Information and Communication Engineering

Research in this area covers a range of applications to communications systems. Examples range from high-speed circuits for next-generation optical systems, to system demonstrations and validation test-beds for wireless sensor networks.

Further information on this Research Theme can be provided by Professor Izzat Darwazeh.

This research theme studies the fundamental principles of information compression, transmission and processing by leveraging tools from information theory, communications theory and machine learning.

Research in this theme includes:

1. Foundations of Information Compression, Transmission, and Security

This work is unveiling the fundamental limits of data transmission and security in current and upcoming communications systems.

Work of the group and its collaborators have contributed to the information-theoretic characterisation of such limits as well as to information-theoretic inspired transceiver designs that achieve such limits [Pérez-Cruz,2010; Rodrigues,2013; Ramos,2013].

Further research from the group has contributed to the foundations and applications of information-theoretic and physical-layer security [Bloch,2008; Reboredo,2013] – this contribution was honoured with the IEEE Information Theory and Communications Societies Joint Paper Award 2011.

2. Foundations and Applications of Compressive Information Processing

Research in this area focuses on the foundations and applications of compressive information processing for future information processing systems.

The group's work has contributed to an information-theoretic inspired kernel design for reconstruction and classification applications [Carson,2012; Chen,2012; Chen,2013; Wang,2013], the foundations of compressive classification [Reboredo,2013] and reconstruction [Renna,2013].

This foundational work in compressive information processing has also been complemented by practical-oriented work in various emerging applications ranging from compressive signal and image processing [Carson,2012; Chen,2012; Renna,2013] to compressive topic modelling [Wang,2013].

This line of research is being funded in part by a Royal Society International Exchange Scheme involving both UCL and Duke University in the USA.

3. Emergent Applications: Unlocking Energy Neutrality in Energy Harvesting Wireless Sensor Networks

Capitalising on compressive sensing and distributed compressive sensing this research looks to unlock energy neutrality in energy harvesting wireless sensor networks [Chen,2013].

It is widely recognized that future deployments of wireless sensor networks infrastructures are expected to be equipped with energy harvesters to substantially increase their autonomy and lifetime.

However, it is also recognized that the existing gap between the sensors’ energy harvesting supply and the sensors’ energy demand is not likely to close in the near future due to limitations in current energy harvesting technology, together with the surge in demand for more data-intensive applications.

With the continuous improvement of energy efficiency representing a major drive in wireless sensor networks research, the major objective of this work is to develop transformative sensing mechanisms, which can be used in conjunction with current or upcoming energy harvesting capabilities, in order to enable the deployment of energy neutral wireless sensor networks with data gathering rates that are substantially higher than the current state-of-the-art.

Members of the group envision a typical (centralised) wireless sensor network architecture where a set of sensor nodes periodically convey data to one or more base stations; in addition, the sensor nodes would be active during a certain period to capture and transmit data and inactive during the remaining period of time to harvest energy from the environment.

By leveraging the emerging paradigms of compressive sensing and distributed compressive sensing as well as energy- and information-optimal data acquisition and transmission protocols [Buranapanichkit,2012; Vittorioso,2012], it will be possible to tightly couple energy demand to the energy supply in wireless sensor networks in order to achieve the proposed breakthroughs [Chen,2013].

This work is being funded by the Engineering and Physical Sciences Research Council, involving both UCL, the University of Cambridge and various industrial partners.

4. Machine learning for analysis and optimization of video representations

Video has been one of the most pervasive forms of online media for some time. Several statistics show that video traffic will dominate IP networks within the next five years. Yet, video remains one of the least-manageable elements of the big data ecosystem. We believe that this difficulty stems primarily from the fact that all advanced computer vision and machine learning algorithms view video as a stream of frames of picture elements. This is despite the fact that pixel-domain representations are known to be notoriously difficult to manage in machine learning systems, mainly due to: their high volume, high redundancy between successive frames, and artifacts stemming from camera calibration under varying illumination. 

In our work, we focus on video representations that go beyond conventional pixel streams and consider spatio-temporal activity information that is directly extractable from compressed video bitstreams or neuromorphic vision sensing (NVS) hardware. In order to understand and optimize video representations, we design deep neural networks (DNNs) that ingest such activity information in order to derive state-of-the-art classification, action recognition and retrieval results within large video datasets. Our recent results have shown that this can be achieved at record-breaking speed and comparable accuracy to the best DNN designs that utilize pixel-domain video representations and/or optical flow calculations. These results can enable for the first time the continuous parsing of large compressed video content libraries and NVS repositories with new & improved versions of crawlers in order to derive continuously-improved semantics or track changes and new content elements, in a manner similar to how search engine bots continuously crawl web content. These outcomes will pave the way for exabyte-scale video datasets to be newly-discovered and analysed over commodity hardware.
This work is funded by EPSRC, projects EP/R025290/1, EP/P02243X/1, and is carried out in collaboration with King’s College London, Queen Mary University of London, and Kingston University. 

5. Graph signal processing for machine learning

We are surrounded by large-scale interconnected systems, from the Internet to the power grid and social networks. While essential, the management of such networked systems is exceedingly hard mainly because of their intrinsic and constantly growing complexity. Currently available controlling strategies provide suboptimal solutions. Hence, the need for novel methods able to deal with this complexity. Answering to this need is the ambitious goal of our research, that blends together reinforcement learning and graph signal processing.
In our group, we focus on developing fundamental learning methodologies targeting data-efficiency in large-scale problems. Theoretically, we tackle the following topics: Graph-Based Bandit Problems and Structural Reinforcement Learning. Some example of use cases of our research are the following: 

• Influence maximization: to minimize advertisement costs while optimizing the evolution of interest over social networks. This is possible by learning how information spreads within social networks to optimally place advertisements/ source information. 
• Traffic prediction: to anticipate the cascade effect of train delays or traffic congestion, modelling the traffic as process evolving over a graph. 
• Detection of malicious behaviours: to identify malicious behaviours among people, modelling social interactions among people on the network as signal evolving on graphs.
• This work is carried out in collaboration with EPFL (Switzerland), Oxford University, and Imperial College.

This research theme develops and analyses the latest communications approaches in the physical layer (PHY).

Current research in this area include:

Massive MIMO

One of the key PHY concepts examined for 5G deployment is that of large-system MIMO or recently known as massive MIMO where there are a large number of antennas at the transmitter and/or receiver.

Members of the group and collaborators have contributed to revealing the fundamental performance limits of a wide class of MIMO fading channels in the very low SNR regime under interference-limited environments [Zhong 2010; Jin 2010; Li 2010; Wen 2010].

Another important direction is to understand the performance of MIMO under practical settings in the presence of such as rank deficiency due to the phenomenon of keyholes. Several major contributions have been made in this regard [Zhong-Wong 2010; Zhong-Jin 2010; Jin 2008].

Finally, recent work has been looking at the performance impacts of deploying high numbers of antennas in fixed physical dimensions [Masouros 2013], along with practical linear and non-linear precoding schemes for large MIMO systems [Masouros-Sellathurai 2013],[Masouros-Sellathurai 2012], [Masouros 2011].

Of particular importance to this line of research, are solutions to address the hardware-efficiency in designing large scale MIMO Transceivers. The group has contributed in these areas with studies of the impact of hardware components [Garcia-Masouros 2016, Garcia 2016, Amadori-Masouros 2017], studies on hybrid analog-digital techniques [Amadori-Masouros 2016, Li-Masouros 2017, Amadori-Masouros 2017]

Increasingly, relaying has emerged as a revolutionary technique for cellular networks, particularly for improving the performance of users at the cell edges. Several significant contributions have been made in this area, not only for understanding the capacity of MIMO relaying channels but also the sum-rate of a multiuser MIMO channel [Jin-McKay 2010; Zhong-Jin-Wong 2010; Wen 2011; Wen-Wong 2010].

Cognitive Cooperative Communications

Future spectrum allocation as proposed by Ofcom and FCC is required to be dynamic, adaptive and self-organised, with a strong need for interference control and management within and across networks.

This breaks the rigid boundaries across networks and demands network coordination, giving rise to several fundamental challenges. In the context of primary-secondary link co-existence for cognitive radio, a number of interference-constrained transmission schemes have been introduced [Khan 2012], [Masouros-Ratnarajah 2012], [Ratnarajah 2013], [Masouros-Ratnarajah-Sellathurai 2013] for the cognitive downlink channel.

In addition, several significant contributions in the optimisation of using cooperative relays for beamforming and some in the context of cognitive radio technologies have been made [Huang 2012; Huang 2011; Zheng 2009; Zheng-Wong 2009; Zheng-Wong-Paulraj 2009].

Wireless Interference as Useful Signal Power

While interference has always been regarded as the main obstacle in wireless systems, recent work has revealed that, on an instantaneous basis, it can contribute constructively to the useful signal energy gleaned at the wireless receiver [Masouros-2011],[Masouros-Ratnarajah-Sellathurai 2013].

Read more about Wireless Interference as a Source of Green Signal Power

This important source of useful energy, is currently ignored by the techniques adopted in the communication standards. The existing work of Green Communications focuses on energy-efficient network planning, smart duty-cycled BSs, heterogeneous cell deployment, integration with renewable energy sources and use of ICT for energy saving, amongst others.

The radical approach of exploiting interference power promises a vast reduction in the power consumption of the network, without the cost of redesigning and redeploying network components as per the solutions above.

While the signal processing work on wireless communications to date has focused on cancelling or minimizing interference, this work focuses on allowing the utilization of signal power from constructive interference to achieve the required link performance with a much reduced actual transmit power.

Integrated Optical/Wireless Networks

Recent technological advances and deployments are creating a new landscape in the access networks, with an integration of wireless and fiber technologies a key supporting technology.

Our research work considers how the optical fibre access network can be best used to support the next generation of radio access networks. It considers front and back-haul networks, including how mm-wave wireless technologies can support 5G wireless (see IPHOBAC-NG).

The group have been involved in system demonstrations for mm-wave back-haul [Omomukuyo 2013] as well as architectures to support dynamic allocations of bandwidth [Milosaclievic 2012], [Attard 2006].

Optical Access Networks

Research in collaboration with BT investigated the deployment of long reach (100km), high aggregate bit rate (10Gbit/s) and high customer number (1024) optical access networks. These use advanced optical components currently only feasible in core networks to provide highly efficient optical access networks [Shea 2007].

The area of long-reach optical access networks is now gaining a great deal of attention [Shea & Mitchell 2007]. We have now extended this idea to consider how multiple short-reach PONs could be consolidated into a multi-wavelength long reach architecture. This was first demonstrated for 2.5Gbit/s PONs using Semiconductor Optical Amplifiers (SOAs) and cross-gain modulation(XGM) [Shea 2009] and has since been expended to 10Gbit/s PONs using an interferometric structure of SOAs and cross-phase modulation (XPM) in [Cao 2013].

As part of this work, collaborations with the UCL optical networks group considered the impact on the burst mode receiver requirement of such technologies [Mendinueta 2011a] and [Mendinueta 2011b].

At the core of the Internet resides the network layer. It’s the glue that connects the millions of networks together to implement the information superhighway. The group has lots of activities that try to re-imagine and re-engineer this interconnection

Resilience

As our lives become more and more dependent on the availability of the Internet it is crucial that our data does not get lost and arrives quickly at the destination. We envision, for example, remote medical surgery where every movement on the controller needs to arrive in milliseconds to the remote operating room without getting lost.

To build routing protocols that deal with resilience efficiently is one of our main goals. No packet left behind.

A Faster Internet

We want faster and low delay communication. Adding capacity to our links is just part of the story. We need to make every link as useful as possible and, in many cases, reduce the delay from the service to the user.

We are designing routing mechanisms that allow much better-tailored decisions depending on the service being used. We are also designing network architectures that place services in the edge close to the users reducing service delay.

Content is King

The Internet is becoming a pure video delivery system. Soon, 90% of the traffic on the net is going to be video. The Internet was not designed for this. We have several projects that redesign the network layer to improve content delivery. Through this, we are evolving new solutions to replace the current IPv6 infrastructure (see for example 5GMEDIA project) and more revolutionary solutions that redefine the network layer as an Information Centric Network (ICN) where IP addresses are replaced by content addresses.

 

Information on this Research Theme can be provided by Professor George Pavlou.

Our ambitious vision is to offer an unprecedented level of immersion and interaction through virtual reality communications.

To make this vision a reality, key challenges behind content delivery need to be addressed: extremely high-resolution video signals that depict the 360-degree surrounding scene (omnidirectional videos) need to be efficiently delivered to the end users and the interactive nature of user-content interaction must be taken into account.

Unlike classical directional videos, omnidirectional imaging involves spherical acquisition, representation and rendering geometries. The visual signal is defined on a sphere and the user, virtually positioned at the centre of the sphere, can navigate the scene by changing viewing direction. At each instant in time, only a portion of the spherical content is displayed. This displayed portion depends on the user’s viewing direction, thus cannot be known in advance by the sender.

Interacting with virtual scenarios is becoming an important way to enhance the way we learn, carry out research and do our work [Royal Society]“Interacting with virtual scenarios is becoming an important way to enhance the way we learn, carry out research and do our work
[Royal Society]

The novel type of content (spherical versus classical planar perspective videos), the high spatio-temporal resolution and low latency required, in addition to the interactive way this media is consumed has raised new engineering challenges throughout the entire end-to-end multimedia chain.

These challenges include signal acquisition, representation, processing, compression, delivery and rendering steps. Our research is aimed at designing an optimal video streaming strategy for omnidirectional videos, based on a user-, content- and application-dependent prediction model of users’ behavior.  

1.    Adaptive Streaming for Virtual Reality

The key goal of this research is to optimise the adaptive streaming platform for interactive communications such as Virtual Reality (VR).

VR applications target high-quality and zero-latency scene navigation to provide users with a full-immersion sensation within a scene.

From a network perspective, this requires transmission of the omnidirectional content in its entirety, at a high resolution, which is not always feasible in bandwidth-limited networks.

In this work, we propose an optimal transmission strategy for virtual reality applications able to fulfill the bandwidth requirements, while optimizing the end-user quality experienced in the navigation.

This line of research is also being funded in part by Adobe, USA. 

2.    Predicting Users Behaviour in Virtual Reality

In Virtual Reality (VR) applications, understanding how users explore the visual content is important in order to optimize content creation and distribution, develop user-centric services, or even to detect disorders in medical applications.

In our work, we seek novel machine learning strategies to both identify common patterns in the way users interact with the content and predict the way user navigate within the VR content.  

This line of research is also being funded in part by a Royal Society International Exchange Scheme involving both UCL and CWI in The Netherlands and is carried out in collaboration with EPFL (Switzerland), CWI, and INRIA.

Our work in the area of the Internet of Things (IoT) is researching and innovating across different layers, from waveform design through to applications and services, and from sensors, through to connectivity and data analytics.

The Internet of Things (IoT) is fuelling innovation in nearly every part of our lives. Billions of connected devices are acquiring and processing data, which in turn is analysed for the purpose of optimising products and services, or informing business decisions.

A key bottleneck in IoT systems are the current communication networks, which have not been designed to handle the unprecedented heterogeneity, mobility and traffic present in IoT networks and systems, as well as the ultra-low latency requirements specified for future, next-generation networks.

Our areas of work include but are not limited to the following:

Waveform design for the IoT

A new signal waveform termed enhanced narrowband IoT (eNB-IoT) has been designed using non-orthogonal modulation techniques. eNB-IoT can double the number of connected devices, quadruple the data rate per device, and support deeper signal coverage compared to existing NB-IoT without the need for additional spectrum resources.

Internet of Visual Things / Internet of Silicon Retinas

Vision intelligence combines conventional image processing with deep learning. While conventional cameras use a frame-based approach, dynamic vision systems (DVS) are designed to mimic the human retina by processing sparse streams of events at microsecond resolution. Research in this area is looking at techniques for representing, compressing and transmitting captured DVS data to the cloud efficiently. This area of research can be seen as “neuromorphic learning” combining the biologically inspired silicon retinas with the biologically inspired neural networks.

Localisation / Wireless Sensor Networks

Research in the area of localisation is a timeless topic with endless applications. The research in this areas is looking at ways of optimising existing solutions in terms of simplifying the network architecture, lowering the cost of the devices used and replacing conventional “mains-powered” devices with very low power smart devices capable of running off a battery for several years and in some cases, enhanced with energy harvesting capabilities. Applications for outdoor localisation include for example, proximity sensing in the construction industry to improve the safety of construction workers. A popular application for indoor localisation is asset management, for example, tracking beds in a hospital and monitoring their real-time location, as they are moved from one ward to another, or to different areas of the hospital.

Smart buildings

Improvement in energy efficiency and optimisation of user comfort levels are the key aims of “smart” buildings. This is achieved through the acquisition and analysis of building data obtained by using multiple sensors deployed throughout the buildings. These sensors should consume minimum power to allow them to be powered by a battery with a lifetime spanning several years. At the same time, the communication between the sensors and the cloud should be robust enough to ensure that the sensor data reaches its final destination with minimum packet losses.

Machine learning for smart cities

Beyond the massive volume of data, requiring a throughput rate of tens of gigabits per second, and the massive quantity of connected devices, amounting to billions, Ofcom has also predicted an unprecedented large service heterogeneity, in terms of latency, throughput, user capacity and so forth, for IoT services. 

This heterogeneity cannot be served any longer by current networks, based on the “one-size-fits-all” approach, using one wireless network for all services and devices. Instead, the general industry consensus is that the next digital revolution can be sustained only by agile and highly flexible networks, where the mobile network adapts its operation based on the actual service requirements and the related context.

To this end, another area of research is looking at implementing flexible, dynamic and automated reconfigurable networks by developing novel theoretical frameworks to design and evaluate decision-making strategies in large-scale networks. This will lead to real-time control and re-configuration of dynamic and large-scale mobile IoT communication networks.

Information on the research theme can be provided by Professor Alex Galis.