To develop a generic learning framework for handling uncertainties in measurements acquired in a networked battlespace environment. Links to L_WP1 through domain knowledge, L_WP3 and L_WP4 in handling incomplete sensor information and achieving robustness to jamming, and L_WP5 in real time implementation. This WP will exploit domain knowledge of a networked battlespace to improve performance and confidence and to reduce uncertainty. Examples are digital maps about terrain and layout of the field, historical data about the site, geometric relations between platforms, and operational conditions such as weather.
L_WP2.1 Reducing uncertainty by incorporating domain knowledge using Bayesian inference, adaptive signal processing and sparse sampling
We will consider how to quantify the information in the world model and express it in a deterministic/probabilistic statement. New signal processing algorithms offering adaptivity to operational environments will be developed by exploiting domain knowledge (e.g. the change of the operation conditions when the sensor platforms move, or what decisions follow from the signal processing results and their consequence), and design parameters in these algorithms (e.g. the threshold for detection) or different types of signal processing models/algorithms will be selected based on domain information. Historical data will be used to build up the priors in Bayesian inference for different objects of interest and different scenarios, which will reduce the reliance on real-time measurements in the battlespace. The Bayesian inference framework will also be extended from a single to multiple sensor platforms operating in a networked environment, by fusing all the information, including sensory capabilities, constraints and geometric relationships between different sensor platforms.
L_WP2.2: Robust signal processing techniques under uncertainty, modelling uncertainty with stochastic dynamic processes, and characterization of uncertainty with a game theoretic framework
Robust signal processing techniques based on convex optimizations will be developed to tackle uncertainty. Mathematical models and approximation techniques will be developed to model an uncertainty region as a convex hull so that low complexity algorithms can be developed. Robust techniques based on both a probabilistic approach and worst case optimizations will be developed. Instead of treating uncertainty as caused by a static collection of events and associated relationship, the uncertainty will be investigated within the framework of dynamically evolving phenomena. In this framework, uncertainty will be considered as caused by dynamic entities having states and transitions from one state to another resulting from actions in the battlespace. Both hidden Markov model and Bayesian networks will be used to characterise uncertainty. To enhance characterization of uncertainty and to understand the underlying mechanisms further, this WP will consider uncertainty as caused by dynamically varying actions created by various players in the battlespace, e.g. coalitional forces and enemies. Hence a game theoretical framework will be developed. The work will start with a non-cooperative game theoretical framework and will be extended to Bayesian games to account for incomplete information. The possible battlespace scenarios that will be considered within this framework will include air formation to ground attack-defence system, defence against jamming in radars (linked to L_WP 4.1) and counteracting uncertainty created by deception by enemies, for example fake RF signal injection.