Towards 6G: How could the future wireless networks look like?
Nowadays, wireless networks are mainly configured in a centralised manner, where the whole traffic is managed from a central point. In fact, a central unit collects and processes all the information, in order to make decisions on resource allocation, a process often introducing extra delays. However, in the future we can imagine having billion devices from phones and tablets to industrial machines and drones, struggling to use the spectrum, causing a lot of interference, but also demanding instantaneous decisions, especially in the case of critical applications. Naturally, we can wonder, could we exploit part of the capabilities of such devices at the edge of the network by deploying a network in a distributed way, instead of relying on a centralised one?
Dr. David Gesbert, professor, head of the Communications Systems Department of EURECOM and holder of the ERC Advanced grant PERFUME on Smart Device Communication, will guide us through the main outcomes of PERFUME project: an emerging distributed network design, featuring intelligence and decision making at the edge nodes.
Q. What is the context and motivation of your recently finished 5-year ERC Advanced project, PERFUME?
DG. Our goal was to design how future wireless networks will operate, going towards the future 6G (beyond5G) technology, making use of all the capabilities of the network closer to the end user. In other words, exploit the memory, the computing power and transmission capabilities of phones, tablets, PCs but also robots, which are all equipment residing at the edge of the network. Currently, the terminal equipment is viewed as passive devices, running applications in a way that does not require local intelligence, under-utilising their potential capabilities. Also, in the future the rapid increase of automated industrial manufacturing, using robotic devices to perform essential tasks, will rapidly increase the need for timely decisions locally, which are extremely hard to achieve in a centralised deployment. Indeed, those devices have computational power used currently to run their applications, but not yet for the benefit of the network. So the idea is, why don’t we use the network in a distributed fashion and exploit part of the devices capabilities for the service of the network, instead of relying on a centralised network, where we need a unit that controls all traffic?
Q. What is the difference between a centralised and a distributed network?
DG. In a centralised network, the central point would have indeed all the information needed to make a good informed decision on how the resources would be allocated to the devices of the network (power transmission, frequency, bandwidth), but it will take much time to collect it. So, by the time the decision is sent back to the user, it could be already outdated, since networks are not static. For example, imagine you are in your car, moving around, with an environment changing constantly, therefore if you take too much time to make a decision, there would be too much latency, causing potential safety issues.
On the other hand, in a distributed network the decisions could be very quick, since there is no need to centralise everything. The downside however, is the access to information. In fact, each terminal typically only has access to its local neighbourhood information, “seeing” maybe a few other devices, without knowing exactly what the other devices are going to do. So, they are going to try and cooperate based on partial information. If I have to give an analogy, imagine you are playing football with a team and you need to score a goal, but suppose that the players are playing blind folded and they don’t actually see each other: they have to operate with hearing and feeling but not seeing. Now, they have to progress to the opposite side of the field, pass the ball to each other and score the goal. Well, it is not easy to cooperate when you don’t know what your teammates around are doing! The same problem applies here. For instance radio devices would like to cooperate with each other to minimise interference, but they have limited inter-communication capability between them, hence how can they coordinate under such uncertainties?
Q. Could a distributed network configuration improve the network performance then?
DG. In fact, between the two network configurations we end up with a trade off: more timely decisions but based on partial information (distributed) versus more complete information but decisions with more delay (centralised). In other words, either we centralise all the information for all the devices, push them to the cloud to make a decision and send it back to the network, or we maintain the decisions at the local level. For a critical mission, needing an immediate response as accurate as possible, can we know which is the best design between the two?
Well, there is no easy answer! For that, there is a need for extensive analysis, but we were able to show that distributed systems would do just as well and even better than the centralised ones. Then, the choice depends on the exact application.
The key idea of the PERFUME project is to develop decision making algorithms for connected devices, made to be robust with respect to the fact that the information is noisy, local and not complete. We developed a new theoretical framework for this problematic, and defined some fundamental limits by using tools from the domain of information theory. We also designed more practical algorithms based on signal processing and also machine learning (ML) and implemented them into simulation software.
Q. So, what is the main outcome of PERFUME project?
DG. Apart from the design of ML algorithms for efficient and robust distributed networks, there was a part in my proposal, which at the time I did not think that it would be central. But things turned out differently as they usually do in research. We experimented with the above described algorithms that we designed, using terminals that are not phones, but in fact flying robots (drones). Imagine that we have drones that carry radio equipment, which would essentially make them flying base stations, creating a flying network. Then, we would need that drones to position themselves and fly at the appropriate positions, so that they can serve the communication needs in the best possible way.
Today when we make a phone call, we receive it through the nearest cell tower coming from our operator, but it is a fixed one. Now potentially, the drones would be able to move around freely, and put themselves always at the optimal position, so people can have the best communication quality. Drones can collect constant measurements and then our algorithms convert those measurements into an optimal path. Essentially the drones will be flying autonomously to the optimum location in the sky, such as we could obtain the best possible coverage at any time. We came up with a lot of interesting problems when using radio equipments and so we formed an experimental lab with drone prototypes and this ended up being an important part of the project. The application to robots drew a lot of positive attention and this work became central, with half of the people on the team working on drones as flying radios, and cooperating drones. We also won the 2019 best project award from the SCS Pole for our work.
Taken together, I would say we were able to provide two main outcomes:
- Information theoretic limits and also new ML algorithms to improve the understanding for efficient and robust distributed systems at the edge of the network.
- Intelligent, cooperating flying radios, through connected drones programmed to optimise network coverage.
Q. Drones are a very interesting application of connected devices at the edge of the network. What are the challenges for actually deploying flying connected drones?
DG. This is new in a sense that there is no product like that, because indeed there are a lot of challenges. First of all, it is not easy for people to accept the idea of having drones in the sky, finding it threatening and untrustworthy. So, the first obstacle to overcome is social acceptance.
However there are some applications for which flying drones would be very useful, such as rescue missions and geolocalisation of missing people. We have developed algorithms for that. Another one is to provide extended coverage in case of people gathering for a sporting event or a concert and there is not enough bandwidth for the service providers to give internet access to everybody there. We could very quickly provide extended coverage capacity by flying drones, acting as a relay between the people of the ground and the fixed infrastructure.
The other challenge concerns current drone regulations that forbid flying drones without a certified pilot, meaning that autonomous flying is prohibited. So, we are waiting for the regulatory aspect to evolve as well. It is analogous to the idea of autonomous cars; they exist, but they are not allowed to be driven in Europe yet.
Another more technical limitation for the moment is the energy autonomy of drones. Currently, they have a flight time of around 30 minutes, which is not sustainable as a solution for a potential product but we have devised practical solutions around it.
Now, regarding the distributed implementation of wireless networks, the main issue is that we need enough consensus among the lead players, in order to change the way we design wireless networks, since it affects fundamentally everything. It is not an add-on that we can propose for a network, it changes radically the way we deploy it. We would need to wait for a political momentum behind this idea, but it is uncertain if it is ever going to happen since there is too much at stake for all the parties already involved. Now things have been changing because of open source, which is acting as a distraction, facilitating small players to enter this market with their own ideas, so maybe things will be changing more rapidly.
Q. And your conclusion remarks at the end of a 5-year ERC project?
DG. First of all, I encourage all my colleagues to try and pursue the ERC funding, which is a great opportunity to do research with very little constraints. For example, what seemed as a small idea in my proposal turned out to be the central point. There is no consortium like in normal EU projects, often lead by industry players, so if you want to change the course of your project you won’t need to ask permission. The ERC gives you a lot of freedom, but I am sure they already know!
Through this project, I discovered key opportunities in the area of the interaction between robotics and communications, an aspect I didn’t consider before submitting this ERC project.
Drones is an example, but it is not the only one. In the future, it is clear that there will be more robots, used in homes, stores, factories, etc. and we will have teams of robots, needing to work together, communicate and interact. Thus, their communication has to be as efficient as possible, meaning that we need to design proper connection protocols and decision making algorithms at the level of the robot, taking into account the limitations of the communications protocols. For example, if we need robots making decisions based on camera/sensors/gps inputs, like in the case of autonomous cars, it will not be possible to transmit such a large amount of collected data. We would have to be able to identify the essential information and choose which data to share with the network. These challenges, at the interface between the decision making protocols and communication protocols, are going to be very exciting to solve.
by Dora Matzakou for EURECOM