Deep Learning-Based Impulsive Noise Detection and Impulsive-Noise-Aware Receiver for DFT-S-OFDM Systems

Wireless communication systems commonly assume noise is additive white Gaussian noise (AWGN). However, in real-world scenarios, noise is often non-Gaussian, rendering AWGN-based algorithms less effective. Impulsive noise, a typical non-Gaussian noise, can significantly degrade system performance. Accurate impulsive noise position detection is crucial, as many suppression methods depend on this information. This paper proposes an impulsive noise detection neural network (INDNet), integrating a convolutional neural network (CNN) and bidirectional gated recurrent unit (Bi-GRU) to detect impulsive noise positions. Leveraging these detection results, an impulsive-noise-aware receiver is further developed for discrete Fourier transform spread orthogonal frequency division multiplexing (DFT-S-OFDM) systems. Simulation results demonstrate that INDNet achieves superior detection performance compared to the conventional threshold-based impulse detection method and other advanced deep learning networks. In particular, at the impulsive noise-to-background noise ratio (INR) of 20 dB and large Eb/N0, INDNet maintains the detection probability of up to 86.51%, achieving a maximum improvement of 44.05% over other methods. Additionally, the proposed INDNet demonstrates robust generalization capability. The proposed impulsive-noise-aware receiver effectively suppresses impulsive noise regardless of its occurrence probability. When this probability is low, the bit error rate (BER) performance of the proposed receiver closely approaches that in an AWGN scenario.