In this Letter, we propose a real-time machine learning scheme of a tracking optical intensity-modulation and direct-detection (IMDD) system's conditional distribution using linear optical sampling and inline Gaussian mixer modeling (GMM) programming. End-to-end conditional distribution tracking enables an adaptive decoding of optical IMDD signals, with robustness to the bias point shift of the optical intensity modulator. Experimental demonstration is conducted over a 20-Gbits/s optical pulse amplitude modulation-4 (PAM-4) modulation system. Optical PAM- 4 signals are optically down-sampled by short pulses to 250 Msa/s. Then, statistical characters of signal distribution can be estimated using inline GMM processing. Due to the real-time learned distribution, intelligent decoding of received signals exhibits a perfect adaptation to the changing bias point of aMach-Zendner intensity modulator, enhancing the communication reliability with bit error rate (BER) below 3.8 . 10-3. In addition, the proposed scheme also provides the possibility of practical implementation to other machine learning signal decoding methods.
ASJC Scopus subject areas
- Atomic and Molecular Physics, and Optics