Spectrum sensing research based on multi-swarm artificial fishswarm algorithm and fuzzy twin SVM

To address the limitations of traditional spectrum sensing algorithms, such as poor robustness in complex channel environments, and the high model training complexity of deep learning-based sensing algorithms, a spectrum sensing method that integrates a multi-population artificial fish swarm algorithm with a fuzzy twin support vector machine (fuzzy twin support vector machine, FTSVM) is proposed. Firstly, the trace of the covariance matrix of the received signal and the mean of its off-diagonal elements are calculated to construct a two-dimensional feature vector, which is then used by FTSVM for training and recognition. Secondly, the fuzzy membership of the samples can adjust the FTSVM hyperplane, thereby making the trained model more inclined to recognize the presence of the primary user's signal. Finally, an improved artificial fish swarm algorithm with a multi-population mechanism is designed to optimize the parameters of the spectrum sensing model. Simulation results show that, when applied to small sample datasets and low signal-to-noise ratio environments, the proposed method achieves a significant improvement in sensing performance compared to other feature extraction and SVM methods. Meanwhile, the multi-population mechanism in the optimization algorithm reduces model training time.