Implementation of an adaptive equalizer based on machine learning for real-time coherent optical receivers
Examensarbete för masterexamen
Please use this identifier to cite or link to this item:
Bibliographical item details
|Type: ||Examensarbete för masterexamen|
|Title: ||Implementation of an adaptive equalizer based on machine learning for real-time coherent optical receivers|
|Authors: ||Zheng, Boyang|
|Abstract: ||An adaptive equalizer is used to compensate for polarization mode dispersion (PMD) in polarization multiplexed (PM) optical communication systems. The PMD is not only influenced by the optical fiber but also affected by the fiber non-linear effects. Furthermore, the non-linear effects will be more and more severe with the increase of the channel’s data throughput. However, traditional algorithms, such as the
constant modulus algorithm (CMA) and decision-directed Least Mean Square (DD LMS), only focus on compensating the PMD, and are not efficient for the non-linear effects. We try to apply machine learning-neural networks (ML-NNs) algorithm in the adaptive equalizer to deal with non-linear effects and more complex PMD.
We also compare the ML-NNs and the CMA algorithm both in the software and hardware performance.
In this project, we improve the traditional ML-NNs structure based on the cascade multiple-input multiple-output (MIMO) filter theory so that the ML-NNs adaptive equalizer could be implemented in a real-time system more efficiently. MATLAB simulations are used to compare the software performance between the two
algorithms via the effective signal-to-noise ratio (SNR). The two algorithms are re-written to the very-high-speed integrated circuit hardware description language (VHDL) format to compare the field-programmable gate array (FPGA) resource utilization. In addition, we extract information on how the number of taps in the
ML-NNs equalizer influences the FPGA resource utilization. We verified that the two types of equalizers have the same MATLAB software per formance in terms of effective SNR. We also conclude that the FPGA is a good choice for implementing the forward propagation of the ML-NNs equalizer. However, it is not suitable for implementing backpropagation because of the high degree of multipliers required and the complexity of calculating derivatives.|
|Keywords: ||Adaptive equalizer;CMA;Machine-learning;Matlab;FPGA;VHDL;Coherent optical receivers|
|Issue Date: ||2021|
|Publisher: ||Chalmers tekniska högskola / Institutionen för data och informationsteknik|
|Collection:||Examensarbeten för masterexamen // Master Theses|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.