Deep Learning-Driven Pilot-Assisted Signal Processing for MIMO Systems: A Step Towards End-to-End Learning

Massive Input Massive Output (MIMO) systems play a vital role in building next-generation wireless communication technologies, however, channel estimation and signal detection are still challenging. To address these tasks, we propose a deep learning-based pilot-assisted channel estimation and signal detection network called DeepPACESigNet for both channel estimation and signal detection in MIMO systems, respectively. We integrated pilot symbols along with the traditional methods such as least square (LS) and minimum mean square error (MMSE) for channel estimation to determine channel characteristics without channel state information (CSI). Later, this traditionally estimated channel, along with the received data symbols, is integrated into our customized deep learning (DL) model to perform the signal detection task. We considered 4x4 MIMO system, and simulation results show that the proposed DeepPACESigNet model based on the traditionally estimated channel for MMSE is performing better than the DeepPACESigNet model based on the traditionally estimated channel for the LS method. This paper sets the foundation for future developments in joint channel estimation and signal detection with zero CSI knowledge. This research study is in its initial phase, and we will replace these traditional methods with advanced DL-based channel estimation methods to form end-to-end learning for next-generation wireless communication networks.