Fading Channel Classification with Walsh-Hadamard Transform and Convolutional Neural Network

Fading channel classification is a useful function in the design of wireless communications because the knowledge of the channel state information can help the subsequent steps in the wireless communication processing including the information symbols extraction from the received signal. This paper proposes the application of the Walsh-Hadamard Transform (WHT) in combination with Convolutional Neural Network (CNN) for the problem of fading channel classification. WHT belongs to the generalized class of Fourier transforms and it is a non-sinusoidal, orthogonal transformation technique that decomposes a signal into a set of Walsh functions. WHT has been used in image processing but less in the wireless communication domain. CNN has been recently used in many wireless communications problems including fading channel classification, where it has shown to outperform ’shallow’ machine learning algorithms. This paper presents the novel combination of WHT with CNN for the problem of channel classification. The approach is applied to a data set of chirp signals derived from the technical specification of the radar altimeter, which is submitted to different fading conditions in a channel emulator implemented with FPGA in a radio frequency laboratory. The results show that the proposed approach is able to significantly outperform (especially in presence of noise) the application of CNN on the original time-based representation of the signal or the spectral domain representation based on the use of the Fourier transform and Wavelet transform.