Research On Signal Detection Classification And Recovery Technology Based On MWC

High sampling rates and storage of large amounts of instantaneous sampled data are major constraints on non-cooperative wideband signal processing.Sub-Nyquist sampling techniques can exploit the sparsity of wideband signals to sample wideband signals at much lower sampling rates and with less sampled data than the Nyquist rate.In this paper,a sub-Nyquist sampling technique based on Modulated Wideband Converter(MWC)is studied in depth,and on this basis,wideband spectrum sensing,signal classification and identification,and signal reconstruction are investigated.The main research elements and innovations in this paper are as follows:1.For the problems of poor detection performance and cumbersome threshold derivation at low signal-to-noise ratio of traditional signal detection methods in broadband spectrum sensing,a deep learning-based broadband signal pre-detection method is proposed to improve the detection probability and resistance to noise uncertainty.To address the problems of insufficient real-time performance and the need for a priori information in broadband spectrum sensing,a sensing algorithm that uses multiple powers of the inverse matrix of the autocorrelation matrix of the sampled samples to approximate the noise subspace is designed by combining the relationship between the array signal processing model and the MWC system structure,which achieved spectrum blind sensing.2.In order to solve the problem of low recognition rate in signal classification and recognition with low signal-to-noise ratio,the inter-class recognition and intra-class recognition of signals are carried out successively under the condition of nonreconstruction.In this paper,a classification algorithm of frequency-hopping signals and fixed-frequency signals is proposed by using the intersection of perception results of time slices.At the same time,a non-reconfigurable carrier frequency estimation method is designed based on the perception algorithm proposed in this paper.For intra-class recognition,an intra-class recognition method based on random forest is proposed.By calculating the energy focusing efficiency of three kinds of lag products of compressed signals,the intra-class recognition of fixed frequency signals is completed by using random forest classifier,which significantly improves the recognition probability under low signal-to-noise ratio.3.For the signal reconstruction problem,firstly,under the condition that the signal sparsity is known,for the problem that the matching tracking class algorithm cannot reflect the relationship between residuals and atoms by inner product under low signalto-noise ratio,a correlation coefficient is selected to replace the inner product in a mathematical sense,and an algorithm is proposed to adapt to the low signal-to-noise ratio environment by combining the characteristics of sub-band signals occupying at most two spectrum segments under MWC system.On the other hand,a fast blind reconstruction algorithm is proposed under the condition of unknown signal sparsity by combining the perception algorithm and the characteristics of subband signals occupying at most two spectrum segments under MWC system.Finally,based on the deficiency of the algorithm that it is easy to adjudicate the support set error of adjacent spectrum fragments at the carrier frequency location under high SNR,an adaptive stopping condition of the greedy iterative algorithm is investigated in combination with the continuous to finite(CTF)module of the MWC system,which breaks the signal sparsity a priori limitation.Moreover,in order to reduce the number of pseudo-nonzero elements,the signal number preestimation is combined with the perceptual algorithm to reduce the reconstruction error and achieve the blind reconstruction.