Neural Network Based Denoising in the Wireless Channel Characterization

Channel denoising is of much importance for further channel characterization and fading analysis. In this paper, we proposed a novel method for the wireless channel impulse response (CIR) denoising based on neural network. Discriminating effective signals from the noise is considered as a binary classification problem, which can be resolved by some machine learning methods. The back propagation neural network (BPNN) model with the amplitude and angle information of CIR as input data is implemented to settle the classification problem. What's more, the Fβ-score, which combines both Precision and Recall together, is selected as the evaluation index to assess the classification performance. Compared with the performance of wavelet transform, the proposed method shows a better denoising result at both high and low signal-to-noise ratio (SNR) due to its full utilization of amplitude and angle information, while the BPNN with only amplitude as input data shows the worst result comparing with the other two methods.