Figure 3
From: Photonic machine learning implementation for signal recovery in optical communications

Signal recovery by training an experimentally implemented photonic reservoir. (a) Schematic that associates the number of bits considered for training with the corresponding bit timeframes of the reservoir response that is taken into account. In order to take into account the consecutive bits in the computation, a time latency of the processing applies, equal to the total duration of the consecutive bits. In this case, the latency is 4-bit timeframes (4τ). (b) BER performance of the recovered test bit stream versus the number of bits considered for training, when handling degraded signals from the short-reach transmission system (Supplementary Fig. 1) and considering different processing methodologies: training on the transmission output as received from the communication channel (black rectangles), training on the output response of the photonic reservoir with minimized optical feedback (ELM approach - blue triangles) and training on the output response of the photonic reservoir with optimized optical feedback conditions (RC approach – red dots). (c) The same as (b), but for the long-haul transmission system (Supplementary Fig. 2). BER statistics emerge from 5 independent repetitions of the experiment that use different bit streams as input.