Residual Neural Networks for Digital Predistortion

Abstract

Tracking the nonlinear behavior of an RF power amplifier (PA) is challenging. To tackle this problem, we build a connection between residual learning and the PA nonlinearity, and propose a novel residual neural network structure, referred to as the residual real-valued time-delay neural network (R2TDNN). Instead of learning the whole behavior of the PA, the R2TDNN focuses on learning its nonlinear behavior by adding identity shortcut connections between the input and output layer. In particular, we apply the R2TDNN to digital predistortion and measure experimental results on a real PA. Compared with neural networks recently proposed by Liu et al. and Wang et al., the R2TDNN achieves the best linearization performance in terms of normalized mean square error and adjacent channel power ratio with less or similar computational complexity. Furthermore, the R2TDNN exhibits significantly faster training speed and lower training error.

Single-channel 1.61 Tb/s Optical Coherent Transmission Enabled by Neural Network-Based Digital Pre-Distortion

“…To linearize RF amplifier, different structures based on feed-forward NN (FFNN) [20] , time-delay NNs (TDNNs) [18], [23], [26] , convolutional NNs (CNNs) [24] were proposed. It was recently shown for an RF amplifier that adding a residual NN (ResNet) structure in TDNNs improves the performance and training rate [19] .…”

Over-the-fiber Digital Predistortion Using Reinforcement Learning

“…1. Similar to [14], [27], [28] , in order to take the memory effect into account, the input layer is fed with 2L + 1 real-valued signals, where one signal corresponds to the current instantaneous input and the remaining 2L signals to the inputs of the previous and future L time steps, respectively. Moreover, a shortcut connection is employed to directly add the current input to the output, which has been shown to allow for better performance and quicker convergence rate [16]- [18], [28] .…”

Deep Neural Network-Based Digital Pre-Distortion for High Baudrate Optical Coherent Transmission

“…The memory effects were included in the DPDs based on time-delay NNs (TDNNs) [20], [25], [28] and on convolutional NNs (CNNs) [26]. Recently, some of the above schemes have been compared in [21] and shown experimentally that adding residual neural network (ResNet) structure improves the nonlinearity mitigation of RF amplifiers. However, NNbased DPDs for optical coherent transmitters are so far not well explored.…”