Eye Diagram Modeling Of High-Speed Channels Using Artificial Neural Networks With An Improved Adaptive Sampling Algorithm

Goay, Chan Hong (2019) Eye Diagram Modeling Of High-Speed Channels Using Artificial Neural Networks With An Improved Adaptive Sampling Algorithm. Masters thesis, Universiti Sains Malaysia.

Download (1MB) | Preview


xxiii EYE DIAGRAM MODELING OF HIGH-SPEED CHANNELS USING ARTIFICIAL NEURAL NETWORKS WITH AN IMPROVED ADAPTIVE SAMPLING ALGORITHM ABSTRACT As data rates increase to the gigabit range and beyond, signal integrity (SI) analysis becomes increasingly difficult and time consuming process. Thus, many researchers have started to look out for artificial neural networks (ANNs) as an alternative to traditional SI modeling tool because ANNs are easy to use and fast. However, large amount of samples need to be generated for the training process of the ANN for the modeling of a complex design, resulting in a high neural model development cost. The adaptive sampling technique is used for the data generation due to its flexibility where it generates samples according to the non-linearity of the regions in the design space. This work proposes an improvement to the original adaptive sampling algorithm and uses it as the sampling method for eye diagram modeling. This reduces the number of training samples by 16.1%, validation samples by 14.7% and neural model development time by 23%. Besides that, the use of the prior knowledge input neural network (PKI-ANN) and the prior knowledge input difference neural network (PKID-ANN) for the modeling of high dimensional SI problem is proposed. The normalized worst-case error for the PKI-ANN is only 6.66% and for the PKID-ANN is only 6.32% as compared to that of the conventional ANN which is 11.44%. Finally, the neural network technique for the modeling of entire bit rate error (BER) contours is proposed which provides engineers with more information, such as the full shape of eye instead rather than just the height and width of the eye. An Average testing performance of R2=0.983 is achieved for the BER contour neural modeling technique.

Item Type: Thesis (Masters)
Subjects: T Technology
T Technology > TK Electrical Engineering. Electronics. Nuclear Engineering
Divisions: Kampus Kejuruteraan (Engineering Campus) > Pusat Pengajian Kejuruteraaan Elektrik & Elektronik (School of Electrical & Electronic Engineering) > Thesis
Depositing User: Mr Mohamed Yunus Mat Yusof
Date Deposited: 08 Feb 2022 07:29
Last Modified: 08 Feb 2022 07:29
URI: http://eprints.usm.my/id/eprint/51457

Actions (login required)

View Item View Item