[Discuss-gnuradio] GOSSIP / gEDA / scheme

['Eric Blossom']

On Thu, Nov 15, 2001 at 11:17:54AM -0800, Ettus, Matt wrote:
> 
> 
> > -----Original Message-----
> > From: 'Eric Blossom' [mailto:address@hidden
> > Sent: Monday, November 05, 2001 6:18 PM
> > 
> > Eventually we'll get to some kind of scripting solution to string the
> > modules together.  Possible Guile (scheme).
> 
> 
> That's interesting.  For quite some time now, I've been using GOSSIP (
> http://savannah.gnu.org/projects/gossip ) for system simulation on several
> projects.  It uses guile for hooking up modules, as well as controlling the
> whole simulation process.  

Thanks for the pointer.  I'd stumbled across it before, but hadn't
really paid any attention to it.

> Last night I finally got some time to take a serious look at gnuradio.  I
> really like that its threaded, and will use multiple processors if
> available.  I also like that its possible to produce and consume variable
> numbers of data items.  GOSSIP doesn't do either of those things.

As I suspect you've inferred, it can also consume and produce
vectors of data at each sample index.  I plan to use it for a general
purpose FFT block.  Basically serial samples in; complex FFT out every
1/N samples, where N is the FFT size.

> GOSSIPs use of scheme is quite powerful.  While its not done yet, it was not
> too difficult for me to write a netlister backend (also in scheme) for gEDA
> ( http://geda.seul.org ), which is a schematic capture program.  I can use
> it to make block diagrams, and netlist straight to GOSSIP simulation files.
> It would take more of a scheme guru than I am, but it would be only a few
> more steps to make gEDA directly run gossip interactively.

Cool.  In the last project I worked on I built a scheme based system
that we used to generate netlists for PCB layout.  Saved us the
trouble of schematic capture and symbol creation.  You could build
scheme functional descriptions and then invoke them (e.g., create 1MB
x 32 of flash and do it using 16-bit wide parts).  We also used a data
base of parts that we searched by a partially specified set of
attributes.  You could say I want a 1K resistor and based on some
project specific selection criteria you would get a 0805 or 0604
resistor.  If you also specified that it needed to be 1W, it would
pick the "best" part that fit all the criteria.  I glued a
conventional schematic capture tool into the flow, and then we were
able to merge the scheme generated netlists with the conventionaly
generated netlists and create a merged netlist.  The two universes
where tied together with "wormholes" which looked like components to
each side.  We ended up doing all the digital design in scheme and
left the analog stuff in traditional schematic capture.  Seemed to
give us the best of both worlds.  The system automatically added
bypass caps, and other stuff.  It could also generate "cap and bead"
bypassing if desired.

> There's another package out there, called SPUC ( http://spuc.sourceforge.net
> ) which is a C++ class template library for signal processing.  There might
> be some stuff of use to GNURADIO in there.

It's a nice package too.

> Anyway, I'm really excited about GNURADIO.  Hopefully I'll be able to
> compile it tonight...

I hope we're getting your compilation problems sorted out.

Eric