Professional Documents
Culture Documents
Research Article
Abstract: Millimeter wave (mmWave) communications have been considered as a key technology for future 5G wireless
networks since it can provide orders-of-magnitude wider bandwidth than current cellular bands. To overcome the severe
propagation loss of the mmWave channel, an economic and energy-efficient analogue/digital hybrid precoding and combining
transceiver architecture is widely used in mmWave massive multiple-input multiple-output (MIMO) systems. The digital
precoding/combining layer offers more freedom than pure analogue one and enables multi-stream transmission. In this study,
the authors consider the problem of codebook-based joint hybrid precoder and combiner design for multi-stream transmission in
mmWave MIMO systems. The authors propose to jointly select an analogue precoder and combiner pair for each data stream
successively, which can maximise the channel gain as well as suppress the interference between different data streams. Then,
the digital precoder and combiner are computed based on the obtained effective baseband channel to further mitigate the
interference and maximise the sum-rate. Both fully-connected and partially-connected hybrid beamforming structures are
investigated. Simulation results demonstrate that the proposed algorithms exhibit prominent advantages in combating
interference between different data streams and offer satisfactory performance improvements compared with the existing
codebook-based hybrid beamforming schemes.
simplified. A popular solution to design the hybrid precoder/ loss in practical mmWave multiplexing systems. This motivates us
combiner based on the spectral efficiency maximisation of point- to reconsider the hybrid precoder and combiner design and find a
to-point mmWave communications is to minimise the Euclidean better way for the multiple data streams transmission in mmWave
distance between the hybrid precoder/combiner and the full-digital MIMO communications.
precoder/combiner. In this paper, we consider the problem of codebook-based joint
The existing hybrid precoding designs typically assume the use hybrid precoder and combiner design for multi-stream transmission
of infinite resolution phase shifters for implementing the analogue in mmWave MIMO systems. We first focus on the fully-connected
beamformers. In [5, 6], the authors proposed to iteratively design structure and propose to jointly select the analogue precoder and
the analogue and digital precoder to approximate the unconstrained combiner pair for each data stream successively, which can
precoder. In [7], a novel full-digital approximation method was maximise the channel gain as well as suppress the interference
introduced, where the analogue precoder was directly chosen to between different data streams. Then, the digital precoder and
match the phase of singular vectors of channel with constant combiner are computed based on the obtained effective baseband
magnitude before optimising the digital precoder. The researches channel to further mitigate the interference and maximise the sum-
on hybrid precoding applied in partially-connected architectures rate. Then this joint hybrid precoder and combiner design scheme
are also studied in [8–10]. A successive interference cancellation is extended to partially-connected architectures, where each
based hybrid precoding scheme is proposed in [8, 9], where the analogue precoder and combiner pair of the transmit and receive
hybrid precoders for each sub-array are successively designed. In sub-array pair is selected one by one to enhance the corresponding
[10], the analogue precoder and combiner were first iteratively sub-channel gain as well as mitigate the inter sub-array
designed, then digital precoder and combiner were optimised based interference. Simulation results demonstrate that our proposed
on the effective MIMO channel. All of these hybrid designs with algorithms exhibit prominent advantages in combating the effect of
infinite phase shifters can achieve satisfactory performance. While interference between different data streams in both fully-connected
the components for realising accurate phase shifters could be very and partially-connected structures and offer satisfactory
complicated and expensive, low-resolution phase shifters with performance improvements compared with the existing codebook-
discrete/quantised phases are cost-effective and typically adopted based hybrid beamforming schemes.
in realistic systems [11–13]. The performance lower bound of the The following notation is used throughout this paper. Bolditalic
hybrid beamformer with quantised-phase shifters is derived in [12]. lower-case letters indicate column vectors and bolditalic face
Furthermore, according to the special characteristic of mmWave upper-case letters indicate matrices; ℂ denotes the set of all
channel, more practical codebook-based hybrid precoder designs complex numbers; ( ⋅ )T and ( ⋅ )H denote the transpose and
have been widely used [14–19], in which the columns of the transpose-conjugate operation, respectively. Following the matrix
analogue precoder are selected from certain candidate vectors, such indexing rule which is widely used in Matlab, we use A(: , i) to
as array response vectors of the channel and discrete Fourier denote the ith column of a matrix A. IL is the L × L identity matrix;
transform (DFT) beamformers, which have constant modulus and E{ ⋅ } represents statistical expectation. Finally, | ⋅ |, ∥ ⋅ ∥, and
discrete phases. In these codebook-based hybrid precoder designs, ∥ ⋅ ∥F are the scalar magnitude, vector norm, and Frobenius norm,
orthogonal matching pursuit (OMP) is a typical algorithm which
can be seen as a sparsity constrained matrix reconstruction respectively.
problem. In [20], the authors proposed a Turbo-like beamforming
scheme to seek the optimal pair of analogue precoder and combiner 2 System model and problem formulation
with low searching complexity but near-optimal performance. 2.1 System model
In the existing codebook-free or codebook-based algorithms
mentioned above, the optimal hybrid precoder and combiner are We consider a single-user mmWave MIMO multiplexing system
individually designed to approximate, in the best Frobenius norm, with hybrid precoder and combiner as illustrated in Fig. 1a. The
the right and left singular vectors of the channel matrix, transmitter employs Nt antennas and NtRF RF chains to
receptively. While the separate design for hybrid precoder and simultaneously transmit Ns data streams to the receiver which is
combiner can provide satisfactory performance in terms of spectral
efficiency, orthogonality of resulting spatial multiplexing channel equipped with Nr antennas and NrRF RF chains. To ensure the
cannot be guaranteed [14]. Therefore, the conventional hybrid efficiency of the communication with a limited number of RF
precoder and combiner designs may cause significant performance chains, the number of data streams is constrained as
x = FRFFBBs (1) where λ is the signal wavelength, and d is the distance between
antenna elements. The transmit array response vector at(θ) can be
where s is the Ns × 1 symbol vector such that E{ssH} = (1/Ns)INs written in a similar fashion.
The total transmit power constraint is enforced by normalising FBB
such that ∥ FRFFBB ∥2F = Ns. 2.3 Problem formulation
For simplicity, we consider a narrowband block-fading We consider the problem of codebook-based hybrid precoder and
propagation channel, which yields a receive signal as combiner design in a mmWave multiplexing system. Specifically,
let ℱ and W denote the beamsteering codebooks for the analogue
y = PHFRFFBBs + n, (2) precoder and combiner, respectively. If BtRF (BrRF) bits are used to
quantise the AoD (AoA), ℱ and W will consist of all the possible
where y is the Nr × 1 received vector, H is the Nr × Nt channel analogue precoder and combiner vectors [17], which can be
matrix, P represents the average receive power, and presented as
n ∼ CN(0, σn2 INr) is the complex Gaussian noise vector corrupting
RF RF
the receive signal. ℱ = {at(2πi/2Bt ): i = 1, …, 2Bt }, (7)
The receiver uses its NrRF RF chains and phase shifters to
process the obtained receive signal which has a form of RF RF
W = {ar(2πi/2Br ): i = 1, …, 2Br } . (8)
H H H H
s^ = PWBB WRF HFRFFBBs + WBB WRF n (3) The columns of analogue precoding (combining) matrix FRF (WRF)
are picked from candidate vectors in ℱ (W), i.e.
where WRF is the Nr × NrRF
analogue RF combiner, WBB is the f RFl ∈ ℱ, ∀l = 1, …, NtRF, wRFl ∈ W, ∀l = 1, …, NrRF.
RF
Nr × Ns digital baseband combiner and normalised to satisfy When Gaussian symbols are transmitted over the mmWave
∥ WRFWBB ∥F2 = Ns. MIMO channel, the achieved spectral efficiency is given by [14]
In this study, we implicitly assume perfect timing and frequency
recovery and the channel state information (CSI) of H is known P −1 H H
R = log2 INs + R W W HF F
perfectly and instantaneously to both transmitter and receiver. In Ns n BB RF RF BB
practice, CSI can be accurately and efficiently obtained by channel (9)
H H
estimation at the receiver [15] and further shared at the transmitter × FBB FRF HWRFWBB ,
with effective feedback techniques.
H H
where Rn ≜ σn2 WBB WRF WRFWBB is the noise covariance matrix
2.2 Millimeter wave MIMO channel model after combining.
Due to high free-space pathloss and large tightly-packed antenna While most existing hybrid precoder and combiner design
arrays, the mmWave propagation in a massive MIMO system is algorithms aim to maximise spectral efficiency in (9), we should
well characterised by a limited spatial selectivity or scattering note that it is actually a performance upper bound for a general
model, e.g. the Saleh–Valenzuela model, which allows us to MIMO system. When we use simple precoder and combiner to
accurately capture the mathematical structure in mmWave channels implement spatial multiplexing, we are more interested in the
[14]. The matrix channel H is assumed to be a sum contribution of signal-to-interference-plus-noise ratio (SINR) or the rate of each
Ncl scattering clusters, each of which provides Nray propagation stream, rather than the spectral efficiency of the MIMO system.
paths to the channel matrix H. Therefore, the discrete-time narrow- Even though precoder and combiner designed by approximating
band mmWave channel H can be formulated as the right and left singular vectors of the channel matrix can provide
satisfactory spectral efficiency performance, they cannot guarantee
N N the orthogonality of the resulting effective spatial multiplexing
NtNr cl ray
Nc1Nray i∑ ∑ αilΛr(θilr )Λt(θilt )ar(θilr )at(θilt )H,
H= (4) channel for the multiple data streams transmission and
=1l=1 consequently have low sum-rate performance. Therefore, for a
practical spatial multiplexing system, the sum-rate of all the data
where αil ∼ CN(0, σα2 , i) is the complex gain of the lth propagation streams is a more reasonable and practical performance metric
path (ray) in the ith scattering cluster, following independent which is described as follows.
Given the received signal in (3), the SINR of the kth data stream
identically distributed (i.i.d.) form. σα2 , i represents the average
is formulated by
power of the ith cluster, and the total power satisfies
N
∑i =cl 1 σα2 , i = Nc1. θilt and θilr are the angle of departure (AoD) and the
angle of arrival (AoA), respectively, which are assumed to be γk
Laplacian-distributed with a mean cluster angle θit and θir as well as (10)
(P/Ns) ∣ W(: , k)HHF(: , k) ∣2
an angle spread of σθit and σθir. The functions Λt(θilt ) and Λr(θilr ) = N
represent the transmit and receive antenna element gain associated (P/Ns)∑i =s 1, i ≠ k ∣ W(: , k)HHF(: , i) ∣2 + σn2 ∥ W(: , k) ∥2
with the corresponding AoD and AoA. If ideal antenna elements
are assumed, Λt(θilt ) can be given by
⋆ ⋆
FRF((k − 1)Mt + 1: kMt, k) = f RF k
,
⋆ ⋆
(27)
WRF ((k − 1)Mr + 1: kMr, k) = wRF k
.
where bk ∈ ℂMt × 1 and dk ∈ ℂMr × 1 play the same roles as pk and qk.
⋆ ⋆
Particularly, when k = 1, let bk = f RF k
and dk = wRF k
; when k > 1,
we utilise the Gram–Schmidt procedure to find the orthogonormal
vectors as
k−1
⋆
bk = f RF k
− ∑ biH f RF bi,
⋆
k
i=1
bk = bk / ∥ bk ∥ , k = 2, …, Ns; (29)
k−1
⋆
dk = wRF k
− ∑ diHwRF di,⋆
k
i=1
dk = dk / ∥ dk ∥ , k = 2, …, Ns . (30)
~
Then we replace Hk in (26) with the updated sub-channel Hk.
After RF precoder and combiner for all sub-antenna-array pairs
⋆ ⋆
are obtained, the digital precoder FBB and combiner WBB can be
computed based on the effective baseband channel using a similar
procedure with (20)–(24). The final joint hybrid precoder and
combiner design for partially-connected architecture is summarised
in Fig. 3.
Fig. 3 Hybrid precoder and combiner design for partially-connected
structure 5 Complexity analysis
In this section, we analyse the complexity of the proposed joint
sub-antenna-array with only Mt antennas associated with the hybrid precoder and combiner designs and provide a comparison
corresponding RF chain. Therefore, analogue precoding matrix FRF on the computation complexity with other codebook-based
has a different structure as follows schemes.
For the fully-connected architectures, we notice that the
f1 0 … 0 required multiplications for every beamformer pair's selection is
0 f2 0 |ℱ | | W | (NtNr + Nr). Moreover, seen from Fig. 2, in the kth
FRF = (25) iteration, the computation complexity mainly comes from the
⋮ ⋱ ⋮ ~ (k + 1)
correction of pk and qk, as well as the update of H which
0 0 … f NtRF
requires
where f k, k = 1, …, NtRF denotes the Mt × 1 steering vector for the 3(k2 − k + 1)(Nt + Nr)
kth sub-antenna-array (or RF chain). The analogue combiner 2
matrix WRF has a similar construction.
The non-convex constraints on FRF and WRF make the multiplications and (NtNr + 1)(Nt + Nr) multiplications,
optimisation problem (12) very difficult to be solved. Therefore, respectively. Finally, the joint design scheme will perform SVD
we adopt a similar idea introduced in the fully-connected hybrid
k2 − k
Mt Mr + (Mt + Mr)
2
~
multiplications to compute Hk. Besides, SVD computation is also
required to obtain the digital precoder and combiner. The total
computational complexity is summarised in Table 1, in which we
also include the complexity of the most widely-used codebook-
based spatially sparse precoding scheme [14] for the comparison
purpose.
6 Simulation results
In this section, we illustrate the simulation results of the proposed
joint hybrid precoder and combiner design. We first consider a
mmWave MIMO multiplexing system with fully-connected hybrid
architecture. Both transmitter and receiver are equipped with a
128-antenna ULA and antenna spacing is d = λ/2. The number of
RF chains at transmitter and receiver is NtRF = NrRF = 4, so is the
number of data streams Ns = 4. The channel parameters are set as
Ncl = 10 clusters, Nray = 10 rays per cluster, and the average power
N
of the ith cluster is σα2 , i = c(7/10)i where c = (∑i =cl 1 (7/10)i)−1Ncl.
The azimuths of the AoAs/AoDs within a cluster are assumed to be
Laplacian-distributed with an angle spread of σθir = σθit = 2.5∘. The
mean cluster AoDs are assumed to be uniformly distributed over
[0, 2π] and the mean cluster AoAs are uniformly distributed over Fig. 4 Spectral efficiency and sum-rate performance in the fully-connected
an arbitrary π/3 sector. Finally, we employ a codebook consists of
structure. Nt = Nr = 128, NtRF = NrRF = 4, Ns = 4
array response vectors with 64 uniformly quantised angle
resolutions for the selection of analogue beamformers.
Fig. 4a shows the spectral efficiency versus signal-to-noise- In Figs. 5a and b, we turn to illustrate how the number of
transmit and receive antennas affects the spectral efficiency and
ratio (SNR) over 106 channel realizations. For a comparison
sum-rate performance. We assume Nt = Nr = N which is varying
purpose, we also include two state-of-the-art algorithms: (i)
spatially sparse precoding (SSP) in [14] which is a classic from 16 to 256. The SNR is set at 20 dB and NtRF = NrRF = Ns = 4.
codebook-based hybrid precoding design. [For a fair comparison, It can be observed from these two figures that the proposed
spatially sparse MMSE combining proposed in [14] is also used in algorithm has significant superiority compared with the SSP
the simulation.] (ii) Alternating minimisation using phase approach in the spectral efficiency as well as sum-rate
extraction (PE AltMin) algorithm in [5] which is a codebook-free performance. Interestingly, we also notice that the SSP scheme
hybrid precoding design. The optimal (OPT) full-digital may exhibit severe performance degradation with an increase in the
beamforming scheme with the unconstrained SVD algorithm is number of antennas when the resolution of codebook is relatively
also plotted as the performance benchmark. It can be observed that low. This is because that the performance of SSP depends heavily
our proposed algorithm outperforms the codebook-based SSP on the resolution of codebook and the number of antennas. When
algorithm. Note that the PE AltMin algorithm has a continuous either of the two factors is not large enough, the SSP algorithm
phase on a RF beamformer and can achieve extremely close may have performance degradation.
performance to the optimal full-digital approach. Therefore, we Figs. 6a and b provide spectral efficiency and sum-rate versus
just consider it as a reference for the codebook-free algorithms but the number of data streams Ns, respectively. The number of
would not compare it with our proposed algorithm. Fig. 4b presents transmit and receive RF chains are also changing along with Ns.
the sum-rate versus SNR with the same system settings as in Fig. We can see that three approaches can achieve comparable
4a. We can notice that the proposed joint hybrid precoder and performance in spectral efficiency. However, only our proposed
combiner design has significant performance advantages over the algorithm can maintain a satisfactory sum-rate achievement with
other two hybrid beamforming designs due to the fact that our joint the increasing number of data streams and the strong interference
analogue precoder and combiner selection approach aims to
mitigate the interference between different data streams.
IET Commun., 2017, Vol. 11 Iss. 17, pp. 2596-2604 2601
© The Institution of Engineering and Technology 2017
17518636, 2017, 17, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-com.2017.0263 by Algeria Hinari NPL, Wiley Online Library on [07/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 5 Effects of the number of transceiver antennas in the fully-connected Fig. 6 Effects of the number of data streams in the fully-connected
structure structure
(a) Spectral efficiency, (b) Sum-rate performance. SNR=20 dB, NtRF = NrRF = 4, (a) Spectral efficiency, (b) Sum-rate performance. Nt = Nr = 128, SNR=20 dB
Ns = 4
numbers of transmitter and receiver RF chains/sub-arrays are also
between different data streams can even cause a performance loss assumed to change to satisfy NtRF = NrRF = Ns. Similar conclusions
when the SSP algorithm is applied. can be drawn that our proposed algorithm does not exhibit obvious
Now we turn to demonstrate the proposed algorithm for the weakness though finite-resolution codebook is applied. However,
partially-connected hybrid architecture and compare it with other the proposed scheme can also perform well in the sum-rate
two classic algorithms: (i) Semi-definite relaxation-based performance, while the achievable sum-rate using SDR or SIC is
alternating minimisation (SDR) in [5], (ii) successive interference almost ruined due to the severe inter-stream interference.
cancellation-based hybrid precoding (SIC) in [8]. Again, these two
algorithms are codebook-free approaches. The spectral efficiency 7 Conclusions
and sum-rate performances with different values of SNR are
illustrated in Figs. 7a and b. We can observe that three algorithms This study considered the problem of codebook-based joint hybrid
have very close performance in achieving a satisfactory spectral precoder and combiner design for multi-stream transmission in
efficiency even though the two schemes are both for codebook-free mmWave MIMO systems. We proposed to jointly select analogue
based designs. The SDR and SIC design the hybrid precoder and precoder and combiner pair for each data stream successively,
combiner separately to maximise the spectral efficiency and the which can maximise the channel gain as well as suppress the
interference cannot be well mitigated. Therefore our proposed interference between different data streams. Then, the digital
algorithm presents significant performance advantages in sum-rate precoder and combiner were computed based on the obtained
for multiplexing transmissions. To the best of our knowledge, all baseband effective channel to further mitigate the interference and
existing hybrid beamformer designs aim to maximise spectral maximise the sum-rate. Both fully-connected and partially-
efficiency. The results obtained from this simulation prompt us to connected hybrid bemaforming structures were investigated.
pay attention to the inter-stream interference mitigation in the Simulation results demonstrated the performance improvements of
multiplexing systems. our proposed algorithms compared with the existing codebook-
The number of antennas versus spectral efficiency is also shown based hybrid beamforming schemes.
in Fig. 8a. We still assume Nt = Nr = N in the simulations. It can
be observed that the proposed algorithm can outperform the SDR 8 Acknowledgments
and SIC schemes when both the transmitter and receiver are This study is supported by the Natural Science Foundation of
equipped with 128-ULAs, which is the most common number of China (grant no. 61671101 and 61601080).
antennas applied during the previously demonstrated researches.
Sum-rate versus N is illustrated in Fig. 8b and our proposed
algorithm can always produce significant performance advantages.
Finally, we demonstrate the effect of the number of data streams
on the spectral efficiency and sum-rate in Figs. 9a and b,
respectively. With the different numbers of data streams, the
2602 IET Commun., 2017, Vol. 11 Iss. 17, pp. 2596-2604
© The Institution of Engineering and Technology 2017
17518636, 2017, 17, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-com.2017.0263 by Algeria Hinari NPL, Wiley Online Library on [07/03/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 7 Spectral efficiency and sum-rate performance in the partially-connected structure. Nt = Nr = 128, NtRF = NrRF = 4, Ns = 4
9 References wireless communication system design’, IEEE Trans. Commun., 2015, 63,
(9), pp. 3029–3056
[1] Pi, Z., Khan, F.: ‘An introduction to millimeter-wave mobile broadband [5] Yu, X., Shen, J.-C., Zhang, J., et al.: ‘Alternating minimisation algorithms for
systems’, IEEE Commun. Mag., 2011, 49, (6), pp. 101–107 hybrid precoding in millimeter wave MIMO systems’, IEEE J. Sel. Top.
[2] Rappaport, T., Sun, S., Mayzus, R., et al.: ‘Millimeter wave mobile Signal Process., 2016, 10, (3), pp. 485–500
communications for 5G cellular: it will work!’, IEEE Access, 2013, 1, pp. [6] Rusu, C., Méndez-Rial, R., González-Prelcic, N., et al.: ‘Low complexity
335–349 hybrid precoding strategies for millimeter wave communication systems’,
[3] Heath, R.W., González-Prelcic, N., Rangan, S., et al.: ‘An overview of signal IEEE Trans. Wirel. Commun., 2016, 15, (12), pp. 8380–8393
processing techniques for millimeter wave MIMO systems’, IEEE J. Sel. Top. [7] López-Valcarce, R., González-Prelcic, N., Rusu, C., et al.: ‘Hybrid precoders
Signal Process., 2016, 10, (3), pp. 436–453 and combiners for mmWave MIMO systems with per-antenna power
[4] Rappapport, T.S., MacCartney, G.R., Samimi, M.K., et al.: ‘Wideband constraints’. IEEE Global Communication Conf.(GLOBECOM), Washington,
millimeter-wave propagation measurements and channel models for future DC, December 2016, pp. 1–6