6G Mitola Radio: Cognitive Brain That Has Collective Intelligence
1 September 2020
Combining game theory and learning-based methods together to achieve the intelligence needed for 6G radios.
Amount £ 470 567
Project Website gow.epsrc.ukri.org
Research topics 6G | AI | Cognitive Radio
While 5G is being launched worldwide, discussion for 6G is already taking shape. One unanimous view is that 6G mobile radios should be empowered by great intelligence, the kind of intelligence that allows each radio to make wise decisions that optimise its quality-of-experience over time and impact the network in a constructive way. In addition, 6G mobile radios will be more than just communication devices, providing also computation, security, energy services and etc. when appropriate. 'Intelligent' radio is not a new concept. In fact, back in 1998, Mitola formalised this concept and coined it cognitive radio, (also known as Mitola radio by many). This concept refers to a futuristic mobile communication device that goes beyond the possession of any hardware flexibility and is gifted the intelligence to access the spectrum anytime anywhere according to the environment and its need. The notion is general in that the term 'need' can include beyond-communication capability, such as computing, security, etc. in today's scenarios.
After 20 years of effort, however, progress has been limited. For dynamic spectrum sharing, 5G has shared spectrum technologies such as LAA and LWA, but the intelligence remains at a very basic listen-before-talk (LBT) level. The deadlock for a genuine Mitola radio appears to be the need to make decisions based on very limited local information (local observations and actions) that should not only benefit itself but the entire network as a whole (global influence), without the overhead of one form of cooperation or another. In other words, the key is collective intelligence (as opposed to individual intelligence), one that enables each radio to evaluate and optimise its action and policy collectively with other coexisting radios without talking to them directly.
There will be several step changes if the ideal Mitola radio is successfully realised in 6G. First, spectrum utilisation will always be at the maximum with abundant spectrum resources available, and resource allocation is literally done in a self-organising fashion without any overhead for coordination. Latency for managing the resources will be significantly reduced as a result. Hidden terminal problem will also be eliminated because Mitola radio should possess the intelligence to identify them through interacting with the radio environment and optimise its action to avoid them. Furthermore, it will also be possible for Mitola radios to share not only the spectrum efficiently but also assist the network as service providers using their energy and computing resources.
Without coordination or cooperation, collective intelligence demands each radio establishing global intelligence of the network by itself. To achieve this, artificial intelligence (AI) may come as a convenient idea but the fact that the best action of one radio (i.e. a learning agent) is dependent on the action of another radio (another learning agent) troubles the state-of-the-art AI algorithms, making them highly ineffective. Different from the entire literature, this project's novelty is to develop an intelligence gathering mechanism that takes the game-theoretic perspective to enrich deep reinforcement learning. Such integration will equip Mitola radio the brain power of collective intelligence (from local action to global influence), and result in a holistic approach to optimise the parameters and essential functionalities for Mitola radio enabled multi-function wireless communications and service networks.
The project team includes BT and Toshiba, both of which have been active in the development of 5G and are keen to lead the research of 6G technologies. They will play an instrumental role in ensuring that the project outcomes are of great relevance, and their expertise will be crucial in the development of the testbed demonstrators of this project. They will also host the PDRAs to carry out tests of the proposed algorithms using their facilities.