How to estimate DSP processor performance
Phil Lapsley, G. Blalock · IEEE Spectrum · 1996
The market for products based on digital signal processing (DSP) technology-wireless communication devices and PC multimedia peripherals, for example-is growing rapidly. Semiconductor manufacturers have responded to this demand with a bewildering array of DSP processors. Selecting the best one for a given application presents a difficult and time-consuming challenge for DSP system designers. Simple, familiar performance measures like MIPS and MOPS are misleading and neglect factors like memory usage, power consumption and application execution time. Complex alternatives-such as application benchmarks-suffer from limitations that virtually preclude fair comparisons. Fortunately, a compromise methodology that combines algorithm kernel benchmarking with application profiling yields good estimates of processor performance weighted to the target application.