| Person in charge of the project: |
Energy-Aware Data-Type Exploration for Signal Processing in Software Defined Radio|
The low energy efficiency (GOPS/mW) of current Software Defined Radio|
(SDR) solutions for personal handhelds is a major roadblock. New processor architectures with major improvements on energy efficiency are emerging. However, this is still not sufficient to meet the continuously increasing complexity of wireless physical layers within the limited energy budget. Beyond an architecture centric approach, careful algorithm-architecture co-design is required in order to conciliate the needs for flexibility and high energy efficiency. The typical worst-case static dimensioning should be avoided both for the architectures and for the algorithm software implementation. Instead, scalable solutions are preferred. The latter are dynamically re-configured at run-time to provide the performance versus energy consumption tradeoff which matches the instantaneous requirements. Particularly, such scalable implementations can exploit data format properties (e.g. fixed-point signal representation) to offer different tradeoffs between accuracy and energy consumption. As a proof of concept, the data format scalability has been applied in the implementation of relevant SDR baseband processing algorithms. Separate implementations with different fixed-point representations are produced at design time. Then, a run time controller selects the appropriate implementation depending on the external conditions. By leveraging on sub-word parallel architectures with support for various sub-word sizes, the reduced bit widths are greatly translated into savings in energy and processing time.
Duration of the project:
01.10.2006 - 01.10.2010
Onderzoek met eigen middelen