Signal processing in resource insufficient environment

Abstract

Most embedded signal processing applications are developed in at least two separate stages: signal-processing design followed by its digital implementation. With such an approach computational tasks that implement the signal processing algorithms are usually scheduled by treating their execution times and periods as unchangeable parameters. Task schedulability therefore is independent of the actual state of the physical environment; it depends only on the amount of computing resources available. In embedded systems, typically due to power and energy constraints, the available computing resources are definitely limited. A better overall performance might be achieved if signal-processing design and task scheduling are linked, and an integrated approach is applied. An attempt is made to handle temporary resource insufficiency by introducing quality-of-service (QoS) adaptation into signal processing. The approach applied can be considered as a "never-give-up" strategy, where the signal processing is performed in any case at the price of lower quality. In the proposed solution different algorithms are available at task execution level, having different execution times and quality. The version to be executed is selected by the ongoing scheduling mechanism. In our experimental setup the earliest deadline first (EDF) algorithm is applied for this purpose, and different-order median-filters are utilized to illustrate the concept of QoS adaptation in signal processing.