Sustainable LSTM-Based Precoding for RIS-Aided mmWave MIMO Systems with Implicit CSI

TL;DR

The paper addresses the high pilot overhead and complexity of RIS-aided mmWave MIMO by proposing a CSI-free, LSTM-based precoding framework that uses uplink pilot observations to implicitly learn channel characteristics. It integrates a practical phase-amplitude coupling model for RIS elements and adopts a multi-label training strategy to handle near-optimal codewords within a predefined DFT codebook, enabling robust performance. Empirically, the approach achieves over 90% of exhaustive-search spectral efficiency with only a small fraction of the computational cost, demonstrates resilience to distribution mismatch, and scales to larger RIS arrays, supporting energy-efficient, real-time deployment for sustainable 6G networks. Overall, the method offers a practical, low-overhead path to near-optimal RIS precoding in dynamic mmWave environments with substantial energy savings.

Abstract

In this paper, we propose a sustainable long short-term memory (LSTM)-based precoding framework for reconfigurable intelligent surface (RIS)-assisted millimeter-wave (mmWave) MIMO systems. Instead of explicit channel state information (CSI) estimation, the framework exploits uplink pilot sequences to implicitly learn channel characteristics, reducing both pilot overhead and inference complexity. Practical hardware constraints are addressed by incorporating the phase-dependent amplitude model of RIS elements, while a multi-label training strategy improves robustness when multiple near-optimal codewords yield comparable performance. Simulations show that the proposed design achieves over 90% of the spectral efficiency of exhaustive search (ES) with only 2.2% of its computation time, cutting energy consumption by nearly two orders of magnitude. The method also demonstrates resilience under distribution mismatch and scalability to larger RIS arrays, making it a practical and energy-efficient solution for sustainable 6G wireless networks.

Figures (5)

• Figure 1: The RIS-assisted mmWave MIMO system.
• Figure 2: LSTM model training flow chart.
• Figure 3: Illustration of one-hot encoding for multi-labeled classification.
• Figure 4: Spectral efficiency versus transmission power, comparing LSTM, CNN, ES, AO, and random selection ($K_t = K_r = 10$, $N_{t} = 10$, $N_{r} = 2$, $N = 64$, and $L = 2$).
• Figure 5: Spectral efficiency versus transmission power, comparing single-labeled and multi-labeled LSTM under ideal and practical RIS models.