RE: [linrad] FFT performance

[edson@xxxxxxxxxxxxxxxx>]

Hi Roger,

Thanks for taking the time with me on this issue.

On Thu, 14 Nov 2002, W3SZ wrote:

> Hi, Edson!
> 
> A similar thing happened to me a long time ago on my Pentium Pro 200.  I
> don't remember the details (more than 2 years ago, I think) but I think
> the following:
> 
> Three things:
> 
> 1.	Does the same thing happen when you type "A" for weak signal Cw
> or "D" for SSB as it does when you type "F"?  
> 

It does happen on the other modes as well. 

> 2.	There may be a parameter file there that is causing problems.
> Look for par_userint in the 'lir' directory where linrad is.  It may be
> that the adspeed is set to 48000 or something that overtaxes your cpu
> (though it shouldn't).  I attached par_userint listing below to show you
> what my parameters are.  But this may be too much for your machine.  Try
> setting the AD speed to 11025 or something even lower.  You can edit
> this file on your machine with a text editor.
> 
> vga mode [12]
> font scale [1]
> mouse speed [100]
> input mode [7]
> rx channels [2]
> ad channels [4]
> ad speed [96000]
> ad device no [0]
> ad device mode [0]
> ad frag [0]
> da device no [7]
> min da speed [5000]
> max da speed [48000]
> max da channels [2]
> max da bytes [2]
> min da channels [1]
> min da bytes [1]
> check [2220024]
> 

This is my par_userint

vga mode [12]
font scale [1]
mouse speed [10]
input mode [20]
rx channels [1]
ad channels [2]
ad speed [8000]
ad device no [32]
ad device mode [0]
ad frag [0]
da device no [32]
min da speed [8000]
max da speed [48000]
max da channels [2]
max da bytes [2]
min da channels [1]
min da bytes [1]
check [2220024]

> My par_wcw file (for mode 'a') is:
> 
> First FFT bandwidth (Hz) [20]	200
> First FFT window (power of sin) [3]  0
> First forward FFT version [3]  may need to make 1 or 2
> First FFT storage time (s) [20] 4
> First FFT amplitude [1]
> Enable second FFT [0]
> First backward FFT version [0]
> First backward FFT att. N [6]
> Second FFT bandwidth factor in powers of 2 [2]
> Second FFT window (power of sin) [0]
> Second forward FFT version [0]
> Second forward FFT att. N [7]
> Second FFT storage time (s) [5]
> Enable AFC/SPUR/DECODE [1]  0
> AFC lock range Hz [500]
> AFC max drift Hz/minute [1000]
> Enable morse decoding [0]
> Max no of spurs to cancel [0]
> Spur timeconstant (0.1sek) [5]
> First mixer bandwidth reduction in powers of 2 [1]
> First mixer no of channels [1]
> Baseband storage time (s) [200]
> Output delay margin (0.1sek) [5]
> Output sampling speed (Hz) [6000]
> Default output mode [1]
> Audio expander exponent [3]
> A/D speed [22050]
> Check [1110038]
> 
> You can carefully make a par_test file for mode "F", or make a par_wcw
> file maybe for mode "A", etc.  That may get you started.  If the files
> aren't made just right, they will not work.
>

What "just right" mean? How can it be found?
 
> For starters, cut the A/D speed, or reduce the powers of sine, and/or
> broaden the FFT bin size, and/or turn off AFC to reduce CPU
> requirements.  I am sure others will give you better information in this
> regard.  I've always had fast computer and so haven't really learned
> what must be changed to speed things up.
>

I have run linrad on a pentium 200 sometime ago without problems. It is 
very unlikely that a P3 1.2 GHz cannot provide enough CPU power. For a 
sampling rate of 8 kHz, the CPU load should very low. 

I have written applications using heavy FFT and sampling at 44.1 kHz on
the same machine without problems.
 
> 3.	Did you really try EVERY combination of parameters on the page?
> If not it would be helpful to know what you did try so we can figure out
> where things are wrong.
> 

No, of course not. What I've tried was to get low values and change them
up a in steps. If I have to try every value, than that is something the
computer should do, not me :-). The paramenters I played the most ware the
1st FFT bandwidth (400, 350, 200, ..., 5, 0 ) and the FFT window (0, 1, 2,
..., 9). I turned off the second FFT.

> And then there is always the possibility that the error has nothing to
> do with anything I said, but that it is a symptom of an audio driver or
> other problem. 
> 

May be. In my experience, there are lots of sound applications that makes
use of some parameters that are not supported by many sound drivers. But
since I have run linrad (two years ago) on few different type of
soundcards, I assume it is quite mature in terms of support of different
sound drivers. I will take a look at the code and see what is actually
causing the "Too slow" messages. Again, 1.2 GHz should not be too slow for
an application that claims to work on a P60.

73,

-- Edson, pu1jte, n1vtn, 7n4ncl




LINRADDARNIL