PaperlayerView exact revision on arXiv

Source-linked AI summary

Impact of General Channel Aging Conditions on the Downlink Performance of Massive MIMO

Anastasios K. Papazafeiropoulos

arXiv:1605.07661v1cs.IT

TL;DR

Channel aging in massive MIMO is difficult to characterize realistically because user mobility and oscillator phase noise jointly degrade the CSI used for downlink transmission. The paper develops a joint model and deterministic SINR equivalents for MRT and RZF, finding that Doppler degradation dominates phase noise except under low mobility, while massive MIMO and RZF remain advantageous.

  • Problem

    The paper addresses the limited characterization of massive-MIMO channel aging when both user mobility and phase noise affect time-varying CSI.

  • Method

    The authors build a joint channel-phase-noise CSI model and derive deterministic downlink SINR equivalents for MRT and RZF using large random-matrix theory.

  • Results

    Doppler shift causes more degradation than phase noise, with phase-noise impact becoming meaningful at low mobility; simulations validate the deterministic approximations.

  • Takeaways & Limitations

    Massive MIMO remains preferable under generalized channel aging, and RZF performs better than MRC across varied Doppler and phase-noise conditions.

Abstract

from arXiv · show

Recent works have identified massive multiple-input-multiple-output (MIMO) as a key technology for achieving substantial gains in spectral and energy efficiency. Additionally, the turn to low-cost transceivers, being prone to hardware impairments is the most effective and attractive way for cost-efficient applications concerning massive MIMO systems. In this context, the impact of channel aging, which severely affects the performance, is investigated herein by considering a generalized model. Specifically, we show that both Doppler shift because of the users' relative movement as well as phase noise due to noisy local oscillators (LOs) contribute to channel aging. To this end, we first propose a joint model, encompassing both effects, in order to investigate the performance of a massive MIMO system based on the inevitable time-varying nature of realistic mobile communications. Then, we derive the deterministic equivalents (DEs) for the signal-to-noise-and-interference ratios (SINRs) with maximum ratio transmission (MRT) and regularized zero-forcing precoding (RZF). Our analysis not only demonstrates a performance comparison between MRT and RZF under these conditions, but most importantly, it reveals interesting properties regarding the effects of user mobility and phase noise. In particular, the large antenna limit behavior depends profoundly on both effects, but the burden due to user mobility is much more detrimental than phase noise even for moderate user velocities ($\approx30$ km/h), while the negative impact of phase noise is noteworthy at lower mobility conditions.

I. INTRODUCTION

The paper studies generalized channel aging in massive MIMO by jointly modeling user mobility and phase noise, then derives downlink SINR approximations for MRT and RZF. It shows that Doppler effects are more damaging than phase noise, while massive MIMO and RZF remain effective under these conditions.

  • Scope and limitations: The analysis focuses on TDD systems and neglects uplink-downlink hardware mismatch, leaving that impairment for future channel-aging studies.The omission is intended to isolate Doppler shift and phase noise.
  • Motivation and contribution: The study jointly models Doppler shift from user mobility and phase noise from local oscillators as contributors to channel aging.The model targets realistic time-varying TDD massive-MIMO channels with imperfect CSI.
  • Motivation and contribution: Large random-matrix theory yields deterministic equivalents for downlink SINRs with MRT and RZF under generalized channel aging.The analysis includes uplink-training imperfections, channel time variation, and different LO arrangements.
  • Main findings: The generalized-aging setting preserves the square root power-law, allowing per-user transmit power to decrease as the BS antenna count increases.The antenna count is scaled to the inverse square root of transmit power.
  • Main findings: Doppler shift has a much more severe impact than phase noise even at low vehicle velocities, while phase-noise effects become meaningful under low mobility.The paper compares these effects through analytical results and simulations.
  • Main findings: Massive MIMO outperforms conventional MIMO under generalized channel aging, with RZF performing better than MRC across various Doppler and phase-noise severities.The comparison is reported for the studied downlink setting.

B. Channel Estimation

The paper estimates the effective channel during TDD uplink training and models its subsequent evolution through user mobility and phase noise. The resulting time-dependent CSI accounts for imperfect estimation and delayed channel knowledge during data transmission.

  • B. Channel Estimation: TDD uplink pilots estimate the effective channel at time 0, while channel variation during data transmission makes the estimate progressively delayed.Training assumes negligible variation over its short duration; the channel can vary symbol by symbol during the remaining coherence period.
  • C. Channel Aging: The model represents the effective channel as a time-varying estimated component plus an uncorrelated error that combines imperfect and delayed CSI effects.The combined error depends on the channel covariance and the aging matrix A_n.
  • C. Channel Aging: Longer delays reduce A_n and therefore make CSI estimated at time 0 less accurate.The same estimate is used at successive times until the current transmission symbol, so aging accumulates with delay.
  • C. Channel Aging: The joint channel model incorporates 2-D isotropic scattering, user mobility, and phase noise through a time-dependent matrix A_n.A_n describes the combined effects at both user equipment and base-station local oscillators.
  • C. Channel Aging: Phase noise adds symbol-by-symbol random phase drift, producing an additional deviation between the actual channel and its time-0 estimate.This contribution further degrades channel aging beyond user mobility alone.

III. DOWNLINK TRANSMISSION

The downlink serves all users simultaneously with linear precoding based on imperfect, delayed CSI. MRT and RZF are formulated within a received-signal and achievable-rate framework that uses instantaneous downlink SINR.

  • III. DOWNLINK TRANSMISSION: The base station transmits simultaneously to all users using SDMA, with the downlink channel obtained from uplink reciprocity.The received signal is modeled during the data-transmission portion of the coherence period.
  • III. DOWNLINK TRANSMISSION: MRT uses the available delayed CSI A_n ĝ_k,0, whereas RZF designs its precoder from the corresponding regularized matrix formulation.Both precoders operate with CSI that was estimated at time 0 and propagated to time n.
  • III. DOWNLINK TRANSMISSION: The RZF regularization is scaled by M so that it converges to a constant as M and K grow, while its optimization is outside the paper’s scope.The formulation permits an arbitrary Hermitian nonnegative definite matrix Z and regularization parameter a.
  • III. DOWNLINK TRANSMISSION: The achievable user rate is defined from the instantaneous downlink SINR under a worst-case uncorrelated additive-noise interpretation.The resulting rate lower bounds the mutual information between the received signal and transmitted symbols.

IV. DETERMINISTIC EQUIVALENT ANALYSIS

The analysis derives deterministic equivalents for downlink SINRs and rates as the numbers of antennas and users grow with a finite user-to-antenna ratio. These equivalents provide asymptotic approximations for MRT and RZF under imperfect and delayed CSI.

  • IV. DETERMINISTIC EQUIVALENT ANALYSIS: Random-matrix-theory deterministic equivalents approximate the downlink SINRs as K and M tend to infinity while K/M = β remains finite.The framework targets imperfect and delayed CSI caused by user mobility and phase noise.
  • IV. DETERMINISTIC EQUIVALENT ANALYSIS: The deterministic SINR converges almost surely to the corresponding random SINR, and deterministic user rates follow from this convergence.The rate construction uses dominated convergence and the continuous mapping theorem.
  • IV. DETERMINISTIC EQUIVALENT ANALYSIS: The paper derives deterministic downlink achievable rates for both MRT and RZF, treating identical-statistics phase noise for SLOs as the main exposition case.A more general setting with independent, non-identically distributed SLO phase noises is examined in an appendix.

A. MRT

For MRT, the paper derives a general deterministic SINR equivalent and examines simplified covariance cases and transmit-power scaling. The analysis shows that mobility and phase noise reduce SINR, while the power-scaling law remains unchanged.

  • A. MRT: Theorem 2 gives the MRT downlink deterministic SINR under imperfect and delayed CSI caused by phase noise and user mobility.The result is the main general asymptotic expression for user k at time n.
  • A. MRT: The MRT deterministic equivalent simplifies when the channel covariance is R_k = I_M and the aging matrix becomes a scalar or scaled identity.These cases correspond to no large-scale component and CLO or identical-LO SLO settings.
  • A. MRT: For transmit powers scaled as 1/M^q, the MRT SINR diverges when q < 1/2 and vanishes when q > 1/2.The proposition identifies q = 1/2 as the critical scaling regime for maintaining finite asymptotic SINR.
  • A. MRT: With uplink and downlink powers scaled by 1/M, the MRT downlink SINR per user can remain finite under phase noise and user mobility.This result is stated for fixed scaling constants E_u and E_d.
  • A. MRT: Phase noise and user mobility reduce the MRT SINR but do not alter the power-scaling law.Similar power-scaling conclusions are stated to hold for RZF, although this subsection focuses on MRT.

B. RZF

The RZF analysis derives a deterministic equivalent for downlink SINR under imperfect and delayed CSI caused jointly by phase noise and user mobility.

  • RZF: The derivation begins with the general-channel-aging SINR formulation and specializes it through definitions and corollaries for particular covariance and oscillator models.Special cases include identical large-scale effects, no large-scale component, and common phase noise or a single common local oscillator.
  • RZF: Theorem 3 gives the downlink deterministic equivalent of user k's SINR with RZF under imperfect CSI and generalized channel aging.The model accounts for delayed CSI caused by both phase noise and user mobility.
  • RZF: The resulting expressions are obtained using deterministic-equivalent analysis with auxiliary quantities defined through fixed-point and matrix expressions.The supplied passages reference the auxiliary definitions but do not provide their complete formulas.

V. NUMERICAL RESULTS

The numerical study evaluates achievable sum-rates under phase noise and Doppler-induced channel aging, comparing MRT and RZF with analytical and simulation results.

  • V. NUMERICAL RESULTS: The study uses LTE-based numerical settings, including a 1 ms coherence time and a 196-channel-use coherence block.The setup assumes fD = 250 Hz, fc = 2 GHz, and W = 20 MHz.
  • V. NUMERICAL RESULTS: RZF outperforms MRT in the considered rate scenarios, while increasing phase noise decreases the achievable sum-rate.The comparison uses static environments with varying BS antenna counts and phase-noise levels.
  • V. NUMERICAL RESULTS: Doppler shift is more detrimental than phase noise, with phase-noise degradation becoming relatively insignificant around 30 km/h while achievable rates can become impractically low.The Doppler comparison uses M = 60 and varies normalized Doppler shift and phase-noise levels.
  • V. NUMERICAL RESULTS: Phase-noise losses remain prominent in static environments even though mobility dominates degradation at higher normalized Doppler shifts.The figure compares RZF and MRT analytical rates with simulation points across channel-aging conditions.

C. Required Transmit Power Comparison between MRT and RZF

The transmit-power analysis compares MRT and RZF as antenna count, Doppler shift, phase noise, and cellular interference vary.

  • C. Required Transmit Power Comparison between MRT and RZF: Doubling the BS antenna count reduces required transmit power by approximately 1.5 dB in static conditions, while more severe phase noise increases the required power.This result concerns achieving 1 bit/s/Hz per user.
  • C. Required Transmit Power Comparison between MRT and RZF: At M = 60, increasing Doppler shift raises required transmit power until it saturates, and higher phase noise causes saturation to occur sooner.The no-channel-aging reference remains independent of normalized Doppler shift and phase noise.
  • D. Extension to Multi-Cell Large MIMO Systems: The multicell analysis models each cell as an instance of the single-cell setting while adding pilot-contamination interference between cells.The supplied passages state that only the inter-cell interference term differs analytically.
  • D. Extension to Multi-Cell Large MIMO Systems: The cellular extension reports simulation results rather than a separate deterministic-equivalent analysis for the L = 7 configuration.The authors describe the omitted analysis as straightforward but distracting from the section's objective.
  • D. Extension to Multi-Cell Large MIMO Systems: In the seven-cell extension, channel aging additionally degrades performance through inter-cell interference and increases sensitivity to normalized Doppler shift.The multicell setting uses pilot contamination and reports central-cell sum-rates from simulation.

APPENDIX A USEFUL LEMMAS

Appendix A collects matrix identities, random-matrix lemmas, and deterministic-equivalent results used in the main derivations.

  • APPENDIX A USEFUL LEMMAS: The appendix states matrix inversion and rank-1 perturbation lemmas for manipulating resolvent expressions.These include two matrix inversion lemmas and a rank-1 perturbation result.
  • APPENDIX A USEFUL LEMMAS: Additional lemmas provide trace limits and deterministic-equivalent tools for Gaussian vectors and independent random matrices.The results assume bounded spectral norms and, in one case, finite moments of matrix entries.
  • APPENDIX A USEFUL LEMMAS: The appendix also records bounded-spectral-norm and independence conditions required by the random-matrix results.These assumptions support the asymptotic analysis used for the MRT and RZF SINR expressions.

APPENDIX B PROOF OF PROPOSITION 1

The proof establishes the MMSE channel-estimation representation used for Proposition 1 and derives deterministic-equivalent components for the normalization and signal-power terms under phase-noise settings.

  • Channel estimation: The training model uses mutually orthogonal pilot sequences, assumes a constant channel during training, and includes additive Gaussian noise and local-oscillator phase noise.The pilots have τ symbols and satisfy ΨΨH = I_K.
  • Channel estimation: Correlating the received pilot signal with user k’s sequence yields the effective channel observation used for MMSE estimation.The construction substitutes g_k,0 = Θ_k,0 h_k,0 and accounts for the projected training noise.
  • Channel decomposition: The estimated channel and estimation error are modeled as jointly Gaussian and statistically independent through the orthogonality principle.The estimate is distributed as CN(0, D_k), while the error covariance is R_k − D_k.
  • Deterministic-equivalent derivation: The proof derives deterministic equivalents for the normalization parameter and desired signal power using trace identities and the stated lemmas.The resulting deterministic-equivalent signal power is denoted by S̄_k,n.
  • Phase-noise cases: The phase-noise contribution differs between common-LO and separate-LO settings through the corresponding diagonal phase-difference traces.The proof specializes the BS-antenna analysis to identical or non-identical local-oscillator statistics as stated.

APPENDIX D PROOF OF THEOREM 3

The proof of Theorem 3 decomposes the SINR-related terms, derives deterministic equivalents for normalization and signal power, and handles dependence between resolvent terms and channel estimates.

  • Normalization parameter: The proof begins by deriving the deterministic equivalent of the normalization parameter using the theorem’s normalization expressions.The derivation proceeds through algebraic manipulation and trace-based identities.
  • Desired signal power: The desired signal power is decomposed into terms whose deterministic equivalents are obtained using the cited lemmas and theorems.The resulting deterministic-equivalent signal power is denoted by S̄_k,n.
  • Matrix calculation: The proof uses the matrix expression involving Z and M a I_M to evaluate one of the deterministic-equivalent terms.This expression appears within the resolvent-based calculation.
  • Dependence handling: Although g_k,n and ĝ_i,0 are mutually independent, Σ_i is not independent of ĝ_k,0, so Lemma 2 is required.Substitution of the resulting relation leads to the subsequent evaluation of Q_ik and its deterministic-equivalent terms.
  • Final result: The derivation concludes by obtaining the deterministic equivalent γ̄_k,n.The final step closes the proof for the theorem’s SINR-related quantity.
Loading 1605.07661v1…