Learning RIS Configuration with Quantized Responses: A Neuroevolution-Trained Multi-Branch Attention Convolutional Neural Network

In this paper, we consider the problem of jointly controlling the configuration of a Reconfigurable Intelligent Surface (RIS) with unit elements of quantized responses and a codebook-based transmit precoder in RIS-empowered multiple-input single-output communication systems. The adjustable elements...

Full description

Saved in:
Bibliographic Details
Published in:IEEE International Conference on Communications workshops pp. 1894 - 1899
Main Authors: Stamatelis, George, Stylianopoulos, Kyriakos, Alexandropoulos, George C.
Format: Conference Proceeding
Language:English
Published: IEEE 08.06.2025
Subjects:
adaptive decision making
attention network
Benchmark testing
Conferences
Convolutional neural networks
deep neuroevolution
MISO
Optimization
Performance evaluation
Precoding
Real-time systems
Reconfigurable intelligent surface
Reconfigurable intelligent surfaces
Vectors
ISSN:2694-2941
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In this paper, we consider the problem of jointly controlling the configuration of a Reconfigurable Intelligent Surface (RIS) with unit elements of quantized responses and a codebook-based transmit precoder in RIS-empowered multiple-input single-output communication systems. The adjustable elements of the RIS and the precoding vector need to be jointly updated in real time to account for rapid changes in the wireless channels, making the application of complicated discrete optimization algorithms impractical. We present a novel multi-branch attention convolutional neural network architecture for this design objective which is optimized using neuroevolution, leveraging its capability to effectively tackle the non-differentiable problem arising from the quantized phase states of the RIS elements. The channel matrices of all involved links are first passed to separate self-attention layers to obtain initial embeddings, which are then concatenated and passed to a convolutional network for spatial feature extraction, before being fed to a per-element multi-layered perceptron for the final RIS phase configuration calculation. Our numerical investigattions showcase the superiority of our approach over both learning-based and classical discrete optimization benchmarks.
ISSN:2694-2941
DOI:10.1109/ICCWorkshops67674.2025.11162282