Deep learning enabled photonic Nyquist folding receiver for wideband RF spectral analysis

Abstract

The need for real-time wideband radio frequency (RF) spectral analysis is driven by continued advances in modern wireless communications and RADAR systems used both for military and civilian applications. However, wideband RF sensing presents a challenge for typical high-speed analog to digital converters (ADC) since ADCs capable of operating continuously are typically limited to monitoring less than 1 GHz bands. Here, we leverage the high bandwidth of photonics to build a Nyquist folding receiver (NYFR) that uses an asymmetric optical frequency comb and a deep convolutional neural network to monitor a ∼5 GHz bandwidth using a 1 GS/s ADC with a 1 MHz update rate. We tested the deep-learning assisted NYFR on several signal classes, including linear chirps, nonlinear chirps, and continuous wave signals. The system presented here tackles many of the limitations of typical NYFR systems, including the ability to recover signals that cross Nyquist zones and the ability to detect multiple signals simultaneously. We also show that using a non-linear encoding to map the RF signal into the optical domain can improve the accuracy of the recovered RF spectrum.