Linearization of microwave power amplifier using multi-port receiver with machine learning techniques in X-band

Modern telecommunication systems require high-efficiency modulations with high PAPR. To have high efficiency while being linear, a linearization technique must be implemented. One of the efficient methods for linearization is the digital pre-distortion method. In this paper, a digital pre-distortion method using different types of neural networks is used to linearize PA. A six-port receiver in the digital pre-distortion loop is used to demodulate the output of the PA to the baseband and linearize the PA in the baseband. Using this receiver has reduced the cost, noise, and complexity of the demodulator used in the pre-distortion circuit. Adjacent channel power ratio (ACPR) has been used as a performance metric. According to the results, the BiLSTM network used in this paper is associated with a severe reduction in complexity and a significant improvement in the ACPR parameter compared to the other types of BiLSTM network previously used for linearization. It is observed that for the three input signals 16QAM, 64QAM, and OFDM with 600 MHz bandwidth, the maximum improvement in ACPR parameter using BiLSTM network is 25.2dB, 23.1dB, and 22.5dB in X-Band, respectively.