1

Analog Versus Hybrid Precoding for Multiuser

Massive MIMO with Quantized CSI Feedback
Yaqiong Zhao, Student Member, IEEE, Wei Xu, Senior Member, IEEE, Jindan Xu,

Student Member, IEEE, Shi Jin, Senior Member, IEEE, Kezhi Wang, Member, IEEE, and

Mohamed-Slim Alouini, Fellow, IEEE

arXiv:2006.00899v1 [cs.IT] 1 Jun 2020

Abstract

In this letter, we study the performance of a downlink multiuser massive multiple-input multiple-output (MIMO)
system with sub-connected structure over limited feedback channels. Tight rate approximations are theoretically
analyzed for the system with pure analog precoding and hybrid precoding. The effect of quantized analog and digital
precoding is characterized in the derived expressions. Furthermore, it is revealed that the pure analog precoding
outperforms the hybrid precoding using maximal-ratio transmission (MRT) or zero forcing (ZF) under certain
conditions, and we theoretically characterize the conditions in closed form with respect to signal-to-noise ratio
(SNR), the number of users and the number of feedback bits. Numerical results verify the derived conclusions on
both Rayleigh channels and mmWave channels.

Index Terms

Massive MIMO, analog precoding, hybrid precoding, limited feedback.

I. I NTRODUCTION

M
ASSIVE multiple-input multiple-output (MIMO) has attracted increasing attention as a key tech-
nology in the fifth-generation (5G) network [1]. With its advantages for inter-user interference

Y. Zhao, J. Xu, and S. Jin are with the National Mobile Communications Research Laboratory, Southeast University, Nanjing 210096,
China (email: {zhaoyaqiong, jdxu, jinshi}@seu.edu.cn).
W. Xu is with the National Mobile Communications Research Laboratory, Southeast University, Nanjing 210096, China, and is also with
the Purple Mountain Laboratories, Nanjing 210000, China ([email protected]).
K. Wang is with the Department of Computer and Information Sciences, Northumbria University, Newcastle, UK (email:
[email protected]).
M. -S. Alouini are with the Computer, Electrical and Mathematical Science and Engineering Division, King Abdullah University of Science
and Technology, Thuwal, Saudi Arabia, 23955 (email: [email protected]).
 2

cancellation and noise supression, massive MIMO can achieve near-optimal performance with simple
linear precoding schemes, such as maximal-ratio transmission (MRT) and zero forcing (ZF) [2]. These
fully-digital precoding schemes need each antenna to be driven by one dedicated radio-frequency (RF)
chain, which imposes prohibitively high cost and power consumption [3]. To address this issue, two kinds
of designs have been introduced, i.e., pure analog precoding and hybrid analog-and-digital precoding [4]
[5].
Although the hybrid precoding has been widely considered in research, evidence has shown that it may
not be the best choice for all cases compared with the pure analog precoding, especially when considering
that analog precoding consumes less power than the hybrid precoding due to some implementation facts [4]
[6]. In [7] [8], the analog precoding approximately transformed the effective channel into a diagonal matrix,
which implies that digital processing is no longer needed for further multiuser interference cancellation.
Specifically in [9], the authors found that pure analog processing could surpass the hybrid processing with
maximal-ratio combination (MRC) or ZF in an uplink channel under the assumption of perfect channel
state information (CSI). For the downlink channel, however, this compromise is still unclear. Moreover
in practice, only quantized CSI, instead of perfect one, is available through limited feedback [10]-[13].
Against the above background, this letter investigates the performance of a downlink massive MIMO
system with low-cost sub-connected architecture and quantized CSI feedback. Tight rate approximations
are derived for the ZF/MRT-based hybrid precodings and the analog precoding in the large base station
(BS) antenna regime. With the derived results, we explicitly characterize the performance comparison
between the hybrid precoding and the analog precoding. In particular, we show that for all SNRs analog
precoding outperforms the ZF-based hybrid precoding when B2 ≤ B20 where B2 denotes the number of
feedback bits in digital precoding and B20 is a constant in closed-form with respect to system parameters.
Similar observation has also been obtained for the comparison of MRT-based hybrid precoding and pure
analog precoding. For other cases, the superiority of different precoding schemes would be complicated
and it depends on specific values of SNR, which has also been derived in closed forms.
The remainder of this paper is organized as follows. System model is introduced in Section II. In
Section III, we derive the achievable rates of the system using various precoding schemes. In Section
IV, we present the conditions under which the pure analog precoding can beat the hybrid precoding.
Simulation results and conclusions are given in Section V and Section VI, respectively.
 3

II. S YSTEM M ODEL

We consider a multiuser massive MIMO downlink channel with sub-connected architecture. Multiple
users are simultaneously served by a BS which is equipped with M antennas and K RF chains. Each RF
M
chain is connected to a subset of N antennas through dedicated phase shifters where N = K
. Considering
that the number of transmit streams should not exceed the number of RF chains, we assume that K single-
antenna users are scheduled from the user pool. Assuming flat Rayleigh fading, the received signal at the
kth user can be expressed by

yk = βk hH
k AWs + nk , k = 1, 2, · · · , K, (1)

where hH
k ∼ CN (0M , IM ) denotes the downlink channel from the BS to the kth user, and s ∈ C
K×1
is
P
the data vector with E[ssH ] = I
K K
where P is the total transmit power at the BS, nk ∼ CN (0, σ 2 ) is
the additive Gaussian noise, βk denotes the path loss of the kth user, and A = [a1 , a2 , · · · , aK ] ∈ CM ×K
and W = [w1 , w2 , · · · , wK ] ∈ CK×K respectively stand for the analog precoder and the digital precoder.
Due to the hardware dissipation caused by the power dividers [14], the analog precoder is written as
A= √1 F where F denotes the equivalent analog precoding implemented by the phase shifter network.
N

To conduct equal power allocation for users, it holds that kFwk k = 1 for k = 1, 2, · · · , K.
In the current design of hybrid precoding for massive MIMO, the design of analog precoder can be
accomplished in two ways, that is, by either utilizing a quantized CSI feedback from users, e.g., precoding
matrix indicator (PMI) feedback, or by using uplink channel estimates, e.g., the sounding reference signal
(SRS), as specified in the 5G New Radio (NR) specifications [15]. Based on this, the BS designs the
analog precoder in terms of discretized phase shifters. Then, an analogly beamformed pilot will be sent by
the BS for further equivalent channel estimation to compensate for the performance degradation caused
by quantization error of the analog precoder. Each user estimates its own effective channel and feeds
back a quantized version of the precoded CSI which is used by the BS to design the digital precoder. In
this paper, phases of the MK entries of A are quantized up to B1 bits, i.e., the codebook is designed
n j2πn o
as A = e 2B1 , n = 0, 1, · · · , 2B1 − 1 . Based on the minimum Euclidean distance criterion, a common
way of analog precoder design in the sub-connected structure follows [9] [10]

 N1 argmax R[h∗k,i ej ϕ̂k,i ], N(k − 1) + 1 ≤ i ≤ Nk

j ϕ̂
ak,i = e k,i ∈A (2)

 0, otherwise
 4

where ak,i and hk,i are respectively the ith element of ak and hk , and R[x] returns the real part of x. For
∆
the kth user, we define its effective channel as gkH = hH
k A. Then, the ergodic rate of the kth user can be

written as   
γβk 2
K
gkH wk
Rk = E log2 1 + 2  , (3)
  
γ P H
1+ K
βj |gk wj |
j6=k

P
where γ , σ2
is the SNR.
The digital part of the system is based on the effective channel. Since the effective channel is always
correlated even for independent and identically distributed (i.i.d.) channels of H, we adopt a channel
statistics-based codebook G [10]-[12]. The kth user quantizes its effective channel gk according to ĝk =
arg maxĝki ∈G |gkH ĝki |, in which ĝki is defined as
1
Rk2 vi
ĝki = 1 , (4)
kRk2 vi k

where Rk denotes the correlation matrix of user k’s effective channel and vi ∈ CK×1, i ∈ {1, 2, · · · , 2B2 }
is an i.i.d. complex Gaussian vector randomly chosen from random vector quantization (RVQ) codebook
of size 2B2 . With the limited feedback of B2 bits, the BS calculates the digital precoder by using the
quantized CSI feedback ĝk .

III. ACHIEVABLE R ATE A NALYSIS

A. MRT/ZF-based Hybrid Precoding with Limited Feedback

For the MRT-based hybrid precoding, the digital precoder of user k is determined as

wk = ĝk . (5)

With the design above, we derive the downlink rate in the following theorem.
Theorem 1: Tight rate approximation of MRT-based hybrid precoding in the large BS antenna regime
is obtained as  B2 
γβk πsinc2 (δ) 2 (δ)
− 2 N )( πN sinc
−
K−1
MRT K
( 4 +KN 4
+ K)
RH,k = log2 1 + πN sinc2 (δ) γ K πsinc 2 (δ)
, (6)
4
+ K + K
β k ( N
+ 2
)

∆ π ∆ sin(x) P
where δ = 2B1
, sinc(x) = x
and β k = βj .
j6=k

Proof. See Appendix A.

 5

From the above theorem, we can also get the achievable rate of the MRT-based hybrid system with
perfect CSI as a special case by letting B1 → ∞ and B2 → ∞ in (6). It yields
K 2
!
γβk N π
MRTp K
( 4
+ N
)
RH,k = log2 1 + . (7)
πN γβ k π K
4
+ K + K 2
( + N
)

Note that this result coincides with the result in [9, Eq. (15)] except for that the signal term and the
1
interference term calculated in [9] have an additional factor of N
if we let βi = 1, i = 1, 2, · · · , K, which
is because we consider the effect of the divider network.
For the ZF-based precoding, the digital precoder equals

WZF = Ĝ(ĜH Ĝ)−1 , (8)

where Ĝ = [ĝ1 , ĝ2 , · · · , ĝK ] and WZF = [w1ZF , w2ZF, · · · , wK

ZF
]. The digital precoder is then normalized
wkZF
as wk = kFw ZF
. A tractable and tight bound of its achievable rate is given in the following proposition.
k k

Proposition 1: The ergodic rate of the ZF-based hybrid precoding is asymptotically lower bounded by
!
2
ZF 4KN + πNγβ k sinc (δ)
RH,k > log2 B2 . (9)
4KN + 4γβ k 2− K−1

Proof. See Appendix B.

B. Pure Analog Precoding

When the pure analog precoding is applied, the digital precoding matrix reduces to an identity matrix,
i.e., W = IK . Then we can easily get the following proposition.
Proposition 2: Tight rate approximation of the pure analog precoding in the large BS antenna regime
is characterized as 2 (δ)
!
γβk ( πN sinc
4
+ 1)
RA,k = log2 1+ . (10)
KN + γβ k
2 πsinc2 (δ) πsinc2 (δ) 2
Proof. With the pure analog precoding, E[ gkH wk ] = E[|gk,k |2 ] = 4
+ 1
N
− 4N
and E[ gkH wj ] =
E[|gk,i |2 ] = 1
N
. By applying [16, Lemma 1], and substituting these two expressions into (3) and using
similar manipulations as that in the last step of (30), the proof completes.

As a special case, we have the achievable rate of the system with perfect CSI by letting B1 → ∞,
which gives !
πN γβk
p 4
RA,k = log2 1 + , (11)
KN + γβ k
 6

which consists with the existing result in [9, Eq. (13)].

IV. P URE A NALOG VS H YBRID P RECODING

To compare the performance of the pure analog precoding and the hybrid precoding, we evaluate the
rate gap as
X
∆R = RH,k − RA,k , (12)

where X ∈ {ZF, MRT} represents the strategy of the digital precoding in the hybrid precoding.

A. MRT-based Hybrid Precoding

By letting △R ≥ 0, and substituting (6) and (10) into (12), we get

B2 !
γβ k K − 2 − 2− K−1 πsinc2 (δ) B2
− > 2− K−1 − K + 1. (13)
K N 4

It implies that the MRT-based hybrid precoding outperforms the pure analog precoding if (13) holds. It is
B2
K−2−2
−
K−1 πsinc2 (δ)
obvious that the right hand side of (13) is negative when K > 1. Checking that N
≥ 4
,
we can rewrite (13) as
πNsinc2 (δ) − K−1
B2
K≥ +2 + 2, (14)
4

under which the MRT-bsed hybrid precoding outperforms the pure analog precoding for all SNR values.
This is due to the fact that multiuser interference becomes a dominant factor when K becomes large, and
the hybrid precoding which utilizes digital processing to eliminate interference enjoys a better performance
than the pure analog precoding.
Moreover from (14), it is revealed that B2 has a marginal impact on the performance of the MRT-based
B2
hybrid precoding since 2− K−1 ≪ 1. Therefore we can focus on the impact of B1 on system performance.
  2
π
2 π 2 π 2B1
By applying Taylor’s expansion to sin (δ) with δ = 2B1
, we get sinc ( 2B1 ) ≈ 1 − 3
, and (14) is
further equivalent to  
3
1 π N ∆
B1 ≤ log2   B2
 = B10 . (15)
2 3πN − 12 K − 2 − 2 − K−1

Otherwise, when B1 > B10 , it is not difficult to get


 > 0, γ < γ0
∆R (16)
 ≤ 0, γ ≥ γ
0
 7

 B2 
−
M 2 K−1 −K+1
∆
where γ0 =  B2 2 (δ)
.
β k K−2−2 K−1 − πNsinc
−
4

Remark 1: When B1 ≤ B10 , the MRT-bsed hybrid precoding always outperforms the pure analog
precoding in terms of the ergodic achievable rate for all SNRs. Otherwise the MRT-bsed hybrid precoding
can beat the pure analog precoding only for low SNRs satisfying γ < γ0 .

B. ZF-based Hybrid Precoding

Similarly by letting △R ≥ 0 for X = ZF in (12) and using (9) and (10), we obtain
 
4(βk + β k ) ∆
B2 > (K − 1)log2 1+ = B20 , (17)
πNβk sinc2 (δ)
 B2 
πN βk sinc2 (δ) 1−2 K−1
−
B2
which uses the fact that 4
> 2− K−1 (βk + β k ). When B2 > B20 , it holds true that

 > 0, γ > γ1
∆R (18)
 ≤ 0, γ ≤ γ
1

B2
∆ −
4M (β k 2 K−1 +βk )
where γ1 =   B2  B2  .
β k πN βk sinc2 (δ) 1−2 K−1 −4(βk +β k )2 K−1
− −

Otherwise, when B2 ≤ B20 , the pure analog precoding always outperforms the ZF-based hybrid precod-
ing for all SNRs. This is because the precoding vector of the kth user is supposed to lie in the null space
of channels of other users and thus the ZF-based hybrid precoding is more dependent on the accuracy of
channel estimation to cancel out interference.
Remark 2: If B2 ≤ B20 , the pure analog precoding beats the ZF-based hybrid precoding for all SNRs.
Otherwise the ZF-based hybrid precoding takes over for high SNRs satisfying γ ≥ γ1 .

V. S IMULATION R ESULTS

In this section, we compare the downlink performance of the analog precoding and hybrid precoding
based on different linear processing schemes. The path loss factor βk , k = 1, 2, · · · , K is chosen uniformly
in [0.5, 1.5].

A. Rayleigh Fading Channels

In Fig. 1(a) we illustrate the spectral efficiency achieved by pure analog precoding and the MRT-based
hybrid precoding. It is observed that when B1 ≤ B10 = 1.05, the MRT-based hybrid precoding outperforms
 8

12 8
Analog numerical, B1=1 Analog numerical
7 Analog in (10)
Analog in (10), B1=1
10 ZF numerical, B2=2

Average sum rate (bps/Hz)

Average sum rate (bps/Hz)

Analog numerical, B1=5 6 ZF in (9), B 2=2

Analog in (10), B1=5
8 ZF numerical, B2=5
MRT numerical, B 1=1 5
ZF in (9), B 2=5
MRT in (6), B1=1
6 4
MRT numerical, B 1=5
MRT in (6), B1=5 3
4
2
2
1

0 0
-10 -8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18 -10 -8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18
SNR (dB) SNR (dB)
(a) M = 120, K = 6, and B2 = 10. (b) M = 60, K = 6, and B1 = 2.
Fig. 1. Achievable rates of analog precoding and MRT-based hybrid precoding in (a) and ZF-based hybrid
precoding in (b)

pure analog precoding for all SNRs, while when B1 > B10 the pure analog precoding performs better at high
SNRs, confirming the conclusions in Section IV. This is because for small B1 , the system performance
is severely affected by the quantization error, i.e., the interference becomes dominant, thus the hybrid
precoding using digital processing to eliminate interference outperforms the pure analog precoding. In
addition, the corresponding threshold calculated from (14) is 17 for B1 = 5, so for K = 6 < 17, the pure
analog precoding outperforms the MRT-based hybrid precoding at high SNRs.
Similar trends can be found when comparing pure analog precoding and the ZF-based hybrid precoding
in Fig. 1(b). When the effective channel is roughly quantized, i.e., B2 ≤ B20 = 4.29, the pure analog
precoding performs better for all SNRs since the accuracy of channel estimation is insufficient to support
effective beamforming for ZF, while the ZF-based hybrid precoding takes over at high SNRs when B2 >
B20 , which verifies our observations in Section IV.

B. Large mmWave Multiuser Channels

Apart from Rayleigh fading channels, hybrid/analog precoding can also be applied to mmWave com-
munications. To capture the nature of high-frequency propagations, we adopt the widely-used geometric
channel model [7] [9]
Np
s
M X k H k
hH
k = α a (φl ), (19)
Np l=1 l
 9

Average sum rate (bps/Hz)

Analog precoding
ZF-based hybrid precoding, B2=3
6
ZF-based hybrid precoding, B2=12
MRT-based hybrid precoding, B2=3
4 MRT-based hybrid precoding, B2=12

0
-10 -5 0 5 10 15 20
SNR (dB)
Fig. 2. Achievable rates over mmWave channels with M = 40, K = 5, B1 = 2, and Np = 10.

where Np is the number of propagation paths from BS to user and αlk ∼ CN (0, 1) denotes the complex gain
of the l-th path. φkl is the azimuth angle of departure drawn independently from the uniform distribution
over [0, 2π]. a(φkl ) is the array response vector of BS. Here we consider a uniform linear array (ULA)
whose array response vector is given by [9, eq. (29)].
Fig. 2 shows that conclusions on the superiority of the pure analog precoding over hybrid precodings
also hold for the mm-Wave channels. Specifically, when B2 is small, i.e., B2 = 3, the analog precoding
outperforms the ZF-based hybrid precoding for all SNRs, while for large B2 , i.e., B2 = 12, the ZF-based
hybrid precoding takes over at high SNRs, which verifies Remark 2.

VI. C ONCLUSION

In this paper, we derive tight rate approximations of a massive MIMO system using pure analog
precoding and hybrid precoding with limited feedback in the large BS antenna regime. The effect of
quantized analog and digital precodings are characterized in the obtained expressions. Furthermore, it is
revealed that pure analog precoding outperforms hybrid precoding in terms of the ergodic achievable rate
under certain conditions, which have been derived in closed forms with respect to the SNR, the number
of users and the number of feedback bits. Numerical results verify the observed.

A PPENDIX A
P ROOF O F T HEOREM 1
h i h i
H 2 H 2
To evaluate (3), we calculate the terms E gk wk and E gk wj . Let gk = kgk k g̃k where g̃k is
h i h i
H 2  2 H 2
the normalized gk , then we have E gk wk = E kgk kF E g̃k wk .
 10

From the definition of gk , its kth element equals

Nk Nk
1 X (a) 1 X
gk,k = hk,l e−j ϕ̂k,l = λl , (20)
N N
l=N (k−1)+1 l=N (k−1)+1

∆ ∆
where (a) uses λl = hk,l e−j ϕ̂k,l . Defining εk,l = ϕk,l − ϕ̂k,l , it yields λl = |hk,l |ejεk,l . Recall hH
k ∼

CN (0M , IM ), where {hk,i }’s are i.i.d. complex Gaussian variables with zero mean and unit variance. It
√
π
implies that |hk,l | follows the Rayleigh distribution with mean 2
and variance 1 − π4 . The quantization
∆ π
error εk,l ∼ U[−δ, δ) is a uniform distribution with δ = 2B1
. Then we have
√
 jεk,l
 (a) (b) πsinc(δ)
E[R[λl ]] = E R[|hk,l |e ] = E[|hk,l | cos(εk,l )] = E[|hk,l |]E[cos(εk,l )] = , (21)
2
√
where (a) utilizes the independence between |hk,l | and εk,l , and (b) is due to E[|hk,l |] = 2π and E[cos(εk,l )] =
1
Rδ
2δ −δ
cos εk,l dεk,l = sinc(δ) because of the given distributions of |hk,l | and εk,l . Analogously, we can also
1+sinc(δ) cos(δ)
get E[(R[λl ])2 ] = E[|hk,l |2 ]E[cos2 (εk,l)] = 2
. Then,

V[R[λl ]] = E[(R[λl ])2 ] − (E[R[λl ]])2 = w1 , (22)

∆ 1+sinc(δ) cos(δ) 2 (δ)

where w1 = 2
− πsinc4 . Applying the Central Limit Theorem to (20) and using (21) and (22),
we get
√ 
πsinc(δ) w1
R[gk,k ] ∼ N , . (23)
2 N

Following similar reasons, it can be readily proved that

w2 1 1
I[gk,k ] ∼ N (0, ), R[gk,i ] ∼ N (0, ), I[gk,i ] ∼ N (0, ), (24)
N 2N 2N
∆ 1−sinc(δ) cos(δ)
where w2 = 2
and I[x] returns the imaginary part of x. Now we have

πsinc2 (δ) w1 +w2

E[|gk,k |2 ] = E[|R[gk,k ]|2 ]+E[|I[gk,k ]|2 ] = + . (25)
4 N
1
E[|gk,i |2 ] = E[|R[gk,i ]|2 ] + E[|I[gk,i ]|2 ] = . (26)
N

According (25) and (26), it gives

K
X πsinc2 (δ) K πsinc2 (δ)
E kgk kF = E[|gk,k | ] + E[|gk,i |2 ] =
2 2

+ − . (27)
i=1,i6=k
4 N 4N
 11

From [10, Lemma 2], Rk can be asymptotically written as Rk = diag[r1,1 , r2,2 , · · · , rk,k ] where

πsinc2 (δ) 1 πsinc2 (δ)

4
+ N
− 4N
, i=k
ri,i = (28)
1

N
, i 6= k

and diag[·] returns a diagonal matrix with the input as its elements. Then we follow the result in [11] and
get
B2
2
h
2
i σk,2 B2 2− K−1
E g̃kH ĝk ≈ 1 − 2 2− K−1 = 1 − πN sinc2 (δ) πsinc2 (δ)
, (29)
σk,1 +1−
4 4

1
where σk,1 and σk,2 respectively stand for the largest and the second largest singular value of Rk2 .
Combining (27) and (29), it yields
B2
2 (a) 2 (b) πsinc2 (δ) K 2− K−1
E[ gkH wk ] = E[kgk k2F ]E[ g̃kH ĝk ] → + − ,
4 N N
πsinc2 (δ) πsinc2 (δ) 2 (δ)
+K K
+N − πsinc
where (a) utilizes wk = gk and in (b) we consider the fact that πsinc2 (δ)
4 N
πsinc2 (δ)
→ 1 and 4
πNsinc2 (δ)
4N
2 →
4
+ K
N
− 4N 4
+1− πsinc4 (δ)
1
N
hold in massive MIMO with large N and fixed K.
1
(Rj2 ) vi,j
j,j
From (4), we get ĝj,j = 1 , where ĝj,j and vi,j are respectively the jth element of ĝj and vi ,
kRj2 vi k
and i denotes the index of the quantization vector of gj . Then we have
 1   1 
(Rj2 )j,j vi,j (Rj2 )j,j
E[ĝj,j ] = E  1
 (a)
= E[vi,j ]E  1
 (b)
= 0, (30)
kRj vi k2
kRj vi k
2

1
(Rj2 )
where (a) uses the independence between vi,j and 1
j,j
, and (b) results from vi ∼ CN (0M , IM ).
kRj2 vi k
Analogously,
 
πsinc2 (δ) πsinc2 (δ)
2
(Rj )j,j |vi,j |2 2
(Rj )j,j (a) 4
+ 1
N
− 4N
E[|ĝj,j | ] = E  1
 = E[|vi,j | ] h 1 i = πsinc2 (δ) K πsinc2 (δ)
, (31)
kRj vi 2
k2 E kRj2 vi k2 4
+ N
− 4N

h 1 K
i P K
2 πsinc2 (δ) K πsinc2 (δ)
(Rj )l,l E[|vi,l |2 ] = and E[|vi,l |2 ] = 1
P
where (a) uses E kRj vi k =
2
(Rj )l,l = 4
+ N
− 4N
l=1 l=1
according to the distribution of vi and (28). Following trivially the above steps, we get E[ĝj,i ] = 0, and
1
E[|ĝj,i |2 ] = πNsinc2 (δ) 2 . Providing that {vi,l }’s are i.i.d. complex Gaussian variables with zero
4
+K− πsinc4 (δ)
∗
mean and unit variance, then ĝj,i and ĝj,l are independent for any i 6= l, which implies E[ĝj,i ĝj,l ] =
 12

h i
2
∗
E[ĝj,i ]E[ĝj,l ] = 0 and E[|ĝj,i |2 |ĝj,l
∗ 2
| ] = E[|ĝj,i |2 ]E[|ĝj,l
∗ 2
| ]. Then E gkH wj can be calculated as
 2

h i h i K K
2 2
X
∗ (a) X
∗ ∗
E gkH wj = E gkH ĝj = E gk,i ĝj,i  = E[gk,i gk,l ]E[ĝj,i ĝj,l ]
i=1 1≤i≤l≤K
K πsinc2 (δ) K
(b) X (c) +
= E[|ĝj,i |2 ]E[|gk,i |2 ] → 2
πN sinc2 (δ)
N
, (32)
i=1 4
+K
∗ ∗
where (a) is due to the independence between gk and ĝj , (b) holds because E[gk,i gk,l ]E[ĝj,i ĝj,l ] = 0 for
any i 6= l, and (c) is obtained by using similar manipulations as that in the last step of (30). Finally, by
applying [16, Lemma 1] and substituting (30) and (32) into (3), the theorem is proved.

A PPENDIX B
P ROOF O F P ROPOSITION 1
√
a.s. πsinc(δ) a.s.
From the distributions of gk,k and gk,i in (23) – (24), we get gk,k → 2
and gk,i → 0 since
w1 1 a.s. πsinc2 (δ)
N
→ 0, wN2 → 0, and 2N
→ 0 for the massive MIMO with large N, thus G → 4
IK . Now the
asymptotic rate of the ZF-based hybrid system with perfect CSI can be represented as

γβk πsinc2 (δ)

 
ZF a.s.
RH,kp → log2 1 + . (33)
4K

On the other hand, it is known from [12] that the rate loss caused by the imperfect CSI can be upper
2 
 h i
γβ k E |gkH wj |
bounded by log2 1 + K
. Thanks to the orthogonality between ĝk and wj in the ZF precoding,
we further have B2
h i h i 2− K−1
2 2
gkH wj kgk k2F E H
 
E ≤E 1 − g̃k ĝk = . (34)
N

Therefore, we have B2
γβ 2− K−1
Rloss ≤ log2 (1 + k ). (35)
KN

Combining (33) and (35), the proposition is directly proved.

R EFERENCES

[1] M. Wang, F. Gao, S. Jin, and H. Lin, “An overview of enhanced massive MIMO with array signal processing techniques,” IEEE J. Sel.
Topics Signal Process., vol. 13, no. 5, pp. 886-901, Sept. 2019.
[2] M. Matthaiou, C. Zhong, and T. Ratnarajah, “Novel generic bounds on the sum rate of MIMO ZF receivers,” IEEE Trans. Signal
Process., vol. 59, no. 9, pp. 4341–4353, Sept. 2011.
 13

[3] X. Yu, J. C. Shen, J. Zhang, and K. B. Letaief, “Alternating minimization algorithms for hybrid precoding in millimeter wave MIMO
systems,” IEEE J. Sel. Topics Signal Process., vol. 10, no. 3, pp. 485–500, Apr. 2016.
[4] A. Alkhateeb, J. Mo, N. Gonzalez-Prelcic, and R. W. Heath, “MIMO precoding and combining solutions for millimeter-wave systems,”
IEEE Commun. Mag., vol. 52, no. 12, pp. 122–131, Dec. 2014.
[5] R. Ghanaatian, V. Jamali, A. Burg, and R. Schober, “Feedback-aware precoding for millimeter-wave massive MIMO systems,” in Proc.
IEEE 30th Annu. Int. Symp. Pers. Indoor, Mobile Radio Commun., Sept. 2019, pp. 1–7.
[6] C. Fang, B. Makki, J. Li, and T. Svensson. (2020, Jan 13). Hybrid precoding in cooperative millimeter wave networks [Online]. Available:
https://arxiv.org/abs/2001.04390v1.
[7] L. Liang, W. Xu, and X. Dong, “Low-complexity hybrid precoding in massive multiuser MIMO systems,” IEEE Wireless Commun.
Lett., vol. 3, no. 6, pp. 653–656, Dec. 2014.
[8] M. Fozooni, M. Matthaiou, S. Jin, and G. C. Alexandropoulos, “Massive MIMO relaying with hybrid processing,” in Proc. IEEE ICC,
Kuala Lumpur, Malaysia, May 2016, pp. 1–6.
[9] J. Du, W. Xu, B. Sheng, and C. Zhao, “Rethinking uplink hybrid processing: When is pure analog processing suggested?” IEEE Trans.
Veh. Technol., vol. 68, no. 5, pp. 5139–5144, May 2019.
[10] J. Du, W. Xu, H. Shen, X. Dong, and C. Zhao, “Hybrid precoding architecture for massive multiuser MIMO with dissipation: Sub-
connected or fully-connected structures?” IEEE Trans. Wireless Commun., vol. 17, no. 8, pp. 5465–5479, Aug. 2018.
[11] B. Clerckx, G. Kim, and S. Kim, “MU-MIMO with channel statistics-based codebooks in spatially correlated channels,” in Proc. IEEE
GLOBECOM 2008, New Orleans, LA, USA, Dec. 2008, pp. 1–5.
[12] W. Shen, L. Dai, Y. Zhang, J. Li, and Z. Wang, “On the performance of channel-statistics-based codebook for massive MIMO channel
feedback,” IEEE Trans. Veh. Technol., vol. 66, no. 8, pp. 7553–7557, Aug. 2017.
[13] H. Wang, W. Wang, V. K. N. Lau, and Z. Zhang, “Hybrid limited feedback in 5G cellular systems with massive MIMO, IEEE Syst.
J., vol. 11, no. 1, pp. 50–61, Mar. 2017.
[14] V. Venkateswaran, F. Pivit, and L. Guan, “Hybrid RF and digital beamformer for cellular networks: Algorithms, microwave architectures,
and measurements,” IEEE Trans. Microw. Theory Techn., vol. 64, no. 7, pp. 2226–2243, Jul. 2016.
[15] A. Zaidi, G. Durisi, F. Athley, J. Medbo, X. Chen, and U. Gustavsson, 5G Physical Layer: Principles, Models and Technology
Components, Academic Press, 2018.
[16] Q. Zhang, S. Jin, K. Wong, H. Zhu, and M. Matthaiou, “Power scaling of uplink massive MIMO systems with arbitrary-rank channel
means,” IEEE J. Sel. Topics Signal Process., vol. 8, no. 5, pp. 966–981, Oct. 2014.