[time-nuts] Soundcard sampling Re: Picking a good HP 10811

[Bruce Griffiths]

Ulrich Bangert wrote:
> James and Stanley,
>
> Bruce is already aware that I am doing sound card experiments since a
> few years but you are not. 
>
> That is why I would like to enter the discussion and encourage you to
> perform some preliminary experiments on the topic that don't need the
> real world existence of a suited sound card because you will generate
> your "input" signals yourself the artificial way. This will enable you
> to give your signals some known technical properties that are to be
> found in real world signals too. 
>
> I encourage you to test your algos for phase/frequency detection on
> these artififial signals because if you discover problems with your algo
> you will have an immediate understanding of the sources of your problems
> (not necessarily a solution) because you made the signal in exactly this
> way.
>
> First generate yourself a sinusoidal signal at 1000 Hz and Amplitude 1
> and a second at 1000.125 Hz and 80 dB down against the first signal. 
>
> Const RunTime=1000;
>       SampleRate=48000;
>       f1=1000.0;
>       f2=1000.125;
> Var   I:Integer;
>       Time:Double=0;
>       Signal1,Signal2,Signal3:Double;
>
> For I:=0 to (RunTime*SampleRate) do
> begin;
>   Time:=Time+1.0/SampleRate;
>   Signal1:=Sin(2.0*Pi*f1*Time);
>   Signal2:=1.0/10000.0*Sin(2.0*Pi*f2*Time);
>   Signal3:=Signal1+Signal2;
> end.
>
> I write this down in a Pascalish style of language but it is easily
> translated into any language of your choice. The thing modelled here are
> two signals coming from a double mixer system with slightly different
> frequencies. Typical cheap sound cards exhibit a crosstalk of -70 to -80
> dB between the two channels. So Signal2 represents what is leaking from
> channel 2 of the sound card into channel 1 of the soundcard and Signal3
> repesents what you really find in channel 1 due to the crosstalk. 
>
> Assumption is that the amplitude in channel 2 is close to "1" too.
> Signal3 may be handled online or stored in a file or whatever you like.
> Apply your algos to Signal3 and feed the results into an Tau-sigma
> diagram, say on a second to second base (Tau0=1). The better your Algo
> is the more you will see why I suggested this experiment.
>
> Even -100 dB of crosstalk does not get you completely rid of this nasty
> effect. You may even change the signals to integer entities and scale
> not to "1" but to 32767 (for 16 bit resolution) to study the influence
> of bit resolution of real world sound cards. After that experiment
> decide yourself whether a cheap sound card with 80 dB channel separation
> does the job.
>
> After having understood that crosstalk is a limiting factor in using a
> sound card for stability measurements I would like to draw your
> attention to a case that has nothing to do with crosstalk at all but is
> a fundamental property of ALL sampled systems with amplitude/time
> quantization. Using something like:
>
> Const RunTime=1000;
>       SampleRate=48000;
>       f1=1000.125;
> Var   I:Integer;
>       Time:Double=0;
>       Signal1:Double;
>
> For I:=0 to (RunTime*SampleRate) do
> begin;
>   Time:=Time+1.0/SampleRate;
>   Signal1:=Round(32767.0*Sin(2.0*Pi*f1*Time)/32767.0;
> end.
>
> generate yourself a SINGLE signal of odd frequency and 16 bit
> resolution. Apply your Algos to that and try to explain the small phase
> modulation (similar to the crosstalk case) that you will find at a
> repetition rate of 8 s. I know, the effect IS SMALL but after putting
> your results into a tau-sigma diagram you notice how nasty it is
> nevertheless.
>
> I am writing you this because the real problems of a method are
> sometimes to be found at places where one would not have expected them
> at all.
>
> Best regards
> Ulrich Bangert
>    
>
>   
Ulrich

If the crosstalk is sufficiently stable to cause a significant problem,
then in principle the crosstalk transfer function can be measured and
used to mathematically cancel the crosstalk before extracting the
desired beat frequency phases.

Bruce