Research on Line Spectrum Enhancement Technology Based on Unsupervised Neural Network

The adaptive line spectrum enhancer is a classic adaptive filter that can be used to filter, reduce noise, and other processing of signals. In this paper, an adaptive filtering algorithm is designed and simulated based on the minimum mean squared error algorithm and the steepest descent method, and the adaptive line spectrum intensifier is designed to continuously process the input signal by using the filter in the form of feedback to gradually adapt to the characteristics of the signal and continuously optimize the coefficient of the filter. However, the parameters of traditional ALE are fixed, that is, the filter in the form of linear feedback is used, which limits its application range and performance. In view of the limitations of traditional ALE, a line spectrum enhancement technique based on unsupervised neural network is attempted, which refers to the princple of ALE, trains the signal in an unsupervised form, designs an unsupervised neural network model by using the correlation characteristics of the input signal and its delay version, simulates and compares the performance differences between the two algorithms, and observes the influence of different parameters on the algorithm. The results show that the coefficient update speed of traditional ALE is slow, so when the frequency of the signal changes, it takes a certain amount of time to adapt to the new frequency characteristics, resulting in poor filtering effect. Line spectrum enhancement technology based on unsupervised neural network has strong robustness to the input signal-to-noise ratio, because the neural network has the ability to fit complex signals, and its output signal is closer to the target signal and has a higher signal-to-noise ratio gain.