Learning of Discrepancy to Validate Decoding Results with Error-Correcting Codes

In the realm of mobile communications, the receiver often encounters undesired situations where it receives only noise, interference, or unintended signals. In such cases, it may incorrectly infer that the decoding result is valid, particularly when the error detection capability, mainly provided by a cyclic redundancy check (CR C) code, falls short. While bounded distance decoding techniques based on the employed error-correcting code have been introduced to address this issue, there is still room for improvement especially when dealing with unintended signals. In this paper, we present a formula to identify the situation of unintended signals and find it expressed in terms of a metric called discrepancy. Utilizing this discrepancy, we interpret the problem of determining the validity of decoding results as a binary classification task. Then, we propose suitable machine learning techniques to improve the accuracy of the binary classification process. Our experimental results, conducted for the 5G New Radio (NR) system, demonstrate significant performance improvement resulting from the application of the proposed methods.