CSI Compression for Massive MIMO-OFDM: Mismatch-Aware Rate-Distortion Trade-offs

Abstract: We study channel state information (CSI) compression for wideband frequency division duplex massive multiple-input multiple-output (MIMO) when the base station (BS) reconstructs CSI using an imperfect covariance model. Under matched second-order statistics, remote rate--distortion theory yields transform coding with reverse water-filling (RWF) over covariance eigenmodes. With decoder-side covariance mismatch, however, this allocation is no longer end-to-end optimal. We derive an achievable mismatched Gaussian rate--distortion characterization based on a Gaussian test channel and a mismatched minimum mean square error (MMSE) reconstruction rule. In a shared-eigenvector regime (common eigenbasis, mismatched eigenvalues), the problem decouples across modes and leads to a robust reverse water-filling (RRWF) allocation computable via bisection and per-mode root finding. Simulations using wideband massive MIMO covariance models show that RRWF consistently improves reconstruction distortion and end-to-end mean square error relative to conventional RWF under mismatch.

• The paper introduces a mismatched rate-distortion framework for CSI compression by accounting for encoder-decoder covariance mismatches in MIMO-OFDM systems.
• It develops a robust reverse waterfilling algorithm that optimally allocates bits per scalar Gaussian mode, mitigating performance losses from statistical mismatch.
• Simulations show that the proposed approach significantly reduces NMSE under practical mismatch conditions, guiding efficient CSI feedback design in future 6G systems.

Mismatch-Aware Rate–Distortion Trade-offs in CSI Compression for Massive MIMO-OFDM

Introduction and Context

The paper "CSI Compression for Massive MIMO-OFDM: Mismatch-Aware Rate-Distortion Trade-offs" (2604.17426) tackles a central bottleneck in FDD massive MIMO-OFDM systems: the compression and feedback of high-dimensional downlink channel state information (CSI) under practical constraints. While transform coding based on the eigenstructure of channel statistics achieves the information-theoretic rate-distortion (RD) limit under matched second-order statistics, real-world deployments inevitably exhibit mismatch between the covariance models used at the encoder (user) and decoder (base station), due to calibration errors, nonstationarity, and estimation inaccuracies.

Recent works have primarily focused on end-to-end learning–based CSI compression, but these methods have substantial complexity not viable for edge devices. This work brings the analysis back to the information-theoretic regime, considering Gaussian sources and channel estimation, but innovatively abandons the fragile joint-covariance assumption, instead rigorously characterizing the mismatch-aware RD trade-off.

CSI Compression under Covariance Mismatch

The core technical advancement is an achievable mismatched RD characterization for compressing Gaussian CSI estimates using a test channel and mismatched MMSE reconstruction at the BS. Rather than assuming that encoder and decoder operate under identical channel statistics, the paper models distinct encoder- and decoder-side covariance matrices ($C$ and $C^{(b)}$). The fundamental RD trade-off is established:

• The encoder designs a rate-constrained Gaussian test channel for the MMSE CSI estimate.
• The decoder reconstructs with an MMSE estimator based on its own (possibly mismatched) covariance.
• The resulting distortion is a function of both the source law and the mismatch, decomposing as

$$D^{\text{E2E}}(R) = D^{\text{mmse}} + D^{\text{quant}}(R; C, C^{(b)})$$

with $D^{\text{mmse}}$ being the pilot-based estimation error and $D^{\text{quant}}$ the rate-limited quantization/distortion under mismatch.

The key achievement is an explicit expression for the achievable distortion under arbitrary covariance mismatch, utilizing the structure of the MMSE filter with the mismatched covariance, and showing that every test channel covariance $C_q \succ 0$ yields an achievable $(R, D)$ pair, where $R = \log_2 |C + C_q|/|C_q|$, and $C_q$ is optimized under the decoder's model.

Shared-Eigenvector Regime and Robust Waterfilling

The analysis is made tractable for large-scale MIMO-OFDM by focusing on the shared-eigenvector regime where encoder and decoder covariances possess the same eigenbasis but different eigenvalues—a realistic assumption in practice due to spatial consistency of array responses. In this regime:

• The transform coding problem decomposes into $MN$ parallel scalar Gaussian modes.
• The optimal bit allocation is generalized from the classic reverse waterfilling (RWF) to a robust reverse waterfilling (RRWF) scheme, where the per-mode allocation solves a cubic equation depending on both the source and mismatched eigenvalues.
• The RRWF allocation is efficiently computable via bisection and per-mode root finding, yielding explicit control over the allocation of rate under mismatch.

Theoretical recoveries show that for matched eigenvalues, RRWF reduces to classic RWF, thus unifying both cases.

Numerical Results and Empirical Impact

Simulation results, based on spatial-frequency statistical MIMO-OFDM channel models, demonstrate substantial reduction of end-to-end NMSE under decoder-side mismatch across a broad spectrum of mismatch severities. RRWF consistently outperforms both:

• RWF tuned to the true (unavailable at BS) statistics, which is non-robust to mismatch,
• and RWF tuned to the (potentially erroneous) BS side statistics.

The performance degradation of conventional RWF with increased mismatch is significant, while RRWF maintains robustness due to reallocation toward modes effective under decoder assumptions. Maximum gain is observed at low and moderate feedback rates, while all schemes meet the MMSE floor at high rates.

Theoretical and Practical Implications

This work articulates, for the first time, the precise rate-robustness–distortion trade-off for linear transform-based CSI compression in the presence of decoder-side statistical model mismatch, addressing a long-standing gap in RD theory for feedback systems. Practically, the proposed RRWF allocation is both tractable and implementable, directly informing optimal design of covariance-based quantization codebooks and transform-coding protocols for FDD massive MIMO and 6G systems.

The results indicate that system designers must anticipate and account for decoder-side mismatch rather than naively applying RWF on encoder-side statistics, especially in regimes with limited feedback, where ignoring mismatch incurs measurable distortion penalties.

Future Directions

This framework enables further exploration into:

• Joint optimization of pilot allocation, feedback allocation, and RRWF bit loading considering dynamic mismatch statistics;
• Extension to nonlinear sources (e.g., GMM, non-Gaussian channels), for which the Gaussian framework serves as an outer bound;
• Analysis and mitigation of performance loss under joint space-frequency eigenbasis drift;
• Implementation in practical codebooks and vector quantization, leveraging RRWF allocations.

Conclusion

This work provides a comprehensive, information-theoretically grounded analysis of CSI compression under encoder–decoder covariance mismatch, offering a robust RD benchmark and a computationally efficient RRWF algorithm. The proposed approach delivers improved NMSE performance under realistic mismatch scenarios, with clear implications for the design of future massive MIMO feedback strategies in FDD and beyond (2604.17426).