VLSI Digital Signal Processing Systems: Design and Implementation

18.2    EVOLUTION OF PROGRAMMABLE DIGITAL SIGNAL PROCESSORS

General purpose DSPs were first developed in the early 1980s for iterative and computation-intensive digital signal processing applications. These applications run mostly under tight real-time constraints and involve large numbers of memory accesses and repetitive executions of the same algorithm. To meet the performance requirements and to accommodate programmability at the same time, earlier DSP processors deviated from the general purpose microprocessors by introducing hardware multipliers for fast multiply-accumulate operation, and by using multiple-access memories. PDSPs do not provide all of the programmer friendly features typically available on general purpose processors. As a result, these achieve higher performance with low cost and low power consumption. The first generation DSPs were developed for high- throughput (millions of operations per second) audio signal processing. Their characteristics are compared with the microprocessor of the day (M68000) in Table 18.1 [2]. As can be seen, the DSP-1 exhibits a 50:1 performance advantage over M68000, for about the same cost and power [2],

DSP processors have experienced performance improvements, and price and power consumption reduction during the last 2 decades. Generally DSP processors exploit parallelism where one instruction can incorporate several operations in parallel. As a result, their MOPS (million operations per second) figures had been much higher than ...