 3 Subframe = 1 ms DL control and data UL control and data DL control DL data UL control DL control UL data UL control DL control DL data UL data UL control Figure 3.45 Different NR TDD-based subframe structures [53]. nrofDownlinkSymbols InrofUplinkSymbols nrofDownlinkSlots enrofUplinkSlots dI-UL-TransmissionPeriodicity nrofDownlinkSymbols nrofUplinkSymbols dl-UL-Transmission Periodicity Specified in tdd-UL-DL-configurationCommon n (SlotIndex) Figure 3.46 TDD UL/DL common and dedicated configurations (15,39]. nrofDownlinkSymbols: Number of consecutive DL symbols in the beginning of the slot following the last full DL slot nrofUplinkSlots: Number of consecutive full UL slots at the end of each DL-UL pattern nrofUplinkSymbols: Number of consecutive UL symbols at the end of the slot preceding the first full UL slot As shown in Table 3.13, a slot format includes a specific downlink, uplink, and flexible symbol configuration. In TDD mode, a slot can be all downlink, all uplink, or a combina- tion of downlink and uplink segments. Data t

ransmission can be scheduled to span one or multiple slots when slot aggregation is supported. For each serving cell, if a UE receives the RRC parameter UL-DL-configuration-common, it must set the slot format per slot over the number of slots indicated by this parameter. If the UE is additionally provided with RRC parameter UL-DL-configuration-dedicated for the slot format per slot over a number of slots, the latter parameter overrides only flexible symbols per slot over the number of slots indicated by UL-DL-configuration-common parameter. The UE determines the duration New Radio Access Physical Layer Aspects (Part 1) 391 One slot k = NRB(H)NRB SCS = 120 kHz PRB 0 PRB 1 SCS = 60 kHz PRB 0 PRB 1 PRB 2 PRB 3 SCS = 30 kHz PRB 0 PRB 1 PRB 2 PRB 3 PRB 4 PRB 5 PRB 6 PRB 7 Resource block SCS = 15 kHz PRBOPRB1PRB2PRB3PRB4PRB5PRB6PRB7PRB8PRB9PRB10PRB11PRB12PRB13PRB14PRB15 Resource element (k,1) in resource block resource element (k,T) in resource grid 1=14x2"-1 Figure 3.47 Illustration of NR resource grids and PRBs [12]. of each slot in the number of slots, in each configured BWP, based on the subcarrier spac- ing value provided by higher layer parameter ref-scs. The UE considers symbols in a slot indicated as downlink by higher layer parameter UL-DL-configuration-common or by higher layer parameter UL-DL-configuration-dedicated as available for receiving control/traffic. The UE further considers symbols in a slot as uplink indicated by higher layer parameter UL-DL-configuration-common or by higher layer parameter UL-DL-configuration-dedicated as available for transmission of control/traffic [14,15]. 3.7 Time-Frequency Resources 3.7.1 Physical Resource Blocks The basic scheduling unit in NR is a physical resource block (PRB) comprising 12 subcar- riers in the frequency domain over one OFDM symbol. All subcarriers within a PRB have the same subcarrier spacing and CP length. When an NR system supports multiple numerol- logies, the corresponding PRBs are multiplexed in the time domain such that the boundaries of PRBs are al

igned. For this purpose, multiple PRBs of the same bandwidth form a PRB grid, as illustrated in Fig. 3.47. A PRB grid formed by subcarriers spaced apart by 392 Chapter 3 PRB N3 PRB N3 BWP(2) NCRB =N 3 +N start BWP (0) PRB 1 PRB 1 PRB 0 PRB 0 PRB N2 PRB N2 BWP(1) PRB 1 Carrier PRB 1 Carrier bandwidth PRB 0 NCRB = 1+N start BWP bandwidth PRB 0 BWP(2) start PRB N1 PRB N1 BWP(0) BWP (1) PRB 1 start PRB 1 PRB 0 PRB 0 in a BWP PRB 0 PRB 0 in a BWP NCRB = 0+N start BWP(0) N BWP(0) start CRB 0 CRB 0 Point A: PRB 0 in reference resource block Point A: PRB 0 in reference resource block Figure 3.48 Mapping between nCRB and NPRB [12,39]. where 1, 4 is a non-negative integer, is a superset of PRB grids with subcarrier spacing 15 kHz. For each numerology and carrier frequency, a resource grid of Ngrid(H)NR. subcarriers and Nsymbol(H) OFDM symbols is defined, starting at a CRB Nstart(H), whose value is signaled via RRC signaling. There is one set of resource grids per link direction (uplink or downlink). There is a single resource grid for a given antenna port p, numerology parameter u and link direction [12,28]. An antenna port is a logical entity which is distinct from a physical antenna. Each antenna port is associated with a specific set of reference signals such that the channel over which a symbol is transmitted on that antenna port can be distinguished from the channel over which another symbol is conveyed on the same antenna port. Two antenna ports are said to be quasi-co-located, if the large-scale properties of the channel over which a symbol is conveyed on one antenna port can be inferred from the channel over which a symbol is conveyed on another antenna port. The large- scale properties include delay spread, Doppler spread, Doppler shift, average gain, average delay, and other spatial parameters [12]. In other words, a UE receiver can assume that the radio chan- nels corresponding to two different antenna ports have the same large-scale properties (e.g., aver- age delay spread, Doppler spread/shift, average delay,

average gain, and spatial receive parameters), if the antenna ports are specified as being quasi-co-located. The UE can assume that two antenna ports are quasi-co-located with respect to certain channel properties either by NR specification or explicitly informed by the network via signaling. New Radio Access Physical Layer Aspects (Part 1) 393 Table 3.14: Minimum and maximum number of resource blocks/transmission bandwidths [12]. Subcarrier min(NPRB) max(NPRB) Spacing (kHz) Minimum Bandwidth (MHz) Maximum Bandwidth (MHz) 17.28 34.56 69.12 397.44 Each element in the resource grid for antenna port p and numerology parameter u is called a resource element and is uniquely identified by pair (k, 1) where k is the index in the fre- quency domain (k is defined relative to point A such that k = 0 corresponds to the subcarrier centered around point A) and l refers to the symbol position in the time domain. A resource block is defined as NRB = 12 consecutive subcarriers in the frequency domain. Point A is a common reference point for resource block grids which is derived from the higher layer parameters [12]. As shown in Fig. 3.48, the resource blocks for each subcarrier spacing configuration are num- bered from 0 to NRB(H)NRB 1 in upward direction in the frequency domain. Table 3.14 pro- vides the minimum and maximum values of NRB(H) and their corresponding transmission bandwidths. The relationship between the CRB number NCRB in the frequency domain and resource elements (k,l for each subcarrier spacing configuration is given by NCRB = k/NRB where index k is defined relative to Point A such that k = 0 corresponds to the subcarrier cen- tered around that point A. Physical resource blocks are defined within BWPs and are numbered from 0 to The relationship between the physical resource block NPRB in the ith BWP and the CRB NCRB is given by NCRB NPRB + start -BWP(i) where NBWP(i) start is the CRB where BWP starts relative to CRB 0 as shown in Fig. 3.48. Similar to LTE, virtual resource blocks are defined within a BWP and are

enumerated from 0 to VRB BWP(i) 1. The virtual resource blocks are resource blocks that are permuted across frequency dimension to take advantage of frequency diversity [12,28]. An interleaved mapping maps virtual resource blocks to physical resource blocks using an interleaver that spans the entire BWP and operates on pairs or quad- ruplets of resource blocks. A block interleaver with two rows is used, with pairs/quadruplets of resource blocks written in columns and read in rows. Whether to use pairs or quadruplets of resource blocks in the interleaving operation is configurable by higher layer signaling. The interleaved resource-block mapping in the frequency domain provides frequency diversity [13-15]. 3.7.2 Bandwidth Part 3GPP NR supports very large operating bandwidths relative to the previous generations of 3GPP standards. Since the UEs in a cell may have different bandwidth capabilities, the use Chapter 3 Carrier bandwidth Carrier bandwidth BWP 2 BWP 1 Supporting UE reduced bandwidth capability Supporting different numerologies in different BWPs carrier bandwidth carrier bandwidth Different BWP 2 BWP 1 Supporting noncontiguous spectrum Supporting UE reduced power consumption Carrier bandwidth Future Legacy technology technology BWP 2 BWP 1 Supporting forward compatibility Figure 3.49 Bandwidth part use cases [42]. of wide bandwidth may cause more power consumption and may increase RF and baseband implementation complexity. Therefore, NR introduces the concept of BWP and allows the UEs with different bandwidth capabilities to operate in the cell with (configurable) smaller instantaneous bandwidth relative to the configured cell bandwidth, making NR more energy efficient solution despite its support of wideband channels (see Fig. 3.49). Alternatively, one may consider scheduling a UE such that it only transmits or receives within a certain fre- quency band. However, the difference of the latter approach with BWP is that the UE is not required to transmit or receive outside of the configured frequency band o

f the active BWP. The granularity of bandwidth allocation in NR is one PRB. For each serving cell, up to 4 downlink/uplink BWPs can be configured separately and independently for paired spectrum; nevertheless, only one BWP can be active at a given time and the UE is not expected to receive downlink/uplink physical signals/channels outside of an active BWP. For paired spec- trum, a downlink BWP and an uplink BWP are jointly configured as a pair and up to four pairs can be configured. One can configure up to four BWPs on a supplemental uplink (SUL) carrier. Different use cases of the BWP are illustrated in Fig. 3.49 [28]. In other words, a BWP is a subset of contiguous CRBs for a given numerology (note that different numerologies may be used in different BWPs) on a given RF carrier. The starting position NBWP(i) and the number of resource blocks N PRB BBP(i) in a BWP satisfy Nerid Norid(u) BWP(i) start Norid size (u), respectively. If a UE is configured with a SUL, then it can additionally be configured with up to four BWPs on the SUL with a single SUL BWP being active at a given time. As we mentioned earlier, the transmit/receive bandwidth of a UE does not need to be as large as the bandwidth of the cell and it can be adaptively adjusted according to UE New Radio Access Physical Layer Aspects (Part 1) 395 PRB N3 BWP 3 (active) PRB 1 PRB 0 BWP 2 BWP 2 PRB N2 (active) (active) PRB 1 BWP 1 BWP 1 (active) (active) PRB 0 PRB N1 PRB 1 PRB 0 Figure 3.50 Illustration of the active bandwidth part adaptation concept [12,19,28]. operational conditions. With Bandwidth Adaptation (BA), a UE's bandwidth can be resized (e.g., reduced during a period of low activity for power saving); its location can be moved in the frequency domain (e.g., to increase scheduling flexibility); and the subcarrier spacing can be changed (e.g., to allow different services). A subset of the total cell bandwidth is called a BWP and the BA is achieved by configuring the UE with different BWP(s) and notifying the UE of the instantaneous active one. Fig.

3.50 shows a scenario where three different BWPs are configured: BWP1 = 50 MHz and subcarrier spacing of 15 kHz; BWP2 = 10 MHz and subcarrier spacing of 15 kHz; and BWP3 = 25 MHz and subcarrier spacing of 60 kHz [19]. Note that at each time, only one BWP is active. An initial active downlink BWP is defined by its location, number of contiguous PRBs, subcar- rier spacing, and CP, for the control resource set corresponding to TypeO-PDCCH common search space. For operation on the primary cell, a UE is provided by higher layer parameter initial-UL-BWP an initial uplink BWP to perform random-access procedure. If the UE is con- figured with a secondary carrier on the primary cell, it can also be configured with an initial BWP for random-access procedure on the secondary carrier [14,15]. A UE can be provided with a timer value by RRC parameter BWP-InactivityTimer for the primary cell. The UE sub- sequently starts the timer each time that it detects a DCI format 1_1 indicating an active down- link BWP, other than the default downlink BWP, for paired spectrum operation or each time the UE detects DCI format 1_1 or DCI format 0_1 indicating an active downlink or uplink BWP, other than the default downlink/uplink BWP, for unpaired spectrum operation. The UE increments the timer every 1 ms for sub-6 GHz carrier frequencies or every 0.5 ms for carrier frequencies above 6 GHz, if it does not detect a DCI format 1_1 for paired spectrum operation 396 Chapter 3 or if it does not detect a DCI format 1_1 or DCI format 0_1 for unpaired spectrum operation. The timer expires when the value is equal to the BWP-InactivityTimer. Upon expiration of the timer, the UE switches from the active BWP to the default BWP [14,15]. In conjunction with an UL/DL carrier pair for an FDD band, or a bidirectional carrier for a TDD band, a UE may be configured with an additional SUL. The SUL differs from the aggregated uplink in the sense that the UE may be scheduled to transmit either on the SUL or on the uplink of the carrier being supplemented, but n

ot on both at the same time. In the case of SUL, the UE is configured with two uplink carriers in conjunction of one downlink carrier of the same cell, and uplink transmissions on those carriers are controlled by the net- work to avoid colliding uplink control and traffic channels in time domain. The colliding transmissions on uplink traffic channel are avoided through scheduling while overlapping transmissions on uplink control channel are avoided via limiting configuration of uplink control channel on only one of the two uplink carriers. In addition, initial access is sup- ported on each of the uplink carriers. To improve uplink coverage for high-frequency sce- narios, a low-frequency SUL carrier can be configured. In NR, the UE can take advantage of the bandwidth adaptation feature to save power while satisfying the requirements of vari- ous services/applications. The network can configure up to four BWPs for each UE, and dynamically send change indications to the UEs as required. The BWP switching for a serving cell is used to activate an inactive BWP or to deacti- vate an active BWP at any given time, as shown in Fig. 3.51. The BWP switching is controlled by the PDCCH indicating a downlink assignment or an uplink grant, by the bandwidthPartInactivityTimer, or by the MAC entity itself upon initiation of random- access procedure. Upon addition of a Special Cell (SpCell)22 or activation of an SCell, one BWP is initially active without receiving PDCCH indicating a downlink assignment or an uplink grant. The active BWP for a serving cell is indicated by either RRC or PDCCH. For paired spectrum, a downlink BWP is paired with an uplink BWP, and BWP switching applies to both uplink and downlink [20]. The BWP switching options are illustrated and compared in Fig. 3.52. 3.7.3 Resource Allocation One of the main design objectives for signaling the resource allocation information, in the form of a set of resource blocks in each slot, to the active UEs in the cell is to find a bal- anced tradeoff between flexibility and

signaling overhead. Indications of localized/distrib- uted resource allocations to different UEs are transmitted via PDCCH. The resource allocation field in PDCCH is interpreted by the UE depending on the PDCCH DCI format. In the context of dual connectivity, the Special Cell refers to the primary cell of the MCG or the primary SCell of the SCG depending on whether the MAC entity is associated with the MCG or the SCG. Otherwise, the Special Cell refers to the PCell. A Special Cell supports PUCCH transmission and contention-based ran- dom-access procedure and is always activated [20]. New Radio Access Physical Layer Aspects (Part 1) 397 BWP switching BWP switching BWP configuration by DCI by DCI by RRC configuration P cell for P cell BWP 1 BWP 1 Initial BWP BWP 2 configuration S cell for S cell BWP 2 BWP 1 BWP 1 BWP switching BWP switching S cell activation by DCI by DCI Initial access Connected state Multiple activated cells Single activated cell Figure 3.51 Illustration of BWP adaptation, activation, and switching [42]. The resource allocation in NR is defined both in time domain and frequency domain. Unlike LTE where the resource allocation in time domain was determined based on a fixed/ predefined rule, in NR the resource allocation in time domain is more flexible, whereas the resource allocation in frequency domain is relatively similar to that of LTE [30]. Resource allocation type specifies the way in which the scheduler allocates physical resource blocks in frequency domain to each user for transmission in the downlink or uplink. 3.7.3.1 Resource Allocation in Time Domain In the downlink, when the UE is scheduled to receive PDSCH via a DCI, that is, the time domain resource assignment field of DCI provides a row index to an allocation table, where the indexed row defines the slot offset K0, the start and length indicator SLIV and the Chapter 3 BWP switch RF tuning and PDCCH MAC CE parsing MAC CE AGC settling monitoring BWP 1 BWP 2 Transition time 1 Transition time 2 BWP switch L1 RF tuning and signaling AGC

settling L1 signaling PDCCH parsing monitoring BWP 1 BWP 2 Transition time Figure 3.52 Comparison of BWP switching options [39]. PDSCH mapping type. As shown in Fig. 3.53, the slot allocated for PDSCH transmis- sion is determined by parameter K of the indexed row n + K where n is the slot with the scheduling DCI, K0 is based on the numerology of PDSCH. The starting symbol S relative to (the start of the slot), and the number of consecutive symbols L counting from the symbol S allocated for the PDSCH are determined from the start and length indicator SLIV [14,15]. The slot allocated for the PDSCH is defined as | where UPDSCH and UPDCCH are the subcarrier spacing config- urations for PDSCH and PDCCH, respectively. If (L <7, SLIV = 14(L-1)+S; otherwise SLIV = 14(14 - L + 1) + (14 - where O<L < 14 - S and the PDSCH mapping type is set to Type A or Type B [15]. The per- missible S and L combinations corresponding to PDSCH allocations are shown Table 3.15. The PDSCH mapping type is related to the relative location of the demodulation refer- ence signal (DM-RS) and the slot boundary as well as the size of the data. The mapping type A is used when the first DM-RS is located in the second or third OFDM symbol of New Radio Access Physical Layer Aspects (Part 1) 399 HPDSCH 2 HPDCCH (Slot n) PDSCH S: Start symbol L: Number of consecutive symbols 2 "PDSCH 2HPDCCH ACK/NACK on PUCCH PDSCH (Slot n) 2 UPDSCH 2 HPDCCH PUSCH (Slot n) S: Start symbol L: Number of consecutive symbols Figure 3.53 Illustration of downlink/uplink time-domain resource allocation [12,15,39]. Table 3.15: Permissible S and L values [15]. Normal Cyclic Prefix Extended Cyclic Prefix Mapping Type S + L PDSCH Type A {0,1,2,3} {3,...,14} {3,...,14} {0,1,2,3} {3,...,12} {3,...,12} Type B {0,...,12} {2,4,7} {2,...,14} {0,...,10} {2,4,6} {2,...,12} PUSCH Type A {4,...,14} {4,...,14} {4,...,12} {4,...,12} Type B {0,...,13} {1,...,14} {1,...,14} {1,...,12} {1,...,12} the slot following a CORESET (i.e., a control region) at the beginning of a slot. The DM-RS is mappe

d relative to the start of the slot boundary regardless of the start of data transmission in the slot. This mapping type is primarily intended for the cases where the data occupies most of the slot. The mapping type B is used when the first DM-RS is located in the first symbol of the data allocation, that is, the DM-RS location is not given relative to the slot boundary, rather relative to where the data is located. This mapping is intended for transmissions over a small fraction of the slot to support very low latency and other transmissions that cannot wait until a slot boundary starts regardless of the transmission duration. 400 Chapter 3 When the UE is configured with pdsch-AggregationFactor > 1, the same symbol allocation is applied across the pdsch-AggregationFactor consecutive slots that have not been defined as uplink by the SFI. Fig. 3.53 illustrates time-domain and frequency-domain resource allo- cation procedure for PDSCH and demonstrates how the above parameters are used to locate the user allocation. In the uplink, when a UE is scheduled to transmit a transport block on PUSCH by a DCI with or without CSI report(s), the time domain resource assignment field of the DCI pro- vides a row index to a table defined in [15], where the indexed row defines the slot offset K2, the start and length indicator SLIV, or directly by the start symbol S and the allocation length L, and the mapping type to be used in PUSCH transmission as shown in Fig. 3.53. The slot where the UE transmits the PUSCH is determined by parameter K2 as K2 where n is the slot with the scheduling DCI, K2 is based on the numerology of PUSCH, UPUSCH and UPDCCH are the subcarrier spacing configurations for PUSCH and PDCCH, respectively. The starting symbol S relative to the start of the slot, and the number of consecutive symbols L counting from the symbol S allocated for the PUSCH are determined from the start and length indicator SLIV of the indexed row as fol- lows. If (L - 1) < then SLIV = 14(L-1) + S; otherwise SLIV = 14(14 - + 1) + (14 -s

-1), where 0 L<14 - S and the PUSCH mapping type is set to Type A or Type B [15]. The permissible S and L combinations corresponding to PDSCH allocations are shown in Table 3.15. When the UE is configured with pusch - AggregationFactor > 1, the same symbol allocation is applied across the pusch - AggregationFacto consecutive slots and the PUSCH is limited to a single transmission layer. The UE repeats the transport block across pusch - Aggregation Factor consecutive slots applying the same symbol allocation in each slot [15]. 3.7.3.2 Resource Allocation in Frequency Domain A UE determines the frequency-domain resources on which it transmits or receives data by examining the resource-block allocation and BWP indicator fields in a DCI. The resource allocation fields determine the resources blocks in the active BWP on which data is transmitted. The gNB can signal the allocated resources to a UE using resource alloca- tion type 0 or type 1, which are conceptually similar to LTE resource allocation type 0 and type 2 with the difference that in LTE, the resource allocation signales the allocations across the carrier bandwidth, whereas in NR, the indication is relevant only for the active BWP. Resource allocation type 0 is a bitmap-based allocation scheme, indicating the set of resource blocks that the UE is supposed to receive in the downlink transmission where the size of the bitmap is equal to the number of resource blocks in the BWP. This would allow an arbitrary combination of resource blocks to be scheduled for the UE at the expense of a large control/signaling overhead and some downlink coverage issues for larger BWP sizes due to limited capacity of a single OFDM symbol. Consequently, there New Radio Access Physical Layer Aspects (Part 1) 401 Table 3.16: Nominal resource block group size P [15]. Bandwidth Part Size NPRB ,BWP(i) Configuration 1P Configuration 2P 37-72 73-144 145-275 Bitmap 1100 1110000000000001100 RBG18 RBG19 RBG20 RBG21 RBG22 RBG23 RBG24 Resource indication value Starting position Length PRB1 18

PRB19 PRB20 PRB21 PRB22 PRB23 PRB24 Length Starting position Figure 3.54 Illustration of frequency-domain type 0 and type 1 resource allocations [15,39]. is a need to reduce the bitmap size while maintaining the allocation flexibility. This can be achieved by addressing a group of contiguous resource blocks as opposed to individual PRBs. The size of the resource block group (RBG) is determined by the size of the BWP (see Table 3.16). Resource allocation type 1 indicates the allocated resources to the UE by means of a starting position and the length of the resource block allocation, thus only supporting contiguous allocations in frequency domain. In order to further reduce the sig- naling overhead, resource allocation type 1 combines the starting position and the length of resource allocation values into a single value referred to as resource indication value [15]. The two resource allocation types are illustrated in Fig. 3.54 (example). The resource allocation scheme is configured using a bit in the DCI. Both resource allocation types refer to virtual resource blocks. For resource allocation type 0, a non-interleaved mapping from virtual to physical resource blocks is used, thus the virtual resource blocks are directly mapped to the corresponding physical resource blocks. For resource allocation type 1, both interleaved and non-interleaved mapping is supported. The VRB-to-PRB mapping bit, if pres- ent, indicates whether the allocation is based on interleaved or non-interleaved mapping. In downlink resource allocation type 0, the resource block assignment information includes a bitmap representing the RBGs that are allocated to the scheduled UE where RBG is a set of consecutive physical resource blocks defined by a higher layer parameter and the size of the carrier BWP. The total number of RBGs NRBG defined for a downlink carrier BWP of 402 Chapter size BWis o/whe = the size of the first RBG is the size of last RBG is BGwn= if modP 0 and P otherwise. The size of all other RBGs is P [15]. The nominal values of P a

used for PUSCH transmission when transform precoding is disabled. The uplink resource allocation type 1 is used for PUSCH transmission regardless of whether transform precoding is enabled or disabled. The UE can assume that when the scheduling PDCCH is received with DCI format 0_0, then uplink resource allocation type 1 is used. If a BWP indicator field is not configured in the scheduling DCI, the RB indexing for uplink type 0 and type 1 resource allocation is determined within the UE's active BWP. If a BWP indicator field is configured in the scheduling DCI, then the RB indexing for uplink type 0 and type 1 resource allocation is determined within the UE's BWP indicated by BWP indicator field value in the DCI. Upon detection of PDCCH intended for the UE, the UE must first deter- mine the uplink BWP and then the resource allocation within the BWP [15]. The uplink resource allocation type 0 is similar to the downlink counterpart, where the resource block assignment information includes a bitmap indicating the RBGs that are New Radio Access Physical Layer Aspects (Part 1) 403 allocated to the scheduled UE. The size of RBGs is given in Table 3.16. The uplink resource allocation type 1 is also similar to the downlink part described earlier, where the resource block assignment information informs the UE of a set of contiguously allocated noninterleaved virtual resource blocks within the active carrier BWP of size PRB BWP except for the case when DCI format 00 is decoded in any common search space in which case the size of the initial BWP NBWP(0) is used [15]. 3.7.3.3 Physical Resource Block Bundling A UE cannot make any assumption about correlation of reference signals between different PDSCH scheduling occasions in the time domain. This is necessary to allow more flexibility in precoder-based beamforming and spatial signal processing. In the frequency domain, however, the UE can assume that there is correlation between reference signals within a precoding resource block group (PRG). Over the frequency span of one PRG

, the UE may assume that the downlink precoder remains the same, and exploit this in the channel estima- tion process. The correlation assumption does hold between PRGs. It can be concluded that there is a tradeoff between the precoding flexibility and the channel estimation performance, that is, a large PRG size can improve the channel estimation accuracy at the cost of precod- ing flexibility. Therefore, the gNB may indicate the PRG size to the device where the possi- ble PRG sizes are two, four or the total scheduled bandwidth in terms of PRBs. It is also possible to dynamically indicate the PRG size through the DCI. In addition, the UE can be configured to assume that the PRG size is equal to the scheduled bandwidth when the scheduled bandwidth is larger than half of the active BWP. A UE may assume that the precoding granularity is PBWP(i) consecutive resource blocks in the frequency domain, where P'BWP(i) values are taken from the limited set of {2, 4, wide- band}. If P'BWP(i) is set to "wideband", the UE can expect to be scheduled with contiguous PRBs with the PRG and the use of the same precoding across the allocated resources by the gNB. If BWP(i) is set to 2 or 4, the ith BWP is partitioned into P'BWP(i) consecutive PRBs to form the PRGs. In practice, the number of consecutive PRBs in each PRG can be one or more. The first PRG size is and the last PRG size given 0; otherwise, the last PRG size is P'BWP(i) [15]. The UE may assume the same precoding is applied for any downlink contiguous allocation of PRBs within a PRG. If a UE is scheduled a PDSCH with DCI format 1_0, it can assume that PBBP(i) is equal to 2 PRBs [15]. 3.7.4 Resource Allocation for Grant-Free/Semi-persistent Scheduling In grant-free uplink transmissions, the UEs can transmit within a set of predetermined resource blocks without any explicit scheduling grants from the base station, resulting in lower control/signaling overhead and latency. However, uplink transmissions require the 404 Chapter 3 UEs to transmit within a given set of resourc

es that are pre-allocated by the base station with a certain periodicity. To avoid resource utilization inefficiency, multiple UEs might be allowed to share the same resources and to operate in a contention-based manner. Hence, collisions are inevitable, affecting connection reliability and latency, which is worsen as the UE traffic increases and retransmissions become necessary. In other words, grant-free trans- mission may not be scalable as the UE density increases in a network. Fig. 3.55 shows a comparison between the contention-based (grant-free) and grant-based performance, where we observe that the efficiency of contention-based transmission degrades with increasing packet size and traffic load, while efficiency of grant-based trans- mission improves by increasing packet size and is stable over different load factors. The markings in the figure indicate that for packet sizes of 20, 30, and 40 bytes, there is a load- ing factor threshold below which contention-based transmission is more efficient than grant- based transmission. When the packet size is sufficiently small, for example, 10 bytes, contention-based transmission is always optimal within certain load factor range. Based on the analysis in [43], keep-alive messages of mobile Internet traffic are better suited to use Contention-based transmission: packet size 10 byte Grant-based transmission: packet size 10 byte Contention-based transmission: packet size 20 byte Grant-based transmission: packet size 20 byte Contention-based transmission: packet size 30 byte Grant-based transmission: packet size 30 byte Contention-based transmission: packet size 40 byte Grant-based transmission: packet size 40 byte Load factor [activation probability of users (%)] Figure 3.55 Comparison of grant-based and grant-free uplink transmissions [43]. New Radio Access Physical Layer Aspects (Part 1) 405 contention-based transmission, while the rest of the mobile Internet packet types (e.g., video or voice) need to be transmitted by grant-based transmission, if not taking into

account the latency constraints. This is because the former is of relatively small packet size and small loading, while packet size of the latter is large to be transmitted by contention (Fig. 3.55). The semi-persistent scheduling (SPS)-based resource allocation refers to a transmission mode in which the serving base station allocates at least a part of resources and transport formats to the UE semi-statically over a certain time interval consisting of a number of TTIs. In the downlink, the semi-persistent scheduling is configured by RRC per serving cell and per BWP. Multiple configurations can be active simultaneously on different serving cells. Activation and deactivation of the downlink semi-persistent scheduling transmissions are independent among the serving cells [20]. In downlink semi-persistent scheduling, a downlink assignment is provided by PDCCH, and stored or cleared based on layer-1 signaling indicating semi-persistent scheduling activation or deactivation. A configured scheduling RNTI (CS-RNTI), which is similar to SPS C- RNTI in LTE, is configured by RRC for activation, deactivation, and retransmission. Furthermore, the number of configured HARQ processes and the periodicity of semi- persistent scheduling are signaled by RRC. When SPS resources are released by upper layers, all the corresponding configurations are subsequently released. In other words, the gNB can allocate downlink resources for the initial HARQ transmissions to UEs. The RRC defines the periodicity of the configured downlink assignments while PDCCH addressed with CS-RNTI can either signal and activate the configured downlink assignment, or deacti- vate it, that is, a PDCCH scrambled with CS-RNTI indicates that the downlink assignment can be implicitly reused according to the periodicity defined by RRC, until deactivated. After a downlink assignment is configured for semi-persistent scheduling, the MAC entity considers that the Nth downlink assignment will occur in a slot whose number meets the following criteria: s10od(1024N ) whe

re Nslot frame denotes the number of slots per frame Nslot is the slot number in the frame, TSPS denotes the SPS period, SFNstart-time and Tslot start-time represent the SFN and slot of the first transmission of PDSCH where the configured downlink assignment was initialized, respectively [20]. In the uplink direction, there are two types of transmission without dynamic grant known as configured grant Type 1, where an uplink grant is provided by RRC, and stored as config- ured uplink grant; and configured grant Type 2, where an uplink grant is provided by PDCCH and stored or cleared as configured uplink grant based on physical layer signaling 406 Chapter 3 indicating configured grant activation or deactivation. Both Type 1 and Type 2 grants are configured by RRC per serving cell and per BWP. Multiple configurations can be active simultaneously on different serving cells. For Type 2 grant, activation and deactivation are independent among the serving cells. When the configured grant Type 1 is used, the RRC configures the following parameters: a CS-RNTI for retransmission; periodicity of the con- figured grant Type 1; the offset of a resource with respect to SFN = 0 in time domain; time-domain parameters which include the start symbol and the length of the assignment; as well as the number of HARQ processes [20]. Alternatively, when the configured grant Type 2 is going to be used, the RRC configures the following parameters: a CS-RNTI for activa- tion, deactivation, and retransmission; the periodicity of the configured grant Type 2; and the number of HARQ processes. Once a configured grant Type 1 semi-persistent allocation is set up in a serving cell by upper layers, the corresponding MAC entity stores the uplink grant provided by upper layers and initializes the configured uplink grant to start in the sym- bol according to the provided parameters. After an uplink grant is set up for a configured grant Type 1 uplink transmission, the MAC entity considers the Nth uplink transmit opportu- nity to occur in the symbol n

umber which satisfies the following equation [20]: where symbol slot , Ms, and toffset denote the number of symbols per slot, symbol number in the slot, and the time domain offset, respectively. Similarly, subsequent to an uplink grant set up for a configured grant Type 2, the MAC entity considers the time-domain location of the Nth uplink grant-free transmission at the symbol for which the following criterion is satis- fied [20]: sFa slot start-time AN 20 where SFNstart-time, Tslot symbol start-time' start-time represent the SFN, slot, and symbol where the first transmission of PUSCH with the configured uplink grant was initialized, respectively. When a configured grant is released by upper layers, all corresponding configurations are cleared. Retransmissions except for repetition of the configured grants use uplink grants with CS-RNTI. In summary, in the downlink, the gNB can allocate downlink resources for the initial HARQ transmissions to UEs with semi-persistent scheduling. The RRC signaling defines the periodicity of the configured downlink assignments while PDCCH addressed with New Radio Access Physical Layer Aspects (Part 1) 407 CS-RNTI can either signal and activate the configured downlink assignment, or deactivate it, that is, a PDCCH scrambled with CS-RNTI indicates that the downlink assignment can be implicitly reused according to the periodicity defined by RRC, until deactivated. In the uplink, the gNB can allocate uplink resources for the initial HARQ transmissions to UEs with configured uplink grants. Two types of configured uplink grants are defined: Type 1, where RRC directly provides the configured uplink grant (including the periodicity) and Type 2, where RRC defines the periodicity of the configured uplink grant while PDCCH addressed to CS-RNTI can either signal and activate the configured uplink grant, or deacti- vate it, that is, a PDCCH addressed with CS-RNTI indicates that the uplink grant can be implicitly reused according to the periodicity defined by RRC, until deactivated [19]. When ca

rrier aggregation is configured, one configured uplink grant can be signaled per serving cell. Thus each serving cell can have one configured uplink grant active at any time. Similarly, when bandwidth adaptation is configured, one configured uplink grant can be signaled per BWP. A configured uplink grant for one serving cell can either be of Type 1 or Type 2. For Type 2, activation and deactivation of configured uplink grants are inde- pendent among the serving cells. When SUL is configured, a configured uplink grant can only be signaled for one of the two uplink carriers of the cell. References ITU-R Specifications Report ITU-R P.2406-0, Studies for Short-Path Propagation Data and Models for Terrestrial Radiocommunication Systems in the Frequency Range 6 GHz to 100 GHz, September 2017. Recommendation ITU-R P.1238-9, Propagation Data and Prediction Methods for the Planning of Indoor Radiocommunication Systems and Radio Local Area Networks in the Frequency Range 300 MHz to 100 GHz, June 2017. [3] Report ITU-R M.2376-0, Technical Feasibility of IMT in Bands Above 6 GHz, July 2015. [4] Report ITU-R M.2412-0, Guidelines for Evaluation of Radio Interface Technologies for IMT-2020, October 2017. [5] Report ITU-R M.2411-0, Requirements, Evaluation Criteria and Submission Templates for the Development of IMT-2020, November 2017. 3GPP Specifications [6] 3GPP TR 21.915, Summary of Rel-15 Work Items (Release 15), March 2019. [7] 3GPP TR 36.873, Study on 3D Channel Model for LTE (Release 12), December 2017. [8] 3GPP TR 37.910, Study on Self Evaluation towards IMT-2020 Submission (Release 15), December 2018. [9] 3GPP TS 38.104, NR, Base Station (BS) Radio Transmission and Reception (Release 15), December 2018. [10] 3GPP TS 38.201, NR, Physical Layer - General Description (Release 15), December 2018. [11] 3GPP TS 38.202, NR, Services Provided by the Physical Layer (Release 15), December 2018. ITU-R specifications can be accessed at the following URL: http://www.itu.int/en/ITU-R/study-groups/rsg5/ rwp5d/imt-2020/. 3GPP specific

ations can be accessed at the following URL: http://www.3gpp.org/ftp/Specs/archive/. 408 Chapter 3 3GPP TS 38.211, NR, Physical Channels and Modulation (Release 15), December 2018. [13] 3GPP TS 38.212, NR, Multiplexing and Channel Coding (Release 15), December 2018. 3GPP TS 38.213, NR, Physical Layer Procedures for Control (Release 15), December 2018. [15] 3GPP TS 38.214, NR, Physical Layer Procedures for Data (Release 15), December 2018. 3GPP TS 38.215, NR, Physical Layer Measurements (Release 15), March 2018. [17] 3GPP TS 36.104, E-UTRA, Base Station (BS) Radio Transmission and Reception (Release 15), June 2018. 3GPP TS 38.133, NR, Requirements for Support of Radio Resource Management (Release 15), December 2018. [19] 3GPP TS 38.300, NR, NR and NG-RAN Overall Description, Stage 2 (Release 15), December 2018. [20] 3GPP TS 38.321, NR, Medium Access Control (MAC) Protocol Specification, (Release 15), December 2018. [21] 3GPP TS 38.331, NR, Radio Resource Control (RRC), Protocol Specification (Release 15), December 2018. 3GPP TR 38.802, Study on New Radio Access Technology Physical Layer Aspects (Release 14), March 2017. [23] 3GPP TR 38.812, Study on Non-Orthogonal Multiple Access (NOMA) for NR; (Release 15) October 2018. 3GPP TR 38.900, Study on Channel Model for Frequency Spectrum Above 6 GHz (Release 15), June 2018. [25] 3GPP TR 38.901, Study on Channel Model for Frequencies From 0.5 to 100 GHz (Release 15), June 2018. Articles, Books, White Papers, and Application Notes [26] P. Marsch, Ö. Bulakci, 5G System Design: Architectural and Functional Considerations and Long-Term Research, Wiley, 2018. [27] A. Zaidi, F. Athle, 5G Physical Layer: Principles, Models and Technology Components, Academic Press, 2018. [28] E. Dahlman, S. Parkvall, 5G NR: The Next Generation Wireless Access Technology, Academic Press, 2018. [29] F.-L. Luo, C.J. Zhang, Signal Processing for 5G: Algorithms and Implementations, Wiley-IEEE Press, 2016. S. Ahmadi, LTE-Advanced: A Practical Systems Approach to Understanding 3GPP LTE Releases 10 and

11 Radio Access Technologies, 2013, Academic Press. [31] T.L. Marzetta, et al., Fundamentals of Massive MIMO, Cambridge University Press, 2016. [32] T.S. Rappaport, R.W. Heath Jr., Millimeter Wave Wireless Communications, Prentice Hall, 2014. [33] B. Sklar, Digital Communications: Fundamentals and Applications, Prentice Hall, 1987. [34] F. Ademaj, et al., 3GPP 3D MIMO channel model: a holistic implementation guideline for open source simulation tools, EURASIP J. Wireless Commun. Netw. (2016) 55. [35] B. Mondal, et al., 3D channel model in 3GPP, IEEE Commun. Mag. 53 (Issue 3) (2015). [36] T.S. Rappaport et al., Overview of millimeter wave communications for fifth-generation (5G) wireless networks-with a focus on propagation models, IEEE Trans. Antennas Propag. (Special Issue on 5G) (2017). [37] M. Zhang, et al., ns-3 Implementation of the 3GPP MIMO Channel Model for Frequency Spectrum above 6 GHz, ePrint of Cornell University Library, February 2017. [38] X. Wang et al., Multicarrier Waveforms for 5G, Institut für Nachrichtenübertragung, WebDemos. <http:// www.inue.uni-stuttgart.de/lehre/demo.html>. [39] 5G New Radio, ShareTechNote. Available from: <http://www.sharetechnote.com [40] D. Bharadia et al., Full duplex radios, in: Proceedings of ACM SIGCOMM, 2013. New Radio Access Physical Layer Aspects (Part 1) 409 [41] A. Sabharwal, et al., In-band full-duplex wireless: challenges and opportunities, IEEE J. Sel. Areas Commun. 32 (9) (2014). [42] J. Campos, Understanding the 5G NR Physical Layer, Keysight Technologies, 2017. [43] I. Chih-Lin, Seven fundamental rethinking for next-generation wireless communications, SIP 6 (2017). [44] J. Fan, et al., Faster-than-Nyquist signaling: an overview, IEEE Access 5 (2017). [45] A. Roessler, 5G Waveform Candidates, Application Note, Rohde & Schwarz, June 2016. [46] B. Farhang-Boroujeny, OFDM VS. filter bank multicarrier, IEEE Signal Process Mag. (2011). White Paper, 5G Waveform & Multiple Access Techniques, Qualcomm Technologies, 2015. GTI 5G Device RF Component Research Report

, 2018. Available from: <http://www.gtigroup.org> [49] 3GPP TSG RAN WG1, R1-165425, f-OFDM Scheme and Filter Design, 2016. [50] Z. Ding, et al., A survey on non-orthogonal multiple access for 5G networks: research challenges and future trends, IEEE J. Sel. Areas Commun. 35 (10) (2017). Z. Wu et al., Comprehensive study and comparison on 5G NOMA schemes, IEEE Access, (2018). [52] Y. Yuan, et al., Non-orthogonal transmission technology in LTE evolution, IEEE Commun. Mag. (2016). [53] MediaTek, A New Era for Enhanced Mobile Broadband, White Paper, March 2018. CHAPTER 4 New Radio Access Physical Layer Aspects (Part 2) In this chapter, we discuss the theoretical and practical aspects of the downlink and uplink physical layer signal processing in NR and highlight the functional and procedural similari- ties and differences with the LTE physical layer processing. The chapter will describe gen- eration, configuration, and beamformed transmission of various physical signals and physical channels as well as the HARQ protocols and power control schemes. Unlike LTE where the downlink and uplink waveforms and multiple access schemes are different, the NR uses OFDM waveform as the basis for both downlink and uplink transmission (except in certain cases where DFT precoding is used in the uplink), resulting in many similarities in functional blocks and their respective operation in the downlink and uplink. The physical channel processing in NR utilizes polar codes for robust coding of the control channels and low-density parity check (LDPC) codes for the data channels, deviating from channel cod- ing schemes that are used in LTE. Massive MIMO is one of the main enabling technologies in 5G wireless communications. A large number of antenna elements at the base station bring extra degrees of freedom for increasing the throughput and considerable beamforming gains for improving the coverage. In practice, a large number of antenna elements can be assembled into multiple antenna panels for the purpose of cost reduction and power savi

ng. Multi-panel MIMO is expected to become promising for mmWave massive MIMO systems. The NR enables multi-panel antenna array operation through introduction of new reference signals, measurement, and reporting procedures. In this chapter, two of the unique NR MIMO features, that is, modular and high-resolution channel state information acquisition and beam management that distin- guish NR from LTE, are described. The modular framework is composed of three compo- nents, namely resource setting, CSI reporting setting, and measurement setting, which associates a resource setting with a reporting setting. These settings serve as building blocks that allow the network to customize the CSI measurement and reporting for a UE. To improve user throughput, a high-resolution dual-stage precoding referred to as Type II CSI is supported to allow more accurate estimation of the channel, thereby improving the efficiency of the NR MU-MIMO schemes. The associated codebook features a frequency non-selective basis subset selection coupled with a frequency-selective linear combination of amplitude and phase of the precoding vectors within the basis subset. As a result, the 5G NR. DOI: https://doi.org/10.1016/B978-0-08-102267-2.00021-X © 2019 Elsevier Inc. All rights reserved. 412 Chapter 4 precoding matrix indicator (PMI) for Type II CSI consists of several components, each with different frequency resolution. Since the NR is primarily geared toward MU-MIMO opera- tion, Type II CSI is complemented by Type I CSI designed for scenarios that do not require high spatial resolution, for example, SU-MIMO transmission. In order to establish and sustain a link for data transmission and reception, beam manage- ment enables the network to perform beam switching using physical-layer measurement and link quality reporting. Beam management is especially relevant for above-6 GHz frequency planning where both gNB and UE employ narrow beams for data transmission and recep- tion. The beam management can further be used for sub-6 GHz multi-TRP scena

rios. When used in conjunction with CSI acquisition, the beam management allows the network to establish a seamless and low-latency link with the UE for data transmission. This is specifi- cally important for over-6 GHz where a large number of narrow analog beams are used for data transmission, which in some scenarios requires frequent beam switching. Once the link is established via beam management, CSI acquisition can assist the network in link adaptation. 4.1 Downlink Physical Layer Functions and Procedures 4.1.1 Overall Description of Downlink Physical Layer The NR downlink physical layer consists of higher layer configurable functional blocks and protocols that are configured according to the downlink physical channel characteristics, use case, deployment scenario, etc. As shown in Fig. 4.1, the downlink physical layer proces- sing generally includes receiving higher layer data [e.g., MAC PDUs in the case of down- link shared channel or master information block (MIB) in the case of physical broadcast channel (PBCH)]; cyclic redundancy check (CRC) calculation and attachment; channel encoding and rate matching; modulation; mapping to physical resources and antennas; multi-antenna processing; and support of layer-1 control and HARQ-related signaling. It was mentioned in Chapter 3 that OFDM was chosen as the default waveform in NR for both downlink and uplink directions due to its robustness to multipath delay spread and frequency-selectivity of wireless channels as well as scheduling flexibility for transmission of different channels and signals. Unlike LTE, the DFT-precoded OFDM is an optional transmission scheme in NR uplink that is used in link-budget-limited use cases. While the use of DFT-precoded OFDM in the uplink has certain advantages in reducing the PAPR (and alternatively the cubic metric) and achieving higher power-amplifier efficiency, it has several drawbacks including limitation in the use of spatial multiplexing, asymmetric downlink/uplink transmissions which would limit the sidelink operation,

and scheduling complexity. New Radio Access Physical Layer Aspects (Part 2) gNB MAC UE MAC DL-SCH/PCH Transport Blocks Information Element CRC Calculation + Attachment Payload Generation + Multiplexing Scrambling LDPC Graph Selection CRC Calculation + Attachment CRC Calculation + Attachment Code Block Segmentation HARQ ACK/NACK Channel Coding (LDPC Encoding) Channel Coding (Polar Encoding) Channel Coding (Polar Encoding) Channel Decoding (LDPC Rate Matching + HARQ Rate Matching Rate Matching Decoding) + HARQ Buffer HARQ ACK/NACK Interleaving + Scrambling + CCE-to-REG Mapping Modulation Scrambling Scrambling De-scrambling + De-interleaving Layer Mapping QPSK Modulation QPSK Modulation Demodulation Antenna Port Mapping SS Block Generation Layer De-mapping CSI Estimation Reference Signal Generation Precoding DM-RS/CSI-RS/PT-RS/TRS MIMO Detection/Equalization Reference Signal Virtual/Physical Resource Configuration Channel Element Mapping Resource De-mapping Estimation OFDM Modulation + CP CP Removal OFDM Insertion PBCH Detection Demodulation Downlink TX Beamforming RX Beamforming Synchronization IQ Modulation/RF Up-conversion/ LNA and RX Filter/IQ PA and TX Filter Demodulator/Down-conversion Figure 4.1 Overall downlink physical layer processing [5]. 4.1.2 Reference Signals To facilitate estimation of the multipath communication channel and reliable, coherent detection of traffic/control channels, an OFDM system makes use of reference signals (or pilot subcarriers). The pilot subcarriers provide estimate of the channel frequency response at the pilot locations over the time-frequency resource grid. It is possible to estimate the channel at other time-frequency locations using interpolation techniques. Using predefined pilot subcarriers to estimate the channel matrix, it is possible to equalize the effects of the channel and to reduce noise and interference effects on the received resource blocks. The NR specifications include several types of reference signals that are configured and trans- mitted in different manner

s, which are used for different purposes by a receiving device. While perfect knowledge of the radio channel can be used to find an upper bound for system performance, such knowledge is not available in practice and the channel needs to be frequently estimated. Channel estimation can be performed in various ways including the use of frequency and/or time correlation properties of the wireless channel, blind or pilot-based channel estimation, and adaptive or non-adaptive channel estimation. Non-parametric methods attempt to estimate the frequency response without relying on a 414 Chapter 4 specific channel model. In contrast, the parametric estimation methods assume a certain channel model and determine the parameters of this model. Spaced-time and space- d-frequency correlation functions, discussed in the previous chapter, are specific properties of channel that can be incorporated in the estimation method, improving the quality of estima- tions. Pilot-based estimation methods are the most commonly used in OFDM systems which are applicable in systems where the sender transmits some known signals to the receiver. A pilot-based channel estimation is defined as the use of the channel samples estimated at the pilot tones to reconstruct channel samples at the remaining data/control-bearing subcar- riers. As a result, the pilot pattern design is essentially a conventional sampling rate selec- tion problem in the two-dimensional signal processing space. To avoid aliasing during reconstruction of the channel time-frequency function, the pilot tone selection should follow the two-dimensional sampling theorem. When multiple antennas are used, the receiver must estimate the channel impulse response (or the transfer function) from each of the transmit antennas to correctly detect the signal. This is achieved through distributing reference sig- nals (or pilot tones) among the transmit antennas. Let Af = 1/Tu and Tsymbol = Tg denote the subcarrier spacing (SCS) and the OFDM symbol duration (inclusive of the guard interval), re

spectively. Let us further assume that the pilot subcarriers are transmitted at integer multiples of subcarrier spacing and OFDM symbol duration in frequency and time directions, respectively (i.e., fp = m/Tu and Tp = nT symbol where m and n are integers). The (m,n) pair represents the pilots' separation in terms of subcarrier spacing and OFDM symbol duration. From the sampling theorem point of view, the channel's two-dimensional delay-Doppler response h(T,v) can be fully reconstructed, if the two-dimensional transform function H(t,f) is sampled greater than or equal to the Nyquist rate across time and fre- quency dimensions. Hence, the time-domain sampling rate must be greater than or equal to the channel's maximum Doppler spread, that is, Tp < 1/Vmax (the sampling rate in time must be less than the coherence time) and the frequency-domain sampling rate must be greater than or equal to the channel's maximum delay spread, that is, fp < 1/Tmax (the sam- pling rate in frequency must be less than coherence bandwidth). Assuming a wide-sense sta- tionary uncorrelated scattering channel model and further assuming the channel to be constant over one OFDM symbol, the frequency response H(t,f) of an L-path channel is given H(t,f) = where V1, V1, and T1 denote the phase, Doppler frequency, and delay of the 1th path, respectively. All these parameters are independent random variables. In general, the pilot signals are oversampled to ensure a good trade-off between performance and overhead. Therefore the choice of (m,n) depends on the channel's maximum delay spread and maximum Doppler spread and must satisfy the following equation according to two-dimensional sampling theorem: ms27" New Radio Access Physical Layer Aspects (Part 2) 415 It should be noted that the pilot density for a regular pattern can be calculated using the pre- ceding equation. For large values of Vmax (i.e., large Doppler spread, or alternatively small channel coherence time means that channel is time-varying), n should be small to appropri- ately track c

hannel time variations. On the other hand, for large values of T max (i.e., large delay spread, or alternatively small coherence bandwidth means that channel is frequency- selective), m should be small to closely follow channel frequency variation. In a regularly spaced pilot pattern, the pilot symbols are evenly spaced in frequency and in time. Cell-specific reference signals (CRS) were originally defined in 3GPP LTE Rel-8 and have continued to be used as downlink wideband always-on power-boosted pilot subcarriers which are scaled with the number of transmit antenna ports that are essential in coherent decoding/detection of the LTE downlink control channels, mobility measurements, etc. The always-on and power-boosted properties of CRS have been the potential cause of inter-cell interference in LTE networks even in the absence of user traffic. The reference signals in NR are different from LTE in the sense that they are only present when the UE has data allocation; thus they are UE specific and confined to the time-frequency region where user data are allocated. Unlike LTE, NR does not utilize cell-specific reference signals, rather exclusively relies on user-specific demodulation reference signals (DM-RSs) for channel estimation, enabling efficient beamforming and other multi-antenna schemes. In NR, the reference signals are not transmitted unless there are data to transmit, thereby improving network energy efficiency and reducing inter-cell interference. Support for low-latency transmission is an important part of NR design. In a front-loaded DM-RS structure, the ref- erence signals and downlink control channel carrying scheduling information are located at the beginning of the slot; thus a device can start processing the received data immediately without prior buffering and time-domain interleaving across OFDM symbols, thereby mini- mizing the decoding delay. There are other types of reference signals such as phase tracking reference signals (PT-RS) which are used to counter phase noise at higher frequencies.

A question may arise that if there are no wideband cell-specific reference signals, how the UEs will measure the reference signal received power (RSRP) during initial access or cell selection (mobility measurements) as they will not have any allocation at that time? The answer lies in the fact that NR uses a primary synchronization signal (PSS)/secondary syn- chronization signal (SSS)/PBCH block structure, where the PBCH has its own set of refer- ence signals which will always be present in the PBCH SO the UE while detecting the PSS/ SSS/PBCH block should be able to measure an RSRP value from the PBCH DM-RS. To support channel tracking, different types of reference signals are transmitted in downlink and uplink. The reference signals in the downlink include the following [6]: UE-specific DM-RS for physical downlink control channel (PDCCH) can be used for downlink channel estimation and coherent demodulation of PDCCH. The DM-RS for PDCCH is transmitted together with the PDCCH and is present only in the resource blocks that are used for PDCCH transmission. 416 Chapter 4 UE-specific DM-RS for physical downlink shared channel (PDSCH) can be used for downlink channel estimation for coherent demodulation of PDSCH. The DM-RS for PDSCH is transmitted together with the PDSCH and is present only in the resource blocks that are allocated for PDSCH transmission. UE-specific PT-RS can be used in addition to the DM-RS for PDSCH for correcting common phase error (CPE) between PDSCH symbols not containing DM-RS. It may also be used for Doppler and time-varying channel tracking. The phase noise of the transmitters increases as the frequency of operation increases. The PT-RS plays a crucial role especially in mmWave frequencies to minimize the effect of the oscillator phase noise on system performance. The phase noise appears as a common phase rotation of all the subcarriers, known as CPE in an OFDM system. The NR system typically maps the PT-RS information to a few subcarriers per symbol because the phase rotation equally affects

all subcarriers over an OFDM symbol but exhibits low correlation from symbol to symbol. The system configures the PT-RS depending on the quality of the oscillators, carrier frequency, SCS, and modulation and coding schemes that are used for the transmission. The PT-RS for PDSCH is transmitted together with the PDSCH on need basis. The PT-RS is denser in time domain but sparser in frequency domain com- pared to the DM-RS, and if configured, occurs only in conjunction with the DM-RS. UE-specific CSI-RS can be used for estimation of CSI to allow CSI measurement and reporting which assists the gNB in modulation coding scheme (MCS) selection, resource allocation, beamforming, and MIMO rank selection. The CSI-RS can be con- figured for periodic, aperiodic, or semi-persistent transmission with a configurable den- sity by the gNB. The CSI-RS also can be used for interference measurement (IM) and fine frequency/time tracking purposes. Specific instances of CSI-RS can be configured for time/frequency tracking and mobility measurements. In the absence of the cell- specific reference signals in NR the CSI-RS can be used for radio resource manage- ment, measurements and mobility management purposes in connected mode. Tracking reference signals (TRS) are sparse set of reference signals, which are intended to assist the device in time and frequency tracking. The TRS does not exist indepen- dently, and a specific CSI-RS configuration is used as TRS. In addition to time and fre- quency tracking, the TRS is used for estimation of delay spread and Doppler spread at the UE side. It is transmitted with a limited bandwidth for a configurable period of time, controlled by the upper layer parameters. Table 4.1 provides the L1 overhead associated with various NR downlink reference signals. In NR, the overhead due to the L1/L2 control signaling depends on the size and periodicity of the configured control resource set (CORESET) in the cell which includes the overhead from the PDCCH DM-RSs. If the CORESET is transmitted in every slot, max