1

Hybrid Precoder and Combiner Design with Low

Resolution Phase Shifters in mmWave MIMO
Systems
Zihuan Wang, Student Member, IEEE, Ming Li, Senior Member, IEEE, Qian Liu, Member, IEEE, and A. Lee
Swindlehurst, Fellow, IEEE
arXiv:1710.06192v1 [[Link]] 17 Oct 2017

Abstract—Millimeter wave (mmWave) communications have vices. This flood of mobile traffic has significantly exacerbated
been considered as a key technology for next generation cellular spectrum congestion in current frequency bands, and therefore
systems and Wi-Fi networks because of its advances in provid- stimulated intensive interest in exploiting new spectrum bands
ing orders-of-magnitude wider bandwidth than current wireless
networks. Economical and energy-efficient analog/digial hybrid for wireless communications. Millimeter wave (mmWave)
precoding and combining transceivers have been often proposed wireless communications, operating in the frequency bands
for mmWave massive multiple-input multiple-output (MIMO) from 30-300 GHz, have been demonstrated as a promising
systems to overcome the severe propagation loss of mmWave candidate to fundamentally solve the spectrum congestion
channels. One major shortcoming of existing solutions lies in the problem [1]-[3].
assumption of infinite or high-resolution phase shifters (PSs) to
realize the analog beamformers. However, low-resolution PSs are However, challenges always come along with opportunities.
typically adopted in practice to reduce the hardware cost and MmWave communications still need to overcome several tech-
power consumption. Motivated by this fact, in this paper, we nical difficulties before real-world deployment. As a negative
investigate the practical design of hybrid precoders and com- result of the ten-fold increase of the carrier frequency, the
biners with low-resolution PSs in mmWave MIMO systems. In propagation loss in mmWave bands is much higher than
particular, we propose an iterative algorithm which successively
designs the low-resolution analog precoder and combiner pair for that of conventional frequency bands (e.g. 2.4 GHz) due to
each data stream, aiming at conditionally maximizing the spectral atmospheric absorption, rain attenuation, and low penetration
efficiency. Then, the digital precoder and combiner are computed [4]. From a positive perspective, the smaller wavelength of
based on the obtained effective baseband channel to further mmWave signals allows a large antenna array to be packed in
enhance the spectral efficiency. In an effort to achieve an even a small physical dimension [5]. With the aid of pre/post-coding
more hardware-efficient large antenna array, we also investigate
the design of hybrid beamformers with one-bit resolution (binary) techniques in massive multiple-input multiple-output (MIMO)
PSs, and present a novel binary analog precoder and combiner systems, the large antenna array can provide sufficient beam-
optimization algorithm with quadratic complexity in the number forming gain to overcome the severe propagation loss of
of antennas. The proposed low-resolution hybrid beamforming mmWave channels. It also enables simultaneous transmission
design is further extended to multiuser MIMO communication of multiple data streams resulting in significant improvements
systems. Simulation results demonstrate the performance ad-
vantages of the proposed algorithms compared to existing low- to spectral efficiency.
resolution hybrid beamforming designs, particularly for the one- For MIMO systems operating in conventional cellular fre-
bit resolution PS scenario. quency bands, the full-digital precoder and combiner are
Index Terms—Millimeter wave (mmWave) communications, completely realized in the digital domain by adjusting both
hybrid precoder, multiple-input multiple-output (MIMO), phase the magnitude and phase of the baseband signals. However,
shifters, one-bit quantization. these conventional full-digital schemes require a large num-
ber of expensive and energy-intensive radio frequency (RF)
I. I NTRODUCTION chains, analog-to-digital converters (ADCs), and digital-to-
analog converters (DACs). Since mmWave communication
The past decade has witnessed the exponential growth of systems operate at much higher carrier frequencies and wider
data traffic along with the rapid proliferation of wireless de- bandwidths, the enormous cost and power consumption of the
Zihuan Wang and Ming Li are with the School of Information and Com- required RF chains and ADCs/DACs make the adoption of
munication Engineering, Dalian University of Technology, Dalian, Liaoning full-digital precoding and combining schemes impractical for
116024, China, (e-mail: wangzihuan@[Link], mli@[Link]). mmWave systems. Recently, economical and energy-efficient
Qian Liu is with the School of Computer Science and Technology,
Dalian University of Technology, Dalian, Liaoning 116024, China (e-mail: analog/digital hybrid precoders and combiners have been
qianliu@[Link]). advocated as a promising approach to tackle this issue. The
A. L. Swindlehurst is with the Center for Pervasive Communications and hybrid precoding approaches adopt a large number of phase
Computing, University of California, Irvine, CA 92697 USA and also with
the Institute for Advanced Study, Technical University of Munich,Munchen shifters (PSs) to implement high-dimensional analog precoders
80333, Germany (e-mail: swindle@[Link]). to compensate for the severe path-loss at mmWave bands,
This paper is supported by the National Natural Science Foundation and a small number of RF chains and DACs to realize low-
of China (Grant No. 61671101 and 61601080) and the Fundamental Re-
search Funds for the Central Universities (Grant No. DUT15RC(3)121 and dimensional digital precoders to provide the necessary flexi-
DUT17JC10). bility to perform advanced multiplexing/multiuser techniques.
 The investigation of hybrid precoder and combiner design ing algorithm is introduced to implement the low-resolution
has attracted extensive attention in recent years because of analog precoder and combiner pair. Then, the digital precoder
its potential energy efficiency for mmWave MIMO communi- and combiner are computed based on the obtained effective
cations. The major challenges in designing hybrid precoders baseband channel to further enhance the spectral efficiency.
are the practical constraints associated with the analog com- Note that the power consumption and cost of the PS are
ponents, such as the requirement that the analog precoding proportional to its resolution. For example, a 4-bit (i.e. 22.5◦)
be implemented with constant modulus PSs. Thus, hybrid resolution PS at mmWave frequencies requires 45-106 mW,
precoder design typically requires the solution of various ma- while a 3-bit (i.e. 45◦ ) resolution PS needs only 15 mW
trix factorization problems with constant modulus constraints. [29]. In an effort to achieve maximum hardware efficiency,
In particular, a popular solution to maximize the spectral we also investigate the design of hybrid beamformers with
efficiency of point-to-point transmission is to minimize the one-bit resolution (binary) PSs. Inspired by the findings in
Euclidean distance between the hybrid precoder and the full- [33], we present a binary analog precoder and combiner
digital precoder [6]-[10]. Hybrid precoder design for partially- optimization algorithm under a rank-1 approximation of the
connected architectures are also studied in [11]-[13]. Due to interference-included equivalent channel. This algorithm has
the special characteristics of mmWave channels, codebook- quadratic complexity in the number of antennas and can
based hybrid precoder designs are commonly proposed [14]- achieve almost the same performance as the optimal exhaustive
[18], in which the columns of the analog precoder are selected search method. Finally, our investigation of low-resolution
from certain candidate vectors, such as array response vectors hybrid precoders and combiners is extended to multiuser
of the channel and discrete Fourier transform (DFT) beam- mmWave MIMO systems. Numerical results in the simulation
formers. Extensions of the hybrid beamformer design to mul- section demonstrate that the proposed algorithms can offer
tiuser mmWave MIMO systems have also been investigated in a performance improvement compared with existing low-
[19]-[27]. resolution hybrid beamforming schemes, especially for the
The aforementioned existing hybrid precoder and combiner one-bit resolution PS scenario.
designs generally assume that infinite or high-resolution PSs Notation: The following notation is used throughout this
are used for implementing the analog beamformers in order paper. Boldface lower-case and upper-case letters indicate col-
to achieve satisfactory performance close to the full-digital umn vectors and matrices, respectively. (·)T and (·)H denote
scheme. However, implementing infinite/high-resolution PSs the transpose and transpose-conjugate operations, respectively.
at mmWave frequencies would significantly increase the en- E{·} represents statistical expectation. Re{·} extracts the real
ergy consumption and complexity of the required hardware part of a complex number; sign(·) denotes the sign operator;
circuits [28], [29]. Obviously, it is impractical to employ angle{·} represents the phase of a complex number. IL
infinite/high-resolution PSs for mmWave systems and real- indicates an L × L identity matrix. C denotes the set of
world analog beamformers will be implemented with low- complex numbers. |A| denotes the determinant of matrix A.
resolution PSs. Consequently, an important research direction |A| denotes the cardinality of set A. |a| and kak are the
is the exploration of signal processing techniques for hybrid magnitude and norm of a scalar a and vector a, respectively.
analog/digital architectures that can mitigate the loss of beam- kAkF denotes the Frobenius norm of matrix A. Finally, we
forming accuracy due to the low-resolution PSs. adopt a Matlab-like matrix indexing notation: A(:, i) denotes
A straightforward approach to obtain the finite-resolution the i-th column of matrix A; A(i, j) denotes the element of
beamformer is to design the infinite-resolution analog beam- the i-th row and the j-th column of matrix A; a(i) denotes
former first, and then directly quantize each phase term to the i-th element of vector a.
a finite set [30]. However, this solution becomes inefficient
when the PSs have very low resolution. An alternative so- II. S YSTEM M ODEL AND P ROBLEM F ORMULATION
lution for hybrid beamforming with finite-resolution PSs is
codebook-based design [14]-[18]. However, for low-resolution A. Point-to-Point mmWave MIMO System Model
PSs, the size of the codebook is very small and the resulting We first consider a point-to-point mmWave MIMO system
performance is not satisfactory. In [31], [32], Sohrabi and using a hybrid precoder and combiner with low-resolution
Yu proposed to iteratively design the low-resolution hybrid PSs, as illustrated in Fig. 1. The transmitter employs Nt
precoder to maximize the spectral efficiency. However, the antennas and NtRF RF chains to simultaneously transmit
performance of this algorithm often suffers when one-bit Ns data streams to the receiver which is equipped with Nr
quantized PSs are applied. antennas and NrRF RF chains. To ensure the efficiency of the
In this paper, we first consider the problem of designing mmWave communication with a limited number of RF chains,
hybrid precoders and combiners with low-resolution PSs for the number of data streams and the number of RF chains are
a point-to-point mmWave MIMO system. The objective of constrained as Ns = NtRF = NrRF .
the proposed algorithm is to minimize the performance loss The transmitted symbols are first processed by a baseband
RF
caused by the low-resolution PSs while maintaining a low digital precoder FBB ∈ CNt ×Ns , then up-converted to the
RF
computational complexity. To achieve this goal, we propose RF domain via Nt RF chains before being precoded with
to successively design the low-resolution analog precoder and an analog precoder FRF of dimension Nt × NtRF . While
combiner pair for each data stream, aiming at conditionally the baseband digital precoder FBB enables both amplitude
maximizing the spectral efficiency. An iterative phase match- and phase modifications, the elements of the analog precoder
Fig. 1. The point-to-point mmWave MIMO system using hybrid precoder and combiner.

FRF , which are implemented by the PSs, have a constant hardware constraint, we aim to jointly design the hybrid
amplitude √1N and quantized phases: FRF (i, j) = √1N ejϑi,j , precoder and combiner for a mmWave MIMO system. When
t t
in which the phase ϑi,j is quantized as ϑi,j ∈ B , { 2πb Gaussian symbols are transmitted over the mmWave MIMO
2B | b =
1, 2, . . . , 2B }, and B is the number of bits to control the phase. channel, the achievable spectral efficiency is given by
We denote the constraint set of the analog precoder as follows:
2πb
FRF (i, j) ∈ F , { √1N ej 2B | b = 1, 2, . . . , 2B }. Obviously, P −1 H H
R = log2 INs + R WBB WRF HFRF FBB ×
Ns n
t
a larger number of bits B leads to finer resolution for the PSs
and potentially better performance, but also results in higher !
hardware complexity and power consumption. FH H H
BB FRF H WRF WBB , (4)
The discrete-time transmitted signal can be written in the
following form √
x = P FRF FBB s (1) where Rn , σn2 WBB H H
WRF WRF WBB is the noise covari-
ance matrix after combining. We aim to jointly design the
where s is the Ns × 1 symbol vector, E{ssH } = N1s INs , P digital beamformers FBB , WBB as well as the low-resolution
represents transmit power and this power constraint is enforced analog beamformers FRF , WRF to maximize the spectral
by normalizing FBB such that kFRF FBB k2F = Ns . efficiency:
We consider a narrow-band slow-fading propagation chan-
n o
nel, which yields the following received signal ⋆
F⋆RF , F⋆BB , WRF ⋆
, WBB = arg max R
y = Hx + n s. t. FRF (i, j) ∈ F , ∀i, j,
√
= P HFRF FBB s + n (2) WRF (i, j) ∈ W, ∀i, j, (5)
where y is the Nr × 1 received signal vector, H is the Nr × kFRF FBB k2F = Ns ,
Nt channel matrix, and n ∼ CN (0, σ 2 INr ) is the complex kWRF WBB k2F = Ns .
Gaussian noise vector corrupting the received signal.
The receiver employs an analog combiner implemented Obviously, the optimization problem (5) is a non-convex NP-
by the PSs and a digital combiner using NrRF RF chains hard problem. In the next section, we attempt to decompose
to process the received signal. The signal after the spatial the original problem into a series of sub-problems and seek
processing has the form a sub-optimal solution with low-complexity and satisfactory
√
b
s = P WBB H
WRFH
HFRF FBB s + WBB H
WRFH
n (3) performance.

where WRF is the Nr × NrRF analog combiner whose

elements have the same constraint as FRF , i.e. WRF (i, j) = III. L OW-R ESOLUTION H YBRID P RECODER AND
√1 ejϕi,j , ϕi,j ∈ B and thus WRF (i, j) ∈ W , C OMBINER D ESIGN
N r
2πb
{ √1N ej 2B | b = 1, 2, . . . , 2B }, WBB is the NrRF ×Ns digital
r To simplify the joint hybrid precoder and combiner design,
baseband combiner and the combiner matrices are normalized
the objective problem is decomposed into two separate op-
such that kWRF WBB k2F = Ns .
timizations. We first focus on the joint design of the analog
precoder FRF and combiner WRF . Then, having the effective
B. Problem Formulation baseband channel associated with the obtained optimal analog
We consider the practical and hardware-efficient scenario precoder and combiner, the digital precoder FBB and com-
in which the PSs have very low-resolution (e.g. B = 1, 2) biner WBB are computed to further maximize the spectral
to reduce the power consumption and complexity. Under this efficiency.
A. Low-Resolution Analog Precoder and Combiner Design Next, we write the analog precoding and combining
We observe that under the assumption of high signal-to- matrices as FRF , [fRF,1 . . . fRF,Ns ] and WRF ,
noise-ratio (SNR), the achievable spectral efficiency in (4) can [wRF,1 . . . wRF,Ns ], respectively, where fRF,l and wRF,l ,
be approximated as l = 1, . . . , Ns , are the analog precoder and combiner pair
for the l-th data stream. Furthermore, we denote FRF,\l
P −1 H H as the precoding matrix excluding the l-th precoder vector
R ≈ log2 R WBB WRF HFRF FBB ×
Ns n fRF,l and WRF,\l as the combining matrix excluding the l-
! th combiner vector wRF,l . Then, the formulation (11) can be
FH H H
. further transformed to (12)-(16), which are presented at the top
BB FRF H WRF WBB (6)
of following page, where α is a very small scalar to assure
invertibility. Thus, the objective in (9) can be reformulated as:
While the per-antenna SNR in mmWave systems is typically
low, the post-combining SNR should be high enough to justify !
 
this approximation. In addition, it has been verified in [32] H H
that for large-scale MIMO systems, the optimal analog beam- log2 WRF HFRF ≈ log2 WRF,\l HFRF,\l
formers are approximately orthogonal, i.e. FH RF FRF ∝ INtRF . !
2
This enables us to assume FBB FH ≈ ζ IN
RF
s when Nt = H
2
BB +log2 wRF,l Ql fRF,l (17)
Ns , where ζ is a normalization factor. Similarly, we have
WBB WBB H
≈ ξ 2 INs and WBBH H
WRF WRF WBB ≈ INs .
2 2 2
Let γ , ζ ξ , then (6) can be further simplified as
! where we define the interference-included channel matrix Ql
P γ2 as
R ≈ log2 WH HFRF FH H
RF H WRF (7)
Ns σ 2 RF
  ! b b bH H b −1 Σ
bVb H.
(a) P γ2 Ql , U(αINs + ΣV FRF,\l WRF,\l U) (18)
H
= Ns log2 + 2 × log2 WRF HFRF (8)
Ns σ 2
According to (17), if FRF,\l and WRF,\l are known, the
where (a) follows since |XY| = |X||Y| when X and Y
problem (9) can be reformulated as finding a corresponding
are both square matrices. Therefore, the analog precoder and
precoder fRF,l and combiner wRF,l pair to conditionally
combiner design with low-resolution PSs can be approximately
maximize the achievable spectral efficiency:
reformulated as:
 
⋆ ⋆ H  ⋆
{FRF , WRF } = arg max log2 WRF HFRF ⋆
fRF ,l , wRF H
,l = arg max wRF,l Ql fRF,l
(9) s. t. fRF ,l (i) ∈ F , i = 1, . . . , Nt , (19)
s. t. FRF (i, j) ∈ F , ∀i, j,
WRF (i, j) ∈ W, ∀i, j. wRF ,l (j) ∈ W, j = 1, . . . , Nr .

Unfortunately, the optimization problem (9) is still NP-hard

RF
and has exponential complexity O(|F |Nt Nt |W|Nr Nr ).
RF This motivates us to propose an iterative algorithm, which
Therefore, we propose to further decompose this difficult starts with appropriate initial RF precoding and combining
optimization problem into a series of sub-problems, in which matrices then successively designs fRF ,l and wRF ,l according
each transmit/receive RF chain pair is considered one by to (19) with an updated Ql as in (18) until the algorithm
one and the analog precoder and combiner for each pair are converges.
successively designed. The complexity of obtaining an optimal solution to (19) for
In particular, we define the singular value decomposition each iteration is now reduced to O(|F |Nt |W|Nr ), which is still
(SVD) of H as too high. To practically solve the problem (19), in what follows
H = UΣVH (10) we present an iterative phase matching algorithm, which
searches the conditionally optimal phase of each element of
where U is an Nr ×Nr unitary matrix, V is an Nt ×Nt unitary the analog precoder fRF ,l and combiner wRF ,l . Specifically,
matrix, and Σ is a rectangular diagonal matrix of singular we first design the analog precoder fRF ,l assuming the analog
values. Due to the sparse nature of the mmWave channel, the combiner wRF,l is fixed. Let ϑl,i be the phase of the i-th
matrix H is typically low rank. In particular, the effective rank element of the analog precoder fRF ,l and let ϕl,j be the
of the channel serves as an upper bound for the number of data phase of the j-th element of the analog combiner wRF,l .
streams Ns that the channel can support. Thus, we assume that If we temporarily remove the discrete phase constraint, the
the channel H can be well approximated by retaining only the optimal continuous phase ϑ̃l,i of the i-th element of the analog
Ns strongest components H ≈ U bΣbVb H , where U b , U(:, 1 :
precoder fRF,l is given by the following proposition, whose
b b
Ns ), Σ , Σ(1 : Ns , 1 : Ns ), and V , V(:, 1 : Ns ). Then, proof is provided in Appendix A.
the objective in (9) can be converted to Proposition 1: Given the phases ϕl,j of the analog combiner
   
H H b b bH wRF,l and the phases ϑl,u , u 6= i, of the analog precoder fRF,l ,
log2 WRF HFRF ≈ log2 WRF UΣV FRF .(11)
the optimal continuous phase ϑ̃l,i of the i-th element of analog
    
H b b bH bVb H FRF WRF
H b
log2 WRF UΣV FRF Σ
= log2 U (12)
! !
  H
= log 2 bV
Σ b H FRF,\l b
fRF,l WRF,\l wRF,l U = log2 bV
Σ b H FRF,\l WRF,\l
H b +Σ
U bVb H fRF,l wRF,l
H b
U (13)
!
 h  −1 i
≈ log 2 bV
Σ b H FRF,\lWRF,\l
H b
U bV
INs + αINs + Σ b H FRF,\l WRF,\l
H b
U bV
Σ b H fRF,l wRF,l
H b
U (14)
! !
h  −1 i
= log 2 bV
Σ b H FRF,\l WRF,\l
H b
U + log 2 bV
INs + αINs + Σ b H FRF,\lWRF,\l
H b
U bV
Σ b H fRF,l wRF,l
H b
U (15)
! !
h  −1 i
H bΣ b H FRF,\l
bV H b αINs + Σ
bVb H FRF,\l WRF,\l
H b bVb H fRF,l
= log 2 WRF,\l U + log 2 1 + wRF,l U U Σ (16)

precoder fRF,l is Algorithm 1 Iterative Phase Matching Algorithm for Low-

  Resolution Analog Precoder and Combiner Design
XNr Nt
X  Input: F , W, H.
ϑ̃l,i = angle ejϕl,j ejϑl,u Ql (j, u) Output: F⋆RF and WRF ⋆
.
 
j=1 u6=i ⋆ ⋆
  1: Initialize FRF = 0, WRF = 0.
X Nr  2: for l = 1 : Ns do
−angle ejϕl,j Ql (j, i) . (20) 3: Obtain FRF,\l from F⋆RF and WRF,\l from WRF
⋆
.
 
j=1 4: b b b H H b −1 b b H
Update Ql =U(αINs+ΣV FRF,\l WRF,\l U) ΣV .

Then, after finding the optimal continuous phase ϑ̃l,i by 5: while no convergence of ϑl,i and ϕl,j do
(20), we reconsider the discrete phase constraint and find the 6: for i = 1 : Nt do
optimal low-resolution phase ϑl,i by quantization: 7: Obtain quantized phase ϑl,i by (20) and (21).
8: end for
ϑl,i = arg min ϑ̃l,i − ϑ̂l,i . (21) 9: for j = 1 : Nr do
ϑ̂l,i ∈B
10: Obtain quantized phase ϕl,j by (22) and (23).
Similarly, if the analog precoder fRF,l is determined, the 11: end for
optimal continuous phase ϕ̃l,j of the j-th element of wRF,l is 12: end while
  13: ⋆
Construct fRF,l ⋆
by ϑl,i and wRF,l by ϕl,j .
X Nt XNt 
14: end for
ϕ̃l,j = angle ejϑl,i ejϕl,u Ql (u, i)
  15: ⋆
Construct F⋆RF by fRF,l ⋆
and WRF ⋆
by wRF,l .
i=1 u6=j ⋆ ⋆
(N ) 16: Goto Step 2 until convergence of FRF and WRF .
X t
jϕl,i
−angle e Ql (j, i) , (22)
i=1

and the optimal low-resolution phase ϕl,j is obtained by B. Digital Precoder and Combiner Design

ϕl,j = arg min ϕ̃l,j − ϕ̂l,j . (23) After all analog precoder-combiner pairs have been deter-
ϕ̂l,j ∈B e as
mined, we can obtain the effective baseband channel H
Motivated by (20)-(23), the iterative procedure to design the e , (W⋆ )H HF⋆ ,
H RF RF (24)
precoder fRF ,l and combiner wRF ,l as in (19) is straightfor-
ward. With appropriate initial ϑl,i , ϕl,j , we design the precoder where F⋆RF , ⋆
[fRF,1 ⋆
, . . . , fRF,N ⋆
] and WRF ,
s
fRF ,l by finding the conditionally optimal phases ϑl,i as in ⋆ ⋆
[wRF,1 , . . . , wRF,Ns ]. For the baseband precoder and
(20) and (21). Then, with the obtained ϑl,i , i = 1, . . . , Nt , combiner design, we define the SVD of the effective
we design the combiner wRF ,l by finding the conditionally baseband channel H e as
optimal phases ϕl,j as in (22) and (23). We alternate the
designs of fRF ,l and wRF ,l iteratively until the obtained e =U
H eΣeV
eH (25)
phase of each element of fRF ,l and wRF ,l does not change
and the convergence is achieved. Note that since in each where Ue and V e are Ns × Ns unitary matrices, Σ e is an
precoder and combiner design step, the objective function of Ns × Ns diagonal matrix of singular values. Then, to further
(19) is monotonically non-decreasing, and thus our proposed enhance the spectral efficiency, an SVD-based baseband digital
algorithm is guaranteed to converge to at least a locally optimal precoder and combiner are employed:
solution.
e
F⋆BB = V, (26)
We summarize the proposed joint low-resolution analog
precoder and combiner design in Algorithm 1. W ⋆ e
= U. (27)
BB
Finally, the baseband precoder and combiner are normalized Now, the joint optimization problem (33) can be decoupled
√
N s F⋆
into individually designing the analog precoder fRF ,l and
F⋆BB = kF⋆ ⋆
BB
, (28) combiner wRF ,l :
RF FBB kF
√ ⋆
⋆ Ns WBB
WBB = ⋆ W⋆
kWRF . (29) ⋆ H
BB kF fRF ,l = arg max fRF,l gl,1 , (34)
fRF ,l ∈ √1 {±1}Nt
Nt

IV. O NE -B IT R ESOLUTION A NALOG P RECODER AND ⋆ H

C OMBINER D ESIGN wRF ,l = arg max
1
wRF,l pl,1 . (35)
wRF ,l ∈ √N {±1}Nr
r
In the previous section, we proposed a novel hybrid beam-
former design for maximizing the spectral efficiency of a These two optimization problems (34) and (35) require only
mmWave MIMO system, in which the analog precoder and the singular vectors pl,1 and gl,1 associated with the largest
combiner are implemented with low-resolution PSs. In order singular value, which can be quickly obtained by the power
to achieve maximum hardware efficiency, in this section we iteration algorithm [34] instead of the complete SVD calcula-
focus on the design of analog precoders and combiners using tion. However, solving (34) and (35) by exhaustive search still
“one-bit” resolution (binary) PSs, which can maximally reduce has exponential complexity in the number of antennas. In order
the power consumption and simplify the hardware complexity. to further reduce the complexity without a significant loss of
Although the iterative phase matching algorithm proposed in performance, we propose to construct a smaller dimension can-
the previous section can also be applied, a simpler approach didate beamformer set, from which the optimal beamformer
is possible in the one-bit case. Therefore, in this section, can be found with linear complexity. In the following, we
we present an efficient one-bit resolution analog beamformer present this algorithm for the precoder design (34) as an
design, which can achieve good performance with much lower example, while the combiner design (35) follows the same
complexity. procedure.
We follow the procedure of the hybrid beamforming de- We introduce an auxiliary variable φ ∈ [−π, π) and we
sign proposed in the previous section, but only modify the reformulate the optimization problem (34) as:
n o n o
optimization problem (19), which attempts to determine the
φ⋆ , fRF,l
⋆
= arg max H
Re fRF,l gl,1 e−jφ (36)
l-th analog precoder and combiner pair. Particularly, we refor- φ∈[−π,π)
mulate this analog beamformer design problem (19) with the fRF ,l ∈ √1 {±1}Nt
Nt
constraint of one-bit resolution PSs as P
Nt
 ⋆ ⋆ H
= arg max fRF,l (i)|gl,1 (i)| cos(φ − ψi ) (37)
fRF ,l , wRF ,l = arg max wRF,l Ql fRF,l . φ∈[−π,π) i=1
fRF ,l ∈ √1 {±1}Nt fRF ,l ∈ √1 {±1}Nt
Nt Nt
1
wRF ,l ∈ √N {±1}Nr
r
(30) where ψi denotes the phase of gl,1 (i). Obviously, given
The optimization problem (30) can be solved through ex- any φ ∈ [−π, π), the corresponding binary precoder that
haustive search with exponential complexity O(2Nt Nr ), which maximizes (37) is
would not be possible with large antenna arrays. Therefore, 1
in the following we attempt to develop an efficient one-bit fRF,l (i) = √ sign (cos (φ − ψi )) , i = 1, . . . , Nt . (38)
Nt
resolution beamformer design with polynomial complexity in
the number of antennas. With the conditionally optimal fRF,l for any given φ shown
We first define the SVD of Ql as in (38), we will now show that we can always construct a set
of Nt candidate binary precoders Fl , {fl,1 , . . . , fl,Nt } and
Ns
X ⋆
Ql = H
λl,i pl,i gl,i , (31)
guarantee fRF ,l ∈ Fl . Then, the maximization in (34) can be
carried out over a set of only Nt candidates without loss of
i=1
performance.
where pl,i and gl,i are the i-th left and right singular vectors We first define the angles ψbi , i = 1, . . . , Nt , as
of Ql , respectively, and λl,i is the i-th largest singular value,   
λl,1 ≥ λl,2 ≥ . . . ≥ λl,Ns . Then, the objective in (30) can be  π 3π
 ψi − π, if ψi ∈ , ,
rewritten as ψbi , 2 2 (39)
 h π π 
Ns
X ψ , if ψi ∈ − , ,
i
H
|wRF,l Ql fRF,l | = H
λl,i wRF,l H
pl,i gl,i fRF,l . (32) 2 2
i=1
so that ψbi ∈ [− π2 , π2 ). Then, we map the angles ψbi to ψei , i =
If we utilize a rank-1 approximation by keeping only the 1, . . . , Nt , which are rearranged in ascending order, i.e. ψe1 ≤
H
strongest term, i.e. Ql ≈ λl,1 pl,1 gl,1 , the optimization func- ψe2 ≤ . . . ≤ ψeNt . Because of the periodicity of the cosine
tion in (30) can be approximated by function, the maximization problem (37) with respect to φ can
 ⋆ ⋆ H H be carried out over any interval of length π. If we construct Nt
fRF ,l , wRF ,l = arg max wRF,l pl,1 gl,1 fRF,l .
fRF ,l ∈ √1 {±1}Nt non-overlapping sub-intervals [ψe1 − π2 , ψe2 − π2 ), [ψe2 − π2 , ψe3 −
Nt
π e π e π ⋆
{±1}Nr 2 ), . . . , [ψNt − 2 , ψ1 + 2 ), then the optimal φ must be located
1
wRF ,l ∈ √N
r
(33) in one of Nt sub-intervals since the full interval is [ψe1 −
π e Algorithm 2 One-Bit Resolution Analog Beamformer Design
2 , ψ1 + π2 ) of length π. Therefore, the optimization problem
(37) can be solved by examining each sub-interval separately. Input: Ql .
⋆ ⋆
Assuming the optimal φ⋆ is in the k-th sub-interval, the Output: fRF ,l and wRF ,l .
corresponding optimal  binary
 precoder can be obtained by (38) 1: Calculate pl,1 and gl,1 by an SVD of Ql .
fl,k (i) = √1N sign cos φ⋆ − ψei , i = 1, . . . , Nt , and
as e 2: Define the angles ψ bi , i = 1, . . . , Nt , by (39).
t b e
3: Map ψi to ψi , i = 1, . . . , Nt , in an ascending order.
has the form
4: for k = 1 : Nt do
e 1
. . 1} −1 . . . − 1]T .
fl,k = √ [1| .{z
| {z }
(40) 5: Obtain e fl,k by (40).
Nt
k Nt −k 6: Obtain b fl,k from e fl,k based on inverse mapping from ψei
b
to ψi , i = 1, . . . , Nt .
After that, given the inverse sorting that maps ψei to ψbi , we 7: Obtain fl,k from b fl,k by (41).
rearrange the corresponding elements of e fl,k and obtain b fl,k . 8: end for
Then, based on the relationship between ψi and ψbi defined in 9: Construct Fl = {fl,1 , . . . , fl,Nt }.
(39), we can achieve the conditionally optimal precoder fl,k 10: Construct Wl by a similar procedure as Steps 2-9.
by 11: Find the optimal fRF ⋆ ⋆
  ,l and wRF ,l by (44).

 b π 3π
 − fl,k (i), if ψi ∈ , , i = 1, . . . , Nt ,
fl,k (i) , 2 2
 h π π 
 bfl,k (i), if ψi ∈ − , , i = 1, . . . , Nt , A. System Model and Problem Formulation
2 2
(41) We consider a multiuser mmWave MIMO uplink system as
for the case that φ⋆ is in the k-th sub-interval. presented in Fig. 2, where a base-station (BS) is equipped with
Since the optimal φ⋆ must be located in one of Nt sub- Nr antennas and NRF RF chains and simultaneously serves
intervals, we can obtain Nt conditionally optimal precoders K mobile users. Due to power consumption and hardware
by examining all Nt sub-intervals and construct a candidate limitations, each mobile user has Nt antennas and a single RF
precoder set Fl as chain to transmit only one data stream to the BS. We further
assume the number of RF chains at the BS is equal to the
Fl , {fl,1 , . . . , fl,Nt }, (42)
number of users, i.e. NRF = K.
⋆
which must contain the optimal precoder fRF,l . Therefore, Let fRF,k be the analog precoder of the k-th user, where
without loss of performance, the problem in (34) can be each element of fRF,k has a constant magnitude √1N and low-
t
transformed to an equivalent maximization task over only the resolution discrete phases, i.e. fRF,k (i) ∈ F , ∀i = 1, . . . , Nt .
set Fl The transmitted signal of the k-th user after precoding can be
⋆ H
fRF ,l = arg max fRF,l gl,1 , (43) formulated as
fRF ,l ∈Fl p
xk = Pk fRF,k sk (45)
which has linear complexity O(Nt ). Similarly, we can also
2
construct a candidate analog combiner set Wl and obtain where sk is the symbol of the k-th user, E{|sk | } = 1, and
⋆
wRF ,l by the same procedure. Pk is the k-th user’s transmit power.
The rank-1 solution returned by (43) is based on the rank-1 Let Hk ∈ CNr ×Nt , k = 1, . . . , K, denote the uplink
approximation of the interference-included equivalent channel channel from the k-th user to the BS. The received signal
Ql . The approximation of Ql may cause a performance degra- at the BS can be written as
dation when we revisit the original problem (30). Therefore,
K p
X
in order to enhance the performance, we propose to jointly
r= Pk Hk fRF,k sk + n (46)
select the precoder and combiner over candidate sets Fl and
k=1
Wl as
n o where n ∼ CN (0, σ 2 INr ) is complex Gaussian noise. The
⋆ ⋆ H
fRF ,l , w RF ,l = arg max wRF,l Ql fRF,l (44) BS first applies an Nr × K analog combining matrix WRF ,
fRF ,l ∈Fl
wRF ,l ∈Wl [wRF ,1 . . . wRF ,K ] to process the received signal, in which the
which may return the rank-1 or a better solution with quadratic analog combiner wRF,k corresponding to the k-th user is also
complexity O(Nt Nr ). This low-complexity analog beam- implemented by low-resolution PSs, i.e. wRF ,k (j) ∈ W, j =
former design with one-bit resolution PSs is summarized in 1, . . . , Nr . Then, a baseband digital combiner wBB,k ∈ CK×1
Algorithm 2. is employed to retrieve the information of the k-th user. Let
wk , WRF wBB,k denote the hybrid combiner corresponding
to the k-th user. After the combining process at the BS, the
V. H YBRID P RECODER AND C OMBINER D ESIGN FOR
estimated symbol of the k-th user can be expressed as
M ULTIUSER MM WAVE MIMO S YSTEMS
In this section, we consider a mmWave multiuser MIMO p K p
X
uplink system and extend the low-resolution hybrid precoder ŝk = Pk wkH Hk fRF,k sk + wkH Pi Hi fRF,i si + wkH n.
and combiner designs proposed in the previous sections to the i=1
i6=k
multiuser system. (47)
Fig. 2. The multiuser mmWave MIMO system using hybrid precoder and combiner.

Given the received signal at the BS in (47), the signal-to- This analog precoder and combiner design problem can be
interference-plus-noise ratio (SINR) of the k-th user can be efficiently solved by the algorithm presented in Sec. III-A
expressed as when low-resolution PSs are utilized, or the algorithm pro-
√ posed in Sec. IV if only one-bit resolution PSs are available.
| Pk wkH Hk fRF,k |2
γk = K Then, the analog precoders fRF,k and combiners wRF ,k ,
P √ (48)
| Pi wkH Hi fRF,i |2 +σ 2 k wk k2 k = 2, 3, . . . , K, for the remaining K − 1 users are succes-
i=1,i6=k sively designed by an iterative procedure. In each iteration, we
and the achievable sum-rate of the multiuser uplink system is attempt to find the analog beamformer pair that suppresses the
K interference from the users whose analog beamformers have
X
Ru = log(1 + γk ). (49) already been determined. To achieve this goal, the channel of
k=1 the user whose combiner is to be calculated is projected onto
We aim to jointly design the analog precoders and combiners the space orthogonal to the collection of previously designed
implemented by low-resolution PSs as well as the digital analog combiners. This approach leads to orthogonal analog
combiners to maximize the sum-rate of the uplink multiuser combiners that suppress the inter-user interference.
system: Specifically, to design the k-th user’s analog beamformer
pair, we first extract the orthonormal components di of the pre-
n o K
X ⋆
⋆ ⋆ ⋆ K viously determined analog combiners wRF ,i , i = 1, . . . , k − 1
wRF ,k , wBB,k , fRF,k k=1
= arg max log (1 + γk )
by the Gram-Schmidt procedure:
k=1
s. t. fRF,k (i) ∈ F , ∀k, i, i−1
X
wRF ,k (j) ∈ W, ∀k, j. ⋆
qi = wRF ,i − dH ⋆
j wRF ,i dj , (52)
(50) j=1

B. Low-Resolution Hybrid Precoder and Combiner Design

Obviously, the optimization problem (50) cannot be di- di = qi /kqi k. (53)
rectly solved. Thus, we adopt an approach similar to [26]
⋆ ⋆
and propose to successively design the low-resolution analog Note that d1 = wRF ,1 and wRF ,1 is the analog combiner
beamformer pair for each user, aiming at enhancing the calculated for the first user. Then, the combiner components
channel gain as well as suppressing the inter-user interference. are removed from the k-th user’s channel to obtain the
Then, the baseband combiner at the BS is calculated to further modified channel H b k as
mitigate the interference and maximize the sum-rate. !
In particular, for the first user, the analog precoder and k−1
X
combiner pair is designed to maximize the corresponding bk =
H INr − di dH Hk . (54)
i
channel gain, which can be formulated as follows: i=1
 ⋆ ⋆ H
wRF ,1 , fRF ,1 = arg max wRF ,1 H1 fRF ,1 b k , the analog beam-
Finally, based on the modified channel H
s. t. fRF ,1 (i) ∈ F , i = 1, . . . , Nt , (51) former pair for the k-th user is found by solving the following
wRF ,1 (j) ∈ W, j = 1, . . . , Nr . optimization using the algorithms proposed in the previous
sections:
120
 ⋆ Full−digital
⋆
wRF ,k , fRF H b
,k = arg max wRF ,k Hk fRF,k 110 PE−AltMin [6] (B=∞)

s. t. fRF,k (i) ∈ F , i = 1, . . . , Nt , (55) 100

CDM [30] (B=2)
CDM [30] (B=1)

Spectral Efficiency (b/s/Hz)

wRF ,k (j) ∈ W, j = 1, . . . , Nr . 90 HBF [32] (B=2)
HBF [32] (B=1)
After finding the analog beamformers for all users, the 80 Proposed (B=2)
Proposed (B=1)
effective baseband
√ channel for each user can be obtained 70
H
as hek , Pk (WRF ⋆
) Hk fRF⋆
,k . Then, a minimum mean
60
square error (MMSE) baseband digital combiner for the k-th 102
user is employed to further suppress the interference: 50
100
h i−1 40 98
⋆
wBB,k = He (He )H + σ 2 (WRF ⋆
)H WRF⋆
hek , (56)
30 96
34 34.5 35 35.5
where He , [he1 , . . . , heK ]. The proposed low-resolution 20
0 5 10 15 20 25 30 35 40
hybrid precoder and combiner design for multiuser mmWave SNR (dB)
systems is summarized in Algorithm 3.
Fig. 3. Spectral efficiency versus SNR (Nt = 64, Nr = 64, NtRF =
Algorithm 3 Low-Resolution Hybrid Precoder and Combiner NrRF = 6, Ns = 6).
Design for Multiuser mmWave Systems
Input: F , W, Hk , k = 1, . . . , K. 65
⋆ ⋆ ⋆
Output: fRF,k , wRF ,k , wBB,k , k = 1, . . . , K.
⋆ ⋆ 60
1: Obtain wRF ,1 and fRF ,1 for user-1 by solving
 ⋆
Spectral Efficiency (b/s/Hz)
⋆ H 55
wRF ,1 , fRF,1 = arg max |wRF ,1 H1 fRF,1 |.
wRF ,1 (i)∈W
fRF,1 (j)∈F 50

⋆
2: d1 = wRF ,1 . 45
3: for k = 2: K do  Full−digital
b P
k−1
H
40 PE−AltMin [6] (B=∞)
4: Hk = INr − di di H k . CDM [30] (B=2)
i=1 CDM [30] (B=1)
⋆ ⋆ 35
5: Obtain wRF ,k and fRF ,k for user-k by solving HBF [32] (B=2)
 ⋆ HBF [32] (B=1)
⋆
max |wRF H b 30
wRF ,k , fRF ,k = arg ,k Hk fRF,k |. Proposed (B=2)
wRF ,k (i)∈W Proposed (B=1)
fRF,k (j)∈F 25
20 40 60 80 100 120 140
P
k−1 Number of Antenna
⋆
6: qk = wRF ,k − dH ⋆
i wRF ,k di ;
i=1
7: dk = qk /kqk k. Fig. 4. Spectral efficiency versus number of antenna (NtRF = NrRF = 4,
Ns = 4, SNR = 20dB).
8: end for
⋆
9: Obtain digital combiners wBB,k , k = 1, . . . , K, by
h i−1 where αi ∼ CN (0, L1 ) are the independent and identically
H
⋆
wBB,k = He (He )H + σ 2 (WRF ⋆
) WRF ⋆
hek . distributed complex gains of the i-th propagation path (ray)
θit and θir ∈ [− π2 , π2 ] are the angles of departure (AoD) and
the angles of arrival (AoA), respectively. Finally, the array
response vectors at (θt ) and ar (θr ) depend on the antenna
VI. S IMULATION R ESULTS array geometry. We assume that the commonly used uniform
linear arrays (ULAs) are employed, and the transmit antenna
In this section, we provide simulation results for the pro- array response vector at (θt ) and the receive antenna array
posed joint hybrid precoder and combiner designs with low- response vector ar (θr ) can be written as
resolution PSs for point-to-point mmWave systems as well as
1 2π t 2π t
multiuser mmWave systems. MmWave channels are expected at (θt ) = √ [1, ej λ d sin(θ ) , . . . , ej(Nt −1) λ d sin(θ ) ]T ,
to be sparse and have a limited number of propagation paths. Nt
(58)
In the simulations, we adopt a geometric channel model with r 1 j 2π d sin(θ r
) j(N −1) 2π
d sin(θ r
) T
L paths [32]. In particular, the discrete-time narrow-band ar (θ ) = √ [1, e λ ,...,e r λ ] ,
Nr
mmWave channel H is formulated as (59)
r L respectively, where λ is the signal wavelength, and d is the dis-
Nt Nr X tance between antenna elements. In the following simulations,
H= αi ar (θir )at (θit )H (57)
L i=1 we consider an environment with L = 6 scatterers between the
 60 20
Exhaustive search
58 18 Proposed (B=1)
56
16

Spectral Efficiency (b/s/Hz)

Spectral Efficiency (b/s/Hz)

54
14
52

50 12

48 10
Full−digital
46
CDM [30] (B=2) 8
44 CDM [30] (B=1)
HBF [32] (B=2) 6
42 HBF [32] (B=1)
40 Proposed (B=2) 4
Proposed (B=1)
38 2
1 2 3 4 5 6 7 8 9 10 −10 0 10 20 30 40
Number of Iterations SNR (dB)

Fig. 5. Spectral efficiency versus number of iteration (Nt = 64, Nr = 64, Fig. 7. Spectral efficiency versus SNR (Nt = 8, Nr = 8, Ns = 1, B = 1).
NtRF = NrRF = 4, Ns = 4, SNR = 20dB).

52 algorithm in [32] achieves the best performance with low-

resolution PSs in the existing literature. The performance
50 of a fully digital approach using SVD-based beamforming
and the hybrid beamforming scheme with infinite-resolution
Spectral Efficiency (b/s/Hz)

48 51.4 (B = ∞) PSs using the phase extraction (PE-AltMin) algo-

rithm in [6] are also included as performance benchmarks.
51.2
46 Fig. 3 illustrates that the proposed algorithm outperforms the
51 competitors, particularly for the case of 1-bit resolution PSs.
44 Moreover, it can be observed that the proposed algorithm
4.95 5 5.05
with B = 2 achieves performance close to optimal full-digital
42 Full−digital
beamforming and hybrid beamforming with infinite-resolution
PE−AltMin [6] (B=∞)
CDM [30] PSs. For additional simulation validation, Fig. 4 illustrates the
40 HBF [32] spectral efficiency versus the number of antennas and similar
Proposed conclusions can be drawn.
38
1 2 3 4 5 6 In order to illustrate the convergence of the proposed
Resolution (bits) algorithm, we show the spectral efficiency versus the number
of iterations in Fig. 5, which also includes other algorithms
Fig. 6. Spectral efficiency versus resolution of PSs (Nt = 64, Nr = 64, for comparison. It is observed that our proposed algorithms
NtRF = NrRF = 4, Ns = 4, SNR = 20dB).
converge faster than the other two iterative schemes, which is
a highly favorable property. In Fig. 6, we show the spectral
efficiency as a function of B to illustrate the impact of the
transmitter and the receiver. The antenna spacing is d = λ2 .
resolution of PSs on the spectral efficiency. As expected,
increasing the PS resolution will improve the system perfor-
A. Simulation Results of a Point-to-Point mmWave System mance, but using only B = 3 bits is sufficient to closely
We first consider a point-to-point mmWave communication approach the performance of the ideal unquantized case.
system, in which the transmitter and receiver are both equipped Beyond B = 3, the additional cost and complexity associated
with 64-antenna ULAs. The number of RF chains at the with using higher-resolution PSs is not warranted given the
transmitter and receiver are NtRF = NrRF = 6, so the number very marginal increase in spectral efficiency. Moreover, our
of data streams is also assumed to be Ns = 6. proposed algorithms outperform the other two low-resolution
Fig. 3 shows the average spectral efficiency versus SNR over beamforming methods for all PS resolutions.
106 channel realizations. We evaluate the spectral efficiency of To examine the impact of the approximations used in
the algorithm proposed in Sec. III for the case of 2-bit (B = 2) deriving the proposed one-bit resolution hybrid beamformer
resolution PSs and the algorithm proposed in Sec. IV for the scheme, in Fig. 7 we compare it with the optimal exhaustive
case of 1-bit (B = 1) resolution PSs. For comparison purposes, search approach. The number of antennas at both transmitter
we also plot the spectral efficiency of two state-of-the-art low- and receiver is chosen to be 8 and the number of data streams
resolution hybrid beamformer designs: the coordinate descent is Ns = 1. A relatively simple case is examined here due to
method (CDM) algorithm in [30] and the hybrid beamforming the exponential complexity of the exhaustive search method.
(HBF) algorithm in [32]. To the best of our knowledge, the We see from Fig. 7 that the spectral efficiency achieved by
 (I-HB) in [26]. All three algorithms are codebook-based ap-
60
Full−digital proaches and the size of the beamsteering codebook is set at 32
TS−HB [24] (i.e. B = 5 quantization bits) for fairness of the comparison.
50 MMSE−HB [25]
I−HB [26] It can be observed from Fig. 8 that our proposed low-
Proposed (B=2) resolution hybrid beamforming design outperforms the other
Proposed (B=1)
Sum−rate (b/s/Hz)

40 three algorithms using only 2-bit resolution PSs. Moreover,

the performance with 1-bit resolution PSs is also comparable.
30 Fig. 9 further shows the sum-rate versus the number of users
K. From Fig. 9, we see that our proposed algorithm with 2-bit
20
resolution PSs always outperforms the other codebook-based
algorithms. Furthermore, even with 1-bit resolution PSs, the
proposed algorithm can still achieve competitive performance
10
compared with the TS-HB and MMSE-HB approaches when
K > 5.
0
−20 −15 −10 −5 0 5 10 15 20
SNR (dB)
VII. C ONCLUSIONS
Fig. 8. Spectral efficiency versus SNR (Nt = 64, Nr = 16, NtRF = 4,
K = 4). This paper considered the problem of hybrid precoder and
combiner design for mmWave MIMO systems with low-
100 resolution quantized PSs. We proposed an efficient iterative
Full−digital algorithm which successively designs the low-resolution ana-
90 TS−HB [24]
MMSE−HB [25]
log precoder and combiner pair for each data stream. Then,
80 I−HB [26] the digital precoder and combiner were computed based on
Proposed (B=2) the obtained effective baseband channel to further enhance the
Proposed (B=1)
Sum−rate (b/s/Hz)

70 spectral efficiency. The design of low-resolution hybrid beam-

60 formers for multiuser MIMO communication systems was also
investigated. Simulation results verified the effectiveness of the
50
proposed algorithms, particularly for scenarios in which one-
40 bit resolution phase shifters are used.

30

20 A PPENDIX A
P ROOF OF P ROPOSITION 1
10
1 2 3 4 5 6 7 8
Number of users The optimization problem (19) can be equivalently formu-
lated as
Fig. 9. Spectral efficiency versus K (Nt = 64, Nr = 16, NtRF = K,
SNR = 20dB). Nr
X Nt
1 l X l
max √ ejϕj ejϑi Ql (j, i) . (60)
Nt Nr j=1 i=1
the proposed algorithm is the same as that of the optimal
√
exhaustive search method, suggesting that the proposed hy- By discarding the constant coefficient 1/ Nt Nr , (60) can be
brid beamforming algorithm with one-bit resolution PSs can further transformed as
provide optimal or near-optimal performance.  
XNr
l
 l
Nt
X l

max ejϕj ejϑi Ql (j, i) + ejϑu Ql (j, u) . (61)
B. Simulation Results of a Multiuser mmWave System  
j=1 u6=i
Next, we evaluate the performance of the proposed low-
l
resolution beamformer algorithm in a multiuser uplink system. Since the term ejϑi does not involve the summation index j,
We assume there are K = 4 users, each of which is equipped it can be put outside the first summation, resulting in
with Nt = 16 antennas and only one RF chain to transmit
a single data stream. The BS has Nr = 64 antennas and Nr
X Nr
X Nt
X
NrRF = 4 RF chains. Fig. 8 illustrates the sum-rate versus jϑli jϕlj jϕlj l
max e e Ql (j, i) + e ejϑu Ql (j, u) .
SNR for various hybrid beamformer designs. In particular, we j=1 j=1 u6=i
include three state-of-the-art multiuser hybrid beamforming (62)
approaches for comparison: i) two-stage hybrid beamform- Obviously, the optimal value of ϑli makes the phases of the
ing (TS-HB) in [24], ii) MMSE-based hybrid beamforming first and second term equal to obtain the largest amplitude,
(MMSE-HB) in [25], and iii) iterative hybrid beamforming and (20) is proved. 
 R EFERENCES [22] A. Li and C. Masouros, “Hybrid precoding and combining design for
millimeter-wave multi-user MIMO based on SVD,” IEEE Int. Conf. on
Commun. (ICC), Paris, France, May 2017, pp. 1-6.
[1] Z. Pi and F. Khan, “An introduction to millimeter-wave mobile broad-
[23] M. Kim and Y. H. Lee, “MSE-based hybrid RF/baseband processing
band systems,” IEEE Commun. Mag., vol. 49, no. 6, pp. 101-107, June
for millimeter-wave communication systems in MIMO interference
2011.
channels,” IEEE Trans. Veh. Technol., vol. 64, no. 6, pp. 2714-2720,
[2] T. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. N.
June 2015.
Wong, J. K. Schulz, M. Samimi, and F. Gutierrez “Millimeter wave [24] A. Alkhateeb, G. Leus, and R. W. Heath Jr., “Limited feedback hybrid
mobile communications for 5G cellular: It will work!” IEEE Access, precoding for multi-user millimeter wave systems,” IEEE Trans. Wireless
vol. 1, pp. 335-349, 2013.
Commun., vol. 14, no. 11, pp. 6481-6494, Nov. 2015.
[3] A. Lee Swindlehurst, E. Ayanoglu, P. Heydari, and F. Capolino, [25] D. H. N. Nguyen, L. B. Le, and T. Le-Ngoc, “Hybrid MMSE precoding
“Millimeter-wave massive MIMO: The next wireless revolution?” IEEE for mmWave multiuser MIMO systems,” IEEE Int. Conf. on Commun.
Commun. Mag., vol. 52, no. 9, pp. 56-62, Sept. 2014. (ICC), Kuala Lumpur, Malaysia, May 2016.
[4] T. S. Rappapport, G. R. MacCartney, M. K. Samimi, and S. Sun, [26] Z. Wang, M. Li, X. Tian, and Q. Liu, “Iterative hybrid precoder
“Wideband millimeter-wave propagation measurements and channel and combiner design for mmWave multiuser MIMO systems,” IEEE
models for future wireless communication system design”, IEEE Trans. Commun. Lett., vol. 21, no. 7, pp. 1581-1584, July 2017.
Commun., vol. 63, no. 9, pp. 3029-3056, Sept. 2015. [27] T. E. Bogale, L. B. Le, A. Haghighat, and L. Vandendorpe, “On the
[5] R. W. Heath Jr., N. González-Prelcic, S. Rangan, W. Roh, and A. M. number of RF chains and phase shifters, and scheduling design with
Sayeed, “An overview of signal processing techniques for millimeter hybrid analog-digital beamforming,” IEEE Trans. Wireless Commun.,
wave MIMO systems,” IEEE J. Sel. Topics Signal Process., vol. 10, no. vol. 15, no. 5, pp. 3311-3326, May 2016.
3, pp. 436-453, April 2016. [28] A. S. Y. Poon and M. Taghivand, “Supporting and enabling circuits for
[6] X. Yu, J.-C. Shen, J. Zhang, and K. B. Letaief, “Alternating minimization antenna arrays in wireless communications,” Proc. IEEE, vol. 100, no.
algorithms for hybrid precoding in millimeter wave MIMO systems,” 7, pp. 2207-2218, July 2012.
IEEE J. Sel. Topics Signal Process., vol. 10, no. 3, pp. 485-500, April [29] R. Méndez-Rial, C. Rusu, N. González-Prelcic, and A. Alkhateeb,
2016. “Hybrid MIMO architectures for millimeter wave communications:
[7] C. Rusu, R. Méndez-Rial, N. González-Prelcic, and R. W. Heath Phase shifters or switches?”, IEEE Access, vol. 4, pp. 247-267, March
Jr., “Low complexity hybrid precoding strategies for millimeter wave 2016.
communication systems,” IEEE Trans. Wireless Commun., vol. 15, no. [30] J.-C. Chen, “Hyrbrid beamforming with discrete phase shifters for
12, pp. 8380-8393, Sept. 2016. millimeter-wave massive MIMO systems,” IEEE Trans. Veh. Technol.,
[8] R. López-Valcarce, N. González-Prelcic, C. Rusu, and R. W. Heath Jr., to Appear.
“Hybrid precoders and combiners for mmWave MIMO systems with [31] F. Sohrabi and W. Yu, “Hybrid beamforming with finite-resolution phase
per-antenna power constraints.” in Proc. IEEE Global Communication shifters for large-scale MIMO systems,” in Proc. IEEE Workshop Signal
Conference (GLOBECOM), Washington, DC, Dec. 2016, pp. 1-6. Process. Adv. Wireless Commun. (SPAWC), Stockholm, Sweden, June
[9] C.-E. Chen, “An iterative hybrid transceiver design algorithm for mil- 2015, pp. 136-140.
limeter wave MIMO systems,” IEEE Wireless Commun. Lett., vol. 4, [32] F. Sohrabi and W. Yu, “Hybrid digital and analog beamforming design
no. 3, pp. 285-288, June 2015. for large-scale antenna arrays,” IEEE J. Sel. Topics Signal Process., vol.
[10] W. Ni, X. Dong, and W.-S. Lu, “Near-optimal hybrid processing for 10, no. 3, pp. 501-513, April 2016.
massive MIMO systems via matrix decomposition,” IEEE Trans. Signal [33] G. N. Karystinos and D. A. Pados, “Rank-2-optimal adaptive design of
Process., vol. 65, no. 15, pp. 3922-3933, Aug. 2017. binary spreading codes,” IEEE Trans. Inf. Theory, vol. 53, no. 9, pp.
[11] X. Gao, L. Dai, S. Han, C.-L. I, and R. W. Heath Jr., “Energy-efficient 3075-3080, Sept. 2007.
hybrid analog and digital precoding for mmWave MIMO systems with [34] G. H. Golub and C. F. Van Loan, Matrix Computations. Baltimore, USA:
large antenna arrays,” IEEE J. Sel. Areas Commun., vol. 34, no. 4, pp. JHU Press, 2012.
998-1009, April 2016.
[12] L. Dai, X. Gao, J. Quan, S. Han, and C.-L. I, “Near-optimal hybrid
analog and digital precoding for downlink mmWave massive MIMO
systems”, in Proc. IEEE Int. Conf. Commun. (ICC), London, UK, June
2015, pp. 1334-1339.
[13] S. He, C. Qi, Y. Wu, and Y. Huang, “Energy-efficient transceiver design
for hybrid sub-array architecture MIMO systems,” IEEE Access, vol. 4,
pp. 9895-9905, 2016.
[14] O. E. Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. W. Heath Jr.,
“Spatially sparse precoding in millimeter wave MIMO systems,” IEEE
Trans. Wireless Commun., vol. 13, no. 3, pp. 1499-1513, Mar. 2014.
[15] A. Alkhateeb, O. El Ayach, G. Leus, and R. W. Heath Jr., “Channel
estimation and hybrid precoding for millimeter wave cellular systems,”
IEEE J. Sel. Topics Signal Process., vol. 8, no. 5, pp. 831-846, Oct.
2014.
[16] C. Rusu, R. Méndez-Rial, N. González-Prelcic, and R. W. Heath Jr.,
“Low complexity hybrid sparse precoding and combining in millimeter
wave MIMO systems,” IEEE Int. Conf. on Commun. (ICC), London,
UK, June 2015, pp. 1340-1345.
[17] J.-C. Chen, “Efficient codebook-based beamforming algorithm for
millimeter-wave massive MIMO systems,” IEEE Trans. Veh. Technol.,
to Appear.
[18] X. Gao, L. Dai, C. Yuen, and Z. Wang, “Turbo-like beamforming based
on Tabu search algorithm for millimeter-wave massive MIMO systems,”
IEEE Trans. Veh. Technol., vol. 65, no. 7, pp. 5731-5737, July 2016.
[19] S. Han, C.-L. I, Z. Xu, and C. Rowell, “Large-scale antenna systems
with hybrid analog and digital beamforming for millimeter wave 5G,”
IEEE Commun. Mag., vol. 53, no. 1, pp. 186-194, Jan. 2015.
[20] A. Li and C. Masouros, “Hybrid analog-digital millimeter-wave MU-
MIMO transmission with virtual path selection,” IEEE Commun. Lett.,
vol. 21, no. 2, pp. 438-441, Feb. 2017.
[21] 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.