improving sweep measurements (long)

I measure a high Q resonance circuit with sweep generator.
(transfer function re/im to be exact)
There is a considerable delay until the output stabilizes.
This makes the measurement quite slow.

If I could use some multifrequency signal like MLS I would need
only the time that impulse response decays to neglible level.

Is there any way to improve sweep measurement speed or to get rid
of the unnecessary delay ?

The hardware is fixed :
Two DDS's generates the stimulus and mixer signal (sine and cosine)
So I have at my disposal intermediate quadrature signal which is then 
digitized.

The limitations are :
I can only change amplitude and frequency.
Changing these is in sync with IF sampling.
In other words I can reprogram DDS's synchronously to sampling.

So finally the question is: is there anything I can do to the speed up
and remove the sweep lag problem ?

With these limitations I have only one idea.
Sinc amplitude modulation of the stimulus carrier.
This should (I think) provide a box shaped spectra.
Mixing it down to zero frequency and using xcorrelation
with stimulus would do something.

----- finally an additional story ------

When experimenting  I tried to use MLS.
With hardware that made it possible.

Operating at > 100 MHz this is basically what I did.

a) Create a mls sequence.
b) band limit it
    Because of periodic nature of mls I think I can just fft it
    and zero out extra bins and ifft. ???
c) modulate carrier with it.

This would provide carrier with twosided sidelobes with
constant spectrum around carrier.

d) Mix down with analog mixer and then digitally mix
down to zero frequency.

The digital mixer can do quadrature mixing because of sync
with generator

e) extract one sideband , positive or negative.
d) use standard MLS xcorrelation to retrieve impulse response.

Effectively I transformed MLS to band pass and restored response back to 
baseband.

It worked but due to slow sampling rate we could not achive
that wideband response we needed.
Othervise seemed to be a working approach.

JR

JR <user@example.net> wrote in message news:<cc4gup$hek$1@phys-news1.kolumbus.fi>...
> I measure a high Q resonance circuit with sweep generator.
> (transfer function re/im to be exact)
> There is a considerable delay until the output stabilizes.
> This makes the measurement quite slow.
> 
> If I could use some multifrequency signal like MLS I would need
> only the time that impulse response decays to neglible level.
>

You seem to understand this quite well. One thing I would point 
out is that in using a MLS stimulus, you are effectively driving
the Device Under Test with a set of M (random phase) sinusoids
spaced at the MLS bit rate. If the system is linear, then 
superposition applies and you obtain the system response in 1/Mth
the time it would take to apply those sinusoids individually. 
But if the DUT (or something else in the signal path) isn't
completely linear, then you get intermodulation between those
sinusoids, which corrupts the result. 

You could decide how serious this is by filtering the MLS with
a notch filter, and looking to see how much the notch fills in
when you try your measurement.

Consider also that a high-Q filter stores a lot of energy, and
may therefore clip much sooner than the rest of your system. This
may possibly limit the drive level, and thus the S/N. You can get
some S/N back by averaging, but that comes at the expense of 
measurement time. 

<snip!> 

> With these limitations I have only one idea.
> Sinc amplitude modulation of the stimulus carrier.
> This should (I think) provide a box shaped spectra.
> Mixing it down to zero frequency and using xcorrelation
> with stimulus would do something.

Also note that you may be able to subsample instead of mix,
if that is more convenient. You can subsample a MLS at any
ratio that is relatively prime with M. The result is another
MLS (which you can find by simulation). Doing this causes
noise aliasing of course; you'll have to decide if you can
stand the higher noise floor.

> When experimenting  I tried to use MLS.
> With hardware that made it possible.
> 
> Operating at > 100 MHz this is basically what I did.
> 
> a) Create a mls sequence.
> b) band limit it
>     Because of periodic nature of mls I think I can just fft it
>     and zero out extra bins and ifft. ???

I worry about this a little. My guess is you'd be well-advised to
use a DFT of the same length as the MLS if you're depending on 
periodicity to hide the sloppy impulse response of your prefilter.
That's a non-power-of-two DFT, but you could do it in MATLAB, since
it doesn't happen in real time. 

> c) modulate carrier with it.
> 
> This would provide carrier with twosided sidelobes with
> constant spectrum around carrier.

Can you pick your MLS so there's an image around carrier
where you want it? Maybe a (reasonably wide) BP filter would be
more linear than a mixer?

> d) Mix down with analog mixer and then digitally mix
> down to zero frequency.
> 
> The digital mixer can do quadrature mixing because of sync
> with generator
> 
> e) extract one sideband , positive or negative.
>
>
> d) use standard MLS xcorrelation to retrieve impulse response.
> 

The last time I tried something like this, I had to do MLS 
crosscorrelation on both quadrature phases and then compute
the desired response using both components. Help me understand
how you get away without doing this.

> Effectively I transformed MLS to band pass and restored response back to 
> baseband.
> 
> It worked but due to slow sampling rate we could not achive
> that wideband response we needed.
> Othervise seemed to be a working approach.

I don't understand what you mean by slow sampling rate. Is it the
sampling rate of the baseband ADC that is too slow, or are you having
trouble with jitter in your MLS modulation?

So far, we've only discussed using MLS for AM modulation. Have you 
considered doing BPSK instead?

David L. Rick
Hach Company
www.hach.com

replies to: davidDOTrickAThachDOTcom