LSTM-Resnet-Assisted Linearization for the Dual Frequency Comb Photonic RF Channelized Receiver
Xin Zhang, Liyuan Zhao, Jianghua Zhang, Jianfeng Zhang, Zhuohang Zhang, Meili Shen, Shuo Chen, Xin Zheng, Tian Jiang
Abstract
Analog photonic channelized receivers based on dual optical frequency combs (DOFC) are widely used in microwave photonic wideband signal processing. It uses photonic means to perform narrowband segmentation of the wideband signal spectrum, thus avoiding the tradeoff between speed, bandwidth, and accuracy of ADC. However, many factors such as link noise and device nonlinearity restrict the dynamic range of the system, thus affecting the signal detection and processing performance. In this paper, a Long-Short-Term-Memory-Residual network (LSTM-Resnet) is designated to realize the linearization of the DOFC parallel channelized receiver. Benefiting from efficient feature extraction and adaptive processing capabilities, the third-order spur-free dynamic range (SFDR2/3) of the system is improved from 84.12 dBm·to 107.19 dBm·.The spurious suppression ratio (SSR) of the single-tone signal and linear frequency modulation (LFM) signal have been improved by 37.54 dB and 33.39 dB. In addition, the mutual interferences between muti-channels caused by distortion are also significantly suppressed. Experimental results on signals of different format types validate the generalization of LSTM-Resnet and demonstrate its advantages in improving the linearity of DOFC channelized receivers.