Research On The Matching Technology Of Signals For The Same Target Based On Doppler And Deep Learning Detection

With the development of wireless communication technology,on the one hand,the number of communication devices can coexist in a certain range.On the other hand,for a single device,it can transmit signals with different working frequencies at the same time,resulting in an increasing amount of data received and stored by the monitoring platform,that is,the received data contains signals of different targets at different working frequencies.However,for the monitoring party,it is impossible to distinguish which targets these signals come from.If these signals are directly used for target tracking and positioning,it will produce huge redundant calculation and consume human and material resources.If we can find a method from the massive signal data,we can classify the signal according to the monitored target,and only process the signal on a single target,so as to reduce the signal processing scale.It is found that in the process of target monitoring,due to the relative motion between the monitoring platform and the monitored target,the signal spectrum at the monitoring end has Doppler frequency offset,and because the Doppler frequency offset changes of different signals on the same target in the same monitoring platform are consistent,it can be judged whether the signal belongs to the same target by comparing the Doppler change information between different signals.If we want to realize the comparison of Doppler information,we need to detect this kind of signal,but the current signal detection algorithm can only capture the rough position of signal distribution,and can’t describe more precise signal distribution information.Aiming at the limitations of existing signal detection algorithms,this thesis proposes a signal detection algorithm based on Doppler depth learning semantic segmentation,which can detect the time-frequency distribution information of the signal at the pixel level in the spectrum waterfall.The Doppler frequency offset information of the signal is captured through the signal detection algorithm,which is characterized on the twodimensional shape in the waterfall.Then a signal matching method with the target based on Doppler is proposed to judge whether the signal comes from the same target by comparing the shape similarity of the signal in the waterfall.The specific research work of this thesis is as follows:(1)A signal detection algorithm based on Doppler depth learning semantic segmentation is proposed.Firstly,this thesis expounds the construction background of Doppler scene,simulates and generates Doppler signal for this scene,uses this simulation data to make Doppler signal spectrum waterfall data set,then extracts the semantic features of Doppler signal in waterfall through deep learning semantic segmentation network,optimizes and improves deep learning semantic segmentation network,and finally uses semantic segmentation network model to detect the time-frequency information of Doppler signal.In this thesis,the detection performance of the algorithm is verified by simulating Doppler signal data under different Signal-to-Noise Ratios.(2)This thesis presents a method of signal target matching based on Doppler.Firstly,the Doppler frequency offset variation characteristics of signals on the same target and the factors affecting its variation are analyzed,and then the Doppler signal spectrum waterfall data processed by the signal detection algorithm is preprocessed for bandwidth and frequency adaptation,so that the signals on the same target are in the same matching environment.Based on the Doppler signal shape feature extraction method,the two signal shape features are matched and the similarity between the two shapes is calculated,and then whether the signal comes from the same target platform is judged according to the similarity.Considering that the distribution of Doppler signal on the waterfall diagram does not have rotation,the matching time of the algorithm is shortened by matching the two shape contour point sets one by one according to the sampling order instead of the matching method of dynamic programming algorithm.Aiming at the situation that the signal Doppler information representation is not obvious due to the short sampling time,this study proposes the same target matching method based on time series.By counting the matching results of different sampling times,it can judge whether the two signals come from the same target platform,so as to avoid the contingency of single shape matching results.In this thesis,the performance of signal to target matching algorithm is verified by simulating Doppler signal data under different Signal-to-Noise Ratio.