AI-enhanced DSP for wireless communications

Wireless systems are designed combining analog front-ends and digital baseband or physical layer (PHY) processing. Traditionally, PHY DSP blocks are created by expert knowledge from communications theory based on how the system behaves and how signals should be processed. This can include filters, FFTs, modulation, coding, multiple-antenna operations, hardware non-ideality compensation and a few other blocks.

More recently, artificial intelligence has been appearing across many domains. In some fields it can lead to solutions un-attainable by expert models, while in other fields it simply cannot improve the performance of traditional solutions. Among many other domains, AI or ML-based solutions are also being investigated for PHY processing.

In this PhD, you will take a critical look at those developments. By extensively reviewing the recent state-of-the-art on ML-based PHY solutions, you will identify the most promising blocks where AI solutions have the potential to out-perform traditional approaches - either in performance or in complexity - but also clarify which components cannot be improved, based on understanding performance and complexity bounds for the different approaches.

Typically, AI-based solutions are more relevant for non-linear problems, hard-to-model behaviors, or when known solutions have excessive complexity due to the problem size. You will refine those criteria, identifying relevant domains for AI-enhanced PHY. They could come from non-ideal hardware effects, interference between multiple systems, complex mobile multi-path environments, or other sub-problems.

In a second phase, you will select a few DSP blocks where AI-based approaches are most promising from this analysis and propose new designs able to out-perform traditional solutions. You will develop and test AI-based solutions for those components, assess their benefits based on extensive and realistic end-to-end simulations, and optimize them for the best performance/complexity trade-offs. By doing so, you will enable hybrid PHY implementations combining traditional and AI-based blocks for the best overall performance.

You will be part of a large imec community working on the research, implementation and prototyping of future communications systems with experts in wireless communication, signal processing, digital, analog and mm-wave design, and machine learning. This is a unique opportunity to develop innovative, multi-disciplinary technology and shape future wireless networks. You will publish your research in top-level journals and conferences.