FPGAs display their processing prowess

摘要

Mention "DSP" in conversation around the water cooler, and what vision will pop into other folks' heads? Probably a picture of a piece of silicon from a company such as Analog Devices, Motorola, or Texas Instruments. This image isn't necessarily wrong, mind you, but it's a bit like (as the old saying goes) "putting the cart before the horse." First and foremost, DSP stands for digital-signal processing; that is, converting analog signals to the digital domain, arithmetically transforming them in some way and then translating them back to the analog domain for human sensory consumption. Digital-signal processors from the aforementioned companies and many others are only one vehicle for implementing digital-signal-processing functions. The earliest DSP chips, after all, were little more than otherwise general-purpose CPUs with Harvard architectures (separate instruction and data buses and separate caches), befitting their algorithms' data-centric nature. As general-purpose CPUs have grown speedier and particularly as they've added onboard arithmetic coprocessors and signal-processing-optimized instruction-set extensions, they're increasingly taking over the calculation burden that might have formerly required a separate compute engine. (This article uses "digital-signal processing" to refer to the function and "DSP" to refer to the processor.)