Development of Electrical Engineering heavily influences advancement of contemporary technologies.
One of the important problems of electrical engineering nowadays is increasing signal transmission
speed, that is directly related to increasing data transmission and processing speed in electronic devices.
In addition, it is important to keep or decrease sizes of microschemes, chips and other electronic
components. In the high-speed design signal integrity is violated, or signal distorted, by different
factors: connections between device chips and connectors, channel length loss, crosstalk between
parallel channels, etc. Therefore, important part of Signal Integrity is devoted to signal restoration, or
equalization algorithms.
Because of constant requirements to increase signal speed, decrease size of components and improve
transmission materials, signal equalization algorithms are constantly being developed and adjusted to
problems of certain devices.
The goal of the project is to develop fast and effective algorithms of choosing optimal parameters for
signal equalizations using machine learning in high-speed channels and devices for two-level, and well
as multi-level signals. Algorithms developed in the project and software modules created based on them
will speed up and improve signal equalization in high-speed electronic devices.
Project aims to investigate effectiveness of nonlinear regression methods based on neural network for
restoring signal in electronic devices. By means of classification algorithms the problems of signal
equalization and detection will be united considering electronic device as a black box. In the first place,
Convolution Neural Networks will be used for adaptive restoration of distorted signal because of nonlinear
effects. Then by means of deep neural network decoder the signal will be detected. It will make possible to
restore and decode signal at the same time, that is especially important for multi-level signals.
Equalization algorithms developed in the project for two-level signals will be generalized for multi-level
signals. Multi-level signals represent alternative innovative approach to increase signal speed and in future
it will be actively used to increase data speed in electronic devices. Nowadays four-level signal (PAM-4 –
Pulse Amplitude Modulation) is well researched and integrated into advanced chips.
In the given project the algorithms will be developed in the way that it will be possible to integrate
them directly into devices, for example chips. On the basis of these algorithms, software modules will
be developed, that might be easily integrated with the existing EDA packages or in house tools
developed by high-tech companies for simulation and testing of their signal integrity problems.