Events

Optimizing communication systems - "if we do not have the entire model"

  • 30

    Aug

    2022


Name of the Speaker: Prof Henk Wymeersch
Name of the Organizer: Dr. Sheetal Kalyani
Venue: ESB 244 (Seminar Hall)
Date/Time: August 30th 2022 at 11 AM

Abstract
Deep learning techniques have been applied to optimize a specific function such as coding, modulation, or equalization in wireless and optical communication systems. Such a modular implementation allows the individual system components to be optimized and analyzed separately and thus presents a convenient way of building the communication link.

Nevertheless, this approach can be sub-optimal, where the optimum receivers or optimum blocks are not known or not available. Therefore, an end-to-end communication system that can optimize the whole system jointly using a channel autoencoder (AE) has been used in several recent works of literature. In this way, the whole system can be optimized jointly for an arbitrary differentiable end-to-end performance metric (such as the bit error rate) without the conventional block structure to achieve global optimization.

However, the unavailability of a channel model in wireless systems or the unavailability of the light source's dynamic response in optical systems will hinder updating the transmitter parameters in AE.

Therefore, in this talk: a) We will discuss the approximate methods available to optimize the AE without the channel transfer functions in a generic communication system (both wireless and optical).
b) As an example, we will discuss how a neural network (NN) can be used to model the optical waveforms of the light source under different operating temperatures. We will also discuss using this NN in an AE to model an end-to-end fiber-optic system and determine the optimized input modulation levels with a superior bit error rate performance.

Speaker bio:
Muralikrishnan Srinivasan is a postdoctoral researcher with Prof Henk Wymeersch at the Dept. of Electrical Engineering, Chalmers University of Technology, Sweden. At Chalmers, he works on the "HOT-OPTICS" project led by Prof Anders Larsson, Prof Lars Svensson, Prof Peter Andrekson, and Prof Henk Wymeersch. The project aims to develop optical interconnects used in data centers and vehicles with very high throughput and efficiency.

In 2021, he worked as a postdoctoral researcher at ETIS, ENSEA, Cergy France, on Physical Layer Security (PLS) with Prof. Arsenia Chorti. He is also currently a part of the IEEE INGR pre-standardization working group for PLS. In 2020, he received his doctoral degree from the Indian Institute of Technology Madras after conducting his doctoral research under Prof Sheetal Kalyani on approximate models for generalized fading channels.

His research interests include machine learning for wireless communication systems, optical interconnects, physical layer security, massive multiple-input-multiple-output systems, aerial base stations, air-corridors, hypergeometric functions, extreme-value theory, and privacy.