An unsupervised machine learning scheme for index-based CSI feedback in Wi-Fi

High-speed wireless data transmission relies heavily on beamforming to improve data rates and signal-to-noise ratios (SNR). However, traditional feedback mechanisms introduce significant overhead and system complexity, challenging the efficiency of modern networks. This paper investigates novel index-based feedback mechanisms designed to reduce beamforming overhead in Wi-Fi links. We propose an unsupervised learning-based framework to generate a set of candidate beamforming vectors, allowing the receiver to transmit only the index of the best candidate instead of the full beamforming matrix. We explore five distinct methods for generating and representing these candidate sets, utilizing channel outer product matrices, matrix serialization, and effective distance metrics. Furthermore, we analyze the impact of using partial information in compressed feedback and compare it with our proposed index-based approaches. Extensive IEEE 802.11 standard-compliant simulations demonstrate that our methods effectively minimize feedback overhead, enhancing throughput while maintaining robust link performance.