Air-Induced Passive Intermodulation Modeling Method Based on CNN-LSTM-FIR Neural Network

In this article, we analyzed the principle of the air- induced passive intermodulation (PIM) signals generated by passive devices in communication systems and modeled the air-induced PIM. Air-induced PIM is the interference caused by nonlinear transformations of the transmitted signals and can severely degrade the quality of the received signals. Therefore, in this article, we designed a neural network model, CNN-LSTM-FIR, to model and compensate for the air-induced PIM signals generated by the system. Firstly, we follow the actual physics process generated by air-induced PIM, and add the leading and lagging terms of the signal as inputs. The memory of the signal is extracted by using the CNN layer, and the nonlinearity of the signal is extracted by the nonlinear activation function. In addition, the fully connected layer is used to distribute the weight of the signal, and then the deeper memory of the signal is extracted through the LSTM layer, and finally the signal is effectively sorted out and output through the FIR filter layer. Experiments show that the above model has excellent modeling and cancellation capabilities, and achieves a cancellation performance of more than 18dB.