Graph-based machine learning for efficient signal processing algorithms (GLEAM)

Today's digital society uses global connectivity to benefit from technological innovations. This is made possible by communication networks that use digital signal processing (DSP) as the fundamental methodology for efficient and reliable data transmission.

Classical DSP requires extensive knowledge of physical processes and is designed for simple linear dynamics. Modern DSP methods use machine learning and neural networks, which require little modeling knowledge but are computationally intensive and data inefficient. Many communication systems are non-linear and require low-complexity solutions.

Dr. Boris Karanov's research project aims to optimize DSP for nonlinear systems through domain knowledge and machine learning. The project is interdisciplinary and includes algorithm development, experimental demonstration and hardware implementation with applications in fiber optic and light communication systems. The new algorithms will be general and applicable to various problems, such as speech and audio processing, so that they can be adopted by both academia and industry.