Re: [Discuss-gnuradio] USRP: FFT on FPGA?

[Jens Elsner]

Matt,

thanks for your reply.

> > is it feasible to implement an FFT directly on the
> > FPGA in the USRP? 
> 
> Possible, yes, but the FPGA does not have any "hard" multipliers, so
> you'll need to use logic cells for that.  But I don't think it is necessary.
> 
> > I would like to use the USRP for spectral analysis, 
> > a 512-point FFT (and windowing) would greatly reduce
> > CPU load. A "hardware" FFT would also be a great plus
> > for OFDM waveforms.
> >   
> 
> The CPU on the host machine is very capable of high-speed FFTs.  We can
> easily do thousand point FFTs on 8 MS/s data streams.

I would like to localize signals in time and frequency in real time.
My CPU can handle the FFT in real time, but additional (and not
optimized) tasks like segmentation and classification take up a fair
share of CPU time as well.

It would just be nice to have a "switch" that tells the USRP to give
frequency domain data, maybe even averaging internally. I understand
that this is a step away from "Software Radio", but this seems to be
necessary with my standard hardware. Once I get to the point I need
optimization, I'll consider this again. Buying a new PC might be
easier. :)

Regards,
Jens