BERT Aided Error Correction with Natural Redundancy for Signal Detection

This paper considers utilizing the widely existed natural redundancy (NR) for error correction to improve the signal detection performance. To exploit the NR in the transmitted sources, the pre-trained models (PTMs) are introduced into signal detection. Intended for uncompressed Chinese text sources, an effective error correction scheme aided by the Bidirectional Encoder Representation from Transformers (BERT) model is proposed for signal detection. Since the BERT model has a certain de noising ability, our signal detection scheme is essentially equivalent to adding an error-correcting decoder to the classical detector, thus the performance gain can be obtained. Simulation results show that performance of the proposed BERT aided signal detector can achieve considerable performance gains over the classical signal detector.