Metacognitive Radar For Emerging Sensing Applications (METSA)
Name of the research project :
U-AGR-7234 - C22/IS/17391632/METSA - OTTERSTEN Björn
Funders :
FNR - Fonds National de la Recherche
Funding number :
C22/IS/17391632
Funding text :
Many of the recent scientific works in signal processing have been increasingly influenced by the most complex and least explored signal processing system – the human nervous system. Cognition refers to the process through which humans and animals sense and interact with their environment and cognitive radar signal processing aims to parallels to neurobiological cognition to learn from the sensed environment and act accordingly e.g., better focus the sensing on particular areas of interest while nulling others. While the radar gets data on the entire surrounding, it’s the cognitive signal processing that brings about the decision on how to act after inferring from the data.
Similarly, metacognition is a well-studied concept in both neurobiology and educational psychology, and can be termed as a higher order thinking which can be summarized as learning about learning or knowing about knowing. It is pertinent to remark the importance of metacognition as separate from plain cognition. Since the cognitive cycle is a closed loop system, without any provision of altering any of the steps once the cycle has kicked in a operational system. This leads to an inherent inflexibility of the system to adapt to drastic change in the channel conditions, change of engineering modules, or the operating objective or all of these. Hence, the radar must include multiple strategies with their own cognitive cycles. The selection of the appropriate strategy is handled by metacognition.
Consider the case of vehicles equipped with Automatic Emergency Braking (AEB) systems. It has been reported that many situations like bridges, railroad tracks and parking garages can trick the vehicles into breaking by making them believe that they are about to crash. Even steam coming out of underground tunnels and ducts in large cities can trigger the system, apparently. A standard operational cognitive radar cannot fully “learn” this situation. In a metacognitive radar, on the other hand, this flaw can be monitored and a learning strategy for this situation devised by selecting the optimization objective and solution. Furthermore, the learned information can be transferred in a connected networked so that other cars do not make the same mistake.
The project formalises such a learning for the different tasks of radar (detection, estimation, classification and tracking) with the transmit waveforms and receiver algorithms as the means of implementing the learned rules. The project aims to create a system that acts in optimal manner based on various possible strategies that can be learned from the environment.
