Deep Learning-Based Multi-Tone Interference Suppression for Short Polar Codes

In this paper, a novel approach to detect and mitigate multi-tone interference (MTI) in a coded orthogonal frequency division multiplexing (OFDM) system based on deep neural networks (DNNs) is presented. The proposed approach utilizes the long short-term memory (LSTM) aiming to detect the indices of the jammed subcarriers within an OFDM symbol. Training data that is generated according to the statistical channel model is used for the offline training phase. The LSTMDNN is utilized as a classifier in the online deployment, where each subcarrier is classified as either jamming-free or jammed. The proposed approach based on LSTM-DNN requires no information about the interference characteristics. Simulation results show that the LSTM-DNN can learn and is able to classify the subcarriers with high accuracy compared to the traditional classification technique based on the energy detector (ED). In addition, simulation results demonstrate that the LSTMDNN is more robust to the variations in the wireless channel conditions compared to the ED classifier. As for the MTI mitigation technique, soft log-likelihood ratios (LLRs) weighting for the corresponding jammed subcarriers based on LSTM-DNN is proposed and compared with the traditional LLR erasing technique. Simulation comparison shows that the proposed LLR weighting can provide considerable performance enhancement.