Adaptive Linearization for the Sub-Nyquist Photonic Receiver Based on Deep Learning

Due to the nonlinear and aliasing effects, sub-Nyquist photonic receiver for radio frequency (RF) signals with large instantaneous bandwidth suffers limited dynamic range noise performance. We designated a deep residual network (Resnet) realize adaptive linearization across 40 GHz bandwidth. In contrast conventional methods, learning method achieves suppression of multifactorial spurious distortions floor simultaneously. It does not require an accurate calculation transfer function or prior signal information. The experiments demonstrated that proposed Resnet could improve spur-free (SFDR) signal-to-noise ratio (SNR) significantly by testing single-tone signals, dual-tone wireless communication modulated radar signals.