Blind Channel Identification with Fractional Fourier Transform and Machine Learning

Blind channel identification is an important function in wireless communication systems. This paper proposes a new approach based on the application of the Fractional Fourier Transform (FRFT) in combination with machine learning and statistical features commonly adopted in the literature. The approach is evaluated against a data set created by the authors using laboratory equipment including a signal generator, channel emulator, and a spectrum analyzer. Signals with six different fading models including the baseline case of absence of fading are generated for the evaluation of the proposed approach. The impact of white gaussian noise is also evaluated. The results show that the application of statistical features in combination with FRFT significantly outperforms the application of statistical features in the time domain as commonly adopted in the literature. The impact of the rotation angle α in the definition of FRFT is also assessed in combination with a Sequential Forward Feature Selection (SFFS) algorithm to decrease the number of features in the application of the machine learning algorithm. The results of the (SFFS) algorithm are coherent with the findings in the literature on the choice of the optimal statistical features.