Communications & Networks

With immense potential in big data analytics, machine learning and artificial intelligence, massive amounts of data traverse inside communication networks. Timely decision making requires timely data delivery through efficient communication networks. Motivated by the aforementioned, cognitive communications and networks are being designed with embedded intelligence, allowing for significant performance gains. For example, autonomous fault detection and recovery for high network availability and low latency, interference and resource management and network balance, consistent quality of experience.

Our research extends in the following areas:
-    Next-generation communication systems and networks (5G and Beyond)
-    Advanced Software-Defined Radio and Networks
-    Communications waveform designs
-    Machine learning and artificial intelligence-enhanced context-aware network management
-    Advanced optimisation techniques
-    Physical and network security
-    Joint radar-communication systems
-    MIMO radar
-    AI-enhanced wireless localisation

Heterogeneous network (HetNet), which enhances capacity and coverage while delivering a better user experience in cellular networks through data off-loading, and Wireless Avionics Intra-Communications (WAIC), which deploys multiple wireless networks in aircraft to replace the current wired networks for passenger services and aircraft monitoring are some of the technologies developed through joint application of the aforementioned research capabilities.

The Networks & Traffic Engineering group primarily looks at carrying out R&D in two main domains; network protocols and traffic engineering, which sometimes share some similarities in the type of optimization approaches, mathematical modelling and analytical methods used to solve large complex problems.

For networking protocols, the group is interested in researching into architecture and concepts that allow the automated orchestration and management of network slices that allow services and applications to operate with the desired QoS over heterogeneous wireless networks. In particular, we are keen to address resource allocation challenges of network slicing in a heterogeneous wireless network with varying communication environment, where the system faces various challenges in terms of isolation, customization, elasticity and end-to-end coordination. Besides this, the group is interested in developing effective approaches to convert the network service requirements to the desired network resources with considerations at different levels, including the control level, data plane level, and network wide level and developing AI-based dynamic slice optimization and policies that caters for optimal real time service level adjustment.

In the areas of Traffic Engineering, the group supports a wide range of projects related to traffic control, route guidance, traffic prediction and bus service reliability. The group analyzes and leverages on real-time datasets from new technologies such as connected vehicle probe data and advance sensors for pedestrians and vehicle tracking to develop micro, meso and macro models to predict movements of vehicles and pedestrians for use in the various transportation related applications. The group also uses AI techniques to develop fast and automated detection of abnormal events such as incidents, congestion build-up, etc. and develop intelligent solutions to control traffic lights as well as support system wide route guidance decisions.  

With these above goals, the group’s research tries to address cross-disciplinary boundaries in these two domains by combining techniques from network optimization, queueing theory, graph theory, network protocols and algorithms and model predictive control.

GPS is widely used for outdoor localisation purposes but is not suitable for indoor applications. I²R's research in indoor localisation technologies is able to provide centimetre-level positioning accuracy in challenging environments. A typical localisation system consists of mobile devices or tags that are tracked by a set of stationary anchors installed on walls or ceiling. In highly cluttered environments, line-of-sight between the tags and anchors can easily get occluded. Installing many anchors to overcome this problem is however impractical due to high setup and maintenance costs. Hence we were motivated to develop a solution that is practical by requiring only minimal additional infrastructure while maintaining acceptable levels of accuracy and robustness

The solution uses a multi-modal cognitive approach that is rooted in extraction of semantic information from various sensor data and then to apply AI to derive accurate and reliable position information. Hence sensor data from both purposefully installed infrastructure such as UWB anchors as well as other existing infrastructure such as BLE and WiFi access points available in the environment are used. Further, where available, information from additional sensors such as cameras and IMUs are also utilized through data fusion to build a robust and accurate multi-modal positioning engine.

The adoption and evolution of Industrial Internet of Things (IIoT) enables manufacturing industry to use real-time data to increase agility, enhance overall productivity, optimize asset usage and achieve cost-savings throughout the value chain. Massive amount of data could be moved across connected machines, computers and humans, and turned into valuable, actionable information, which delivers transformational business outcomes. This signifies that IIoT has to support diverse use cases and applications with different communication and computational needs, e.g., predictive maintenance, dynamic schedule and resource optimisation, asset tracking, quality inspection, assistive manufacturing, etc. This is challenging given that every application requires different data rate, latency and bandwidth.

Therefore, it is of utmost importance that multiple wireless network protocols can co-exist to connect edge devices, such as video cameras, RFID readers, sensors, machines, robots and etc. It is also essential for a reliable communication system to span across multi-tier networks using various radio access technologies and standards. The IIoT network management needs to make sure that multiple wireless networks can co-exist securely and guarantee the required quality of service with minimal human intervention.

Our IIoT technologies and recent research aims to address these challenges in the IIoT domain:

•   Low-latency Self-Organizing Network (SON) and Wireless Time Synchronization for synchronized
    real- time monitoring, high availability / reliability and efficiency
•   Large scale wireless sensor network deployment (e.g., mobile IoT)
•   Wireless network security and anomaly detection
•   In-network learning and information processing for industrial IoT (e.g., automated intelligence for
    advanced manufacturing)
•   Automatic network fault detection and recovery
•   Joint optimization of communication, computation, and control with new inter-disciplinary approaches

Intelligent Sensor Platform focusing on implementing real-time intelligence at the edge. Edge Analytics will minimise the transmission of raw sensor data to the server, hence reducing communication bandwidth and latency. The key focuses here include developing deployable algorithms on Sensor Drift Detection and System Anomaly Detection at the edge, reducing false positives and false negatives. The Edge Intelligence will be developed on commercial off-the-shelf processors, such as FPGA (Field Programmable Gate Array) or AI (Artificial Intelligence) ASIC (Application Specific Integrated Circuit). The Intelligent Sensor Platform developed is a scalable end to end solution which provides sensor data fusion with standard and reliable communication, backend intelligence, a database for further data processing, and storage in both private and public cloud.

The fifth generation (5G) network offers enhanced mobile broadband (eMBB), critical machine type communications (cMTC), massive machine type communications (mMTC), and fixed wireless access services. Promising to provide much higher data volume, network energy efficiency and spectral efficiency, and much lower latency than the fourth generation (4G) network, 5G aims to use one network to support multiple industries and use cases. Real-time automation, enhanced video services, monitoring and tracking, connected vehicles, hazard and maintenance sensing, smart surveillance, remote operations, autonomous robots, augmented reality (AR), top the list. Mission-critical drones, smart factory/manufacturing, next-generation virtual and augmented reality (VR/AR) glasses, low-complexity massive IoT devices such as sensors and meters, have been identified as the leading killer applications enabled by 5G.

Demanding QoS needs to be guaranteed in the above killer 5G applications. Many challenges need to be addressed to achieve this, e.g., the wireless hostile environment with a lot of operating and moving machines, blockages, environment noises and interferences, the high dimensionality and coupled interaction of the optimizing parameters, the fragmented radio resources and spectrum access, etc. This motivates us to research in applying cognitive learning and AI to learn and adapt to the dynamically changing environment and deliver the requested QoS with the optimal spectrum and power efficiency. Specifically, we work on radio access network, network slicing and virtualization, and build a research test bed to validate our proposed advanced features.

While one of our research focus is to optimize the 5G and beyond systems and networks, we also study how 5G and beyond systems and network enable further intelligence and additional services. Our current work in the latter includes 5G-based fine-grain localization and tracking, joint radar-communication design, etc.