A team of researchers at the Technology Innovation Institute (TII) in the UAE and the University of Athens in Greece has demonstrated how new AI algorithms could organize a collection of RISs (reconfigurable intelligent surfaces) to create a smart radio environment.

The key innovation is a new deep reinforcement learning (DRL) algorithm for dynamically tuning a fleet of RIS devices to improve radio coverage for multiple users.

“This could help unlock opportunities to use reinforcement learning to enable environmental artificial intelligence,” said Prof Mérouane Debbah, Chief Researcher, AI Cross-Center Unit and Digital Science Research Center at TII.  

The team started with a thorough analysis of prior approaches to scaling RIS implementations. Previous researchers have explored relatively simple cases. They realized that a more scalable approach would have to be dynamically returnable, support multiple antennas, and be efficient enough to run on energy-constrained devices.

RISs can provide some value by being programmed once to account for the reflection in buildings and other structures. But in the real world, the movement of people and other factors can also affect the radio environment. The hope is that more dynamic algorithms could adaptively retune each RIS to adjust for environmental changes.

Next, the research team wanted to develop algorithms to support multiple RISs that could be coordinated in parallel. Prior research has looked at how to reprogram one RIS to improve the signal between an individual base station and a single user. But large-scale deployment will need to include dozens of reflectors and perhaps thousands of users. So, the researchers looked at how DRL algorithms could be extended to support multiple users and multiple reflectors. “We were among the first to extend to multiple RIS when we submitted this proposal,” Dr. George C. Alexandropoulos. Principal Researcher at the Digital Science Research Center at the Technology Innovation Institute (TII) and Assistant Professor at the University of Athens, said.

Another challenge was figuring out how to structure the optimization problem into a form that could work with the online learning capabilities of the DRL algorithms. DRL algorithms need to break a problem into one or more agents, objectives, and environments. The team decided to focus on one approach in which a single agent, such as a base station, could coordinate the settings of the collectives.

They also developed a new algorithm that increased the efficiency of previous approaches. The improved approach could enable the algorithms to run on more energy-constrained processors and reduce energy requirements for the RIS. Previous researchers have considered deep-Q-network (DQN) algorithms for orchestrating smart wireless environments. “We can achieve similar results to sophisticated DQN algorithms, but with at least 50% less computational complexity,” Alexandropoulos said.

Preparing for 6G
While it is still early days for this novel approach that will optimize the airwaves, Prof. Mérouane Debbah and Dr. George Alexandropoulos hope they can prepare the foundation for these kinds of techniques to be included as part of upcoming work on 6G standards. Although 6G specifications are still years away, developing practical implementations to support the standardization process is essential.

Follow-up works will explore different approaches for modelling agents, objectives, and environments for such algorithms. This work could also play a role in improving other wireless architectures. Down the road, they are keen to explore different ways to improve the efficiency of such algorithms.

The team hopes to incorporate these new algorithms into experimental wireless hardware over the next few years. However, it may take another five to seven years before these new techniques are incorporated into next-generation wireless devices, a time plan which follows the 6G roadmap. – TradeArabia News Service