Publications

Abstract: We experimentally demonstrate an arbitrarily biased electro-optic Mach-Zehnder modulator-based pulse picker with variable repetition rate (from 50 MHz to 500kHz) by using a fast synchronized FPGA. We report a maximum pulse extinction ratio of 34 dB.

Abstract: We experimentally demonstrate a novel deep learning-aided time-stretch photonic front end architecture to overcome device-dependent distortions to improve the signal-to-noise and distortion ratio by more than 24 dB, and reduce the bandwidth requirements of the back-end electronic ADC by three times.

Abstract: In this work, we demonstrate a dither-free automatic bias controller for dense wavelength division multiplexing-based multichannel analog radio over fiber systems. The analog optical transport is implemented as an optical doubling scheme using a two-stage external modulation setup, emulating a mmWave 5G fronthaul scenario.

Abstract: We present an automatic polarization aligner using the SPGD algorithm to achieve polarization-independent gain in an optical parametric amplifier (OPA). The polarization state of the input pump is adjusted automatically and a gain error of < 0.07 dB between orthogonal signals is achieved within 25 seconds.

Abstract: We demonstrate bidirectional transmission over a four-core fiber using a recirculating loop to emulate a 480 km transmission distance. Performance degradation due to intercore crosstalk is experimentally analyzed and validated through simulations, confirming accurate long-distance emulation using a compact multi-core fiber span.

Abstract: We present a photonic instantaneous frequency measurement system where frequency-shifted optical tones beat with modulated unknown RF. Using a single laser and frequency-shifting loop, we identified RF frequencies within tens of microseconds.

Abstract: Continuous time stretched photonic analog-to-digital converters (CTS-PADC) have emerged as a promising solution to meet the requirements of large analog bandwidths with high sampling rates, without compromising the vertical resolution of traditional electronic Analog to Digital Converters (eADCs). However, the photonic front end introduces nonlinear distortions which are dependent on input RF signal power to be sampled and quantized. It is proposed to equalize these distortions using deep neural networks (DNNs). This paper proposes three neural architectures for equalization of fiber distortions in PADC: (i) normalized-input feed-forward neural network (NFFNN), (ii) parallel waveform and power predictor (PWPP), and (iii) hybrid feed-forward neural network (HyFFNN). Among these, the HyFFNN demonstrates the ability to significantly correct both frequency and power errors in the time-stretched RF signal, achieving a signal-to-noise-and-distortion ratio (SNDR) of approximately 67.6 dB and an effective number of bits (ENOB) of around 10.93 bits in a 1 GSa/s eADC, for input RF frequencies of up to 1.4 GHz, thus reducing the bandwidth requirements of the eADC by a factor of three. Furthermore, the trained model effectively equalizes distortions due to the photonic frontend, regardless of whether the system operates in the linear or nonlinear regime. The proposed system achieves an ENOB of 10.93 up to 1.4 GHz, surpassing the performance of state-of-the-art commercial ADCs.