Sarah Masaad (IMEC)

Neuromorphic photonic systems can be used to perform signal equalization in telecommunication systems. The results shown here are experimentally obtained from a distorted 16 QAM signal after passing through 20 km of fiber. The signal is coupled onto our photonic integrated chip for processing, which equalizes the signal distortions and retrieves the undistorted form of the signal. This allows replacing the equalization steps, standardly performed on digital computers, by post-digital solutions that can perform the same equalization but at faster rates and with lower energy consumption.

 

The goal of Sarah’s project was to investigate the potential of utilizing photonic reservoir computing in optical communication systems. This stems from the heavy reliance of today’s communication systems on digital signal processing (DSP), which is a pipeline of steps implemented to remove the distortions on transmitted data induced from different components in the system. The processing pipeline forms a speed bottleneck since it is digital and thus operates at a lower speed than that of the optical transmission system. It also requires a significant amount of power to operate which makes it unfeasible for usage in all network types, especially the short-haul ones since they are normally deployed in large numbers.

To this end, Sarah studied the use of an integrated photonic chip that can replace some of the blocks in the digital signal processing pipeline. The use of photonics, as opposed to electronics, allows us to process data at the same speed as that of its transmission which alleviates the speed bottleneck introduced by DSP. Moreover, photonics offers the potential of parallelizing processes through the use of multiple wavelengths that can simultaneously pass through the reservoir. Additionally, photonics can be more power conservative as opposed to electronics. On the other hand, digital processing is more mature than photonic processing and as such offers the possibility of performing a wide range of processing steps. Therefore, our strategy was to migrate some of the processing steps to photonics while keeping other blocks in the digital domain because it is not yet efficient to implement them optically.

Sarah’s work focused on short and medium haul systems deploying single mode and single wavelength optical fibers. The results were obtained by simulating a transmission system deploying coherent transmitters and self-coherent receivers. Different self-coherent receiver schemes were simulated, including the Kramers Kronig (KK) receiver, a linear approximation of the KK receiver, and a photonic linearization receiver. Different system impairments were equalized using a passive reservoir composed of multi-mode interferometers interconnected with waveguides. This architecture is entirely linear and relies on nonlinearities of the receiver or other components to enable targeting nonlinear system issues. Impairments like dispersion, Kerr effects, and receiver nonlinearity were addressed separately first and later a subset of them were addressed in an aggregated solution. The effectiveness of the photonic reservoir in addressing these issues was benchmarked against similar linear equalizers both optically and digitally. Moreover, different machine learning optimization techniques were employed to aid convergence to suitable solutions and mitigate known machine learning problems like overfitting and long training times.

In addition to simulations, Sarah performed experimental work both at our labs and at Aston University’s lab during my secondment there. She learned handling and measurements with photonic integrated chips during experimental work at UGent. This involved the characterization of two different generations of our reservoir chips as well as the characterization of two heater-based readout subsystems. Aspects including insertion loss, polarization sensitivity, coupling losses, and frequency response were characterized. After characterization, Sarah performed high-speed (28 Gbauds) experiments on both intensity modulated signals at UGent as well as coherently modulated signals at Aston University. The experiments utilized a reservoir chip to perform equalizations for fiber-originating distortions. This involved building a transmission system and coupling light into the photonic chip to perform the equalization and then detecting on a receiver. Due to several experimental considerations at the time of these measurements, the readout subsystem was bypassed and instead reservoir states were recorded to be processed offline. Measurements during my secondment at Aston were pivotal in developing my understanding on how to operate coherent transmitters and receivers as well as perform the necessary DSP.