Variational autoencoder-enhanced deep neural network-based detection for MIMO systems

Abstract

Lately, there has been a substantial surge of interest in artificial intelligence (AI) as a promising technology to tremendously elevate the efficiency of multiple-input multiple-output (MIMO) detection within wireless communication networks. AI-aided methodologies like deep neural networks (DNNs) have empowered MIMO receivers to understand intricate channel dynamics and interference contexts, ultimately leading to noteworthy enhancements in throughput and the performance of error rates. However, the techniques in existing literature face challenges in effectively reconciling accuracy and efficiency across a spectrum of channel conditions. Therefore, this study presents a state-of-the-art DNN-centric architecture for MIMO signal detection called variational autoencoder-enhanced DNN-based detection (VAE-DNN-Det), which harnesses the power of variational autoencoders (VAEs) to efficiently capture underlying data distributions, thereby enhancing the DNN’s ability to adapt to complex channel scenarios and achieve near-optimal performance. Simulations are carried out to contrast the bit error rate (BER) performance of the proposed scheme with traditional methods, where the proposed VAE-DNN-Det method attains a signal-to-noise ratio gain of nearly 1 dB compared to the traditional detection methods.