As the standardization of 5G wireless networks progresses, the research community has started focusing on what 6G will be. Motivated by the need of ensuring high data-rates, while at the same time saving spectrum, a major technology that has been proposed for 6G is the integration of communication and sensing services in the same infrastructure. This enables wireless networks to perceive the surrounding environments, triggering new services and leading to a more efficient use of resources. The INTEGRATE project focuses on the theoretical, algorithmic, and architectural foundations of integrated communication and sensing networks, developing the first open access network-level simulator for joint communication and sensing. To this end, a new implementation of wireless transceiver is proposed, which leverages the use of reconfigurable holographic surfaces and allows the integration of communication and sensing with remarkable performance while at the same time reducing the energy consumption. Specifically, INTEGRATE will: 1) Develop reconfigurable holographic surfaces capable of supporting joint communication and sensing tasks and that can be integrated in wireless transceivers with minimal cost and energy requirements. 2) Characterize the fundamental performance limits of integrated communication and sensing networks, developing an algorithmic framework and protocol suite to approach these limits. 3) Build the first open access software simulation platform for joint communication and sensing networks.

This project will develop multi-antenna full-duplex technology to achieve highly efficient spectrum usage in HetNets (heterogeneous networks). Full-duplex radios are much more than just doubling the capacity as perceived in current literature. More explicitly speaking, we believe that it is making communication technologies impossible become possible. The implications of full-duplex communications are transformative. Full duplex, which permits simultaneous transmission and reception, motivates a fundamental rethinking of the ways wireless networks are designed and optimised. However, the fact that the power of the transmitting signal is so much larger (over 100 dB) than that of the receiving signal has fuelled the belief that it is impossible to separate them on a single channel. Recently, this picture has begun to change, following the pioneering work by Choi et al. from Stanford University who demonstrated a working full-duplex radio on a single channel. We envisage that full-duplexing can transform the operations of wireless networks and is expected to have massive benefits in HetNets. HetNets are widely regarded as one key wireless technology for the provision of future wireless communications (including 5G) by complex interoperation between macrocells and small cells. The enormous interest for HetNets is due to their capability of providing high regional capacities and flexible coverage, and more importantly their low infrastructure costs. In HetNets, a mosaic of wireless coverage is obtained by a variety of wireless coverage zones from macrocell to small cell such as, pico- and femtocell. Of increasing interest to mobile operators are the customers' installed femtocells that can greatly improve indoor coverage but share the same frequency band as the macrocells. There is a huge scope of research in resource allocation and physical-layer design and optimisation in HetNets. This project will exploit the full potential of MIMO full-duplexing in HetNets by designing a holistic solution that interconnects antenna design, physical-layer signal processing, and network resource allocation to address the inherent challenges of full-duplexing and realise its massive end-to-end benefits. To achieve this goal, UPFRONT first proposes new antenna design specially for wideband MIMO full-duplexing, which is substantially more challenging than the existing narrowband single-antenna case. Next we leverage the powerful MIMO signal processing to handle the overlooked in-tercell interference and new interference introduced by the full-duplex operation, which are critical to deliver the end-to-end benefits of full-duplex HetNets but were not well studied before. Furthermore, UPFRONT will explore the unexplored full-duplexing opportunities to address the networking-wide resource allocation challenges associated with the adoption of full-duplexing small cells under the greater macrocell structure sharing the same mobile spectrum. The outcomes of UPFRONT will elucidate the importance of a holistic approach to full-duplexing design and have impact in fundamental and practical research of future wireless networks.

Inspired by neuroscience, informed by information-theoretic principles, and motivated by modern wireless systems architectures integrating artificial intelligence (AI) and communications, this Fellowship sets out to develop a paradigm-shifting framework for networked machine learning (ML) that is centred on the following ideas. 1. Free energy minimisation: According to the free energy principle, agents optimise internal models so as to minimise their information-theoretic surprise vis-a-vis the available data and prior information. This principle offers a basis to reason about epistemic uncertainty ("know when you don't know") in AI agents that is grounded in information-theoretic analyses of out-of-sample generalisation - away from the current narrow focus on point-wise accuracy, towards uncertainty quantification and calibration. A well-calibrated agent can make informed decisions about when to refrain from acting, about when and how to collect or request more data from the environment or other agents, and about how to guard against anomalies or malicious agents. 2. Networked meta-learning: In meta-learning, agents do not share an ML model in full as in conventional, centralised, solutions. Rather, only a meta-model is shared as a means to transfer knowledge across agents, while enabling the optimisation of personalised local models. As advocated by FreeML, meta-models can naturally implement the engineering principle of modularity by encompassing a common repository of functions that can be combined to suit the cognitive needs of each agent. This framework bridges the gap between the dominant centralised or joint learning approaches - including also federated learning - and the individual learning baseline, by means of limited model sharing, while still enabling meaningful cooperation with a controlled privacy loss. 3. Native integration of wireless communication and learning: Conventional wireless systems are based on the principle of separation between computing and communications. In contrast, the native integration of communications and learning advocated by FreeML embeds wireless communication primitives as part of the data generating and processing model. Like state-of-the-art integrated solutions, the proposed approach aims at fully utilizing radio channel capacity by avoiding inefficiencies due to separate processing. Unlike existing methods, however, the FreeML framework moves away from the standard problem of communicating under uncertainty (on the communication channel) to the novel problem of communicating uncertainty (on the solution of the cognitive task) under uncertainty (on the communication channel) in order to support networked meta-learning. Overall, FreeML sets out to study a novel, theoretically principled, paradigm for ML that moves away from the current centralised, accuracy-focused, state of the art in ML to embrace decentralization via wireless connectivity, uncertainty quantification, personalisation, modularity, privacy preservation, and the right to erasure. FreeML will involve three industrial partners -- Intel, InterDigital, and Samsung AI -- that will provide guidance and feedback on aspects related to implementation efficiency, communications, and integration with wireless networks, respectively. This Fellowship proposal builds on the PI's unique inter-disciplinary expertise in information theory, ML, and communications, and is intended to enable a step change in the applicant's career towards a leadership position at the intersection of the fields of engineering and ML/AI. Through this programme, the PI will reach out to a diverse community of STEM students, public, regulators, journalists, and academic colleagues across the two fields to advocate for the central role of engineering for reliable and sustainable ML/AI.

Wireless communications have enabled a plethora of novel applications in recent years thanks to the continuous research efforts to increase the spectral efficiency (SE) and energy efficiency (EE) of wireless networks. Multi-antenna (MIMO) processing plays a central part towards harnessing those gains. MIMO has grown much beyond the original point-to-point channel and can nowadays refer to a diverse range of centralized and distributed deployments (e.g. multi-cell MIMO, cooperative/coordinated MIMO, distributed MIMO, massive MIMO, network MIMO). The fundamental bottleneck towards enormous spectral and energy efficiency benefits in multiuser MIMO networks lies in a huge demand for accurate channel state information at the transmitter (CSIT). This has become increasingly difficult to satisfy due to the increasing number of antennas and access points in next generation wireless networks relying on very dense heterogeneous networks and transmitters equipped with a very large number of antennas. CSIT inaccuracy results in a multi-user interference that significantly degrades the network performance. Looking backward, the problem has been to strive to apply techniques designed for perfect CSIT to scenarios with imperfect CSIT. The motivation behind this project is the following: wouldn't it be wiser to design wireless networks from scratch accounting for imperfect CSIT? In this project, we leverage recent progress in information theory and initial results by the PIs to address the above fundamental CSIT problem (and its resulting multi-user interference) by introducing a rate-splitting (RS) network architecture. Contrary to current approaches where transmission is operated in a broadcast manner with one private message per user, the approach considered consists in splitting one receiver's message into a common and a private part and superposing this common message on top of all users' private messages. The common message is decoded by all users but intended to only one of the users. Such approach has recently been found to be optimal from an information theoretic perspective in a multiuser deployment with imperfect CSIT and significant enhancements over conventional approaches in terms of spectral efficiency and power utilization have been demonstrated by the PIs. This visionary project conducted at Imperial College London and University of Edinburgh by leading experts in wireless communication theory aims at leveraging those recent findings to design and demonstrate the suitability of an RS-based MIMO wireless network architecture in a multitude of scenarios. To put together this novel wireless network solution in a credible fashion, this project focuses on designing 1) RS for a single transmission point, 2) RS for a large number of co-localized antennas (also called Massive MIMO) in microwave and millimeter-wave bands, 3) RS for a large number of distributed antennas representative of dense heterogeneous networks, 4) RS for multi-antenna relay channel and finally 5) evaluating the system level performance of RS-based networks. The project will be performed in partnership with leaders in equipment manufacturing and standardization (Toshiba and InterDigital) and in defence and emergency services (Qinetiq). The project demands a strong track record in wireless communication, MIMO signal processing, optimization, information theory and it is to be conducted in a unique research group with a right mix of theoretical and practical skills. With the above and given the novelty and originality of the topic, the research outcomes will be of considerable value to transform the future of wireless and give the industry a fresh and timely insight into the development of robust MIMO wireless networks, advancing UK's research profile of both wireless communication in the world. Its success would radically change the design of the physical layer of wireless communication systems and have a tremendous impact on standardization.

The complexity of wireless communication networks has grown considerably in recent years. This has been driven in part by academic research that has started to define the information theoretic boundaries and advantages of certain complex networking topologies and protocols. On the other hand, the demands from consumers and industry have pushed wireless networks towards more sophisticated architectures and solutions, primarily in order to ensure a broad range of services can be delivered using a common infrastructure. This is particularly true of 4/5G technologies, which many believe should support all things for all people, including voice, data, public safety, distributed sensing and monitoring, etc. However, similar beliefs and trends can be found in other sectors, such as smart grid networks and even satellite networks. It is important that engineers understand the global properties of complex networks, and how these properties arise from local structure. Such information can be fed into models and optimisation routines so that practical networks can be designed to perform as well as possible. A common approach to tackling complex problems is to exploit randomness and statistical properties of the underlying system. Probabilistic approaches to network modelling are not without their difficulties, and some of the main problems that researchers have struggled with over the years arise from the fact that networks are finite entities with physical boundaries. Recent research by the investigators has focused on the effects that boundaries have on connectivity when networks are embedded in some finite spatial domain. Analytic expressions for the overall connection probability have been obtained. These formulae quantify the intuitive phenomenon that nodes near the boundary are more likely to disconnect, and thus they explain how the network outage probability behaves at high node densities. This work has been extended considerably to explore notions of resilience (k-connectivity), the effects of node directivity, diversity and power scaling laws, complicated geometric bounding domains (both convex and non-convex), and even the interplay between higher layer trust protocols and the physical network set-up and spatial domain. In this project, the probabilistic formalism alluded to above will be exploited further to study several key concepts that influence the structure of spatially embedded networks. The following four topics will be treated: - continuum models of spatially embedded networks, including the investigation of spectral and centrality properties of random networks; - mobility models in spatially embedded networks, including random waypoint and Levy flight processes; - trust models in spatially embedded networks, including trust dynamics and protocol design; - temporal models of spatially embedded networks, including dynamical node and link (edge) models. The work will take a mathematical approach, but will always maintain a focus on practical implications and designs.