Forums

lockin amplifier question

Started by jeremy December 4, 2007
hi
i am using a lockin amplifier for the 1st time other than a canned college  
lab demo
i have a 5mV signal from a strain gage that is driven with 10Vpp at 1kHz  
or whatever i choose.
the 5mV is increased to 5.01mV when i add some weight to the arm the
gage is weighing.
i wanted to increase sensitivity of this reading with the LIA since there  
is noise in the next digit, i.e.
i read 5.01mV +- 0.001mV.

But i seem to go out of range (overload) before reading any weight change.
do i need to get a 0.000V average signal first, and put this into
the LIA?

The 4 strain gages are in a full wheatstone bridge configuration.
The 10V 1kHz signal is sent to the bridge and to the LIA 'reference' input,
5mV 1kHz bridge output sent to to LIA signal in. I adjust the LIA phase  
for max output,
zero the signal with the internal LIA offset, and increase sensitivity  
until i overload.
then i back off the sensitivity so theres no overload. at this point i see  
no change in output
with change in weight. the LIA sensitivity is generally 30mV at this  
point, whereas
the signal I want to measure is 0.01mV or less, so it seems natural i wont  
get a change in
output. the only thing i can think of is i must 'balance the bridge' to  
get 0.00V output, and
amplify this to higher levels e.g. at the 0.01mV level.
is this right?

Finally, having read a primer that suggests using 'offset' and 'expand'  
for low-noise small-change signals
such as mine, do I need an LIA with 'expand', which seems to be an output  
amplifier (after the lockin stage)?
My old analog lockins have only offset and  input amplification  
(sensitivity), but no 'expand' function.

Thanks
JR
On Tue, 04 Dec 2007 15:14:46 +0200, jeremy <spam@spam.com> wrote:

>hi >i am using a lockin amplifier for the 1st time other than a canned college >lab demo >i have a 5mV signal from a strain gage that is driven with 10Vpp at 1kHz >or whatever i choose. >the 5mV is increased to 5.01mV when i add some weight to the arm the >gage is weighing. >i wanted to increase sensitivity of this reading with the LIA since there >is noise in the next digit, i.e. >i read 5.01mV +- 0.001mV. > >But i seem to go out of range (overload) before reading any weight change. >do i need to get a 0.000V average signal first, and put this into >the LIA? > >The 4 strain gages are in a full wheatstone bridge configuration. >The 10V 1kHz signal is sent to the bridge and to the LIA 'reference' input, >5mV 1kHz bridge output sent to to LIA signal in. I adjust the LIA phase >for max output, >zero the signal with the internal LIA offset, and increase sensitivity >until i overload. >then i back off the sensitivity so theres no overload. at this point i see >no change in output >with change in weight. the LIA sensitivity is generally 30mV at this >point, whereas >the signal I want to measure is 0.01mV or less, so it seems natural i wont >get a change in >output. the only thing i can think of is i must 'balance the bridge' to >get 0.00V output, and >amplify this to higher levels e.g. at the 0.01mV level. >is this right? > >Finally, having read a primer that suggests using 'offset' and 'expand' >for low-noise small-change signals >such as mine, do I need an LIA with 'expand', which seems to be an output >amplifier (after the lockin stage)? >My old analog lockins have only offset and input amplification >(sensitivity), but no 'expand' function. >
What happens when you change the lock-in offset? Depending on how that is applied, that may be all you need. But I'm not especially hopeful, since it probably just applies to the metered output... essentially a zero-adjust for the meter. You need gain after that, which is "expand". For those who haven't encountered lock-in amps before, what they do is essentially chop the input at the applied reference frequency, then amplify, low-pass filter, and meter it. (The "lock-in" part of the name is a red herring.) The reference frequency is derived from the same source that excites the sensor (or whatever). The chopping action essentially multiplies the input signal by the reference frequency. (Modern units do an actual multiply.) This produces a DC signal proportional to any input component at the reference frequency, while all the product sum-and-difference components are at higher frequencies that are easily removed via the low-pass filter. Lock-ins can give stupendous noise rejection if you set the filter bandwidth narrow enough, such that you can extract signals that are more than 100 dB below the noise. But the lock-in is handling the full signal range. If you want better resolution you must either balance the strain gage bridge first (as suggested) and increase the sensitivity, OR you *might* be able to use an external meter with more resolution. You might need to increase the time constant for that, to further reduce last-digit noise, assuming it is signal-related noise and not from the lock-in itself. I assume you have already tried increasing the TC to reduce last-digit noise in the lock-in's onw meter, but this is still worth a try with an external meter. If the lock-in offset adjust applies to the external output, then an external meter with no more digits than the lock-in can still be used by cranking up its sensitivity... essentially adding "expand". Best regards, Bob Masta DAQARTA v3.50 Data AcQuisition And Real-Time Analysis www.daqarta.com Scope, Spectrum, Spectrogram, FREE Signal Generator Science with your sound card!
"Bob Masta" <NoSpam@daqarta.com> wrote in message 
news:4756ab11.3122477@news.sysmatrix.net...

[snip]

> > For those who haven't encountered lock-in amps before, what > they do is essentially chop the input at the applied reference > frequency, then amplify, low-pass filter, and meter it. (The > "lock-in" part of the name is a red herring.) The reference frequency > is derived from the same source that excites the sensor (or whatever). > The chopping action essentially multiplies the input signal by the > reference frequency. (Modern units do an actual multiply.) > This produces a DC signal proportional to any input component at > the reference frequency, while all the product sum-and-difference > components are at higher frequencies that are easily removed via > the low-pass filter. Lock-ins can give stupendous noise rejection > if you set the filter bandwidth narrow enough, such that you can > extract signals that are more than 100 dB below the noise. >
[snip]
> > Best regards, > > > Bob Masta >
This multiplication to isolate an individual frequency is essentially how the Fourier transform works. What advantage do these "lockin" amps have, today, vs. doing the A->D conversion and then an FFT? Thanks, Bob. Bob
On Wed, 5 Dec 2007 07:47:47 -0800, "BobW"
<nimby_NEEDSPAM@roadrunner.com> wrote:

> >"Bob Masta" <NoSpam@daqarta.com> wrote in message >news:4756ab11.3122477@news.sysmatrix.net... > >[snip] > >> >> For those who haven't encountered lock-in amps before, what >> they do is essentially chop the input at the applied reference >> frequency, then amplify, low-pass filter, and meter it. (The >> "lock-in" part of the name is a red herring.) The reference frequency >> is derived from the same source that excites the sensor (or whatever). >> The chopping action essentially multiplies the input signal by the >> reference frequency. (Modern units do an actual multiply.) >> This produces a DC signal proportional to any input component at >> the reference frequency, while all the product sum-and-difference >> components are at higher frequencies that are easily removed via >> the low-pass filter. Lock-ins can give stupendous noise rejection >> if you set the filter bandwidth narrow enough, such that you can >> extract signals that are more than 100 dB below the noise. >> >[snip] >> >> Best regards, >> >> >> Bob Masta >> > >This multiplication to isolate an individual frequency is essentially how >the Fourier transform works. What advantage do these "lockin" amps have, >today, vs. doing the A->D conversion and then an FFT? >
Not much for many applications, and some modern lock-ins actually do use FFT methods. But notice that the reference frequency in a lock-in is exactly the frequency of interest. You can do that with an FFT, but it's only really practical if the same system is generating the test signal as well. Otherwise, you end up with response leakage "skirts" that cause energy to appear at adjacent frequencies. (Though window functions help this, they are never as good as the synchronous reference approach.) The lock-in typically uses a single-pole filter, the same as an FFT (effectively). The FFT's filter gets narrower by using a bigger sample set, which takes more computing power. (No big deal these days.) But in the lock-in all that's needed is to increase the R and/or C, so there is essentially no cost to cranking it way up. (Other than the fact that you have to wait for it to settle, just as the FFT requires that all N samples be acquired.) Until fairly recently, it was hard to get A/Ds with the specs to rival lock-in dynamic range. But with 24-bit converters now readily available that's not an issue any more. By the way, my earlier explanation was simplified in that it assumed that the reference and response were exactly in phase. That's rarely the case, so lock-ins have 2 channels set 90 degrees apart, plus magnitude/phase readouts. Otherwise, an input that was at the exact reference frequency but 90 degrees out of phase would give 0 from a single lock-in channel. There is exactly the same issue in FFTs, which do all computations in real and imaginary (cosine and sine) phases and combine the results to get magnitude and phase. Interested parties can investigate the benefits of synchronous reference frequencies using my Daqarta software. You don't need to buy anything, because the signal generator portion is free. (The signal inputs won't work after the trial period expires, but the generator will keep working indefinitely.) There is a built-in spectrum analyzer that can monitor the generator output. If you set any arbitrary generator frequency, you will clearly see the spectral "skirts" mentioned above, and as you change the frequency you will see the response change from a "tight skirt" when the frequency is nearly synchronous (ie same as one of the intrinsic spectral lines of the FFT), to a billowing skirt when it is halfway between lines. At that point the spectrum will show nearly half the energy at each of these adjacent lines, plus smaller amounts at higher and lower lines. Now go to the Frequency Step dialog (in the Tone Freq dialog) and set Step Lines to 1.000. When you change frequencies after this, they will only fall exactly on FFT spectral lines. The spectrum of a single sine will show a single peak at that frequency, with no skirts at all. Enjoy! Best regards, Best regards, Bob Masta DAQARTA v3.50 Data AcQuisition And Real-Time Analysis www.daqarta.com Scope, Spectrum, Spectrogram, FREE Signal Generator Science with your sound card!
"Bob Masta" <NoSpam@daqarta.com> wrote in message 
news:4757f9ad.2378055@news.sysmatrix.net...
> On Wed, 5 Dec 2007 07:47:47 -0800, "BobW" > <nimby_NEEDSPAM@roadrunner.com> wrote: > >> >>"Bob Masta" <NoSpam@daqarta.com> wrote in message >>news:4756ab11.3122477@news.sysmatrix.net... >> >>[snip] >> >>> >>> For those who haven't encountered lock-in amps before, what >>> they do is essentially chop the input at the applied reference >>> frequency, then amplify, low-pass filter, and meter it. (The >>> "lock-in" part of the name is a red herring.) The reference frequency >>> is derived from the same source that excites the sensor (or whatever). >>> The chopping action essentially multiplies the input signal by the >>> reference frequency. (Modern units do an actual multiply.) >>> This produces a DC signal proportional to any input component at >>> the reference frequency, while all the product sum-and-difference >>> components are at higher frequencies that are easily removed via >>> the low-pass filter. Lock-ins can give stupendous noise rejection >>> if you set the filter bandwidth narrow enough, such that you can >>> extract signals that are more than 100 dB below the noise. >>> >>[snip] >>> >>> Best regards, >>> >>> >>> Bob Masta >>> >> >>This multiplication to isolate an individual frequency is essentially how >>the Fourier transform works. What advantage do these "lockin" amps have, >>today, vs. doing the A->D conversion and then an FFT? >> > > Not much for many applications, and some modern lock-ins > actually do use FFT methods. > > But notice that the reference frequency in a lock-in is exactly the > frequency of interest. You can do that with an FFT, but it's > only really practical if the same system is generating the test > signal as well. Otherwise, you end up with response leakage > "skirts" that cause energy to appear at adjacent frequencies. > (Though window functions help this, they are never as good > as the synchronous reference approach.) > > The lock-in typically uses a single-pole filter, the same as an FFT > (effectively). The FFT's filter gets narrower by using a bigger > sample set, which takes more computing power. (No big deal > these days.) But in the lock-in all that's needed is to increase > the R and/or C, so there is essentially no cost to cranking it > way up. (Other than the fact that you have to wait for it to > settle, just as the FFT requires that all N samples be acquired.) > > Until fairly recently, it was hard to get A/Ds with the specs to rival > lock-in dynamic range. But with 24-bit converters now readily > available that's not an issue any more. > > By the way, my earlier explanation was simplified in that it > assumed that the reference and response were exactly in > phase. That's rarely the case, so lock-ins have 2 channels > set 90 degrees apart, plus magnitude/phase readouts. > Otherwise, an input that was at the exact reference frequency > but 90 degrees out of phase would give 0 from a single > lock-in channel. There is exactly the same issue in FFTs, > which do all computations in real and imaginary (cosine and > sine) phases and combine the results to get magnitude and phase. > > Interested parties can investigate the benefits of synchronous > reference frequencies using my Daqarta software. You don't need > to buy anything, because the signal generator portion is free. > (The signal inputs won't work after the trial period expires, but > the generator will keep working indefinitely.) > > There is a built-in spectrum analyzer that can monitor the generator > output. If you set any arbitrary generator frequency, you will > clearly see the spectral "skirts" mentioned above, and as you > change the frequency you will see the response change from > a "tight skirt" when the frequency is nearly synchronous (ie same as > one of the intrinsic spectral lines of the FFT), to a billowing skirt > when it is halfway between lines. At that point the spectrum will > show nearly half the energy at each of these adjacent lines, plus > smaller amounts at higher and lower lines. > > Now go to the Frequency Step dialog (in the Tone Freq dialog) > and set Step Lines to 1.000. When you change frequencies > after this, they will only fall exactly on FFT spectral lines. > The spectrum of a single sine will show a single peak at > that frequency, with no skirts at all. > > Enjoy! > > Best regards, > > Bob Masta >
Ahhh...yes. Knowing the target frequency ahead of time, and using synchronous filtering, makes it much more accurate (as compared with a general purpose FFT). Thanks for the great information, Bob. Bob
>> [snip] >>> >>>> >>>> For those who haven't encountered lock-in amps before, what >>>> they do is essentially chop the input at the applied reference >>>> frequency, then amplify, low-pass filter, and meter it. (The >>>> "lock-in" part of the name is a red herring.)
I think the idea of the name is that the device 'locks in' to the reference frequency, (even if its varying), massively rejecting any other frequency [to the 100db levels or more mentioned]
>>>> The reference frequency >>>> is derived from the same source that excites the sensor (or whatever). >>>> The chopping action essentially multiplies the input signal by the >>>> reference frequency. (Modern units do an actual multiply.)
Yeah, the old analog ones available here are also doing analog multiply. That multiply is the heart of the device if u ask me.
>>>> This produces a DC signal proportional to any input component at >>>> the reference frequency, while all the product sum-and-difference >>>> components are at higher frequencies that are easily removed via >>>> the low-pass filter. Lock-ins can give stupendous noise rejection >>>> if you set the filter bandwidth narrow enough, such that you can >>>> extract signals that are more than 100 dB below the noise. >>>> >>> This multiplication to isolate an individual frequency is essentially >>> how >>> the Fourier transform works. What advantage do these "lockin" amps >>> have, >>> today, vs. doing the A->D conversion and then an FFT? >>> >> >> Not much for many applications, and some modern lock-ins >> actually do use FFT methods.
I'd say the main advantage over FFT is again the noise rejection. A->D and FFT doesnt reject noise to the same extent - try to find a a 10microvolt sine signal buried in 1V broadband white noise using a power spectrum, I don't think you'll have too much success. Add in the fact that the reference signal may have some drift and you begin to appreciate the power of the thing.
>> Until fairly recently, it was hard to get A/Ds with the specs to rival >> lock-in dynamic range. But with 24-bit converters now readily >> available that's not an issue any more.
yes although the settling time of a 24bit AD gets up there, so analyzing at e.g. 100kHz may still be easier with an old analog lockin. Anyways I haven't had access to my system for a while but am still a little stumped that a good dvm would see a signal I can't pick up with the lockin, making me think I must have something set wrong. BTW I saw a site mentioning you can make a noise-frequency map with a lockin (like a spectrum analyzer) , does anyone know how?