Re: [Discuss-gnuradio] math range error?

[Eric Blossom]

On Tue, Feb 21, 2006 at 05:12:06PM +0100, Martin Dvh wrote:
> Eric Blossom wrote:
> > On Mon, Feb 20, 2006 at 01:09:17PM -0800, joe j wrote:
> > 
> > You're probably trying to plot some NaN's or Inf's.  Not sure where
> > they might come from (division?).  On the other hand, if what you're
> > trying to get is the angle, just run the output the multiplier into
> > gr.complex_to_arg().  No need for mag or div.

> I also had a lot of these when I was doing things like this.
> I think it has the same cause as why the scopesink autoscaling started at 
> +1e31 many times.
> For some reason the first few samples of usrp_source_c sometimes has strange 
> values (very big, Nan, inf)
> I don't know if it is some uninitialized buffer or something else.
> 
> As a workaround I sometimes use 
> gr.skiphead(gr.sizeof_gr_complex,numberofsamplestoskip) to skip the first few 
> thousand samples of the usrp.
> 
> Although I noticed that you get errors when numberofsamplestoskip is greater 
> then 4000.
> (Probably the proper use of history has to be updated in gr.skiphead)
> 
> Another workaround is capturing to file and using dd to skip the first 32 
> ksamples

Martin, good observation.  

Perhaps logging the data to a file and then inspecting with octave
will reveal something.  If it's in usrp1_source_c, it shouldn't be too
hard to track down.  There's not that much code.

Eric