4.5 Mitigation
4.6.1 Performance Evaluation Under Channel Non-Reciprocity
The simulation scenario considered for evaluating the performance of ZF and MRT
precoded multi-user massive MIMO downlink transmission consists of K = 20 UEs while
T
coh= 250. The non-reciprocity matrices A and C are constructed based on their
essential statistical properties in Table 4.1. Since the UEs are assumed to be single-
antenna devices, off-diagonal entries of A are zero and thus σ
2aod
= 0. The variance of the
m-th diagonal element in A
, σ
2amm
, is assumed to be equal for all the values of m and is
set to σ
2ad
. Next, for each channel non-reciprocity realization, the diagonal elements of A
and off-diagonal elements of C
are assumed to be i.i.d.
CN
0, σ
2 adand
CN
0, σ
2 cod,
respectively. The diagonal elements of C
are assumed to have Gaussian distribution with
zero mean, variance set to σ
2cd
and cross-correlation equal to δ
c2d.
The spectral efficiency and relative SINR degradation of the example scenario are plotted
against downlink SNR in Figure 4.2 and Figure 4.3, respectively, where the lines are
representing the results based on the closed-form analysis and star markers,∗, are the
corresponding simulation results. The perfect alignment between analytical and simulation
results proves the accuracy of the closed-form analysis and the used approximations. As
can be seen in Figure 4.2 and Figure 4.3, the impact of non-reciprocity in the channel is
much more severe in high SNR region as the performance in low SNR region is mainly
limited by noise and not the channel non-reciprocity. While both precoding schemes,
namely ZF and MRT, are suffering from channel non-reciprocity mainly in high SNR
region, it is clear that ZF precoding scheme is much more sensitive to such non-idealities,
as expected based on the analysis in Section 4.3.4.
The impacts of increasing the number of BS antennas on the spectral efficiency are
evaluated in Figure 4.4. As can be seen, the results are perfectly inline with the implications
made by analytical asymptotic performance study in Section 4.3.3 where it was concluded
that under non-reciprocal channels i) both ZF and MRT precoded systems have a finite
achievable sum-rate saturation level; ii) the saturation level is identical for both precoding
schemes. The results clearly confirm that the spectral efficiency of both ZF and MRT
52
Chapter 4. Analysis and Mitigation of Channel Non-Reciprocity in Massive MIMO
Systems
Figure 4.3: Relative SINR degradation vs. downlink SNR (ρd) forN = 100, Mtot=K = 20,
τu=Mtot,ρu= 0 dB.
Figure 4.4: System spectral efficiency vs. the number of antennas at BS (N) for Mtot=K = 20,
ρd= 20,τu=Mtot,ρu= 0 dB,Tcoh= 250. Saturation levels based on (4.21) are plotted in
green horizontal lines for the two indicated channel non-reciprocity parameter settings.
precoded systems saturate towards the value derived in Section 4.3.3. As discussed
earlier, increasing the number of antennas in the BS side, which increases the number
of non-reciprocal transceiver chains and antenna units, causes reduction in advantage
of ZF over MRT to the point that the performance of both systems saturates to the
same value. Note that as opposed to, e.g., pilot contamination [94] problem where the
performance saturation happens with having around 10
5BS antennas [102], the number
of BS antennas that causes performance saturation due to channel non-reciprocity is
around 10
3or even lower which shows the practical relevance of the results.
In Figure 4.5, the optimal number of simultaneously scheduled single-antenna UEs is
calculated as a function of channel non-reciprocity for two example values of downlink
Figure 4.5: Optimal number of single-antenna UEs to maximize system spectral efficiency vs.
non-reciprocity level (σ2a d=σ
2
cd = channel non-reciprocity level, whileσ2cod =δ2cd = channel
non-reciprocity level−10 dB) for N = 100, Mtot=K = 20, ρd= 20,τu=Mtot,ρu= 0 dB,
Tcoh= 250.
SNR, namely, ρ
d= 20 dB, 0 dB. This optimal number of simultaneously scheduled
single-antenna UEs, K
opt, refers to the K which maximizes the spectral efficiency and
is derived by evaluating (4.10) for all the values of N≥ K ≥ 1. Based on the obtained
results, K
optdrops for both precoding schemes as the channel non-reciprocity level, and
subsequently interference in the UE side, increases. Such drops are more visible in high
SNR region, i.e., ρ
d= 20 dB, while in low SNR region, i.e., ρ
d= 0 dB, the performance is
limited mainly by the receiver thermal noise. Comparing ZF and MRT precoded systems,
such drops can be experienced in the case of MRT precoding scheme in fairly severe
non-reciprocity levels, e.g., σ
2ad
>
−15 dB, whereas ZF precoded systems suffer from
a significant drop already with moderate channel non-reciprocity levels, e.g.,
−30 dB
< σ
2ad
<−20 dB. It is interesting to note that as opposed to ideal reciprocal scenario,
the optimal number of simultaneously scheduled single-antenna for MRT is higher than
that of ZF for high SNR values under moderate channel non-reciprocity levels.
Figure 4.6 illustrates the maximum tolerable channel non-reciprocity level as function of
downlink SINR at the UE side based on the derived SINR expressions in (4.13) and (4.18).
The considered example scenario covers two downlink SNR values, namely, ρ
d= 20 dB, 0
dB. As an example point in the obtained results, in order to guarantee 15 dB for downlink
SINR at the UE side in ZF precoded system with ρ
d= 20 dB, channel non-reciprocity
level cannot be higher than around−20 dB. Such analysis can be used in practical system
design and deployments, e.g., to calculate the required channel non-reciprocity calibration
levels for a given target downlink transmission performance, which proves the applicability
of the provided analytical results in practical systems.
In document
Analysis and Mitigation of Channel Non-Reciprocity in TDD MIMO Systems
(Page 65-67)