 |
Jerry McGuire Wanna Trade?
Algorithms developed in the "infinite precision" of a host workstation environment are best (and most quickly) ported to processors that support the dynamic range of IEEE format floating-point math. And while the benefits of floating-point-specific processors are well defined (dynamic range, accuracy of an extended-bit mantissa, ease of achieving numerical accuracy during algorithm development, and dramatically faster time-to-market), the tradeoff has always and forever been B.O.M. cost. "All things being equal, any engineer would certainly rather use a proper floating-point DSP for floating-point math!"The engineer's frustrating choice was between an elegant floating-point processor with fast time to market, OR the low component cost of a fixed-point processor. Penny margins have certainly warranted the long and arduous task of developing and characterizing the accuracy of frequency-domain processing on cheaper integer chips for a wide variety of products. But, all things being equal, any engineer would certainly rather use a proper floating-point host for floating-point math! Among the many bits of knowledge I've acquired in my career, one thing that really stands out is that all things technical do eventually come to those who wait. And we at Analog Devices are awfully pleased to bring this floating-point tradeoff nonsense to a happy end with the advent of our $5 SHARC floating-point DSP. Floating point at a fixed-point price - you're already thinking of how to use it, aren't you? As they say in negotiated tradeoff circles, "this changes everything."
Jerry McGuire is a Vice President of Analog Devices' General Purpose DSP Group. His column appears here regularly. |
Hell-bent on improving the speed and precision of every application that pays homage to Jean-Baptiste Joseph Fourier, engineers have built system-, board- and chip-level floating-point math engines throughout the modern DSP era. Floating-point performance defined the earliest "mine's bigger" DSP chip leadership, but there is also no question that the steady advance of floating-point DSPs has literally created markets and applications that previously did not exist at all (premium digital audio being the most obvious example).