ItemArticle

BORCAE: Bayesian Optimized Residual Convolutional Autoencoder for Efficient Feedback Compression in RIS-Assisted Time-Varying IoT Networks

Title
BORCAE: Bayesian Optimized Residual Convolutional Autoencoder for Efficient Feedback Compression in RIS-Assisted Time-Varying IoT Networks
Creator
Bhardwaj, Vikash Kumar; Singh, Abhishek; Sharma, Shourya; Shukla, Mahendra K.; Pandey, Om Jee
Description
Reconfigurable Intelligent Surfaces (RIS) have strong potential to improve the performance of time-varying Internet of Things (IoT) networks. However, a major challenge in operating RIS effectively is the need for frequent Quantized Phase Configuration (QPC) feedback bits from the Base Station (BS) to the controller. This challenge becomes more serious asthe RIS size grows, since the feedback bandwidth is limited. As a result, efficient compression of control signals is crucial for the practical deployment of RIS. In this work, we propose Bayesian Optimized Residual Convolutional AutoEncoder (BORCAE), a lightweight and noise-resilient feedback compression framework based on a 1D Convolutional Autoencoder with residual connections. The model is designed to reduce QPC feedback size while preserving high reconstruction fidelity. To ensure adaptability across varying deployment conditions, we employ Bayesian hyperparameter optimization using Optuna, which enables automatic tuning of key architectural hyperparameters. This optimization ensures that the architecture generalizes effectively across a wide range of operating scenarios. Additionally, we integrate the Limited Memory Broyden Fletcher Goldfarb Shanno (LBFGS) optimizer during the final training epochs, which accelerates convergence and improves stability. For performance evaluation, we use Normalized Mean Squared Error (NMSE) as the reconstruction metric. Extensive testing across different Signal-to-Interference-plus-Noise Ratio (SINR) levels demonstrates that BORCAE consistently achieves lower NMSE compared to DL-CsiNet and CsiNet. The results highlight the practical viability of BORCAE for RIS-assisted communication, offering improved efficiency, and scalability for real-world IoT and Sixth-Generation (6G) applications. 2020 IEEE.
Source
IEEE Transactions on Artificial Intelligence;
Date
01-01-2025
Publisher
Institute of Electrical and Electronics Engineers Inc.
Subject
autoencoder; Bayesian optimization; Internet of Things (IoT); Limited Memory Broyden Fletcher Goldfarb Shanno (LBFGS).; Reconfigurable Intelligent Surface (RIS)
Coverage
Bhardwaj V.K., Indian Institute of Technology (BHU), Department of Electronics Engineering, Varanasi, 221005, India; Singh A., Indian Institute of Information Technology Bhagalpur, Department of Computer Science and Engineering, Bhagalpur, 813210, India; Sharma S., Christ University, Department of Electronics and Communication Engineering, Bengaluru, 560029, India; Shukla M.K., Christ University, ABVIIITM Gwalior, Department of Information Technology, 474015, India; Pandey O.J., Indian Institute of Technology (BHU), Department of Electronics Engineering, Varanasi, 221005, India
Rights
Restricted Access; Hardcopy may be available in the library
Relation
ISSN: 26914581;
Format
online
Language
English
Type
Article

Citation

Bhardwaj, Vikash Kumar; Singh, Abhishek; Sharma, Shourya; Shukla, Mahendra K.; Pandey, Om Jee, “BORCAE: Bayesian Optimized Residual Convolutional Autoencoder for Efficient Feedback Compression in RIS-Assisted Time-Varying IoT Networks,” CHRIST (Deemed To Be University) Institutional Repository, accessed June 18, 2026, https://archives.christuniversity.in/items/show/22962.

Output Formats