Training Beam Design for Channel Estimation in Hybrid mmWave MIMO Systems

Chengwen Xing; Jianping An; Lian Zhao; Wenqian Shen; Xiaochun Ge

REVIEW 3 major objections 5 minor 33 references

Training Beam Design for Channel Estimation in Hybrid mmWave MIMO Systems

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper's claim: hybrid training beams can be designed so that OMP channel estimation approaches full-digital accuracy even with 1-3 bit phase shifters.

desk verdict Useful incremental design for hybrid mmWave sensing with low-resolution phase shifters, but the headline gains rest on exact on-grid sparsity and need an off-grid test before I'd believe the full-digital equivalence. read the letter →

arxiv 2506.00913 v1 pith:BFX6C2QR submitted 2025-06-01 eess.SP

Xiaochun Ge , Wenqian Shen , Chengwen Xing , Lian Zhao , Jianping An This is my paper

classification eess.SP

keywords hybridmmWaveMIMOchannelestimationtrainingbeamdesignsensingmatrixoptimizationcompressivelow-resolutionphaseshiftersorthogonalmatchingpursuitRiemannianmanifold

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that the training beams used to sound an mmWave MIMO channel can be computed, not chosen randomly, so that the equivalent dictionary seen by a sparse-recovery estimator has low mutual coherence. It proposes alternating hybrid analog-digital algorithms that design the phase-shifter matrix on a complex-circle manifold and the digital block-diagonal matrix by convex optimization or adaptive-stepsize gradient descent, for infinite-resolution and low-resolution phase shifters. The intended payoff is that cheap hardware does not force a large estimation penalty: with infinite-resolution phase shifters the design gives orthogonal-matching-pursuit (OMP) estimation an NMSE almost identical to a full-digital system, and with 1-3 bit quantized phase shifters it outperforms direct-quantization and random beam training. A sympathetic reader would take the contribution as a generalized sensing-matrix design recipe that works when the number of training beams is smaller than the number of antennas.

What carries the argument

The carrying object is the equivalent dictionary $Q = (F_{BB}^T F_{RF}^T A_{T,G}^*) \otimes (W_{BB}^H W_{RF}^H A_{R,G})$, a $T_tT_r \times G_tG_r$ matrix whose columns are the atoms OMP chooses among; its mutual coherence $\mu(Q) = \max_{m<n} |q_m^H q_n| / (\|q_m\|_2 \|q_n\|_2)$ is the figure of merit. Since $Q$ splits as a Kronecker product of transmit and receive factors, the design reduces to shaping $W_{BB}^H W_{RF}^H A_{R,G}$ at the receiver and its counterpart at the transmitter. The algorithm family drives $A_{R,G}^H W_{RF} W_{BB} W_{BB}^H W_{RF}^H A_{R,G}$ toward $I_{G_r}$ using three components: Riemannian conjugate gradient with retraction on the complex circle manifold $M_{cc}^{N_r M_r} = \{x: |x| = 1\}$ for the constant-modulus analog part, a lifted semidefinite variable for the block-diagonal digital part in the infinite-resolution case, and block-wise quantization with monotone acceptance plus adaptive-stepsize gradient descent in the low-resolution case.

What would settle it

Run Algorithm 2 and Algorithm 3 with channel paths whose AoAs/AoDs are drawn from a continuous distribution without snapping them to the $G_t/G_r$ grid, and compare the resulting NMSE against random beams and the full-digital scheme. If the gap to full-digital sensing widens sharply, or the advantage over random and AltMin-DQ baselines shrinks or disappears as angles fall between grid points, the claimed near-full-digital performance is an artifact of the on-grid approximation.

Watch

Extended reading notes

Core claim

The central discovery being advanced is that the hybrid sensing matrix can be optimized for the mutual incoherence property even under constant-modulus and discrete-phase constraints. Writing the channel on a quantized angle grid and vectorizing the received pilots turns estimation into $y = \sqrt{P} Q \bar{h}_{a,G} + \tilde{n}$, where $Q = (F_{BB}^T F_{RF}^T A_{T,G}^*) \otimes (W_{BB}^H W_{RF}^H A_{R,G})$ is the equivalent dictionary; the paper replaces the hard coherence objective with the smooth surrogate $\|A_{R,G}^H W_{RF} W_{BB} W_{BB}^H W_{RF}^H A_{R,G} - I_{G_r}\|_F^2$, driving the normalized Gram matrix toward identity. For infinite-resolution phase shifters, alternating between a Riemannian-manifold conjugate-gradient update of $W_{RF}$ and a semidefinite-lifted convex solve for the block-diagonal $W_{BB}$ is shown numerically to reach nearly the NMSE of full-digital sensing. For low-resolution phase shifters, a block-wise variant optimizes each analog submatrix with continuous phases on the manifold, quantizes that block, and re-optimizes the digital submatrix with gradient descent, accepting the update only when the block objective decreases; the paper reports that this yields lower NMSE than AltMin-DQ and random quantized-phase baselines at 1-3 bits, and spectral efficiency after hybrid precoding close to the full-digital benchmark.

Load-bearing premise

The load-bearing premise is that the true channel is exactly sparse on the predefined quantized angle grid, because Eq. (8) replaces the channel by $A_{R,G} \bar{H}_{a,G} A_{T,G}^H$ and says the grid quantization error is tentatively ignored; real arrival and departure angles are continuous, so off-grid paths create basis mismatch that OMP cannot fully correct.

Editorial extensions

If this is right

With infinite-resolution phase shifters, the proposed beam design brings OMP channel estimation NMSE close to that of a full-digital system, while using fewer training beams than antennas.
With 1-3 bit phase shifters, the block-wise alternating design gives lower NMSE than AltMin-DQ and random quantized beams, with performance improving monotonically as the number of bits grows.
Channels estimated with the designed beams yield higher spectral efficiency after hybrid precoding than those from baseline training beams, and the infinite-resolution design approaches the full-digital precoding spectral efficiency.
The block-wise acceptance rule keeps the objective function from increasing despite phase quantization, so the low-resolution algorithm converges in the simulated settings.
The design remains functional in partial-training regimes ($T_t < N_t$, $T_r < N_r$) and in the generalized $N_s \le N_{RF}$ case where earlier square/DFT-constrained designs do not apply.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Beyond the paper: the same manifold-optimization plus block-wise quantization recipe should transfer to other constant-modulus discrete-phase design problems, such as intelligent-reflecting-surface phase profiles or wideband beam training, wherever the objective is coherence-based.
Beyond the paper: because coherence is a worst-case proxy rather than an average-case one, the optimized dictionaries are likely to benefit other sparse estimators such as basis pursuit or iterative hard thresholding, although the paper only demonstrates OMP.
Beyond the paper: the on-grid assumption suggests a direct extension: alternate the proposed sensing-matrix design with an off-grid refinement of the active angles, which could preserve the performance gain in realistic continuous-angle channels.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

This is a solid, incremental design paper for hybrid mmWave sensing, and the main thing I'd press on is one assumption the authors openly flag but never test. They approximate the channel as exactly sparse on a quantized angle grid (Eq. 8) and then draw all simulated AoAs/AoDs on that same grid. Real angles are continuous, so their optimized sensing matrices, which are tailored to the grid dictionary, may lose their advantage under basis mismatch; the near-full-digital NMSE claim in Fig. 5 would then be overstated. That is the load-bearing soft spot.

What is genuinely new: they handle partial training (Tt < Nt, Tr < Nr) with flexible 1-3 bit phase shifters via a block-wise alternating algorithm, which I don't think appears in [29]/[30]; and they design the block-diagonal digital combiner jointly, then per-block with quantization. The Riemannian gradient derivations are spelled out, complexity orders are given, and the baseline comparisons (AltMin, AltMin-DQ, Random, full-digital) are the right ones. The simulation results are consistent and show monotonic gains with PS resolution. No constants were fitted to the reported NMSE, so circularity is low.

Softer issues, in order. First, the off-grid concern above is real and should be testable: just run the same algorithms with off-grid paths and report NMSE. Second, Fig. 3's caption says Nt=32, Nr=16 while the rest of the paper uses Nt=64, Nr=32; that looks like a typo but needs fixing. Third, there are no error bars or code, so I can't tell how much of the gap is luck. Fourth, the SDP in (23) is solved, then eigen-truncated with no quantified error, and the Gram-Frobenius surrogate is a known heuristic; both are standard practice, so I wouldn't call them fatal. Algorithm 3 lacks a formal convergence proof, but the acceptance step in Step 13-15 guarantees the objective does not increase, which is reasonable.

This paper deserves a serious referee. The problem is relevant to people working on hybrid beamforming and channel estimation, the algorithms are concrete, and the evaluation is fair to the baselines. The reviewer should push for off-grid simulations and a cleaner Fig. 3. If that comes back OK, it's a publishable contribution to mmWave channel estimation.

Referee Report

3 major / 5 minor

Summary. The paper considers training beam design for channel estimation in hybrid analog-digital mmWave MIMO systems. It formulates the problem as compressive-sensing sensing-matrix design: the hybrid combiner and precoder are optimized so that the equivalent dictionary Q in Eq. (15) has low mutual coherence, using the Frobenius-norm deviation of the Gram matrix from the identity as a surrogate. For infinite-resolution phase shifters, Algorithm 2 alternates between a convex SDP for the block-diagonal digital combiner, Eqs. (23)-(24), and a Riemannian-manifold gradient-descent update for the analog combiner, Algorithm 1. For low-resolution phase shifters, Algorithm 3 applies a block-wise alternating optimization with phase quantization and an acceptance test. Orthogonal matching pursuit is used for channel recovery. Simulations compare NMSE, spectral efficiency, and Gram-coherence histograms against full-digital, AltMin, AltMin-DQ, and random sensing schemes, reporting near-full-digital NMSE performance for infinite-resolution PSs and improved NMSE and spectral efficiency for 1-3 bit PSs.

Significance. If the reported results hold, the paper makes a useful contribution: it demonstrates numerically that optimized hybrid training beams with low-resolution phase shifters can approach full-digital channel-estimation performance, and it extends prior hybrid sensing designs to flexible numbers of training beams and partial-training regimes. The algorithms are specified in detail, including gradients, stepsize rules, quantization steps, and complexity estimates, and the simulation study is broader than in several previous works. The strength of the central claim is, however, conditional: the simulations draw AoAs/AoDs exactly on the quantized angle grid used in the dictionary, so the grid-mismatch approximation that the paper itself flags in Eq. (8) is never stress-tested. In addition, the optimization pipeline contains a rank-relaxed SDP followed by eigen-truncation without a quantified error bound, and the connection between the unnormalized Gram objective and the normalized-coherence histograms deserves clarification.

major comments (3)

[Section VI-A and Eq. (8)] The paper states in Section II-C that 'the grid quantization error is tentatively ignored here' in Eq. (8), but Section VI-A draws AoAs/AoDs exactly from the quantized grid defined by cos(theta)=2(g-1)/G-1. Consequently, all NMSE and spectral-efficiency results, including the near-full-digital behavior in Fig. 5, are obtained under a perfectly on-grid channel model. Since the proposed algorithms optimize coherence with respect to this same fixed grid dictionary, while random sensing matrices are more universal by RIP-type arguments, the claimed advantages over AltMin-DQ and Random may shrink or disappear for off-grid paths. This is a load-bearing concern because the abstract and conclusions assert near-full-digital performance and significant gains over baselines. Please add simulations with continuous-valued or perturbed AoAs/AoDs, or with a separate test grid, and report the resulting NMSE and spectral efficiency; if the gains erode, the claims should be tempered accordingly.
[Section IV-A, Eqs. (23)-(24)] The convex problem (23) drops the rank constraint rank(X_BB,q) <= N_s that is implicit in the substitution X_BB,q = W_BB,q W_BB,q^H, and Eq. (24) then replaces the optimal X_BB,q by its best rank-N_s eigen-truncation. No bound is given for the increase in the original objective (20a) caused by this truncation, so the outer iterations of Algorithm 2 are not guaranteed to be non-increasing in the original objective. Please provide an error bound, a modified projection step that ensures descent, or a numerical check showing that the original objective is non-increasing at every outer iteration.
[Section III, Eq. (19)] The text says that 'by normalizing the columns of the equivalent dictionary matrix... then we can obtain the normalized Gram matrix', but the optimization objective in Eq. (19) is written with the unnormalized Gram matrix A_R,G^H W_RF W_BB W_BB^H W_RF^H A_R,G - I, and Algorithms 2 and 3 apply only a global power scaling (Step 8 of Algorithm 2 and Step 18 of Algorithm 3) rather than a per-column normalization. Because the Frobenius surrogate is a coherence proxy only for unit-norm columns, please clarify whether and where column normalization is enforced, and how the optimized objective relates to the normalized-Gram histograms reported in Fig. 4.

minor comments (5)

[Fig. 3 caption] The Fig. 3 caption refers to methods [27] and [28], while the text of Section VI-A compares with [29] and [30]; please reconcile these references.
[Throughout] There are several typographical errors, including 'milimeter' in the abstract, 'Hermittian' in Section V-A, 'diagnonal' in the Fig. 4 caption, and 'spectral frequency' in Fig. 6 and its caption, which should read 'spectral efficiency'.
[Section VI-A] The phrase 'AoAs/AoDs are assumed to be non-uniformly distributed in [0,pi] as defined in (8)' is ambiguous; since Eq. (8) defines a grid that is uniform in cosine, please state explicitly whether the angles are drawn uniformly from the quantized grid indices.
[Section V-A, Eqs. (44)-(45)] The expansion in Eq. (44) is long and difficult to verify; please double-check the terms involving Gamma_4 and consider moving the detailed expansion to an appendix or suppressing it in favor of a compact expression.
[Algorithms 1-3] The stopping criteria are described as 'a stopping criterion triggers' or 'no longer decreases', but no tolerance or maximum-iteration count is specified; please add practical termination parameters so that the algorithms can be reproduced exactly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the sensing matrices are optimized against a Gram-coherence surrogate and the central NMSE/rate claims are produced by an external OMP estimator, not fitted from the reported curves.

full rationale

The paper's derivation chain is self-contained rather than circular. The design objective, Eq. (19a), is the Frobenius-norm mismatch between the normalized Gram matrix of the hybrid dictionary W_H_BB W_H_RF A_R,G and the identity; the reported performance metric, NMSE, is obtained by applying the OMP-based estimator of [30] to synthetic channels. No constant appearing in the NMSE or spectral-efficiency plots is inserted into the sensing-matrix optimization, and the proposed algorithms are not evaluated by reusing their own optimized objective as the recovery metric. Flagged limitation: Section II-C states that for Eq. (8) "the grid quantization error is tentatively ignored here," and Section VI-A draws AoAs/AoDs exactly on that quantized grid, so off-grid basis mismatch is never tested; this is a robustness/correctness concern that could erode the claimed gains, but it is not a circularity, because the on-grid channel realizations are not used to fit the reported NMSE curves. The closest thing to a self-reference is Fig. 7 and Fig. 8, which plot the very objective functions (37a) and (49) being optimized; these figures therefore act as consistency checks of the algorithms' convergence behavior rather than independent validation, while the independent evidence in Fig. 5 and Fig. 6 stands separately. No load-bearing self-citation or imported uniqueness theorem appears: the cited prior work [24], [25], [30], [12] supplies generic tools and baselines, not the paper's conclusion.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central algorithm uses no invented physical entities and no data-fitted constants; it relies on standard compressive sensing dictionary-sparsity assumptions and a Gram-coherence surrogate. The two most load-bearing inputs from outside the paper are the exact on-grid sparsity model and the surrogate objective linking the design to OMP performance.

assumptions (5)

domain assumption The true mmWave channel is exactly sparse on the predefined quantized angle grid: H approximately equals A_R,G H_a,G A_T,G^H with grid quantization error ignored.
Invoked in Section II-C, Eq. (8); the dictionaries A_T,G and A_R,G are built from quantized angles, and off-grid paths are not modeled.
domain assumption Minimizing the Frobenius norm of the Gram matrix minus identity is a valid surrogate for reducing mutual coherence and therefore for improving OMP recovery.
Used to turn problem (17) into (19), following [17] and [22]; worst-case coherence is replaced by an average Gram-distance objective.
domain assumption The mmWave channel is sparse enough for compressive sensing recovery with TtTr less than GtGr, with L dominant paths in the simulation.
Section II-B, Eq. (4); the underdetermined formulation in (13) relies on this sparsity.
standard math OMP recovers the sparse channel coefficients when the equivalent dictionary Q has low mutual coherence, per the mutual incoherence property.
Sections II-C and III; standard compressive sensing result used to justify designing Q.
standard math The Kronecker-product factorization and coherence equality in Eqs. (15) and (16) hold for the constructed Q.
Uses the identity (AB) tensor (CD) equals (A tensor C)(B tensor D) and the max-coherence property for Kronecker products; standard algebra.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Training Beam Design for Channel Estimation in Hybrid mmWave MIMO Systems." pith.science (2026). https://pith.science/paper/BFX6C2QR

@misc{pith2026250600913,
  author       = {Pith},
  title        = {Pith review of: Training Beam Design for Channel Estimation in Hybrid mmWave MIMO Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BFX6C2QR}},
  note         = {Machine review of arXiv:2506.00913}
}

read the original abstract

Training beam design for channel estimation with infinite-resolution and low-resolution phase shifters (PSs) in hybrid analog-digital milimeter wave (mmWave) massive multiple-input multiple-output (MIMO) systems is considered in this paper. By exploiting the sparsity of mmWave channels, the optimization of the sensing matrices (corresponding to training beams) is formulated according to the compressive sensing (CS) theory. Under the condition of infinite-resolution PSs, we propose relevant algorithms to construct the sensing matrix, where the theory of convex optimization and the gradient descent in Riemannian manifold is used to design the digital and analog part, respectively. Furthermore, a block-wise alternating hybrid analog-digital algorithm is proposed to tackle the design of training beams with low-resolution PSs, where the performance degeneration caused by non-convex constant modulus and discrete phase constraints is effectively compensated to some extent thanks to the iterations among blocks. Finally, the orthogonal matching pursuit (OMP) based estimator is adopted for achieving an effective recovery of the sparse mmWave channel. Simulation results demonstrate the performance advantages of proposed algorithms compared with some existing schemes.

Figures

Figures reproduced from arXiv: 2506.00913 by the authors.

**Figure 2.** The process of iterative optimization on a Riemannia [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** NMSE performance against PNR using several types of tr [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Histograms of the magnitudes of off-diagonal entrie [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

**Figure 5.** Figure 5: NMSE performance against PNR using six types of train [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

**Figure 7.** Figure 7: Value of objective function against the number of alt [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

**Figure 8.** Figure 8: Value of objective function against the number of tra [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

33 extracted references · 31 canonical work pages

[29]

Exploiting spatial sparsit y for estimating channels of hybrid MIMO systems in millimeter wave communic ations,

J. Lee, G. Gil and Y . H. Lee, “Exploiting spatial sparsit y for estimating channels of hybrid MIMO systems in millimeter wave communic ations,” in IEEE Globecom 2014 (2014) , Austin, TX USA, Dec. 2014, pp. 3326– 3331

work page 2014
[30]

Channel estimation via orth ogonal matching pursuit for hybrid MIMO systems in millimeter wave com- munications,

J. Lee, G. Gil and Y . H. Lee, “Channel estimation via orth ogonal matching pursuit for hybrid MIMO systems in millimeter wave com- munications,” IEEE Trans. Commun. , vol. 64, no. 6, pp. 2370–2386, Jun. 2016

work page 2016
[1]

Massive MIMO for next generation wireless systems,

E. G. Larsson, O. Edfors, F. Tufvesson and T. L. Marzetta, “Massive MIMO for next generation wireless systems,” IEEE Commun. Mag. , vol. 52, no. 2, pp. 186–195, Feb. 2014

2014
[2]

Millimet er wave mobile communications for 5G cellular: It will work!

T. S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y . Azar, K. Wan g, G. N. Wong, J. K. Schulz, M. Samimi, and F. Gutierrez, “Millimet er wave mobile communications for 5G cellular: It will work!” IEEE Access , vol. 1, pp. 335–349, 2013

work page 2013
[3]

An overview of signal processing techniques for mi llimeter wave MIMO systems,

R. W. Heath, N. Gonz´ alez-Prelcic, S. Rangan, W. Roh and A . M. Sayeed, “An overview of signal processing techniques for mi llimeter wave MIMO systems,” IEEE J. Sel. Topics Signal Process. , vol. 10, no. 3, pp. 436–453, Apr. 2016

work page 2016
[4]

Broadband millimeter-wave propagation me asurements and models using adaptive-beam antennas for outdoor urban c ellular communications,

T. S. Rappaport, F. Gutierrez, E. Ben-Dor, J. N. Murdock, Y . Qiao and J. I. Tamir, “Broadband millimeter-wave propagation me asurements and models using adaptive-beam antennas for outdoor urban c ellular communications,” IEEE Trans. Antennas Propag. , vol. 61, no. 4, pp. 1850–1859, Apr. 2013

work page 2013
[5]

Low RF-complexity techno logies to enable millimeter-wave MIMO with large antenna array for 5G wireless communications,

X. Gao, L. Dai and A. M. Sayeed, “Low RF-complexity techno logies to enable millimeter-wave MIMO with large antenna array for 5G wireless communications,” IEEE Trans. Wireless Commun. , vol. 56, no. 4, pp. 211–217, Apr. 2018

work page 2018
[6]

Scaling up MIMO: Opportunities and chall enges with very large arrays,

F. Rusek, D. Persson, B. Lau, E. G. Larsson, T. L. Marzetta , O. Edfors and F. Tufvesson, “Scaling up MIMO: Opportunities and chall enges with very large arrays,” IEEE Commun. Mag. , vol. 30, no. 1, pp. 40–60, Jan. 2013

work page 2013

Show all 33 references

[7]

An introduction to millimeter-wave mo bile broad- band systems,

Z. Pi and F. Khan, “An introduction to millimeter-wave mo bile broad- band systems,” IEEE Commun. Mag. , vol. 49, no. 6, pp. 101–107, Jun. 2011. 14

2011
[8]

A framework o n hybrid MIMO transceiver design based on matrix-monotonic o ptimiza- tion,

C. Xing, X. Zhao, W. Xu, X. Dong, and G. Y . Li, “A framework o n hybrid MIMO transceiver design based on matrix-monotonic o ptimiza- tion,” IEEE Trans. Signal Process. , vol. 67, no. 13, pp. 3531–3546, Jul. 2019

2019
[9]

Majori zation- minimization aided hybrid transceivers for MIMO interfere nce chan- nels,

S. Gong, C. Xing, V . K. N. Lau, S. Chen and L. Hanzo, “Majori zation- minimization aided hybrid transceivers for MIMO interfere nce chan- nels,” IEEE Trans. Signal Process. , vol. 68, pp. 4903–4918, Aug. 2020

2020
[10]

Trai ning optimization for hybrid MIMO communication systems,

C. Xing, D. Liu, S. Gong, W. Xu, S. Chen and L. Hanzo, “Trai ning optimization for hybrid MIMO communication systems,” IEEE Trans. Wireless Commun., vol. 19, no. 8, pp. 5473–5487, Aug. 2020

2020
[11]

Hybrid MIMO architectures for millimeter wave comm unica- tions: Phase shifters or switches?

R. Mendez-Rial, C. Rusu, N. Gonz´ alez-Prelcic, A. Alkh ateeb and R. W. Heath, “Hybrid MIMO architectures for millimeter wave comm unica- tions: Phase shifters or switches?” IEEE Access , vol. 4, pp. 247–267, 2016

2016
[12]

Hybrid digital and analog beamfor ming design for large-scale antenna arrays,

F. Sohrabi and W. Y u, “Hybrid digital and analog beamfor ming design for large-scale antenna arrays,” IEEE J. Sel. Topics Signal Process. , vol. 10, no. 3, pp. 501–513, Apr. 2016

2016
[13]

Spatially sparse precoding in millimeter wave MIMO system s,

O. E. Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi and R. W. He ath, “Spatially sparse precoding in millimeter wave MIMO system s,” IEEE Trans. Wireless Commun. , vol. 13, no. 3, pp. 1499–1513, Mar. 2014

2014
[14]

Hybrid beamforming with discrete phase shift ers for millimeter-wave massive MIMO systems,

J. Chen, “Hybrid beamforming with discrete phase shift ers for millimeter-wave massive MIMO systems,” IEEE Trans. V eh. Technol. , vol. 66, no. 8, pp. 7604–7608, Aug. 2017

2017
[15]

Hybrid pre coder and combiner design with low-resolution phase shifters in mmWa ve MIMO systems,

Z. Wang, M. Li, Q. Liu and A. L. Swindlehurst, “Hybrid pre coder and combiner design with low-resolution phase shifters in mmWa ve MIMO systems,” IEEE J. Sel. Topics Signal Process. , vol. 12, no. 2, pp. 256– 269, May 2018

2018
[16]

Chann el estimation and hybrid precoding for millimeter wave cellul ar systems,

A. Alkhateeb, O. E. Ayach, G. Leus and R. W. Heath, “Chann el estimation and hybrid precoding for millimeter wave cellul ar systems,” IEEE J. Sel. Topics Signal Process. , vol. 8, no. 5, pp. 831–846, Oct. 2014

2014
[17]

Simon and R

F. Simon and R. Holger, A mathematical introduction to compressive sensing. Birkh¨ auser Press, 2013

2013
[18]

Greed is good: Algorithmic results for spa rse approxima- tion,

J. A. Tropp, “Greed is good: Algorithmic results for spa rse approxima- tion,” IEEE Trans. Inf. Theory , vol. 50, no. 10, pp. 2231–2242, Oct. 2004

2004
[19]

Orthogonal matching pursuit for s parse signal recovery with noise,

T. T. Cai and L. Wang, “Orthogonal matching pursuit for s parse signal recovery with noise,” IEEE Trans. Inf. Theory , vol. 57, no. 7, pp. 4680– 4688, Jul. 2011

2011
[20]

Learning to sen se sparse signals: Simultaneous sensing matrix and sparsifying dictionary op timization,

J. M. Duarte-Carvajalino and G. Sapir, “Learning to sen se sparse signals: Simultaneous sensing matrix and sparsifying dictionary op timization,” IEEE Trans. Image Process. , vol. 18, no. 7, pp. 1395–1408, Jul. 2009

2009
[21]

Alterna ting optimization of sensing matrix and sparsifying dictionary for compressed sensing,

H. Bai, G. Li, S. Li, Q. Li, Q. Jiang and L. Chang, “Alterna ting optimization of sensing matrix and sparsifying dictionary for compressed sensing,” IEEE Trans. Signal Process. , vol. 63, no. 6, pp. 1581–1594, Mar. 2015

2015
[22]

On projection matrix optimization for compressive sensing systems,

G. Li, Z. Zhu, D. Y ang, L. Chang and H. Bai, “On projection matrix optimization for compressive sensing systems,” IEEE Trans. Signal Process., vol. 61, no. 11, pp. 2887–2898, Jun. 2013

2013
[23]

Sensing matrix optimization for block-sparse decoding,

L. Zelnik-Manor, K. Rosenblum and Y . C. Eldar, “Sensing matrix optimization for block-sparse decoding,” IEEE Trans. Signal Process. , vol. 59, no. 9, pp. 4300–4312, Sept. 2011

2011
[24]

Alternating m inimization algorithms for hybrid precoding in millimeter wave MIMO sys tems,

X. Y u, J. Shen, J. Zhang and K. B. Letaief, “Alternating m inimization algorithms for hybrid precoding in millimeter wave MIMO sys tems,” IEEE J. Sel. Topics Signal Process. , vol. 10, no. 3, pp. 485–500, Apr. 2016

2016
[25]

Absil, R

P .-A. Absil, R. Mahony, and R. Sepulchre, Optimization algorithms on matrix manifolds . Princeton, NJ, USA: Princeton Univ. Press, 2009

2009
[26]

Ma and Y

Y . Ma and Y . Fu, Manifold learning theory and applications . Boca Raton, FL, USA: CRC Press, 2012

2012
[27]

Low-rank matrix com pletion via Riemannian pursuit for topological interference manageme nt,

Y . Shi, J. Zhang, and K. B. Letaief, “Low-rank matrix com pletion via Riemannian pursuit for topological interference manageme nt,” in IEEE Int. Symp. Inf. Theory (ISIT) , Hong Kong, Jun. 2015, pp. 1831–1835

2015
[28]

Ma nopt, a Matlab toolbox for optimization on manifolds,

N. Boumal, B. Mishra, P .-A. Absil, and R. Sepulchre, “Ma nopt, a Matlab toolbox for optimization on manifolds,” J. Mach. Learn. Res. , vol. 15, pp. 1455–1459, Jan. 2014

2014
[31]

Channel estima tion using low-resolution PSs for wideband mmWave systems,

Y . Dong, C. Chen, N. Yi, G. Lu, and Y . Jin, “Channel estima tion using low-resolution PSs for wideband mmWave systems,” in Proc. VTC Spring, Sydney, NSW, Australia, Jun. 2017, pp. 1–5

2017
[32]

Mumtaz, J

S. Mumtaz, J. Rodriguez and L. Dai, MmW ave massive MIMO: A paradigm for 5G . London Wall, LON, UK: Academic Press, 2017

2017
[33]

A gradient- based alternat- ing minimization approach for optimization of the measurem ent matrix in compressive sensing,

V . Abolghasemi, S. Ferdowsi, and S. Sanei, “A gradient- based alternat- ing minimization approach for optimization of the measurem ent matrix in compressive sensing,” Signal Process. , vol. 92, pp. 999–1009, 2012. Xiaochun Ge received the B.S. degree in com- munication engine...

2012

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [29]

Exploiting spatial sparsit y for estimating channels of hybrid MIMO systems in millimeter wave communic ations,

J. Lee, G. Gil and Y . H. Lee, “Exploiting spatial sparsit y for estimating channels of hybrid MIMO systems in millimeter wave communic ations,” in IEEE Globecom 2014 (2014) , Austin, TX USA, Dec. 2014, pp. 3326– 3331

work page 2014

[2] [30]

Channel estimation via orth ogonal matching pursuit for hybrid MIMO systems in millimeter wave com- munications,

J. Lee, G. Gil and Y . H. Lee, “Channel estimation via orth ogonal matching pursuit for hybrid MIMO systems in millimeter wave com- munications,” IEEE Trans. Commun. , vol. 64, no. 6, pp. 2370–2386, Jun. 2016

work page 2016

[3] [1]

Massive MIMO for next generation wireless systems,

E. G. Larsson, O. Edfors, F. Tufvesson and T. L. Marzetta, “Massive MIMO for next generation wireless systems,” IEEE Commun. Mag. , vol. 52, no. 2, pp. 186–195, Feb. 2014

2014

[4] [2]

Millimet er wave mobile communications for 5G cellular: It will work!

T. S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y . Azar, K. Wan g, G. N. Wong, J. K. Schulz, M. Samimi, and F. Gutierrez, “Millimet er wave mobile communications for 5G cellular: It will work!” IEEE Access , vol. 1, pp. 335–349, 2013

work page 2013

[5] [3]

An overview of signal processing techniques for mi llimeter wave MIMO systems,

R. W. Heath, N. Gonz´ alez-Prelcic, S. Rangan, W. Roh and A . M. Sayeed, “An overview of signal processing techniques for mi llimeter wave MIMO systems,” IEEE J. Sel. Topics Signal Process. , vol. 10, no. 3, pp. 436–453, Apr. 2016

work page 2016

[6] [4]

Broadband millimeter-wave propagation me asurements and models using adaptive-beam antennas for outdoor urban c ellular communications,

T. S. Rappaport, F. Gutierrez, E. Ben-Dor, J. N. Murdock, Y . Qiao and J. I. Tamir, “Broadband millimeter-wave propagation me asurements and models using adaptive-beam antennas for outdoor urban c ellular communications,” IEEE Trans. Antennas Propag. , vol. 61, no. 4, pp. 1850–1859, Apr. 2013

work page 2013

[7] [5]

Low RF-complexity techno logies to enable millimeter-wave MIMO with large antenna array for 5G wireless communications,

X. Gao, L. Dai and A. M. Sayeed, “Low RF-complexity techno logies to enable millimeter-wave MIMO with large antenna array for 5G wireless communications,” IEEE Trans. Wireless Commun. , vol. 56, no. 4, pp. 211–217, Apr. 2018

work page 2018

[8] [6]

Scaling up MIMO: Opportunities and chall enges with very large arrays,

F. Rusek, D. Persson, B. Lau, E. G. Larsson, T. L. Marzetta , O. Edfors and F. Tufvesson, “Scaling up MIMO: Opportunities and chall enges with very large arrays,” IEEE Commun. Mag. , vol. 30, no. 1, pp. 40–60, Jan. 2013

work page 2013