An UWB Cyclostationary Detection Algorithm Based on Nonparametric Cusum

Abstract

Ultra-wideband (UWB) positioning has become a hot method for indoor navigation and positioning due to its features of high positioning accuracy, good resolution, strong penetration and strong anti-interference. Ultra-wideband signal detection is of great significance to realize ultra-wideband positioning. Ultra-wideband signals are difficult to be detected because of their very low power spectral density. In order to solve this problem, the ultra-wideband signal detection based on cyclic stationary feature is often used. However, there are still many problems in this method, such as low detection probability and large detection delay at low SNR(signal-to-noise ratio). In view of the above problems, this paper proposes an uwb detection algorithm based on non-parametric cumulative sum (NCUSUM). Firstly, the three-dimensional cyclic spectrum of the signal is solved, the 3-dimensional map is normalized to two-dimensional grayscale, and grayscales with signal and noise have large difference, then by using the method of image classification, signal detection is achieved, such as putting the gray images into the convolution Neural Network (CNN) to train and extract feature. After extracting two types of probabilities output from network softmax layer, the signal probability is regarded as statistic. In this way, the previous observation data can be saved and the detection accuracy is higher. The simulation results show that the detection probability of the proposed algorithm is significantly improved and the detection delay is significantly reduced at low SNR compared with the cyclostationary detection based on gray images.