On-site Parallel Signal Processing

Parallel processing is a critical technique to realize real-time signal processing for computationally demanding applications for the control of multi-input multi-output (MIMO) systems. The Structures, Pointing, and Control Engineering (SPACE) laboratory testbed is one such system. Decentralized control algorithms that cannot be handled in real-time by a single processor need to be implemented across multiple processors in order to achieve real-time performance. While the decentralization approach helps in reducing the computational requirements, a sequential system still does not meet the required throughput rates. The objective of this study is the implementation of real-time decentralized control algorithms through parallel processing that can be accomplished using the SPACE testbed. Key technologies in this scenario include parallel programming paradigms; decentralized control algorithms; task re-mapping and re-scheduling in processor failures; and the application program and its user interfaces and visualization. The following figure illustrates the concept.


Architecture for Parallel Signal Processing and Data Interfaces

This research activity involves developing coarse-grain parallel programs with optimization of task mapping and scheduling, which is advisable based on a realistic computational model. A systematic approach is used based on the computational model and the following parameters:

       Complexity and degree of parallelism of the signal processing algorithm;

       Communication latency;

       Bandwidth of the interconnection network; and

       Real-time performance requirements.

Technologies used to support fault tolerance include:

        Pipelined scheduling;

        Fault detection and fault handling; and

        Task re-mapping and re-scheduling.