XAI algorithms applied to long range 5G networks

• Grant number: 22/14431-2
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: February 01, 2023
• End date: January 31, 2026
• Field of knowledge: Engineering - Electrical Engineering - Telecommunications
• Agreement: MCTI/MC
• Principal Investigator: Luciano Leonel Mendes
• Grantee: Pedro Henrique Carneiro de Souza
• Host Institution: Instituto Nacional de Telecomunicações (Inatel). Santa Rita do Sapucaí , SP, Brazil
• Associated research grant: 20/05127-2 - SAMURAI: smart 5G core and multiran integration, AP.TEM

Abstract

In this research project, artificial intelligence/machine learning (AI/ML) algorithms proposals will be analyzed for the physical and link layers, in the context of the explainable artificial intelligence (XAI) frameworks. Novel algorithms and architectures for AI/ML are expected as a result of the research, considering the specialized requirements for the access networks of the fifth generation of mobile network (5G) with long range links. This will be accomplished taking into account the state-of-the-art of XAI algorithms and also the ray tracing (RTR) method for realistic simulation of communication channels. XAI algorithms will be employed for simplifying systems at the physical layer of receivers for long range mobile networks and on the configuration and operation of intelligent reflective surfaces (IRS), for example, aiming to increase the signal coverage of the long range links. Indeed, AI/ML frameworks are often times inaccessible, given the black box nature of the algorithms and mechanisms involved. This can become a critical problem, especially concerning communications systems, where analytic models are largely adopted. Therefore, with the proposed XAI algorithms resulted from the research project, then potentially the requirements of the 5G network could be precisely fulfilled, even when the AI training conditions are not ideal. In this way, the unpredictability of AI/ML algorithms could be mitigated, also for scenarios where some statistics and patterns may not be present at training. In other words, one of the major setbacks for large scale adoption of AI in communication system would be overcome. (AU)