BiLSTM Using Wiener Filter Online Training for Spectral Efficiency Enhancement in IRS-Assisted MU-MISO Systems

Syeda Meher Unnisa, Nazia Parveen

Abstract: These Intelligent Reflecting Surface (IRS)-assisted Multi-User Multiple-Input Single-Output (MU-MISO) communication has gained considerable attention due to its capability to enhance spectral efficiency by intelligently reshaping the wireless propagation environment through passive signal reflection. Despite recent progress, most existing deep learning-based IRS optimization techniques rely on offline training, which limits their robustness under time-varying and noisy channel conditions. To address this limitation, this paper proposes an online adaptive Bidirectional Long Short-Term Memory (BiLSTM) framework combined with a Wiener filter to improve noise robustness and spectral efficiency performance. The Wiener filter enables effective noise suppression, while the Bi-LSTM model continuously updates its parameters in real time to track dynamic channel variations. Extensive simulations are conducted in Python using the Google Colab platform under three different scenarios, including two baseline models for performance comparison. Quantitative results demonstrate that the proposed method achieves a spectral efficiency of 26.13 bits/s/Hz and a training loss of 0.62 at 30 dB SNR, consistently outperforming baseline approaches across the entire SNR range. These results confirm that the integration of online learning with Wiener filtering significantly enhances system stability and adaptability, making the proposed approach a promising solution for future IRS-assisted wireless communication systems.

Keywords: Intelligent Reflecting Surface (IRS), MU-MISO Systems, Spectral Efficiency (SE), BiLSTM, Wiener Filter

How to Cite?: Syeda Meher Unnisa, Nazia Parveen, "BiLSTM Using Wiener Filter Online Training for Spectral Efficiency Enhancement in IRS-Assisted MU-MISO Systems", Volume 15 Issue 1, January 2026, International Journal of Science and Research (IJSR), Pages: 1369-1376, https://www.ijsr.net/getabstract.php?paperid=SR26117142516, DOI: https://dx.doi.org/10.21275/SR26117142516