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

Re: [linrad] Sound card clock accuracy

Part of what you see is likely due to the fact that you are
running the test process in user-space in a multitasking
environment.  Cron jobs, page faults, flushing data to disk,
all kinds of other things are happening on your PC at the same
time as your test.  This is complex enough that you way as
well concider it to be random and un-controllable.
We hope, however that the sample rate is
controled by hardware on the sound card.  Possibly it is on some
sound cards but is software controled on other cards.
This could explain the differing results we see.  The best way
to find out how your sound card really works is to read the
driver source code.

I would expect the better sound cards would have a xtal controlled
sample rate and then a large hardware FIFO buffer and then a DMA
interface to the bus.  Cheaper cards might have less of this
hardware.  Again the device driver souce code would be a good
place to look.  Phone modems are this way.  In the "good" modems
all the work is done in hardware on the card the cheap modems let
the CPU do much of the DSP work.  Same applies with Ethernet
and SCSI cards.  

--- Edson Pereira <edson@xxxxxxxxxxxxx> wrote:
> 
> Hi Chris,
> 
> I have the system clock already syncronized via NTP. So the system
> clock 
> error should be small.
> 
> Your suggestion of suing an external generator and analysing the
> frequency 
> and spectrum would be better than what I am doing, but I don't have a
> good 
> signal generator in hand.
> 
> So far, I am finding mixed results with different sound cards using
> the 
> measurement methos I've described. For example:
> 
> ! 124.999779
> ! 124.999832
> ! 125.000076
> ! 124.999847
> ! 125.000076
> ! 124.999863
> ! 125.000069
> ! 124.999878
> ! 124.999901
> ! 125.000046
> ! 126.248909 <- ?
> ! 123.751038 <- ?
> ! 126.020233 <- ?
> ! 125.228722
> ! 124.761238
> ! 125.010040
> ! 124.986298
> ! 124.997009
> ! 125.001129
> ! 124.998131
> ! 125.000763
> ! 124.998695
> ! 125.000549
> ! 124.999054
> ! 124.999420
> ! 125.000259
> ! 124.999535
> ! 125.000206
> ! 124.999619
> ! 125.000175
> ! 124.999672
> 
> The figures above is provided by a machine (P-III 1.2 GHz) using the
> a cheap sound card and the OSS-free C-Media CMI9738 sound driver.
> 
> ! 126.023323
> ! 123.976028
> ! 125.000076
> ! 125.000092
> ! 125.000107
> ! 125.000130
> ! 125.000160
> ! 126.012695
> ! 125.485916
> ! 123.501068
> ! 125.000084
> ! 125.000099
> ! 125.000122
> ! 125.000145
> ! 126.008385
> ! 125.021347
> ! 123.969788
> ! 125.000084
> ! 125.000092
> ! 125.000114
> ! 125.000130
> ! 126.006180
> ! 125.008499
> ! 123.984879
> ! 125.000076
> ! 125.000084
> ! 125.000099
> ! 125.000122
> ! 126.004822
> ! 125.004662
> ! 125.032608
> ! 123.957573
> 
> The figures above is provided by a machine (P-III 1.2 GHz) using the 
> SB-Live sound card and the alsa emu10k1 sound driver. 
> 
> Each figure represents the ammount of time to sample /dev/dsp
> 1,000,000
> times at 8 kHz. The results are strange. The SB card, if the
> measurements 
> are acceptable, is very inaccurate.
> 
> What I'm trying to do is to see which of my card is best suited for
> linrad
> and JT44, which will be depended on good sampling accuracy. Small
> variations that averages out over time should not cause much
> problems, but
> if the error is cumulative, then there may be a problem.
> 
> 73,
> 
> -- Edson
> 
> On Wed, 4 Dec 2002, Chris Albertson wrote:
> 
> > 
> > The average sample rate is likey quite stable.  (Or at least
> > stable if the temperature inside the PC chassis is constant.)
> > However, I wonder if the interval between samples is stable?
> > The interval could be quite unstable but yet the time to
> > make 1E6 samples would be always the same.  After all, the
> > average of 1,000,000 small random errors would ted to be
> > a constant.
> > 
> > 
> > possably a good test would be to first find a very good
> > signal generator and sample a 1Khz sine wave for a few
> > seconds.  Next compute the spectrum of the samples using DFT.
> > If you see frequency other then 1khz you can then try and track
> > down the cause.  noise on the analog signal, noise in the sample
> > period,...  Linrad itself could do this.  Put in a known good
> > signal and see what comes out.
> > 
> > If you are very concerned about the speed of your PC's clock
> > you can servo it to GPS.  No, I don't mean re-setting the time
> > to match GPS, I mean keep the _rate_ adjusted to match the 
> > GPS signal.  You need a GPS unit, data cable and some software
> > You can see www.ntp.org for the software but NTP comes with
> > most linux/unix systems.  The error in my PC's clock rate  before
> > it was corrected using NTP was about 83PPM.  Depends on the
> > xtal on the main board.
> > 
> > 
> > 
> > --- Edson Pereira <edson@xxxxxxxxxxxxx> wrote:
> > > 
> > > I'm trying to measure the accuracy of sound card clocks running
> under
> > > Linux and would like to discuss with people on the list about a
> > > posible
> > > method of measurement.
> > > 
> > > An indirect measurement I am using is to sample /dev/dsp at 8 kHz
> > > 1,000,000 times and compare time stamps (with millisecond
> resolution
> > > --
> > > ftime()) taken before and after the samplings. 1 million samples
> at 8
> > > kHz
> > > should take exactly 125 seconds. I am measuring how off from 125
> > > seconds 1
> > > million samples takes and from the difference I calculate the
> > > relative
> > > clock error. I am relying on the Linux system clock, so there
> will be
> > > some
> > > additional error, but it should not interfere much with the
> > > measurement.
> > > 
> > > Does any one see a problem with the way the measurement is being 
> > > performed? Any suggestion for improvements?
> > > 
> > > 73,
> > > 
> > > -- Edson, 7n4ncl
> > > 
> > > 
> > > 
> > 
> > 
> > =====
> > Chris Albertson 
> >   Home:   310-376-1029  chrisalbertson90278@xxxxxxxxx
> >   Cell:   310-990-7550
> >   Office: 310-336-5189  Christopher.J.Albertson@xxxxxxxx
> > 
> > __________________________________________________
> > Do you Yahoo!?
> > Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
> > http://mailplus.yahoo.com
> > 
> 


=====
Chris Albertson 
  Home:   310-376-1029  chrisalbertson90278@xxxxxxxxx
  Cell:   310-990-7550
  Office: 310-336-5189  Christopher.J.Albertson@xxxxxxxx

__________________________________________________
Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
http://mailplus.yahoo.com
LINRADDARNIL