Narrowband Interference Cancellation for OFDM Based on Deep Learning and Compressed Sensing

Abstract

The orthogonal frequency division multiplexing (OFDM) technology has been widely used in modern wireless communication systems. Under the hostile wireless propagation channels, the transmitted signal may be corrupted by narrowband interference (NBI), resulting in the loss of data in part of the system band. To address this challenging problem, we propose a joint deep learning (DL) and compressed sensing (CS) approach to estimate and eliminate multiple NBIs. With unknown interfering sources, we first propose an NBI detection network (NDNet) trained with a new loss function to identify the number of NBIs. Different from existing networks, NDNet is designed to cope with both synchronous NBI (S-NBI) and asynchronous NBI (A-NBI). Based on the output of NDNet, an orthogonal matching pursuit (OMP) and improved dichotomous search (IDS) based NBI cancellation scheme, which is referred to as the OMP-IDS algorithm, is proposed to accurately estimate NBIs at a modest complexity. Furthermore, an enhanced OMP-IDS (eOMP-IDS) algorithm is devised to reduce the errors in estimating the frequencies interfered especially by multiple adjacent NBIs. The estimated NBIs can then be effectively cancelled. Theoretical analysis, simulations and experiments validate the feasibility and competitiveness of the proposed schemes.