Deep neural network based anti-jamming processing in the GNSS array receiver

Signal anti-jamming has always been a difficult problem in GNSS (global navigation satellite system) signal processing. There are many GNSS anti-jamming techniques in the existing research, which can achieve good results if the interferences are sparsely distinguishable in some signal feature domains. Specifically, the single antenna based anti-jamming techniques cannot deal with wideband Gaussian noise interference because it is not sparse in time or frequency domain, while the only effective method currently is using multiple antennas to apply the space array processing (SAP) technique since the wideband Gaussian noise interference is sparse in the spatial domain. However, when the incoming directions of the different interferences are not less than that of antennas, the interferences are not sparse to the array anymore, and the SAP anti-jamming performance would decrease drastically. In this paper, a LSTM (long short-term memory) deep neural network (DNN) based algorithm is proposed to enhance the array anti-jamming performance in this situation. The proposed network estimates the interferences as an integrity by exploring the non-linear relationship of the array data received by antennas. Especially, a new loss function is designed exclusively for GNSS anti-jamming problem. The proposed DNN method is verified in the simulation that two wideband Gaussian interferences with JSR (jamming to signal ratio) 50 dB can be eliminated by using two antennas’ data, and the interference cancellation ratio improvement is about 24 dB compared to some other widely used classical SAP algorithms.