Enhanced Low-Complexity FDD System Feedback with Variable Bit Lengths via Generative Modeling

Nurettin Turan, Benedikt Fesl, Wolfgang Utschick

Abstract

This paper introduces an enhanced Gaussian Mixture Model (GMM) framework for limited feedback in Frequency Division Duplex (FDD) MIMO systems. The proposed method enables a single trained model to support multiple feedback bit lengths through component merging and pruning, addressing the rigidity of traditional feedback schemes. By incorporating structural constraints such as Toeplitz and circulant matrices into GMM covariances, the framework significantly reduces data overhead for mobile terminal initialization while maintaining high system throughput.

Introduction

In FDD wireless systems, the lack of channel reciprocity necessitates that the Mobile Terminal (MT) measures the downlink channel and feeds this information back to the Base Station (BS). This feedback process is a critical bottleneck, as it consumes uplink bandwidth. Traditional feedback schemes are often optimized for a fixed resolution; changing the number of feedback bits typically requires re-training or deploying entirely new codebooks. Furthermore, high-performance machine learning models often suffer from significant "offloading overhead," where the BS must transmit large sets of parameters to the MT before communication can begin.

Methodology

The proposed framework utilizes a generative modeling approach where a GMM represents the underlying distribution of the wireless channel. Instead of explicit channel estimation, the MT identifies the GMM component (cluster) that most likely corresponds to the received pilot signals.

Variable Bit Lengths via Mixture Reduction

To support dynamic feedback resolutions without maintaining multiple models, the authors propose a mixture reduction strategy. A large GMM is trained once and then reduced to a target size $K = 2^B$ (where $B$ is the number of bits) using:

  • Merging: Statistically similar components are combined using a moment-preserving approach.
  • Pruning: Components with the lowest mixing coefficients are discarded.

Structured Covariances

To reduce parameter overhead and incorporate physical domain knowledge, the GMM covariance matrices are constrained to Toeplitz or circulant structures. This reflects the symmetries inherent in antenna geometries like Uniform Linear Arrays (ULAs) and dramatically reduces the number of parameters to be offloaded to the MT.

Figure 1: Flowchart of the proposed GMM-based feedback scheme, illustrating the offline fitting at the BS and the online feedback loop at the MT.

Performance Evaluation

The framework was evaluated using the QuaDRiGa channel simulator. Results indicate that the merging strategy (mkGMM) not only matches but often exceeds the performance of GMMs trained directly for smaller bit lengths. This is attributed to the larger model's ability to capture finer channel nuances, which are preserved during the merging process more effectively than a direct fit on a smaller scale.

Figure 2: Normalized spectral efficiency comparison for different bit lengths and reduction techniques. The merging approach (mkGMM) consistently performs well across various resolutions.

Furthermore, the transition to structured covariances (e.g., Toeplitz) reduces the parameter count by over 99% with negligible loss in spectral efficiency. In multi-user (MU-MIMO) scenarios, the GMM-based approach significantly outperforms traditional Lloyd-based codebooks and compressed sensing baselines, particularly under low pilot overhead conditions.

Conclusion

The integration of generative modeling with structural constraints and mixture reduction provides a flexible, low-complexity solution for FDD feedback. By allowing a single model to adapt to varying bit lengths and reducing offloading overhead through physics-informed constraints, this framework provides a robust path toward efficient and adaptive future wireless communication systems.

Summary updated

Abstract

Recently, a versatile limited feedback scheme based on a Gaussian mixture model (GMM) was proposed for frequency division duplex (FDD) systems. This scheme provides high flexibility regarding various system parameters and is applicable to both point-to-point multiple-input multiple-output (MIMO) and multi-user MIMO (MV-MIMO) communications. The GMM is learned to cover the operation of all mobile terminals (MTs) located inside the base station (BS) cell, and each MT only needs to evaluate its strongest mixture component as feedback, eliminating the need for channel estimation at the MT. In this work, we extend the GMM-based feedback scheme to variable feedback lengths by leveraging a single learned GMM through merging or pruning of dispensable mixture components. Additionally, the GMM covariances are restricted to Toeplitz or circulant structure through model-based insights. These extensions significantly reduce the offloading amount and enhance the clustering ability of the GMM which, in turn, leads to an improved system performance. Simulation results for both point-to-point and multi-user systems demonstrate the effectiveness of the proposed extensions.
BibTeX
@inproceedings{turan2023enhanced,
  author = {Nurettin Turan and Benedikt Fesl and Wolfgang Utschick},
  title = {{Enhanced Low-Complexity FDD System Feedback with Variable Bit Lengths via Generative Modeling}},
  booktitle = {ACSSC},
  year = {2023},
  doi = {10.1109/ieeeconf59524.2023.10477075},
}
Figure