Modified parallel FFT energy detection using machine learning based spectrum sensing

Abstract

The research presents an enhanced energy detector using windowing groups, machine learning, and parallel Fast Fourier Transforms to alleviate spectrum congestion in fifth-generation wireless services. Specifically designed for non-stationary signals with low signal-to-noise ratios, this technique addresses key challenges by improving Detection Probability (Pd) and augmenting FFT resolution. By applying specific weighting factors to samples within the sensing frame, the Probability of Detection (Pd) is increased. The Machine Learning algorithm dynamically adjusts the weighting factor multipliers based on the prevailing signal-to-noise conditions. Implementing parallel FFTs for sample groups further enhances resolution. Diverse windowing methods and grouping strategies significantly boost detection probability, especially for non-stationary signals under low SNRs. Compared to the conventional energy detector with 56% detection probability, the proposed method achieves 76–97% probability at − 15 dB SNR proving its efficiency in improving signal detection under challenging conditions.