Photonic machine learning implementation for signal recovery in optical communications

Apostolos Argyris, Julián Bueno, Ingo Fischer
2018 Scientific Reports  
Machine learning techniques have proven very efficient in assorted classification tasks. Nevertheless, processing time-dependent high-speed signals can turn into an extremely challenging task, especially when these signals have been nonlinearly distorted. Recently, analogue hardware concepts using nonlinear transient responses have been gaining significant interest for fast information processing. Here, we introduce a simplified photonic reservoir computing scheme for data classification of
more » ... rely distorted optical communication signals after extended fibre transmission. To this end, we convert the direct bit detection process into a pattern recognition problem. Using an experimental implementation of our photonic reservoir computer, we demonstrate an improvement in bit-error-rate by two orders of magnitude, compared to directly classifying the transmitted signal. This improvement corresponds to an extension of the communication range by over 75%. While we do not yet reach full real-time postprocessing at telecom rates, we discuss how future designs might close the gap. Recent developments in neuro-inspired information processing using recurrent neural networks (RNNs), cognitive computing approaches, machine learning techniques and deep learning 1,2 architectures have had a major impact on solving classification and pattern recognition tasks with remarkable efficiency 3-7 . However, there are hardly any solutions available if the task is time-dependent, the speed requirements are very demanding and the signals to be processed are of high complexity. To this end, analogue hardware implementations of these information processing tools have been gaining increasing recognition 8 . In recent years, implementations of feed-forward and recurrent NNs based on extreme learning machines (ELM) 9,10 and reservoir computing (RC) approaches [11] [12] [13] [14] have been presented in optoelectronic 15-18 and photonic 19-25 hardware. These implementations were in some cases assisted by field programmable gate array (FPGA) modules 25,26 . So far, they have only been employed for standard benchmark tasks such as pattern classification, speech recognition, nonlinear time series prediction and wireless channel equalization. Evolving these hardware implementations to minimal conceptual complexity and to maximal speeds would enable to address signal processing tasks in critical technological fields. An excellent example with ultra-fast post-processing requirements can be found in the contemporary fibre-optic communication networks that now operate even beyond the Tb/s scale 27 . The technological advances in this field target on the highest data throughput over the longest distances with energy efficient and low complexity designs. However, transmission impairments 28 , such as chromatic dispersion, Kerr effect and four-wave mixing, put strict limitations on communication speed and distance in fibre-optic communication systems. Current research aims at extending these limits, by focusing mainly on the two ends of the communication links. At the transmitter side, major efforts target on optimizing the emitter 29,30 , as well as the encoding communication scheme, by using multi-level formats and signal shaping 28,31-34 . At the receiver side, high-speed digital signal processing (DSP) algorithms 35-39 with low-complexity have improved signal recovery by mitigating linear and nonlinear signal distortions. The aforementioned approaches in fibre-based communication systems currently shape the status quo of the field, but they are also facing challenges for future trends. For example, the current DSP methods are efficient as long as nonlinear signal distortions do not become too complicated. For this reason, optimal designs of various transmission systems dictate that the launched optical power in standard single mode fibres (SSMF) should be always restricted to moderate levels (around or below 1 mW). Inevitably though, these power levels limit the signal-to-noise ratio (SNR) of the received signal, given the standard detection capabilities of fast photoreceivers. In a reasonable consideration, one could suggest to increase the launched optical power into the fibre. There are numerous semiconductor laser emitters available ready to offer tens of mW of emission at telecom wavelengths.
doi:10.1038/s41598-018-26927-y pmid:29855549 pmcid:PMC5981473 fatcat:ivhquahxznhczk2m3z2irybtau