The Computer Engineering Lab focuses on how to use field programmable gate array (FPGA), VLSI and parallel processor technologies to solve difficult computing problems. Our strengths include:
- Binary deep neural networks (DNNs) for high-speed inference
- On-chip training of DNNs
- Low-latency time series prediction
- Real-time, high dynamic range object detection*
- Integration of software defined radio with AI technology**
- Spectral weather prediction***
- Enhancement of military working dog performance****
- Plasma receiver for CubeSats
We develop novel architectures, applications and design techniques to solve problems in signal processing and machine learning. In collaboration with Xilinx Research Labs, we pioneered binarised neural network implementations which allow higher performance than that of GPUs.
This has enabled applications such as our NGTF project, “High Speed Machine Learning using FPGAs” project where we demonstrated the feasibility of 500K classifications per second for the radio frequency modulation classification problem. To achieve this, we combined a software defined radio and a machine learning inference engine on a single device. Radio frequency machine learning is an important building block in Electronic Warfare and surveillance applications.
Successful past and present projects include a DIH Bio-inspired high dynamic range imagery (2019), DSTG/Data 61 NGTF ”High speed machine learning using FPGAs”( 2017/2019), TAS DCRC Distributed Autonomous Spectrum Management (DUST) (2019-2022)