Bedrock Models in Communication and Sensing: Advancing Generalization, Transferability, and Performance

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Deep learning (DL) has emerged as a powerful tool for addressing the intricate challenges inherent in communication and sensing systems, significantly enhancing the intelligence of future sixth-generation (6G) networks. While substantial research has demonstrated the potential of DL-based techniques, challenges remain in ensuring robustness, generalization, and interpretability under highly dynamic and unpredictable environments. To address these limitations, this paper introduces a family of mathematically grounded and modularized models, termed bedrock models, designed for seamless integration into communication and sensing systems. Unlike traditional black-box neural architectures, each bedrock module inherits its structure from classical physical-layer operators, enabling a key rollback capability: when the environment becomes adverse or uncertain, the trainable model can deterministically revert to a closed-form classical solution by simply resetting its parameters. This ensures that the system does not perform worse than the well-understood baseline, while retaining the flexibility of AI enhancements under favorable conditions. In communication systems, bedrock models achieve notable performance gains and exhibit strong transferability, supporting direct parameter reuse across tasks and modulation schemes. In sensing applications, the integration of bedrock models significantly improves performance, reducing delay and Doppler estimation errors by an order of magnitude. Additionally, a transmitter-side pre-equalization strategy is proposed, enabling real-time waveform adaptation using sensing information, while preserving the structure of a pretrained communication model. This approach allows the system to mitigate doubly dispersive channels and maintain near-optimal performance. Extensive simulations validate the effectiveness, robustness, and deployment readiness of the proposed bedrock models across diverse scenarios in both communication and sensing domains.
Funding Information
  • Jiangsu Major Project on Fundamental Research (BK20243059)
  • Gusu Innovation Project (ZXL2024360)
  • High-Tech District of Suzhou City (RC2025001)
  • Fundamental, in part by Interdisciplinary Disciplines Breakthrough Plan of the Ministry of Education of China (JYB2025XDXM118)
  • National Natural Science Foundation of China (62301122)
  • Xiaomi Young Scholar Award

This publication has 37 references indexed in Scilit: