Sascha Pypke wrote:
-------------------------------
>The main difference between a >controller and a DSP is the fast >working floating point unit from a DSP.

Hi, Sascha. If you look around, see that most of the DSPs available are fixed-point ones (ahhh I remember the 96000 from Motorola, which died years ago).

And the main diff isn't only the fast ALU, but even more the MAC (multiply & accumulate) unit coupled with simultaneous data move. A typical DSP can, in one cycle, fetch coefficient, multiply it, add with accumulated result and move data (ie. TI DSPs, DMAC) or increase pointers. On a 16-bit fixed-point DSP, the accumulator uses to be 24-bit (not true in old TI DSPs, which had a 16-bit acc).
The MAC instruction is so important because is the basis of most filtering routines (FIR filters).

>You need often floating point >calculation for more precision in some >filter types.

Programming DSPs -> a big bag of tricks to avoid overflow, limit cycles and other nasty issues related to fixed point arithmetic. Keeping dynamic range is one of the main concerns.

>A very fast controller is always to >slow to calculate floating point in >filter-routines. I do some expiriences >with 8052 (33MHZ) but its to slow.
>In the next time i will check the >Cygnal chip. When you using Fixed >Point the filter has a high level of >noise !

All GSM mobile phones have a 16-bit fixed-point DSP inside. And the sound quality isn't too bad !!. And comfort noise is added to avoid full silence !

Many uCs are moving into the DSP bandwagon... I can think now of Hitachi.

Regards,
Txip.