Towards Efficient Channel Compression with Spiking Neural Networks for MIMO Systems

In the context of next-generation wireless communications, Multiple-Input and Multiple-Output (MIMO) systems play a crucial role in enhancing data rates and reliability. A critical component of these systems is Channel Sounding, which requires the transmission and feedback of Channel Frequency Response (CFR) information. However, this process can incur high overhead, especially in dense networks and high-frequency bands. This thesis investigates efficient CFR compression techniques to reduce the feedback burden without sacrificing reconstruction quality. In particular, we explore the potential of Spiking Neural Networks (SNNs) as a low-power, biologically inspired alternative to traditional neural networks for this task. We leverage Sionna, a TensorFlow-based open-source library, to simulate a realistic MIMO Wi-Fi communication scenario and generate custom datasets. Using this data, we design and evaluate two novel approaches for CFR compression, both of which represent original contributions to the scientific literature.