WP2.1 Resource constrained smart sensor systems
WP2.1 Summary presentation slides
Our aim is to co-design algorithmic hardware and software approximations and optimizations to maximise throughput and minimise power consumption for signal and image analysis on single sensor and multi-sensor networks.
Approximation techniques can be roughly divided into algorithmic (e.g. stochastic optimization, sketching, dimensionality reduction), which have theoretical characterizations, and software/hardware (e.g. decreasing data resolution, voltage and frequency scaling, memoisation), which are usually applied ad-hoc without guarantees.
We propose to formalise the error impact of low-level hardware and software approximations, integrating this knowledge in high-level analytics and applying synergistic approximations throughout the implementation stack. The exemplary applications will include 3D scene mapping through complex media and target localisation within a sensor network.
- Y. Wu, A. M. Wallace, A. Aßmann and B. Stewart, "Mixed Precision ℓ1 Solver for Compressive Depth Reconstruction: An ADMM Case Study," 2021 IEEE Workshop on Signal Processing Systems (SiPS), 2021, pp. 70-75, doi: 10.1109/SiPS52927.2021.00021 - View video
- Approximate Proximal-Gradient Methods A. Hamadouche, Y. Wu, A. M. Wallace, J. F. C.- Mota - View Video
WP2.2 Reconfigurable signal processing
WP2.2 Summary presentation slides
Most signal processing systems dispose of information along the processing chain. The benefits include: reduced communications, computational complexity and memory requirements. However, this “lost” information may be valuable and worth recovering: (i) when the system must be rapidly reconfigured beyond its intended use to address an unforeseen and imminent threat; (ii) for post-engagement, as part of a forensic analysis, to extract specific information tailored to evolving mission goals. In this WP we will consider the fundamental limits of such information recovery, active steps that can be put in place to facilitate it and the developments of algorithms to perform the necessary secondary inference.
- E. Ward and B. Mulgrew, "Memory NLEQ Techniques to Mitigate Cross-modulation Effects in Radar," 2021 IEEE Radar Conference (RadarConf21), 2021, pp. 1-6, doi: 10.1109/RadarConf2147009.2021.9455301. - View video
- Learning a Secondary Source From Compressive Measurements for Adaptive Projection Design F. K. Coutts, J. Thompson, and B. Mulgrew - View video
- Detecting LFM Parameters in Joint Communications and Radar Frequency Bands K. Zhang, F. K. Coutts, and J. Thompson - View video