Comparison of Neural Network and Statistical Approaches to the Problem of Signal Detection

With the growing interest in deep learning technologies and empirical data confirming the effectiveness of neural networks in solving tasks, including signal detection problems, the effectiveness of neural networks under the influence of non-Gaussian interferences, a factor that significantly complicates the detection process, remains understudied. Thus, the paper considers the possibility of using neural network technologies to detect signals in scenarios characterized by complex non-Gaussian disturbances with inherent uncertainty in the probability distribution. The research carried out a synthesis and comparative analysis of two algorithms: a statistical adaptive algorithm and a neural network with Fourier transform, designed to ascertain a harmonic signal characterized by an undetermined phase against the background of interference, simulated using an autoregressive model, caused by non-Gaussian interference. According to the simulation results, implementing of neural network technologies for signal detection in the conditions of complex non-Gaussian interferences of the impulse type characterized by a priori uncertainty of probability distributions is characterized by high resistance to the effects of disturbances. However, the application of statistical procedures enables the synthesis of optimal algorithms, robust to intricate interference.