Research On Single Antenna ADS-B Overlapping Signal Separation Method Based On Deep Learning

Automatic Dependent Surveillance-Broadcast(ADS-B)technology,as the main monitoring technology of air traffic control,will become an important support for future aviation safety and operational efficiency.However,the potential problem with this technology is that ADS-B receivers have multiple interleaved signals when receiving airplane message signals.With the wide application of single antenna ADS-B receiver in practical engineering,the interweaving problem of single antenna ADS-B needs to be solved urgently.In view of the above problems,this thesis makes a detailed analysis of the characteristics of ADS-B signals,and based on deep learning,focuses on the single antenna ADS-B interleaved signal separation based on deep learning.The main work is as follows:(1)By studying the principle of ADS-B signal,a mathematical model of single antenna ADS-B interwoven signal is established.The characteristics of ADS-B signal structure and ADS-B signal interleaving are analyzed,which indicates the direction for the selection of subsequent neural networks and the construction of datasets.(2)To solve the overlap problem of ADS-B signals under single antenna,a deep learning separation algorithm based on Temporal Convolutional Network(TCN)is studied.The algorithm first extracts the mixed characteristics of ADS-B interleaved signals in the time domain by using the encoder,then extracts the source ADS-B signal characteristics from the encoder output by using TCN,and finally reconstructs the ADS-B signal’s time domain waveform by using the decoder according to the source ADS-B signal characteristics.The experimental results show that the algorithm validates the feasibility of in-depth learning for ADS-B interleaved signal separation.It can achieve end-to-end signal separation in time domain without manual feature extraction,and the separation accuracy of the algorithm is higher than that of the traditional algorithm——PASA algorithm.(3)In order to make the deep learning separation algorithm easier to be used in engineering,a lightweight deep learning separation algorithm based on Dual Path RNN(DPRNN)is studied for micro-resource platforms.The algorithm does not need to preprocess the signal data,and can directly separate the input one-dimensional ADS-B interlaced signal in time domain.The model size is small,and it is suitable for deployment on small resource platforms and easy to implement in engineering.