 Research
 Open Access
Channel estimation and analog beam selection for uplink multiuser hybrid beamforming system
 YoungChai Ko^{1}Email author and
 MyeongJin Kim^{1}
https://doi.org/10.1186/s1363801606180
© Ko and Kim. 2016
 Received: 19 November 2015
 Accepted: 21 April 2016
 Published: 2 July 2016
Abstract
This paper considers an uplink multiuser hybrid beamforming system where a base station (BS) communicates with multiple users simultaneously. The performance of the uplink multiuser hybrid beamforming system depends on the effective channel which is given by the product of channel matrix and the analog beams. Therefore, to maximize the performance, we need to acquire information of the channels and select the appropriate analog beams from the set of predefined analog beams. In this paper, we propose the channel estimation methods and analog beam selection algorithm for the uplink multiuser hybrid beamforming system. First, we design the estimation methods to exploit the channel information of the users by considering Rayleigh fading and millimeter wave (mmWave) channel models. Then, using the estimated channel information, we propose a lowcomplexity analog beam selection algorithm for the uplink multiuser hybrid beamforming system. We compare the complexity to show that the proposed analog beam selection algorithm has much less complexity than the exhaustive searchbased optimum analog beam selection while the performance loss of the proposed analog beam selection algorithm is not significant compared to the optimum analog beam selection, which is shown by the numerical results.
Keywords
 Uplink multiuser hybrid beamforming
 Analog beamforming
 Digital beamforming
 Low complexity
1 Introduction
Multiuser multiple input multiple output (MUMIMO) is a promising technique for exploiting the spectral efficiency and multiuser diversity gain of wireless channels. Thus, it has become an important part of 3GPP Long Term Evolution and IEEE 802.11 wireless local area network (WLAN) standards [1–3]. By utilizing the simple digital beamforming processing such as zero forcing (ZF) and minimum mean squared error (MMSE), we can achieve the capacity over MUMIMO channels. Recently, massive MUMIMO employing largescale antenna array at base station (BS) has received huge interest since it significantly improves the spectral efficiency and power efficiency [4–8]. The potential of massive MUMIMO can be fully exploited with the digital beamforming where the phase and amplitude adjustment of beams are controlled precisely [9]. However, since the same number of analogtodigital converters (ADCs) and radio frequency (RF) chains as the number of antennas is required to realize the digital beamforming, it is not efficient in terms of cost, complexity, and energy consumption [10]. Thus, realizing digital beamforming for massive MUMIMO appears to be impractical.
On the other hands, increase of smart devices requiring a large amount of data traffics has caused demands to exponentially expand wireless data traffic [11]. As one of solutions to satisfy the increasing demand for wireless data traffic, use of larger spectrum (or bandwidth) and its reuse have attracted tremendous interest while improvement of spectral efficiency has certain limitation to dramatically increase data rate. Due to this reason, millimeter wave (mmWave) communication is a potential technology for future outdoor cellular systems to support high data rate [12–15]. However, to overcome high path loss which is the unfavorable channel characteristic of mmWave frequency while achieving reasonable link budgets, beamforming using a number of antennas to form directive beams with large array gain is absolutely necessary [16]. Unfortunately, digital baseband beamforming is impracticable on account of the high cost of mixed signal components, which also occurs in massive MUMIMO as aforementioned.
To overcome such disadvantages that both massive MUMIMO and mmWave communication have in common, analog beamforming implemented just by employing the phase shifters at each RF chain and the combiner to coherently sum the phasecorrected signals by the phase shifters has been considered [17, 18]. Unfortunately, the performance of analog beamforming is suboptimal due to the limitation of controlling the signal phase. To achieve the larger beamforming gains and support multiuser by enabling precoding for multistream transmission, a hybrid (i.e., a mixture of analog and digital) beamforming was suggested in [19, 20]. While most of studies on hybrid beamforming have considered singleuser MIMO system, we consider an uplink multiuser hybrid beamforming system where multiple users transmit data to BS on the same frequency band simultaneously.
In this paper, we design the channel estimation and analog beam selection algorithm for an uplink multiuser hybrid beamforming system. Since the effective channel, which is defined as the product of the channel matrix and analog beams, determines the performance of the uplink multiuser hybrid beamforming system, the analog beams for both BS and users need to be selected depending on the channel state information to improve the sum rate of the uplink multiuser hybrid beamforming system. Therefore, first, we present the proposed channel estimation methods for uplink multiuser hybrid beamforming system. Since each RF chain observes a combining signal from multiple antennas, BS cannot exploit the channel coefficients of each antenna. Due to this reason, we propose the channel estimation method that exploits the full channel information of the users by repeatedly transmitting the pilot sequence over multiple pilot periods. Also, we discuss the length of multiple pilot sequences needed to acquire the full channel knowledge by using the proposed channel estimation method. However, although such channel estimation method can be applied to Rayleigh fading channel to obtain the channel coefficients, it may not precisely estimate the channel coefficients due to the low beamforming gain in mmWave channel. Therefore, we also discuss on the channel estimation in mmWave channel. After the channel information of the users are obtained based on the proposed channel estimation method according to the kind of channel model, BS selects the analog beams among the predefined analog beams for both BS and users. As an optimum solution that maximizes the sum rate, we can consider the exhaustive search algorithm for the analog beams of both BS and users. However, the complexity of the exhaustive search is too high for practical applications since it calculates the sum rate values for all possible analog beams. For practical applications of the hybrid beamforming system, we propose a lowcomplexity analog beam selection algorithm for the uplink multiuser hybrid beamforming system. We first consider twouser hybrid beamforming system and derive a metric analytically which is a factor of measuring the orthogonality between the effective channels. Then, we extend it to the case where three or more users exist by combining the effective channel vectors of the users whose analog beams are chosen as the weighted effective channel vector. In other words, based on the metric, the proposed analog beam selection algorithm selects the analog beams by considering the orthogonality between “the weighted effective channel vector” and “the effective channels of the users whose analog beams are not determined.” In addition, we characterize the complexity of our proposed analog beam selection algorithm and the exhaustive search. From the numerical results, we confirm that the proposed analog beam selection algorithm provides negligible performance degradation under the equal system configuration while its complexity is much less than the complexity of the optimum analog beam selection algorithm.
The remainder of this paper is organized as follows. Section 2 presents the system model under consideration. Section 3 describes the concept of analog beam set. In Section 4, we design the channel estimation methods for Rayleigh fading and mmWave channel models in uplink multiuser hybrid beamforming system. In Section 5, we describe the lowcomplexity analog beam selection algorithm. Also, we show the computational complexity and discuss on procedure for our proposed analog beam selection algorithm. Simulation results are then given in Section 6. Finally, we draw some conclusions in Section 7.
The following mathematical notations will be used throughout the paper. Upper case and lower case boldfaces are used to denote matrices and vectors, respectively. (·)^{ T },(·)^{−1},(·)^{ H }, and \(\mathbb {E}[\!\cdot ]\) represent the transpose, inverse, conjugate transpose, and expectation operator, respectively. We also indicate ∥a∥ as the norm of vector a and · as the absolute value of complex number. A _{ ij } is the element in the ith row and the jth column of matrix A.
2 System model
where \(\mathbf {H}_{i} \in \mathbb {C}^{N_{\text {BS}} \times N_{\text {user}}}\) is the channel matrix from the ith user to BS and \(\mathbf {n} \in \mathbb {C}^{N_{\text {BS}}\times 1}\) is the additive white Gaussian noise (AWGN) vector at BS with zero mean and variance σ ^{2} per entry.
where \(\bar {\mathbf {n}}=\mathbf {B}\mathbf {n}\) and \( \mathbf {h}_{i}^{\text {eff}}\), defined as \( \mathbf {h}_{i}^{\text {eff}}=\mathbf {B}\mathbf {H}_{i}\mathbf {t}_{i}\), is the effective channel vector between the ith user and BS.
Note that all the entries of the analog beamforming can be written as e ^{ j θ } where θ∈[0,2π] is the phase of the phase shifter.
which can be regarded as the effective channel gain for the ith user. We note from (9) that the performance of the system depends on the effective channels.
respectively, where λ is the signal wavelength and d is the distance between antenna elements.
The channel model in (10) turns out to be Rayleigh fading channel when L is very large.
Based on the channel state information, BS selects the analog beams for both BS and users. However, since we assume that BS has no a priori knowledge of the channels, we develop the channel estimation methods the two different channel estimation methods for Rayleigh fading and mmWave channel models in the uplink multiuser hybrid beamforming system in Section 4 of this paper. Then, in Section 5, the proposed analog beam selection algorithm to determine the analog beams for both BS and users is described based on the estimated channels.
3 Analog beam set
The fundamental idea of analog beamforming is to control the phase of the signal transmitted/received from each antenna via a circuit network of analog phase shifters such that each of the analog beamforming coefficients has constant modulus. Therefore, both BS and users are assumed to operate with the selected analog beams among the predefined analog beams.
where \(\mathbf {c}_{i}\in \mathbb {C}^{N_{\text {user}} \times 1}\) is a candidate for the transmit analog beamforming vector and the nth element in c _{ i } is the value of the phase shifter on the nth antenna of each user.
where \(\mathbf {G}_{i} \in \mathbb {C}^{N_{\mathrm {s}} \times N_{\text {BS}}}\) is a candidate for the receive analog beamforming matrix and the element in the nth row and the jth column of G _{ i } is the value of the jth antenna’s phase shifter on the nth RF chain.
We assume that all users use the same predefined set of analog beams and BS knows it.
4 Channel estimation of uplink multiuser hybrid beamforming system
As shown in (8) and (9), since the performance of uplink multiuser hybrid beamforming system depends on the effective channel, BS calculates the effective channels in advance of determining the analog beams for both BS and users. To do that, BS needs information of the channels. In this section, we propose the two different channel estimation methods for Rayleigh fading and mmWave channel models in uplink multiuser hybrid beamforming system.
4.1 Rayleigh fading channel
 1.
The pilot sequences of users are orthogonal to each other, which means that the number of users should be no larger than the length of pilot sequence, that is, τ≥K. Also, BS has information about the pilot sequences of the users.
 2.
During the channel estimation process, both BS and users use the predetermined analog beams and BS has information of predetermined analog beams which users use for the channel estimation.
 3.
To estimate the channel of each user, we adopt the least squares (LS) estimator.
since each channel matrix has N _{BS}×N _{user} channel coefficients. Therefore, the length of multiple pilot sequences needed to acquire the full channel knowledge by using the proposed channel estimation method is \(N_{\text {BS}}^{\text {RF}} \times N_{\text {user}}\times \tau \).
At the end of the channel estimation periods, BS calculates the effective channels based on the estimated channels and determines the analog beams both BS and users, which is introduced in the next section.
4.2 MmWave channel
After the estimated effective channels are computed, BS selects the analog beams for both BS and users.
5 Analog beam selection for uplink multiuser hybrid beamforming system
It is noticed that we need to select the analog beams among the predefined analog beams for both BS and users to improve the sum rate of the uplink multiuser hybrid beamforming system since the effective channel is the product of channel matrix and analog beams. In this section, we analyze the sum rate of the uplink twouser hybrid beamforming system to introduce the metric which is used for the proposed analog beam selection algorithm. Then, we explain the proposed analog beam selection algorithm and compare the complexity of the proposed analog beam selection algorithm with the exhaustive search.
5.1 Performance analysis
Since the log function is a monotonic increasing function, the sum rate of the uplink twouser hybrid beamforming system in (26) is proportional to \(f\left (\mathbf {h}_{1}^{\text {eff}},\mathbf {h}_{2}^{\text {eff}}\right)\). Therefore, using the metric in (27), our proposed analog beam selection algorithm operates to choose the appropriate analog beams for both BS and users with the predefined analog beam sets in the way of maximizing \(f\left (\mathbf {h}_{1}^{\text {eff}},\mathbf {h}_{2}^{\text {eff}}\right)\) resulting in the optimum solution in term of the sum rate. Since the metric in (27) is based on the analysis of the uplink twouser hybrid beamforming system, it cannot be directly applied to the case where three or more users exist. For K>3 case, we can apply the metric in (27) by combining K−1 effective channel vectors into a weighted effective channel vector. The details of our proposed analog beam selection algorithm will be discussed in the next subsection.
Note that, though the analog beams can be selected by using the sum rate of the uplink multiuser hybrid beamforming system in (8) and (9) for K>3 case, it is not efficient in terms of complexity since the complexity for calculating the sum rate exponentially increases with the number of users.
5.2 Proposed lowcomplexity analog beam selection algorithm
 (1)Computation of the sum rate for each receive analog beam: Assuming that the receive analog beam is fixed as G _{ m }, we select the analog beams for all users. Since each user has the predefined \(N_{\text {user}}^{\text {BSS}}\) analog beams, there are \(K\times N_{\text {user}}^{\text {BSS}}\) possible effective channels. At first, we select one user and its analog beam which provides the maximum effective channel gain among \(K\times N_{\text {user}}^{\text {BSS}}\) possible effective channels, which can be expressed as$$ (i,\mathbf{x}_{i})=\arg\max_{\left(k\in\Omega,\mathbf{c}_{n}\in\mathcal{C}\right)} \left\\mathbf{h}_{k}^{\text{eff}}\left(\mathbf{G}_{m},\mathbf{c}_{n}\right)\right\^{2}. $$(28)Let us denote the weighted effective channel over the users whose analog beams are determined, \(\bar {\mathbf {h}}\), as$$ \bar{\mathbf{h}}=\sum_{j\in S}\alpha_{j}\mathbf{h}_{j}^{\text{eff}}\left(\mathbf{G}_{m},\mathbf{x}_{j}\right), $$(29)where \( \alpha _{j}=\left \\mathbf {h}_{j}^{\text {eff}}\left (\mathbf {G}_{m},\mathbf {x}_{j}\right)\right \^{2}/\sum _{j\in S}\mathbf {h}_{j}^{\text {eff}}\left (\mathbf {G}_{m},\mathbf {x}_{j}\right)\) is the weighted factor for the jth user whose analog beam is chosen as x _{ j }. The analog beams for the rest K−1 users whose analog beams are not determined are selected by measuring the orthogonality between “the effective channels of the users whose analog beams are not chosen” and “the weighted effective channel over the users whose analog beams are determined.” Then, it can be formulated as$$ (i,\mathbf{x}_{i})=\arg\max_{\left(k\in\Omega,\mathbf{c}_{n}\in\mathcal{C}\right)} f\left(\bar{\mathbf{h}},\mathbf{h}_{k}^{\text{eff}}\left(\mathbf{G}_{m},\mathbf{c}_{n}\right) \right). $$(30)
After the analog beams for all users are chosen, we can calculate the sum rate.
 (2)
Determination of the analog beams: We obtain the sum rate values for the predefined \(N_{\text {BS}}^{\text {BSS}}\) analog beams by executing (1). Then, by comparing the sum rate values for each of receive analog beams, we decide the analog beams for both BS and users which provide the maximum sum rate.
As we can see, the proposed analog beam selection algorithm does not need to search over all possible analog beams, which can reduce the complexity compared to the optimum analog beam selection while the loss of sum rate performance is not significant.
5.3 Complexity analysis
 1.We first quantify the complexity to select the transmit analog beams of the users and calculate the sum rate value when the receive analog beam at BS is fixed.

Calculating Forbenius norm of the estimated channel requires 4N _{s} floating point flops. Since there are K users and each user has \(N_{\text {user}}^{\text {BSS}}\) possible analog beams, we need to consider \(KN_{\text {user}}^{\text {BSS}}\) effective channels to find the first user and select its analog beam which provides the maximum vector norm among \(KN_{\text {user}}^{\text {BSS}}\) effective channels, which requires \(4 N_{\mathrm {s}}KN_{\text {user}}^{\text {BSS}}\) flops.

Since the analog beams for the other K−1 users are selected by using the metric in (27), the number of flops to select the analog beams for the other K−1 users is \((N1)N_{\text {user}}^{\text {BSS}}\beta +(N2)N_{\text {user}}^{\text {BSS}}\beta +\cdots +N_{\text {user}}^{\text {BSS}}\beta =\frac {\left (K1\right)K}{2}N_{\text {user}}^{\text {BSS}}\beta \).

Therefore, when the receive analog beam at BS is fixed, the complexity to select the transmit analog beams of the users and calculate the sum rate value is \(N_{\mathrm {s}}KN_{\text {user}}^{\text {BSS}}+\frac {\left (K1\right)K}{2}N_{\text {user}}^{\text {BSS}}\beta +\alpha \).

 2.Since BS selects the analog beam which provides the maximum sum rate value among the sum rate values for \(N_{\text {BS}}^{\text {BSS}}\) possible receive analog beam, the complexity for the proposed lowcomplexity analog beam selection algorithm is estimated as$$ \mathcal{O} \left(N_{\text{BS}}^{\text{BSS}}\left(4 N_{\mathrm{s}}KN_{\text{user}}^{\text{BSS}}+\frac{\left(K1\right)K}{2}N_{\text{user}}^{\text{BSS}}\beta+\alpha\right)\right). $$(31)
5.4 Procedure for analog beam selection

Step 1: The users transmit the multiple pilot sequences so that BS can obtain the effective channel information.

Step 2: BS estimates the effective channels of the users by using the proposed channel estimation methods in Section 4.

Step 3: With the obtained effective channels, BS selects the analog beams for both BS and users by using the proposed analog beam selection algorithm.

Step 4: BS feeds back the indices of the analog beams for the users via errorfree feedback link.
6 Numerical results
In this section, we present simulation results to validate the proposed channel estimation methods and lowcomplexity analog beam selection algorithm. We consider Rayleigh fading and ULA channel models. The proposed different channel estimation methods are applied according to the kind of channel model. Throughout simulations, we consider the uplink multiuser hybrid beamforming system with a case of [(4,6,24),(4,2)]. The signaltonoise ratio (SNR) is defiend as \(\frac {P}{\sigma ^{2}}\). The ULA channel environment is assumed to have L=5 scatters with uniformly random angle of arrival and departure. The complex path gains of ULA channel model are generated using Gaussian random distribution with equal variance. The interelement spacing in both BS and user antenna arrays is set to half a wavelength. The phase of each entry in the predefined analog beams is generated using a random variable distributed uniformly between 0 and 2π. The covariance value of AWGN is 1 per entry. We consider the exhaustive search under the perfect channel knowledge as a benchmark, which is the optimum solution, and the random beam selection, which randomly selects the analog beams among the predefined analog beams, as a baseline algorithm for comparison with our proposed analog beam selection algorithm.
From Figs. 5 and 7, we can notice that the performance loss of the proposed analog beam selection algorithm is not significant compared to the optimum analog beam selection while the proposed analog beam selection algorithm has much less complexity than the exhaustive search based optimum analog beam selection. Therefore, the proposed analog beam selection algorithm is more efficient to be realized on the uplink multiuser hybrid beamforming system.
7 Conclusions
In this paper, we considered an uplink multiuser hybrid beamforming system and investigated the design of channel estimation and analog beam selection algorithm. First, we not only developed the channel estimation method for Rayleigh fading which exploits the full channel information over multiple pilot periods by obtaining the partial channel knowledge during one pilot period and but also considered the channel estimation method for mmWave channel model. Then, we proposed the analog beam selection algorithm for the uplink multiuser hybrid beamforming system that efficiently reduces the complexity over the optimum analog beam selection algorithm. Based on the metric derived from the sum rate performance analysis of the uplink twouser hybrid beamforming system, the proposed analog beam selection algorithm selects the analog beams for both BS and users. By analyzing the complexity in terms of flops, we showed that the proposed analog beam selection algorithm provides much less complexity than the optimum analog beam selection algorithm. We verified by numerical results that the proposed analog beam selection algorithm has little performance degradation in spite of much less complexity than the optimum analog beam selection algorithm under the equal system configuration.
Declarations
Acknowledgements
This work was supported in part by ICT R&D program of MSIP/IITP [2014044006004, Next Generation WLAN System with High Efficient Performance].
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Authors’ Affiliations
References
 D Tse, P Viswanath, Fundamentals of Wireless Communication (Cambridge University Press, Cambridge, 2005).View ArticleMATHGoogle Scholar
 3GPP Physical Channels and Modulation, TS 36.211 (release 10) (2009).Google Scholar
 IEEE Std 802.11ac, Wireless LAN medium access control (MAC) and physical layer (PHY) specifications–amendment 4: enhancements for very high throughput for operation in bands below 6 GHz (2013).Google Scholar
 F Rusek, D Persson, BK Lau, EG Larsson, TL Marzetta, O Edfors, F Tufvesson, Scaling up MIMO: opportunities and challenges with very large arrays. IEEE Signal Process. Mag. 30(1), 40–60 (2013).View ArticleGoogle Scholar
 J Hoydis, S ten Brink, M Debbah, in Proceedings of 2011 49th Annual Allerton Conference on Communication, Control, and Computing. Massive MIMO: How many antennas do we need? (IL, USA, 2011), pp. 545–550.Google Scholar
 HQ Ngo, EG Larsson, TL Marzetta, Energy and spectral efficiency of very large multiuser MIMO systems. IEEE Trans. Commun. 61(4), 1436–1449 (2013).View ArticleGoogle Scholar
 F Rusek, D Persson, BK Lau, EG Larsson, TL Marzetta, O Edfors, F Tufvesson, Scaling up MIMO: opportunities and challenges with very large arrays. IEEE Signal Process. Mag. 30(1), 40–60 (2013). doi:10.1109/MSP.2011.2178495.View ArticleGoogle Scholar
 RC de Lamare, Massive MIMO systems: signal processing challenges and future trends. Union RadioScientifique Internationale (URSI) Bulletin (2013).Google Scholar
 TL Marzetta, Noncooperative cellular wireless with unlimited numbers of base station antennas. IEEE Trans. Wireless Commun. 9(11), 3590–3600 (2010).View ArticleGoogle Scholar
 F Gholam, J Via, I Santamaria, Beamforming design for simplified analog antenna combining architectures. IEEE Trans. Veh. Technol. 60(5), 2373–2378 (2011).View ArticleGoogle Scholar
 International Telecommunication UnionRadiocommunication (ITUR), in Tech. Rep., ITUR M.2243. Accessment of the global mobile broadband deployments and forecast for international mobile telecommunications.Google Scholar
 Z Pi, F Khan, An introduction to millimeterwave mobile broadband systems. IEEE Commun. Mag. 49(6), 101–107 (2011).View ArticleGoogle Scholar
 IEEE 802.11ad standard draft D0.1. [Online] (2013), Available: www.ieee802.org/11/Reports/tgadupdate.htm.
 Standard ECMA387, High rate 60 GHz PHY, MAC and HDMI PAL. [Online] (2009), Available: http://www.ecmainternational.org/.
 S Sun, TS Rappaport, RW Heath, A Nix, S Rangan, MIMO for millimeterwave wireless communications: beamforming, spatial multiplexing, or both?IEEE Commun. Mag. 52(12), 110–121 (2014). doi:10.1109/MCOM.2014.6979962.View ArticleGoogle Scholar
 S Hur, T Kim, DJ Love, JV Krogmeier, TA Thomas, A Ghosh, Millimeter wave beamforming for wireless backhaul and access in small cell networks. IEEE Trans. Commun. 61(10), 4391–4403 (2013).View ArticleGoogle Scholar
 J Wang, Z Lan, CW Pyo, T Baykas, CS Sum, MA Rahman, J Gao, R Funada, F Kojima, H Harada, S Kato, Beam codebook based beamforming protocol for multiGBPS millimeterwave WPAN systems. IEEE J. Sel. Areas Commun. 27(8), 1390–1399 (2009).View ArticleGoogle Scholar
 YM Tsang, ASY Poon, S Addepalli, in Proceedings of IEEE GLOBECOM 2011. Coding the beams: improving beamforming training in mmWave communication system (Houston, Texas, USA, 2011), pp. 1–6.Google Scholar
 X Zhang, AF Molisch, SY Kung, Variablephaseshiftbased RFbaseband codesign for MIMO antenna selection. IEEE Trans. Signal Process. 53(11), 4091–4103 (2005).MathSciNetView ArticleGoogle Scholar
 V Venkateswaran, AJ van der Veen, Analog beamforming in MIMO communications with phase shift networks and online channel estimation. IEEE Trans. Signal Process. 58(8), 4131–4143 (2010).MathSciNetView ArticleGoogle Scholar
 QH Spencer, BD Jeffs, MA Jensen, AL Swindlehurst, Modeling the statistical time and angle of arrival characteristics of an indoor multipath channel. IEEE J. Sel. Areas Commun. 18(3), 347–360 (2000).View ArticleGoogle Scholar
 O El Ayach, S Rajagopal, S AbuSurra, Z Pi, RW Heath, Spatially sparse precoding in millimeter wave MIMO systems. IEEE Trans. Wireless Commun. 13(3), 1499–1513 (2014).View ArticleGoogle Scholar
 GH Golub, CF Van Loan, Matrix Computations, vol. 3 (JHU Press, Baltimore, 2012).MATHGoogle Scholar