Interview Pr. Sofie Pollin
It will be a new chip, a new concept where the radio and the analog domain will take care of some of the processing.
Pr. Pollin can you introduce yourself?
I am Sofie Pollin, professor at KU Leuven in the department of electrical engineering. I work on wireless communication with a focus on the physical layer. I did my PhD at IMEC, one of the largest research centers in electrical engineering from 2002 to 2006. The subject was already at that time called " Cross-Layer Resource Allocation for Quality of Service and Energy Optimization in Wireless Networks" but I was taking into account the impact of hardware and technology because I was working with some of the best analog and digital designers IMEC had at the time. We were working on low power sensor networks, and the first generations with multiple antennas, such as Wi-Fi IEEE 802.11n, and at that time, 4G.
After IMEC, I went to Berkeley for a postdoc where I continued my research on wireless communication systems. After Berkeley, I came back to IMEC and became a principal scientist. After four years, I decided to make the jump to KU Leuven and became a professor, while continuing to work for imec. I have always been interested in the most demanding technologies. I've always tried to be one of the first to explore new directions and see what that means for the physical layer.
Can you introduce us to HERMES?
The HERMES project is a very interesting FET-Open research project. With a FET-Open project, we try to do something disruptive with high risk. The ambition here is to build a very advanced THz link. We want to build all the components that matter for the link, from baseband to THz.
It will be a complete radio and this radio is not just a THz radio. It will be a new chip, a new concept where the radio and the analog domain will take care of some of the processing. This is done to simplify the conversion from analog to digital domain as it is the bottleneck of every system.
What does KUL do for HERMES?
KUL's task in HERMES is to take care of the baseband part. In order to send bits over this THz link, you need to map them to symbols that you send over the analog radio. In HERMES, symbols are different compared to the traditional analog or digital communication modulations we are familiar with, because there is analog processing. Thus, we send a symbol in the form of Walsh coefficients. What KUL does is create methods to optimally map bits to these new Walsh symbols. We don't do this using traditional mathematical tools, but using data-driven or deep learning models.
What should HERMES bring to European citizens?
If HERMES is a success, the end user will be able to have access to a very high speed communication system which can be used in new types of applications. These systems would be inexpensive because they are designed in CMOS technologies with hopefully a low power consumption. This would open up many future applications such as the metaverse or virtual reality. We will use online presence more and more and because of that, the data rate will continue to explode. Thus, we need more efficient communication systems and HERMES technology is a very important catalyst for this.
What do you think is the best indicator of HERMES' success?
I think the HERMES project will be a success if we manage to make all the components we need to build this THz link and make sure we meet the specifications. It would be fantastic.
HERMES will be a success in any case thanks to everything we have learned and thanks to the new generation of researchers we have trained working on this project.
What differentiates HERMES from other European 6G projects?
It is really very different from another European project. First of all, it's very subject-oriented with a strong thematic coherence. We are all working together on this very ambitious system with one goal: to make this system work. It's very focused. It's very revolutionary. We are building something that does not exist and even with some of the best researchers in Europe it is very difficult to do.
Everyone says AI is the answer to everything, in this project, why is AI the answer to communication?
We are using artificial intelligence here with and for this brand-new communication link that we are building. Everyone thinks AI is the answer to everything. It's like a low hanging fruit to use AI to map bits to our symbols. This is something we are exploring to see if we can get better or even comparable solutions to what we would have gotten using mathematical tools. It is always a big challenge to find an AI-powered baseband solution that will work in all possible scenarios and also in scenarios that have not been seen or measured before. With a data-driven approach, does it work with a scenario you don’t have measured data from? But in the world of wireless communication, it's impossible to measure every possible wireless scenario. It is impossible to train a model for all possible future operations incorporating all possible noise levels. It won't work in the end because you can't make a general enough model even if you take millions of measurements, trained for a few million scenarios and not billions of scenarios. This is the biggest challenge we have to solve.
Is the AI really smart in this RFIC design?
It would be smart to extrapolate the billion possible wireless scenarios from the million measured points. So far, it is difficult to build such models. So no, the AI is not yet intelligent. But sometimes, instead of deep learning, I use the term data-driven models. For me, this term is more accurate than deep learning because deep learning implies a form of intelligence. But we can debate it for a very long time. Ultimately, data-driven modeling is more about what we do than making things artificially intelligent.
Do you think that HERMÈS could provide a means for a real AI to communicate and then be a danger to humanity?
Why would a radio be a danger to humanity? In HERMES, we send bits. The human says which bits to send, the radio maps them into symbols. There may be two things that could go wrong.
First the radio might send bits but the receiver would receive other bits, without humans being aware of it. Thus, the radio could hide information and/or tell us unwanted things.
Or, we could have the good bits and the communication we want, but at the same time, with such high bandwidth, the radios communicate with their own bits in the artificial world of which man is unaware. Because we don't control how bits are transferred to symbols, there would be two languages. Radio could then create its parallel internet where artificial bits are constantly being sent to artificial radios that humans are unaware of. It's highly unlikely, but if you really want to debate how radio might endanger humans, this might be a place to start. It would already be a feat if we could create a radio as good as a mathematical model understood and proposed by a human, so I'm not afraid at the moment.
We are building something that does not exist and even with some of the best researchers in Europe it is very difficult to do