Real-Time Cross-Layer Semantic Error Correction Using Language Models and Software-Defined Radio

2026-04-09

AI Summary

This work addresses the limitations of existing semantic error correction approaches, which often prioritize intent understanding over verbatim recovery accuracy and lack real-time validation. We propose a cross-layer semantic error correction framework that integrates physical-layer log-likelihood ratios (LLRs) from software-defined radio (SDR) platforms with semantic context derived from encoder-decoder language models to enable efficient, real-time correction. For the first time, we validate the real-time performance of such a method on a real-world SDR testbed, introducing middleware that supports FPGA-based LLR extraction and a universal interface for language model inference. Experimental results demonstrate that our cross-layer fusion approach significantly outperforms baseline methods relying solely on either the physical or semantic layer in terms of verbatim recovery accuracy.

Abstract

As Language Models (LMs) advance, Semantic Error Correction (SEC) has emerged as a promising approach for reliable network designs. Yet existing methods prioritize intent over accuracy, falling short of verbatim recovery. Our recent work, Cross-Layer SEC (CL-SEC), addressed this by fusing physical-layer Log-Likelihood Ratios (LLRs) with semantic context, but its real-time feasibility remained unvalidated. This paper demonstrates CL-SEC on a live Software-Defined Radio (SDR) testbed, resolving implementation barriers with: 1) an SDR middleware enabling real-time LLR extraction from FPGA hardware, and 2) a generalized inference interface supporting modern encoder-decoder LMs. Real-world experiments confirm that the cross-layer fusion significantly outperforms either source alone.

Problem

• Semantic Error Correction
• Real-Time Feasibility
• Cross-Layer Fusion
• Verbatim Recovery
• Software-Defined Radio

Innovation

• Cross-Layer Semantic Error Correction
• Software-Defined Radio
• Log-Likelihood Ratios
• Language Models
• Real-Time Inference