Abstract:Deep learning is making a profound impact in the physical layer of wireless communications. Despite exhibiting outstanding empirical performance in tasks such as MIMO receive processing, the reasons behind the demonstrated superior performance improvement remain largely unclear. In this work, we advance the field of Explainable AI (xAI) in the physical layer of wireless communications utilizing signal processing principles. Specifically, we focus on the task of MIMO-OFDM receive processing (e.g., symbol detection) using reservoir computing (RC), a framework within recurrent neural networks (RNNs), which outperforms both conventional and other learning-based MIMO detectors. Our analysis provides a signal processing-based, first-principles understanding of the corresponding operation of the RC. Building on this fundamental understanding, we are able to systematically incorporate the domain knowledge of wireless systems (e.g., channel statistics) into the design of the underlying RNN by directly configuring the untrained RNN weights for MIMO-OFDM symbol detection. The introduced RNN weight configuration has been validated through extensive simulations demonstrating significant performance improvements. This establishes a foundation for explainable RC-based architectures in MIMO-OFDM receive processing and provides a roadmap for incorporating domain knowledge into the design of neural networks for NextG systems.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is in the physical layer of wireless communication, how to use the principles of signal processing to systematically incorporate domain knowledge (such as channel statistical information) into the design of Recurrent Neural Networks (RNNs) in order to improve the performance of MIMO - OFDM reception processing tasks (such as symbol detection). Specifically, the authors focus on using Reservoir Computing (RC), which is a method within the RNN framework and shows better performance than traditional and other learning - based methods in MIMO reception processing tasks. ### Main problems and objectives 1. **Application of Explainable AI (xAI)**: Although deep learning has demonstrated excellent empirical performance in MIMO reception processing tasks, the reasons behind it are still unclear. This paper aims to promote the research of Explainable AI (xAI) in the physical layer of wireless communication by combining the basic principles of signal processing. 2. **Systematic introduction of domain knowledge**: Based on the basic understanding of RC operations, the authors are able to systematically configure the domain knowledge of wireless systems (such as channel statistical information) directly into the untrained RNN weights, thereby achieving more efficient MIMO - OFDM symbol detection. 3. **Verification and improvement**: Through extensive simulation experiments, the effectiveness of the proposed RNN weight configuration method is verified, and significant performance improvements are demonstrated. This provides a theoretical basis for MIMO - OFDM reception processing based on the RC architecture and a roadmap for incorporating domain knowledge into the neural network design of next - generation systems (NextG). ### Specific contributions - **Time - domain weight configuration**: A new "time - domain" method for configuring the untrained weights of ESN is introduced, expanding the frequency - domain method introduced in previous work. - **Theoretical explanation**: A theoretical connection between symbol detection on non - minimum - phase wireless channels and the WESN structure is provided, explaining the effectiveness of the WESN architecture. - **Extended configuration method**: The frequency - domain and time - domain weight configuration methods are extended to be applicable to OFDM and MIMO - OFDM symbol detection, and a parameterized MIMO channel representation is utilized. - **Simulation verification**: Through extensive simulations, the effectiveness of the OFDM and MIMO - OFDM WESN weight configuration procedures is verified in 5G/5G - Advanced scenarios. ### Summary This paper systematically improves the RNN weight configuration method by combining signal processing principles and domain knowledge, thereby improving the performance of MIMO - OFDM reception processing and providing theoretical basis and practical guidance for future research.

Towards xAI: Configuring RNN Weights using Domain Knowledge for MIMO Receive Processing

M-Net: A Lightweight Network Based on Multilayer Perceptron for Massive MIMO CSI Feedback

Towards Explainable Machine Learning: The Effectiveness of Reservoir Computing in Wireless Receive Processing

Realizing Over-the-Air Neural Networks in RIS-Assisted MIMO Communication Systems

Low-Complex AI-Empowered Receiver for Spatial Media-Based Modulation MIMO Systems

Joint Deep Reinforcement Learning and Unfolding: Beam Selection and Precoding for Mmwave Multiuser MIMO with Lens Arrays

Learning at the Speed of Wireless: Online Real-Time Learning for AI-Enabled MIMO in NextG

Reconfigurable AI Modules Aided Channel Estimation and MIMO Detection

Deep Reinforcement Learning Based on Location-Aware Imitation Environment for RIS-Aided Mmwave MIMO Systems

A Robust Deep Learning-Based Beamforming Design for RIS-Assisted Multiuser MISO Communications with Practical Constraints

RISnet: A Domain-Knowledge Driven Neural Network Architecture for RIS Optimization with Mutual Coupling and Partial CSI

Rectangular Differential Reflecting Spatial Modulation: A Noncoherent Joint Index-Modulation of RIS-assisted MIMO System

A Deep Learning-Based Intelligent Receiver for Improving the Reliability of the MIMO Wireless Communication System

DMCNet-Pro: A Model-Driven Multi-Pilot Convolution Neural Network for MIMO-OFDM Receivers

Reconfigurable Intelligent Surface Assisted Multiuser MISO Systems Exploiting Deep Reinforcement Learning

AI Empowered RIS-Assisted NOMA Networks: Deep Learning or Reinforcement Learning?

Adaptive and Flexible Model-Based AI for Deep Receivers in Dynamic Channels

Application of Reinforcement Learning and Deep Learning in Multiple-Input and Multiple-Output (MIMO) Systems

Application of deep neural network and deep reinforcement learning in wireless communication

Deep Learning Based MIMO Communications

Spatio-Temporal Representation with Deep Neural Recurrent Network in MIMO CSI Feedback