[Flexradio] [dttsp-linux] Re: [hpsdr] DttSP ovsv.c possible optimization?

[Jim Lux]

At 09:33 AM 8/14/2007, Frank Brickle wrote:
>Jim Lux wrote:
>
> > One could put arbitrarily complex processing in an FPGA, and for that
> > matter, many FPGAs today include a CPU core (or more than one.. e.g.
> > Virtex 4 has multiple PowerPC cores).
>
>Yes, but...why?

More to address the fact that complexity of processing alone doesn't 
necessarily drive you to a DSP CPU vs FPGA or General purpose CPU.


>Choosing to use or not use FPGAs is surely an actuarial decision, not a
>technical one.

Perhaps programmatic decision might be a better term, but that's exactly it.
(Actuarial usually refers to statistical calculations used to assess 
risk, as for insurance).

The original poster suggested implementing FFTs and the filters in a 
FPGA, and you responded that some of the radio algorithms (e.g. AGC, 
noise processing) have non-simple implementations in logic, leading 
you to suggest a TI DSP.  I merely wanted to point out that there are 
some very complex radios with lots of nonlinear logic (in both 
frequency and time domain) implemented entirely in FPGAs, as well as 
entirely in software.  As a developer, I much prefer working on 
general purpose or DSP CPUs, particularly ones with floating point, 
rather than FPGAs, but that's because I come from a software oriented 
signal processing background. Someone who comes from a hardware 
signal processing background might be more comfortable in logic.

   There's also the programmatic issues you allude to above. For 
example, there's a distinct paucity of radiation tolerant, high 
reliability CPUs, DSP or otherwise, while there ARE rad tolerant, 
space qualified FPGAs (e.g. the Xilinx Virtex II, and some Actel 
products like the RTAX series).

  Or, one might be developing a prototype that will eventually be 
turned into a high volume, low power, low cost ASIC for a consumer 
product, in which case the FPGA>ASIC process is fairly low risk.

For ham hacking, on the other hand, we're exceedingly low volume, 
tend to not care about power dissipation, and prefer environments 
with free tools, which makes the ubiquitous PC platform very 
attractive for a signal processing environment, if it can handle the 
load, even though it may be "inefficient" or "suboptimal" in a 
commercial DSP sense.


Jim, W6RMK