Variational Bayesian Inference for Time-Varying Massive MIMO Channels: Estimation and Detection

Duy H. N. Nguyen; Sajjad Nassirpour; Toan-Van Nguyen

REVIEW 3 major objections 5 minor 44 references

Variational Bayesian Inference for Time-Varying Massive MIMO Channels: Estimation and Detection

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A variational Bayesian receiver jointly estimates fast-varying massive MIMO channels and data without knowing the users' time-correlation coefficients or noise variance, and outperforms LMMSE, Kalman, and EP baselines.

desk verdict Plausible VB-based JED receiver for time-varying massive MIMO, but the 'tracks unknown correlation' claim needs robustness tests beyond one favorable initialization. read the letter →

arxiv 2505.10673 v1 pith:DH5NHOEY submitted 2025-05-15 eess.SP

Sajjad Nassirpour , Toan-Van Nguyen , Duy H. N. Nguyen This is my paper · ORCID

classification eess.SP

keywords variationalBayesianinferencemassiveMIMOtime-varyingchannelsjointchannelestimationanddatadetectionunknowntimecorrelationGauss-Markovmodelsymbolerrorratenormalizedmeansquared

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

This paper addresses the problem of keeping accurate channel state information at a massive MIMO base station serving fast-moving users, whose channels change every symbol period. It proposes a variational Bayesian (VB) inference framework that jointly estimates the channel and detects data, treating as unknown two quantities that competing methods assume known: the noise variance and each user's time-correlation coefficient $\eta_i$ in a first-order Gauss-Markov channel model. The paper develops an online processing strategy that uses only current-slot statistics and cites per-slot complexity $O(I_{\mathrm{tr}}(M^3K+|\mathcal{S}|K))$, and a block processing strategy that stacks all received signals to reduce channel estimation error. Simulations claim both strategies track $\eta_i$, yield nearly the same symbol error rate as if $\eta_i$ were known, and outperform LMMSE, Kalman filter, and expectation propagation in symbol error rate and channel normalized mean squared error.

What carries the argument

The machinery is the coordinate-ascent variational Bayesian (CAVI) loop over a mean-field factorization of the posterior into independent variational densities for symbols, channels, correlation coefficients, and noise precision. Each latent variable is updated by taking the expectation of the log-joint distribution with respect to the other variables; because the priors are conjugate, every update is closed form. The two equations carrying the argument are the $\eta_i$ update, where the Gaussian variational posterior for the correlation coefficient has variance and mean given by the quadratic-form terms involving the estimated channel, and the predictive-covariance update, which uses $E[\eta_i^2]=\hat{\eta}_{i,t-1}^2+\tau^{\eta}_{i,t-1}$ to propagate correlation uncertainty into the channel prior. Lemma 1 supplies the quadratic-form expectations that make the symbol and noise-precision updates tractable.

What would settle it

Run the online algorithm on a simulated Gauss-Markov channel with true $\eta_i = 0.5$, initializing $\hat{\eta}_{i,0}=0.95$ with prior variance $10^{-3}$, and record the trajectory of $\langle\eta_i\rangle$ plus the mass of the variational density outside $[0,1]$. If $\langle\eta_i\rangle$ leaves $[0,1]$ or the symbol error rate no longer matches the known-$\eta_i$ oracle, the untruncated Gaussian assumption is the failure point; if it stays inside and tracks, the assumption is adequate.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that a mean-field variational approximation with conjugate priors is enough to make joint channel estimation and data detection work under first-order Gauss-Markov time variation without outside knowledge of the channel dynamics. The online coordinate-ascent updates produce closed-form Gaussian posteriors for channels and for each $\eta_i$, a Gamma posterior for noise precision, and a discrete posterior for symbols; the predictive covariance update replaces $\eta_i^2$ with its second moment so that uncertainty about the correlation coefficient is carried into the channel prior. The block version adds a Gamma variable $\nu_i=(1-\eta_i^2)^{-1}$ and couples adjacent time slots in each channel update. The reported consequence is that per-user correlation coefficients can be tracked from the received signal alone, and the resulting symbol error rate is nearly identical to the oracle case where $\eta_i$ is known, while channel normalized mean squared error is the lowest among the compared methods.

Load-bearing premise

The load-bearing premise is that the unknown time-correlation coefficient $\eta_i$, which by definition lies in $[0,1]$, can be represented by an untruncated Gaussian variational distribution whose mean is initialized close to the true value; if the variational mean drifts outside $[0,1]$ during iterations, the channel prediction and all subsequent updates rest on an invalid correlation coefficient.

Editorial extensions

If this is right

A base station can run a joint estimation and detection receiver for high-mobility users without estimating Doppler frequency or noise variance in advance, because the online VB loop learns both from the received signal.
The online strategy's per-slot complexity makes it a candidate for latency-sensitive uplinks where Kalman-filter and expectation-propagation updates would be computationally heavier.
When the correlation coefficient drifts slowly with time, the random-variable variant of $\eta_i$ shows that the framework can follow Doppler changes rather than assuming a fixed coefficient.
The block strategy trades delay for channel accuracy, substantially lowering channel normalized mean squared error at high SNR, which matters for downlink beamforming that reuses uplink channel estimates.
The interleaved structure divides pilots across the communication block, giving a practical way to stop error propagation in long data frames.

Reading between the lines

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

The untruncated Gaussian prior on $\eta_i$ is the point most worth stress-testing: replacing it with a truncated Gaussian or a Beta prior should preserve closed-form CAVI updates while removing the need for a carefully chosen initialization, turning the reported tracking behavior into a general property rather than an initialization-dependent one.
The same VB treatment could be extended to estimate the spatial correlation matrix $R_i$ or the Doppler spread itself, since the framework already estimates scalar covariance parameters; the block update for $\nu_i$ shows how a related parameter can be inferred from the whole frame.
The interleaved structure suggests an adaptive pilot-insertion rule in which new pilot blocks are scheduled whenever the estimated $\eta_i$ drifts enough that prediction error would grow; this is a testable extension the paper does not pursue.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

Short version: this is a legitimate extension of the same group's variational Bayesian MIMO detector to time-varying channels, with solid derivations, a real complexity win, and simulations that beat LMMSE, KF, and EP. The weak point is the treatment of the time-correlation coefficient eta_i, and I think the paper needs a robustness pass before I'd trust the 'tracks unknown eta' headline.

What's actually new: applying VB-JED to a first-order Gauss-Markov channel with unknown eta_i and noise precision, in both online and block forms. The CAVI updates look internally consistent; I checked the expansions around (18)-(23) and they match the mean-field model. The complexity claim O(Itr(M^3K+|S|K)) versus EP's O(Itr(M^3K^3+...)) is credible and useful. The simulation comparisons are fair to the baselines, and the interleaved structure is a sensible way to stop error propagation.

The soft spots are real but not fatal. The eta_i prior is Gaussian on a parameter that lives in [0,1]; Remark 1 acknowledges this but only resets the initial value. Nothing in the CAVI updates prevents the variational mean or variance from drifting outside the valid range during iterations. If hat_eta^2 + tau_eta > 1, the predictive covariance in (15) has a negative coefficient on R_i—that's an invalid covariance and the rest of the updates are built on sand. All experiments start at hat_eta_0 = 0.95 with tau = 1e-3, against true eta's of 0.97–0.985. So the 'nearly identical to known eta' result may just be a favorable initialization. They don't test a distant start or a truncated/transformed parameterization. The block-processing independence of nu_i and eta_i is also an unproven approximation, though their simulations do suggest it works. Minor point: Algorithm 1 doesn't explicitly carry the variational variance of eta_i to the next slot, and there's no code or error bars.

If I were refereeing, I'd ask for robustness experiments (init eta_0 = 0.5, 0.7), a truncated Gaussian or logit parameterization, and a check that the predictive covariance stays positive semidefinite. The core idea is sound and the paper deserves referee time; it's just not ready to accept as-is.

Referee Report

3 major / 5 minor

Summary. The paper develops variational Bayesian (VB) inference for joint channel estimation and data detection (JED) in an uplink massive MIMO system with high-mobility users and time-varying channels modeled by a first-order Gauss-Markov process. The time-correlation coefficients η_i and the noise precision γ_t are treated as unknown random variables. Two processing strategies are proposed: an online strategy with a prediction phase and an estimation phase, optionally enhanced by an interleaved pilot/data structure, and a block strategy that processes all received signals jointly and introduces a variational parameter ν_i for the innovation covariance. The authors compare their methods against LMMSE, Kalman filtering, and expectation propagation in terms of symbol error rate and channel NMSE, and claim that the VB framework outperforms these benchmarks while tracking unknown η_i nearly as well as when η_i is known.

Significance. If the reported results hold, the paper offers a practically useful receiver for fast-moving users: the online VB strategy has lower complexity than KF and EP, does not require a priori knowledge of the noise variance or the time-correlation coefficients, and the block strategy improves channel NMSE at the cost of delay and complexity. The comparisons are made against external baselines rather than fitted to the algorithm's own outputs, so the central claim is not circular. The main significance risk is that the unknown-η tracking claim rests on an untruncated Gaussian model for a parameter supported on [0,1] and on a favorable initialization in all experiments; without a robustness study, the claimed near-equality with the known-η case is conditional rather than established.

major comments (3)

[Section III, Remark 1, Eq. (15), Eqs. (22)-(23), Algorithm 1] The variational family for η_i is an untruncated Gaussian supported on the whole real line even though η_i must lie in [0,1]. Remark 1 acknowledges this and says the initial value is reset if it falls outside the range, but nothing in the CAVI updates (22)-(23) or in Algorithm 1 constrains the variational mean or variance during iterations. If an iterate satisfies η̂_i^2 + τ_i^η > 1, the coefficient (1 - η̂_i^2 - τ_i^η) in Eq. (15) becomes negative and Σ̂_{t|t-1} is no longer a valid covariance matrix; the channel prediction in (13) and the subsequent updates (19)-(20) are then built from an invalid object. All reported experiments initialize η̂_{i,0}=0.95, τ_{i,0}^η=10^{-3} with true η_i equal to 0.97, 0.985, or N(0.97,5×10^{-5}), i.e., within 0.02-0.035 of the true value. No experiment with a distant initialization (e.g., η̂_{i,0}=0.7 or 0.5) is reported, so the claimed near-identical performance to the known-η case may be an artifact of the favorable initialization rather than a general property of the variational estimator. Please either reparameterize η_i to respect its support (e.g., a probit or logistic transformation, or a truncated Gaussian) or provide robustness experiments with distant initializations and explicitly verify that the predictive covariance in (15) remains positive definite at every iteration.
[Section IV-B, Eqs. (42)-(51), Algorithm 2] The block processing strategy assumes ν_i=(1-η_i^2)^{-1} and η_i are independent in the mean-field factorization q(η_i)q(ν_i), yet the text states the goal of estimating ν_i such that ⟨ν_i⟩=(1-⟨η_i⟩^2)^{-1}. This is internally inconsistent: under q(η_i)q(ν_i), the deterministic relation between ν_i and η_i is not preserved, and the update (51) uses ⟨η_i⟩ and ν_i as if they were independent quantities. The paper says that simulation results validate this assumption, but no experiment isolating the accuracy of this independence approximation is shown. Please either derive the block updates from a consistent hierarchical model in which ν_i is a separate latent variable whose Gamma prior is matched to the prior on η_i, or provide an explicit numerical validation of the independence approximation across different η_i values, SNR levels, and numbers of users.
[Algorithm 2, lines 11-17 and 20-23] The pseudocode for the block processing strategy is not reproducible as written. Inside the loop over t for updating q(h_i,t), lines 13 and 17 say "Follow Algorithm 1 to compute h_i,t+1," but the block strategy has no online prediction phase and h_i,t+1 is itself a variational variable updated by Eq. (43); substituting an online-prediction step would introduce a different estimator. In addition, lines 20-23 update τη_i,t and ⟨η_i⟩ inside a loop over t even though the block updates (46)-(47) are block-wide sums with no dependence on t. Please rewrite the pseudocode so that all h_i,t are initialized consistently and the block-wide updates are performed once per CAVI sweep; as written, the implementation cannot be checked against the equations.

minor comments (5)

[Remark 1] The sentence "we reset the initial value of η̂_i,t-1 if it is less than 0 or exceeds 1, which ensures that η_i always remains within its acceptable range" is inaccurate: resetting the initial value does not constrain subsequent CAVI iterates. Please reword and, if a projection or clipping step is intended, state it explicitly in Algorithm 1.
[Table II] The prior distribution for p(γ_t) is written as Γ(a0,b,b0,b), which appears to be a typo for Γ(a0,b0). Please correct this entry.
[Eq. (51)] Several terms in (51) contain Tr{R_i^{-1}} multiplying outer products of channel vectors, but Tr{R_i^{-1}} is a scalar and the expression does not match the quadratic form in (49). Please correct the trace notation or define the intended operation explicitly.
[Algorithm 2, line 23] The update on line 23 refers to τη_i,t even though the block update in (46) does not depend on t; please use τη_i and update it once per CAVI sweep to avoid confusion.
[Section V-A, Figs. 3-6] The claim that the unknown-η VB method performs "nearly identically" to the known-η case would be more convincing if the paper reported the quantitative SER gap (or the estimated η_i values and their variances) rather than only plotting the curves together.

Circularity Check

0 steps flagged · score 2.0 of 10

No derivation step reduces to its own input; the unknown-η tracking claim is conditional on a favorable hand-set prior (Remark 1), and the only self-citations are routine VB lemmas that are not load-bearing.

full rationale

The paper's central chain is a standard mean-field CAVI derivation from the Gauss-Markov likelihood in (11) and (41). The predictive channel statistics in (13)-(15) are not an input: they combine the previous posterior with the current variational estimate of η_i, while η_i itself is updated from the received data through (22)-(23). No equation is equivalent to a fitted parameter by construction, and the reported SER/NMSE advantages are measured against external LMMSE, Kalman-filter, and EP baselines; the algorithm's outputs are not fitted to those baselines. The only self-citations of note are Lemma 1 from [32] and Lemma 2 from [41], both from the same group. Lemma 1 is proved inside the paper, and Lemma 2's proof is omitted only because it is 'similar to the proof of Lemma 1' (Section IV-C); both are algebraic expectation identities used inside the CAVI updates, so the self-citation is method reuse rather than a load-bearing circular premise. The paper itself flags the main conditionality in Remark 1: the Gaussian prior on η_i 'may result in η_i being less than 0 or exceeding 1,' and 'it is crucial to select the initial value of ˆη_i,t−1 carefully, ensuring it is close to the true value.' In the simulations the initial mean 0.95 with variance 1e-3 is within 0.02–0.035 of the tested values (0.97, 0.985), so the 'nearly identical' tracking result is experimentally conditional on a favorable prior; this is a robustness and parameterization caveat, not a circular reduction. Similarly, Eq. (15) can produce an invalid predictive covariance if ˆη^2+τ^η exceeds 1, but that is a correctness concern about the approximation, not an equivalence of prediction and input. I therefore find no circular step; the score of 2 reflects only a minor non-load-bearing self-citation and the acknowledged initialization dependence.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central algorithm rests on standard variational inference machinery plus two domain-specific approximations: an untruncated Gaussian model for η_i and an independence assumption between η_i and ν_i in block processing. No new physical entities are introduced. The main hand-tuned quantities are the initial η_i prior (close to truth) and vague Gamma hyperparameters.

free parameters (3)

Initial time-correlation mean and variance (ˆη_i,0, τ^η_i,0) = 0.95, 1e-3
Hand-chosen to be close to the simulated true η_i values (0.97, 0.985). The authors state in Remark 1 that choosing an initial value close to the true value is crucial. The tracking claim therefore depends on an informative prior.
Gamma prior hyperparameters for noise precision (a0, b0) and ν_i (¯a_i,0, ¯b_i,0) = 1e-4 for all four
Chosen broadly (vague priors). Not fitted to data, but values affect the variational updates in Eqs. (37) and (49)-(51).
Number of interleaved sections L = 2
The interleaved-structure improvement in Fig. 8 is demonstrated only for L=2; the paper does not study sensitivity to this design choice.

assumptions (5)

standard math Mean-field variational family and coordinate ascent variational inference converge to a local optimum (Bishop Ch. 10; Wainwright et al.)
Used throughout, e.g., Eq. (9)-(10), to justify the update rules and local convergence.
domain assumption First-order Gauss-Markov channel model with known per-user covariance R_i and Rayleigh fading (Eq. (1))
The entire JED problem is built on this model; if real channels deviate from Gauss-Markov, the tracking and SER claims may not carry over.
ad hoc to paper Gaussian prior on η_i supported on the real line, not restricted to [0,1] (Remark 1, Table I)
The paper acknowledges the mismatch and only resets the initial value; later variational means are not clipped or truncated.
ad hoc to paper Block processing treats ν_i=(1−η_i^2)^−1 as independent of η_i (Section IV-B)
The authors state they 'hope' to estimate ν_i such that the relation holds and rely on simulations to validate it; no derivation enforces the relation during updates.
standard math Expectation identities from Lemma 1 [32] and Lemma 2 [41] are taken as given
Used to simplify variational expectations; proofs not repeated (Lemma 2 proof omitted).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Variational Bayesian Inference for Time-Varying Massive MIMO Channels: Estimation and Detection." pith.science (2026). https://pith.science/paper/DH5NHOEY

@misc{pith2026250510673,
  author       = {Pith},
  title        = {Pith review of: Variational Bayesian Inference for Time-Varying Massive MIMO Channels: Estimation and Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DH5NHOEY}},
  note         = {Machine review of arXiv:2505.10673}
}

read the original abstract

Massive multiple-input multiple-output (MIMO) stands as a key technology for advancing performance metrics such as data rate, reliability, and spectrum efficiency in the fifth generation (5G) and beyond of wireless networks. However, its efficiency depends greatly on obtaining accurate channel state information. This task becomes particularly challenging with increasing user mobility. In this paper, we focus on an uplink scenario in which a massive MIMO base station serves multiple high-mobility users. We leverage variational Bayesian(VB) inference for joint channel estimation and data detection(JED), tailored for time-varying channels. In particular, we use the VB framework to provide approximations of the true posterior distributions. To cover more real-world scenarios, we assume the time correlation coefficients associated with the channels are unknown. Our simulations demonstrate the efficacy of our proposed VB-based approach in tracking these unknown time correlation coefficients. We present two processing strategies within the VB framework: online and block processing strategies. The online strategy offers a low-complexity solution for a given time slot, requiring only the knowledge of the parameters/statistics within that time slot. In contrast, the block processing strategy focuses on the entire communication block and processes all received signals together to reduce channel estimation errors. Additionally, we introduce an interleaved structure for the online processing strategy to further enhance its performance. Finally, we conduct a comparative analysis of our VB approach against the linear minimum mean squared error(LMMSE), the Kalman Filter(KF), and the expectation propagation(EP) methods in terms of symbol error rate(SER) and channel normalized mean squared error(NMSE). Our findings reveal that our VB framework surpasses these benchmarks across the performance metrics.

Figures

Figures reproduced from arXiv: 2505.10673 by the authors.

**Figure 2.** Block diagram of (a) online processing and (b) block processing strategies. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** An SER comparison between LMMSE, KF, EP, and our online [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: The SER performance comparison between LMMSE, KF, EP, and [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

**Figure 7.** Figure 7: The channel NMSE performance comparison between LMMSE, KF, [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

**Figure 8.** Figure 8: The SER performance of the proposed online processing using the [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

**Figure 9.** Figure 9: The SER and channel NMSE performance of the proposed online and block processing strategies using the VB framework with 16QAM when [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 38 canonical work pages

[1]

Joint channel estimation and data detection for time-varying MIMO channels in UA V networks,

S. Nassirpour and D. H. Nguyen, “Joint channel estimation and data detection for time-varying MIMO channels in UA V networks,” in Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops) . IEEE, 2024, pp. 1592–1597

work page 2024
[2]

B5G and 6G: Next generation wireless communications technologies, demand and challenges,

E. A. Kadir, R. Shubair, S. K. A. Rahim, M. Himdi, M. R. Kamarudin, and S. L. Rosa, “B5G and 6G: Next generation wireless communications technologies, demand and challenges,” in Proc. Int. Congr. Adv. Technol. and Eng. (ICOTEN) . IEEE, 2021, pp. 1–6

work page 2021
[3]

Five disruptive technology directions for 5G,

F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, “Five disruptive technology directions for 5G,” IEEE Commun. Mag. , vol. 52, no. 2, pp. 74–80, 2014

2014
[4]

Spectral efficiency of DFT- based processing hybrid architectures in massive MIMO,

W. Tan, M. Matthaiou, S. Jin, and X. Li, “Spectral efficiency of DFT- based processing hybrid architectures in massive MIMO,” IEEE Wireless Commun. Letters, vol. 6, no. 5, pp. 586–589, 2017

work page 2017
[5]

An overview of massive MIMO: Benefits and challenges,

L. Lu, G. Y . Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang, “An overview of massive MIMO: Benefits and challenges,” IEEE J. Select. Topics in Signal Process. , vol. 8, no. 5, pp. 742–758, 2014

work page 2014
[6]

GreenMO: Enabling virtualized, sustainable massive MIMO with a single RF chain,

A. Gupta, S. Nassirpour, M. Dunna, E. Patamasing, A. Vahid, and D. Bharadia, “GreenMO: Enabling virtualized, sustainable massive MIMO with a single RF chain,” in Proc. Annual Int. Conf. Mobile Comput. and Netw. (MobiCom) , 2023, pp. 1–17

work page 2023
[7]

Blind channel estimation and symbol detection for multi-cell massive MIMO systems by expectation propagation,

K. Ghavami and M. Naraghi-Pour, “Blind channel estimation and symbol detection for multi-cell massive MIMO systems by expectation propagation,” IEEE Trans. Wireless Commun. , vol. 17, no. 2, pp. 943– 954, 2017

work page 2017
[8]

Blind multiuser detection in massive MIMO channels with clustered sparsity,

L. Chen and X. Yuan, “Blind multiuser detection in massive MIMO channels with clustered sparsity,” IEEE Wireless Commun. Letters , vol. 8, no. 4, pp. 1052–1055, 2019

work page 2019

Show all 44 references

[9]

Cramer-Rao bounds for semi-blind, blind and training sequence based channel estimation,

E. De Carvalho and D. T. Slock, “Cramer-Rao bounds for semi-blind, blind and training sequence based channel estimation,” in Proc. IEEE Signal Process. Workshop on Signal Process. Advances in Wireless Commun. IEEE, 1997, pp. 129–132

1997
[10]

Asymptotic performance of ML methods for semi-blind channel estimation,

E. de Carvalho and D. Slock, “Asymptotic performance of ML methods for semi-blind channel estimation,” in Proc. Asilomar Conf. Signals, Systems, and Comp. , vol. 2. IEEE, 1997, pp. 1624–1628

1997
[11]

Semi-blind channel estimation for multiuser massive MIMO systems,

E. Nayebi and B. D. Rao, “Semi-blind channel estimation for multiuser massive MIMO systems,” IEEE Trans. Signal Process. , vol. 66, no. 2, pp. 540–553, 2017

2017
[12]

Semi-blind pilot decontamination for massive MIMO systems,

D. Hu, L. He, and X. Wang, “Semi-blind pilot decontamination for massive MIMO systems,” IEEE Trans. Wireless Commun., vol. 15, no. 1, pp. 525–536, 2015

2015
[13]

Semi-blind detection in hybrid massive MIMO systems via low-rank matrix completion,

S. Liang, X. Wang, and L. Ping, “Semi-blind detection in hybrid massive MIMO systems via low-rank matrix completion,” IEEE Trans. Wireless Commun., vol. 18, no. 11, pp. 5242–5254, 2019

2019
[14]

Double-sparsity learning-based channel-and-signal estimation in massive MIMO with generalized spatial modulation,

X. Kuai, X. Yuan, W. Yan, H. Liu, and Y . J. Zhang, “Double-sparsity learning-based channel-and-signal estimation in massive MIMO with generalized spatial modulation,” IEEE Trans. Commun. , vol. 68, no. 5, pp. 2863–2877, 2020

2020
[15]

Expectation-maximization- based channel estimation for multiuser MIMO systems,

S. Park, J. W. Choi, J.-Y . Seol, and B. Shim, “Expectation-maximization- based channel estimation for multiuser MIMO systems,” IEEE Trans. Commun., vol. 65, no. 6, pp. 2397–2410, 2017

2017
[16]

Performance analysis of FDD massive MIMO systems under channel aging,

R. Chopra, C. R. Murthy, H. A. Suraweera, and E. G. Larsson, “Performance analysis of FDD massive MIMO systems under channel aging,” IEEE Trans. Wireless Commun. , vol. 17, no. 2, pp. 1094–1108, 2017

2017
[17]

Impact of general channel aging conditions on the downlink performance of massive MIMO,

A. K. Papazafeiropoulos, “Impact of general channel aging conditions on the downlink performance of massive MIMO,” IEEE Trans. Veh. Technol., vol. 66, no. 2, pp. 1428–1442, 2016

2016
[18]

Effects of channel aging in massive MIMO systems,

K. T. Truong and R. W. Heath, “Effects of channel aging in massive MIMO systems,” J. Commun. and Netw. , vol. 15, no. 4, pp. 338–351, 2013

2013
[19]

Sparse Bayesian learning for the time-varying massive MIMO channels: Acquisition and tracking,

J. Ma, S. Zhang, H. Li, F. Gao, and S. Jin, “Sparse Bayesian learning for the time-varying massive MIMO channels: Acquisition and tracking,” IEEE Trans. Commun. , vol. 67, no. 3, pp. 1925–1938, 2018

1925
[20]

Bayesian approach to chan- nel interpolation in massive MIMO receiver,

A. Osinsky, A. Ivanov, and D. Yarotsky, “Bayesian approach to chan- nel interpolation in massive MIMO receiver,” IEEE Commun. Letters , vol. 24, no. 12, pp. 2751–2755, 2020

2020
[21]

Experimental comparison of fast-fading channel interpolation methods for the LTE uplink,

M. Lerch, “Experimental comparison of fast-fading channel interpolation methods for the LTE uplink,” in Proc. Int. Symp. ELMAR. IEEE, 2015, pp. 5–8

2015
[22]

Channel prediction in time-varying massive MIMO environments,

W. Peng, M. Zou, and T. Jiang, “Channel prediction in time-varying massive MIMO environments,” IEEE Access, vol. 5, pp. 23 938–23 946, 2017

2017
[23]

Performance analysis of (TDD) massive MIMO with Kalman channel prediction,

S. Kashyap, C. Moll ´en, E. Bj ¨ornson, and E. G. Larsson, “Performance analysis of (TDD) massive MIMO with Kalman channel prediction,” in Proc. IEEE Int. Conf. Acoustics, Speech and Signal Process. IEEE, 2017, pp. 3554–3558

2017
[24]

Quasi-static and time-selective channel estimation for block- sparse millimeter wave hybrid MIMO systems: Sparse Bayesian learning (SBL) based approaches,

S. Srivastava, A. Mishra, A. Rajoriya, A. K. Jagannatham, and G. As- cheid, “Quasi-static and time-selective channel estimation for block- sparse millimeter wave hybrid MIMO systems: Sparse Bayesian learning (SBL) based approaches,” IEEE Trans. Signal Process. , vol. 67, no. 5...

2018
[25]

Deep CNN-based channel estimation for mmWave massive MIMO systems,

P. Dong, H. Zhang, G. Y . Li, I. S. Gaspar, and N. NaderiAlizadeh, “Deep CNN-based channel estimation for mmWave massive MIMO systems,” IEEE J. Select. Topics in Signal Process. , vol. 13, no. 5, pp. 989–1000, 2019

2019
[26]

Deep learning based channel estimation algorithm for fast time-varying MIMO-OFDM systems,

Y . Liao, Y . Hua, and Y . Cai, “Deep learning based channel estimation algorithm for fast time-varying MIMO-OFDM systems,”IEEE Commun. Letters, vol. 24, no. 3, pp. 572–576, 2019

2019
[27]

Semi-data- aided channel estimation for MIMO systems via reinforcement learning,

T.-K. Kim, Y .-S. Jeon, J. Li, N. Tavangaran, and H. V . Poor, “Semi-data- aided channel estimation for MIMO systems via reinforcement learning,” IEEE Trans. Wireless Commun. , vol. 22, no. 7, pp. 4565–4579, 2022

2022
[28]

Semi-blind chan- nel estimation and data detection for time-varying massive MIMO system,

M. Naraghi-Pour, M. Rashid, and C. Vargas-Rosales, “Semi-blind chan- nel estimation and data detection for time-varying massive MIMO system,” in Proc. IEEE Int. Conf. Commun. (ICC) . IEEE, 2021, pp. 1–6

2021
[29]

Expectation propagation for approximate Bayesian infer- ence,

T. P. Minka, “Expectation propagation for approximate Bayesian infer- ence,” arXiv preprint arXiv:1301.2294 , 2013

2013 arXiv
[30]

Window-based expectation propagation for adaptive signal detection in flat-fading channels,

Y . Qi and T. P. Minka, “Window-based expectation propagation for adaptive signal detection in flat-fading channels,” IEEE Trans. Wireless Commun., vol. 6, no. 1, pp. 348–355, 2007

2007
[31]

Variational Bayes’ joint channel estimation and soft symbol decoding for uplink massive MIMO systems with low resolution ADCs,

S. S. Thoota and C. R. Murthy, “Variational Bayes’ joint channel estimation and soft symbol decoding for uplink massive MIMO systems with low resolution ADCs,” IEEE Trans. Commun. , vol. 69, no. 5, pp. 3467–3481, 2021

2021
[32]

A variational bayesian perspective on massive MIMO detection,

D. H. N. Nguyen, I. Atzeni, A. T ¨olli, and A. L. Swindlehurst, “A variational bayesian perspective on massive MIMO detection,” arXiv preprint arXiv:2205.11649, 2022

2022 arXiv
[33]

Joint CFO, gridless channel estimation and data detection for underwater acoustic OFDM systems,

L. Wan, J. Zhu, E. Cheng, and Z. Xu, “Joint CFO, gridless channel estimation and data detection for underwater acoustic OFDM systems,” IEEE J. Oceanic Eng. , vol. 47, no. 4, pp. 1215–1230, 2022

2022
[34]

Grid-less variational Bayesian channel estimation for antenna array systems with low resolu- tion ADCs,

J. Zhu, C.-k. Wen, J. Tong, C. Xu, and S. Jin, “Grid-less variational Bayesian channel estimation for antenna array systems with low resolu- tion ADCs,” IEEE Trans. Wireless Commun. , vol. 19, no. 3, pp. 1549– 1562, 2019

2019
[35]

Iterative variational Bayesian inference based channel estimation for TWR mmWave sys- tems,

S. Bera, S. Chakraborty, D. Sen, and A. K. Dutta, “Iterative variational Bayesian inference based channel estimation for TWR mmWave sys- tems,” IEEE Trans. Veh. Technol. , vol. 72, no. 8, pp. 10 330–10 344, 2023

2023
[36]

Uplink-aided high mo- bility downlink channel estimation over massive MIMO-OTFS system,

Y . Liu, S. Zhang, F. Gao, J. Ma, and X. Wang, “Uplink-aided high mo- bility downlink channel estimation over massive MIMO-OTFS system,” IEEE J. on Select. Areas in Comm., vol. 38, no. 9, pp. 1994–2009, 2020

1994
[37]

On the throughput of large MIMO beamforming systems with channel aging,

R. Chopra, C. R. Murthy, and H. A. Suraweera, “On the throughput of large MIMO beamforming systems with channel aging,” IEEE Signal Process. Letters, vol. 23, no. 11, pp. 1523–1527, 2016

2016
[38]

C. M. Bishop and N. M. Nasrabadi, Pattern recognition and machine learning. Springer, 2006, vol. 4, no. 4

2006
[39]

Graphical models, exponential families, and variational inference,

M. J. Wainwright, M. I. Jordan et al. , “Graphical models, exponential families, and variational inference,” Foundations and Trends® in Ma- chine Learning, vol. 1, no. 1–2, pp. 1–305, 2008

2008
[40]

Semi-blind chan- nel estimation and data detection for multi-cell massive MIMO systems on time-varying channels,

M. Naraghi-Pour, M. Rashid, and C. Vargas-Rosales, “Semi-blind chan- nel estimation and data detection for multi-cell massive MIMO systems on time-varying channels,” IEEE Access , vol. 9, pp. 161 709–161 722, 2021

2021
[41]

A variational Bayesian perspective on MIMO detection with low-resolution ADCs,

L. V . Nguyen, A. L. Swindlehurst, and D. H. N. Nguyen, “A variational Bayesian perspective on MIMO detection with low-resolution ADCs,” in Proc. Asilomar Conf. Signals, Systems, and Comp. IEEE, 2022, pp. 22–26

2022
[42]

Channel capacity of MIMO architecture using the expo- nential correlation matrix,

S. L. Loyka, “Channel capacity of MIMO architecture using the expo- nential correlation matrix,” IEEE Commun. Letters , vol. 5, no. 9, pp. 369–371, 2001

2001
[43]

Multi-beam multi-hop routing for intelligent re- flecting surfaces aided massive MIMO,

W. Mei and R. Zhang, “Multi-beam multi-hop routing for intelligent re- flecting surfaces aided massive MIMO,” IEEE Trans. Wireless Commun., vol. 21, no. 3, pp. 1897–1912, 2021. 15

1912
[44]

Beamforming design in reconfigurable intelligent surface-assisted IoT networks based on discrete phase shifters and imperfect CSI,

S. Nassirpour, A. Vahid, D.-T. Do, and D. Bharadia, “Beamforming design in reconfigurable intelligent surface-assisted IoT networks based on discrete phase shifters and imperfect CSI,” IEEE Internet Things J. , vol. 11, no. 3, pp. 5301–5315, 2023

2023

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [1]

Joint channel estimation and data detection for time-varying MIMO channels in UA V networks,

S. Nassirpour and D. H. Nguyen, “Joint channel estimation and data detection for time-varying MIMO channels in UA V networks,” in Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops) . IEEE, 2024, pp. 1592–1597

work page 2024

[2] [2]

B5G and 6G: Next generation wireless communications technologies, demand and challenges,

E. A. Kadir, R. Shubair, S. K. A. Rahim, M. Himdi, M. R. Kamarudin, and S. L. Rosa, “B5G and 6G: Next generation wireless communications technologies, demand and challenges,” in Proc. Int. Congr. Adv. Technol. and Eng. (ICOTEN) . IEEE, 2021, pp. 1–6

work page 2021

[3] [3]

Five disruptive technology directions for 5G,

F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, “Five disruptive technology directions for 5G,” IEEE Commun. Mag. , vol. 52, no. 2, pp. 74–80, 2014

2014

[4] [4]

Spectral efficiency of DFT- based processing hybrid architectures in massive MIMO,

W. Tan, M. Matthaiou, S. Jin, and X. Li, “Spectral efficiency of DFT- based processing hybrid architectures in massive MIMO,” IEEE Wireless Commun. Letters, vol. 6, no. 5, pp. 586–589, 2017

work page 2017

[5] [5]

An overview of massive MIMO: Benefits and challenges,

L. Lu, G. Y . Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang, “An overview of massive MIMO: Benefits and challenges,” IEEE J. Select. Topics in Signal Process. , vol. 8, no. 5, pp. 742–758, 2014

work page 2014

[6] [6]

GreenMO: Enabling virtualized, sustainable massive MIMO with a single RF chain,

A. Gupta, S. Nassirpour, M. Dunna, E. Patamasing, A. Vahid, and D. Bharadia, “GreenMO: Enabling virtualized, sustainable massive MIMO with a single RF chain,” in Proc. Annual Int. Conf. Mobile Comput. and Netw. (MobiCom) , 2023, pp. 1–17

work page 2023

[7] [7]

Blind channel estimation and symbol detection for multi-cell massive MIMO systems by expectation propagation,

K. Ghavami and M. Naraghi-Pour, “Blind channel estimation and symbol detection for multi-cell massive MIMO systems by expectation propagation,” IEEE Trans. Wireless Commun. , vol. 17, no. 2, pp. 943– 954, 2017

work page 2017

[8] [8]

Blind multiuser detection in massive MIMO channels with clustered sparsity,

L. Chen and X. Yuan, “Blind multiuser detection in massive MIMO channels with clustered sparsity,” IEEE Wireless Commun. Letters , vol. 8, no. 4, pp. 1052–1055, 2019

work page 2019