Wireless Networks
I've worked a good deal on 802.11/WiFi networks, mainly related to MAC analysis and design. My current work is focussed around small cell networks more generally, especially small cell, mixed LTE/WiFi networks and unlicensed LTE. This includes algorithms for collision-free decentralised scheduling, scrambling code allocation, antenna tilt angle adjustment, energy management and multipath scheduling/routing in mixed LTE/WiFi networks.

Multipath Transport and Coded TCP
TCP congestion control involves a delicate balancing act. Recent work has focussed around improving performance over lossy wireless links by use of network coding at the transport layer. Coded TCP, developed jointly with Muriel Medard, is currently being commercialised via Code On LLC and Speedy Packets. Currently, I'm interested in multi-path overlay transports that build on coded TCP (this is joint work with Nokia Bell Labs and CONNECT)

Predictive Analytics & Decision Making Under Uncertainty
I'm currently working on the links between decision problems and convex optimisation, how we can take a joined up approach to predictive analytics and decision making, and privacy-enhanced algorithms. Past work includes fully decentralised (no message-passing) algorithms for constraint satisfaction, which we have applied to collision-free scheduling, scrambling code allocation and channel allocation in wireless networks. Also a good deal of work on distributed algorithms for optimal resource allocaton in wireless networks.

Velocity-Based Theory (Off-equilibrium Linearisation).
Nonlinear systems are generally difficult to analyse and design. In contrast, we have a nearly complete theory of linear systems and a wealth of design methods. It is therefore very attractive to try to adopt a divide and conquer approach where we decompose a difficult nonlinear design task into simpler linear design tasks, and such approaches underly many popular engineering approaches. My interest is in mathematically establishing to what extent nonlinear analysis and design really can be carried out using linear ideas.

Data-Intensive Modelling of Dynamic Systems.
The trend is for software systems to move off the desktop and into the environment, e.g. to appliances with embedded software which can sense and manipulate its environment. Examples range from mobile appliances and modern intelligent network infrastructures to modern cars, which are packed full of software. Decision-making software relies upon the availablility of suitable models describing the environment and the consequences of different decisions. We need tools for constructing appropriate mathematical/statisical models for real-time decision-support.

Modelling and Control of Wind Turbines.
In pitch regulated wind turbines, the power output is regulated by adjusting the angle of the turbine blades to compensate for wind speed variations. However, the active regulation of wind turbines presents a challenging control problem as (i) the aerodynamic characteristics of the rotor are highly nonlinear; (ii) a primary control objective is to alleviate loads throughout the turbine in order to minimise fatigue damage - this is nonlinear requirement which places little penalty on normal operating loads but a great penalty on occasional high loads; (iii) the actuator bandwidth is low with tight rate limits. For details see online publications. Also our benchmark wind turbine simulink model for evaluating control performance.