A reinforcement learning (RL)-driven control framework that operates over a bank of pretrained multi-rate AEs, each corresponding to a distinct compression ratio (CR), aiming to dynamically optimize the trade-off between reconstruction fidelity and signaling overhead.
Abstract
This paper proposes a system-aware adaptive channel state information (CSI) feedback framework for massive multiple-input multiple-output (mMIMO) systems, aiming to dynamically optimize the trade-off between reconstruction fidelity and signaling overhead. While deep learning-based autoencoders (AEs) have enabled significant CSI compression, conventional fixed-ratio schemes fail to adapt effectively to non-stationary channel conditions. To address this limitation, we develop a reinforcement learning (RL)-driven control framework that operates over a bank of pretrained multi-rate AEs, each corresponding to a distinct compression ratio (CR). At each time step, a centralized RL agent selects the most suitable CR for each user based on observed channel conditions and system performance indicators. Distinct from conventional mean squared error (MSE)-centric designs, we introduce a system-aware reward formulation that jointly accounts for spectral efficiency via signal-to-interference-plus-noise ratio (SINR), feedback overhead constraints, and the computational cost of model adaptation. Simulation results on high-dimensional delay-domain CSI datasets demonstrate that the proposed RL-guided framework effectively balances the overhead-accuracy tradeoff and adapts to dynamic channel environments. The proposed method improves spectral efficiency and feedback efficiency compared with fixed compression schemes and adaptive baselines, while maintaining a modest computational and memory footprint. Averaged over different numbers of users and across all considered baselines, the proposed RL framework reduces the CSI feedback cost by more than 53.4%, improves the average downlink sum rate by 53.64%, and reduces the NMSE by 22.38%. These results demonstrate its ability to achieve a more efficient rate-accuracy-feedback tradeoff under dynamic wireless conditions.
: Massive multiple-input multiple-output (MIMO) technology is a key enabler for 5G and beyond wireless networks, offering significant improvements in spectral efficiency and link reliability. However, conventional beamforming techniques such as Zero Forcing (ZF) and Minimum Mean Square Error (MMSE) require complex matrix computations and fail to adapt efficiently to dynamic channel variations. To address these challenges, this paper proposes a Deep Deterministic Policy Gradient (DDPG)-based beamforming framework that formulates beamforming optimization as a continuous-action deep reinforcement learning problem. The proposed model directly generates complex-valued beamforming weight vectors to jointly maximize spectral efficiency (SE) and energy efficiency (EE) while minimizing the bit error rate (BER) and decision latency. An adaptive state representation incorporating channel state information (CSI), previous beamforming vectors, and performance metrics enables real-time policy learning under time-varying channel conditions. Simulation results demonstrate that the proposed method outperforms conventional and heuristic beamforming schemes, achieving up to 25% higher SE, 45% lower BER, 20% improvement in EE, and 30% latency reduction. The results validate the effectiveness of the proposed framework for energy-efficient, low-latency beamforming in next-generation massive MIMO wireless networks.
Nilakshee Rajule, Mithra Venkatesan, Harshada Magar et al.· Proceedings of the 1st Inter...· 0 citations
An autoencoder-based solution to compress CSI at the UE side and reconstruct it at BS and outperforms the conventional CSI feedback scheme by approximately 21% in terms of throughput is proposed.
A. F. Chabi, João Vitor, D. S. Campos et al.· 0 citations
Artificial intelligence (AI) and machine learning (ML) are increasingly applied to wireless and cellular networks. With sixth-generation (6G) systems envisioned as AI-native, reinforcement learning (RL) offers a natural approach to complex network management and operation. This paper focuses on user admission control in multi-cell massive multiple-input multiple-output (MIMO) systems, where naive selfish strategies aiming to maximize local sum-rate can trigger a tragedy of the commons, degrading per-user performance and generating severe inter-cell interference (ICI). To address these challenges, we introduce a structured RL framework for massive MIMO systems. In particular, the policy is structured to introduce physical inductive bias terms, such as an interference-sensitive attenuation factor, which enables interference-aware learning through the open radio access network (O-RAN) architecture. Through stability analysis, we show that such physical inductive bias terms can guarantee network-wide stability. Experimental results demonstrate that the proposed approach balances aggregate spectral efficiency with per-user performance and maintains robustness during traffic surges, whereas selfish strategies suffer from degraded per-user performance.
Jinho Choi· IEEE Transactions on Communi...· 0 citations
Channel estimation provides the channel state information (CSI) required for coherent detection and precoding in multiple-input multiple-output (MIMO) systems. Accurate CSI is particularly critical under strong noise, limited pilot overhead and model mismatch, where conventional estimators often exhibit significant performance degradation. This work introduces a hybrid residual U-Net-bidirectional long short-term memory with attention (ResUNet-BiLSTM-Attention) channel estimator that learns a nonlinear mapping from noisy pilot observations to MIMO channel coefficients. The architecture combines a ResUNet encoder-decoder for multi-scale spatial feature extraction, a BiLSTM module for capturing structured dependencies in the unfolded feature sequence and a self-attention layer that emphasizes globally informative channel components before reconstruction. The model is trained on a synthetically generated [Formula: see text] MIMO dataset with 20, 000 training and 2, 000 validation samples over a wide SNR range, using a normalized mean square error (NMSE) loss for stable convergence. Extensive simulations show that the proposed estimator consistently outperforms both conventional and learning-based baselines. The comparison includes LS, LMMSE, orthogonal matching pursuit (OMP), simultaneous OMP (SOMP), beamspace-based dynamic support detection with windowing (BSP-DSDW), CNN-CE, U-Net-CE, and lightweight attention-based CE. At 25 dB SNR, it attains an NMSE of about [Formula: see text] dB, corresponding to an NMSE gain of about 9-10 dB over BSP-DSDW and a clear improvement over the added learning-based baselines. A module-wise ablation study further verifies the individual contribution of the ResUNet, BiLSTM, and attention blocks, while the runtime evaluation is conducted under a common GPU-enabled benchmarking setup using repeated inference trials. Additional studies on training set size, different user channels, computational complexity, parameter count, FLOPs, memory requirement, inference latency, pilot length, noise factor, and spatial correlation further confirm the robustness and practical feasibility of the proposed design. With a fixed computational structure, the proposed estimator achieves an observed inference latency range of approximately 2.5-4.0 ms per channel realization under the considered compact MIMO setup.
Mohammad Zubair Khan, Ibrahim Aljubayri, C. Prabha et al.· Scientific Reports· 0 citations
Accurate channel state information (CSI) is essential for multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) systems, yet fast time-varying channels pose significant prediction challenges. Traditional approaches fail under high mobility, while deep learning methods rely heavily on large labeled datasets, limiting generalization with scarce training data. Although large language models (LLMs) show promise, their massive parameter count hinders deployment on resource-constrained edge devices. This paper proposes a lightweight, end-to-end CSI prediction framework built upon a general LLM. A time-frequency dual-domain feature extraction module captures subcarrier correlations and temporal dynamics from historical CSI, overcoming single-domain limitations. The end-to-end design maps historical CSI directly to future states, avoiding error propagation inherent in explicit channel estimation. Parameter efficiency is achieved through low-rank adaptation (LoRA) combined with knowledge distillation from a pre-trained LLM, enabling effective few-shot learning at low computational cost. Simulations demonstrate that the proposed scheme delivers robust prediction accuracy across diverse mobility scenar-ios, maintains strong performance under limited training data, and exhibits zero-shot cross-scenario generalization, significantly outperforming both conventional and deep learning baselines in TDD and FDD modes.
Keywords: Channel prediction, multiple-input multiple-output (MIMO), orthogonal frequency division multiplexing (OFDM), large language model (LLM), knowledge distillation, low-rank adaptation (LoRA)
Accurate channel state information (CSI) prediction is essential for mitigating channel aging and feedback delay in communication systems. This letter proposes an enhanced masked autoencoder (MAE) framework for CSI prediction. Specifically, singular value decomposition (SVD) is first applied to CSI reconstruction and noise suppression, preserving dominant signal components while reducing noise interference. Then, a time–frequency hopping sampling strategy is designed to refine the MAE random masking mechanism, improving uniform subcarrier coverage over the time–frequency grid and enabling the model to learn representative channel features. Furthermore, a multi-scale aligned fusion mechanism is employed to aggregate information across resolutions for capturing diverse multipath dynamics with varying time–frequency scales. These three modules act complementarily on input enhancement, observation coverage, and multi-scale representation learning. Experimental results demonstrate that the proposed model achieves higher prediction accuracy while maintaining an improved accuracy–complexity tradeoff compared with baselines. Our code is publicly available at https://github.com/OpenCommAI/Enhanced_MAE
