DSP Stuff: floating point or fixed point

[Ben Stuyts]

On Sat, 9 Jan 1999, Roman Sowa wrote:

> Say, what?...
> Not satisfied with IEEE floating point format? Why not use
> 40-bit floats in SHARC. And you can also use it as 32-bit
> fixed poit DSP

I think one of the points being made was that you need to have a high  
precision accumulator. Like the Motorola 56k series: fixed point 24 bit data,  
but 56 bit results. That gives you lots of headroom. With the SHARC I think  
you are limited to 32 or 40 bits, right?

Best regards,
Ben