discuss-gnuradio
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Discuss-gnuradio] Re: Mode S and ADS-B


From: Martin Dvh
Subject: Re: [Discuss-gnuradio] Re: Mode S and ADS-B
Date: Mon, 16 Apr 2007 23:04:44 +0200
User-agent: Debian Thunderbird 1.0.2 (X11/20070113)

Eric Blossom wrote:
> On Mon, Apr 16, 2007 at 08:24:46PM +0200, Martin Dvh wrote:
> 
>>Hew How Chee wrote:
>>
>>>Hi Martin,
>>>
>>>
>>>
>>>>I have been using the code with an external tuner
>>>>frontend and did some experiments with fm-receiving,
>>>>which just works fine.
>>>
>>>
>>>Interested to know what is the speed of the CPU that
>>>you are using and how many percent of the CPU power
>>>has been used up.
> 
> 
>>I have a dual-core AMD athlon 64 X2 3800+ (2x2 Ghz)
>>I think only one CPU is used by gnuradio.
>>I haven't looked at the processor percentages  in detail.
> 
> 
> Excellent!  Glad to hear you got a new machine ;)
Yes, I am very happy with it.
But as allways, within no time I am working on the limits of the new machine.
I am still considering a Cell-based machine (PS3).
Did you even got around to playing with that?
> 
> 
> Are you using gr.fir_filter_ccf or gr.fft_filter_ccc?
I am using gr.freq_xlating_fir_filter_fcf
which uses:
gr_fir_fcc

The cx2388x TV capture card is a real source so I don't use filters with 
complex input like ccf or ccc.

> 
> (There's no fft_filter_ccf since the cost for complex taps is the same
> as float taps in the FFT version.  _ccc will accept real taps in the
> argument list and convert them to complex automagically.)
But I can't connect a float input stream to gr_fft_filter_ccc directly.
I would have to put a float2complex block in front.
Even better a unsigned_char to complex or short to complex, (also not yet there 
yet, I know)
I would also have to do the freq translation manually in front by manually 
multiplying with a sinus sig_source with the right freq.

If might be more efficient to make a gr.freq_xlating_fir_filter_fft_fcf which 
does all in one step

But I don't know if an FFT version will also be much more efficient  very high 
decimation factors.
(I use a decimation of 100). The number of taps might run the cpu out-of its 
cache, which is something fftw doesn't like.
On the other hand log(N) is much better then N with high N.

I would have to try and do some profiling.

But for now just first get the tv-card tuner working.

No premature optimization. (I learned this from you)


Greetings,
Martin
> 
> Eric
> 





reply via email to

[Prev in Thread] Current Thread [Next in Thread]