Managing memory requirements in the synthesis of real-time systems from processing graphs

In the past, environmental restrictions on size, weight, and power consumption have severely limited both the processing and storage capacity of embedded signal processing systems. Today, however, as increases in processor speed and capabilities continually out pace increases in memory densities and performance, processor capacity is no longer a major concern for many signal processing applications-memory usage is now the primary concern. We present techniques for managing the memory requirements of signal processing applications in the synthesis of a real time uniprocessor system from processing graphs. To demonstrate the effectiveness of our memory management techniques, we compare the memory requirements of a statically scheduled implementation of an INMARSAT (International Maritime Satellite) mobile receiver, with our dynamic scheduling techniques. The case study demonstrates that state of the art, static schedulers use over 300% more memory than our simple, preemptive, EDF scheduler for a large class of signal processing applications.