Compressed Nonlinear Equalizers for 112-Gbps Optical Interconnects: Efficiency and Stability

Abstract

Low-complexity nonlinear equalization is critical for reliable high-speed short-reach optical interconnects. In this paper, we compare the complexity, efficiency and stability performance of pruned Volterra series-based equalization (VE) and neural network-based equalization (NNE) for 112 Gbps vertical cavity surface emitting laser (VCSEL) enabled optical interconnects. The design space of nonlinear equalizers and their pruning algorithms are carefully investigated to reveal fundamental reasons of powerful nonlinear compensation capability and restriction factors of efficiency and stability. The experimental results show that NNE has more than one order of magnitude bit error rate (BER) advantage over VE at the same computation complexity and pruned NNE has around 50% lower computation complexity compared to VE at the same BER level. Moreover, VE shows serious performance instability due to its intricate structure when communication channel conditions become tough. Moreover, pruned VE presents more consistent equalization performance within varying bias values than NNE.

Electronic Dispersion Compensation and Functional Link Neural Network Equalization for IM/DD Links

“…In contrast, the activation function in the output layer is application specific. In the context of IM/DD systems, the authors in [18], use a multilayer perceptron (MLP) for Rx nonlinear equalization, with multiple hidden layers and an output layer. The authors considered the rectified linear unit ReLU(x) , hyperbolic tangent tanh(x) and sigmoid σ(x) activation functions for the hidden layers, while a softmax(x) function is used to convert the results of output layer to probability distribution for each level of the PAM signal [18].…”

“…In the context of IM/DD systems, the authors in [18], use a multilayer perceptron (MLP) for Rx nonlinear equalization, with multiple hidden layers and an output layer. The authors considered the rectified linear unit ReLU(x) , hyperbolic tangent tanh(x) and sigmoid σ(x) activation functions for the hidden layers, while a softmax(x) function is used to convert the results of output layer to probability distribution for each level of the PAM signal [18]. In the absence of hidden layers, and exhibiting a single neuron structure, as with the FLNN proposed in this article, the activation function can only be confined to the FLNN output.…”

“…To that end, two distinct approaches for nonlinear equalization at the receiver are introduced. The first is one of the widely used approaches for compensating nonlinear effects in optical communication systems, namely the Volterra nonlinear equalizer (VNLE) [15,16,17,18,3,19], which acts as a reference performance point, albeit at an increasing complexity with an increasing memory length. The second approaches utilizes a functional link neural network (FLNN), which is uniquely positioned to meet the aforementioned objectives as it allows for both linear equalization and the nonlinear expansion of the input samples using a functional expansion block (FEB) with orthogonal basis functions [20].…”

Light Field Optimization for Optical Wireless Power Transfer

“…The fast development of OWPT technology is attributed not only to the innovations of systemlevel design, but also to various key devices, such as semiconductor lasers and photovoltaic (PV) cells [12], [13]. In the recent years, high efficient semiconductor lasers have been widely used in the fields of high-speed communications and intelligent manufacturing [13], [14]. The electrooptical conversion efficiency have reached 68% for the continuous-wave 100 W-class 808 nm laser diode arrays [13].…”

Special Issue on Enabling Technology in Optical Fiber Communications: From Device, System to Networking

“…The proposed free-space optics communication system used a dual electro-optical modulator design, which could improve the transmission rate. For direct-detection optical communication systems, Zhang, W. et al compared the complexity, efficiency, and stability performance of pruned Volterra series-based equalization (VE) and neural network-based equalization (NNE) for 112 Gbps vertical-cavity surface-emitting laser (VCSEL)-enabled optical interconnects [ 8 ]. The experimental results showed that NNE has more than one order of magnitude bit error rate (BER) advantage over VE at the same computation complexity, and pruned NNE has around 50% lower computation complexity compared to VE at the same BER level.…”