Neural Network-Based Wiener-Hammerstein Model for the Cancellation of Self-Interference in FDD Systems

The limited isolation of the duplexer in frequency division duplex (FDD) transceivers leads to the leakage of transmitted signals from the transmitter (TX) into the receiver (RX), which consequently induces self-interference generated by components within the RX that falls into the receiving operational frequency band, ultimately resulting in receiver desensitization. For the cancellation of self-interference in the RX of the FDD system, a neural network-based Wiener-Hammerstein model is proposed in this paper. The proposed model is composed of two linear blocks and one nonlinear block, which employs finite impulse response filters to fit the linear parts of the model, with neural network serving as the nonlinear block. Compared to the most popular memory nonlinearity model, the proposed cascaded model achieves higher modeling accuracy and provides a superior solution for cancellation. In addition, the neural network-based nonlinear block possess outstanding nonlinear fitting capabilities while maintaining lower complexity.