Homebrew Lock-In Amplifier

[Picuino]

Some time ago I bought the necessary chips to assemble a homebrew lock-in amplifier and I would like to assemble it and experiment with it.
Has anyone assembled the lock-in amplifier based on the AD630 that is circulating on the internet?
With which PCB can it be assembled with guarantees that it will work well?
Has anyone experimented with it and what interesting experiments can be done?
I know there are a lot of questions, but they can be summed up in just one. Can anyone tell us about your experiences with this instrument?


Homebrew Lock-In amplifier:
https://www.holographyforum.org/data/pdf/aa-Collection_a_k/aa-Laser/aa-lockin/Homebrew_lockin_amplifier.html
https://www.instructables.com/Lock-in-Amplifier/

More professional version:
https://physicsopenlab.org/2019/08/20/lock-in-amplifier/
https://www.ebay.com/itm/275596051897?epid=2282365081


What is a Lock-In Amplifier:
https://en.wikipedia.org/wiki/Lock-in_amplifier
https://www.zhinst.com/en/resources/principles-of-lock-in-detection
https://www.thinksrs.com/downloads/pdfs/applicationnotes/AboutLIAs.pdf
https://www.liquidinstruments.com/digital-lock-in-amplifier/

[Kleinstein]

I have no experiance with this plan / kit, but I have quite a bit experiance with using  lock-in amplifiers and I have build my own version of a lock-in amplifier. My version looks quite different though and has plenty mistakes to learn from.

There are a few issue with this plan:

1) The outputs are directly from the amplfiers. To avoid capacitive loading to the amplifiers it is a good idea to have some 50 or 100 ohm in series to the outputs.
2) The usualy way is to used the output DC coupled. So the capacitor for the output is a thing to drop - a good place for one of the 50 ohm resistors.
3) The amplifier is rather basic. The higher gain setting of more than about x 1000 have a limited bandwidth. This may be just a little much gain for one stage. It limits the use to relatively low frequencies (e.g. 1 kHZ or less). This can be OK for some uses, but could be an issue for others.  In many cases one may have a custom pre-amplifier for the signal source anyway. A protected power source may be substitude a better amplifier to some degree.
4) a point missing are a way to detect clipping, like 2 comparators or simple diode / capacitor detectors for the peak values. Not absolutely needed but a nice to have point.
5) Another usefull and commonly found part with the input is a mains frequency notch filter, possibly also the 2x mains.
6) The output filter has settings for rather low time constants - these make little sense with an amplifier that is limited BW. It is more that a longer time scale could make sense, though today a medium time scale like 100 ms and than digital averaging low pass filtering at the putput would be a good way. So the longer time constants may not be that important. Using electrolytic capacitors with the filter is not ideal. At least the 100 ms range should get way without electrolytic capacitors.
7) depending on how the output is read / displayed one may want more output gain as on option (e.g. with a signal with low SNR)
The reference input is directly to the chip - this is kind of a start to add a reference section there. The reference section ideally includes quite some circuitry, so a start, but yet quite ready. One usually wants a bit of protection and signal forming, so that one can start with a crude signal and still get a reasonable sine or optionally square wave with defiened amplitude. Ideally one would also get a fine phase adjust with a precise 90 deg. jump option (e.g. via a PLL). The 90 deg. phase shift part is very handy to adjust the phase. Quite often the ref. side also include some simple generator option (e.g. comes easy with a PLL or so).

Old style analog lockin amplifiers are to a large part replaced with digital solutions. So digitize the signal with enough dynamic range (e.g. a sound card or similar) and than do much of the LI technique digitally.

For the experiments that can be done, there are a lot of optics experiments (e.g. look at absorbtion or reflection). This can be with modulated LEDs as a light source or a mechical chopper.
Another nice experiment can be a field mill to measure electic fields. Utrasonic , acoustics is possible to, though it may need a higher BW amplifier to not get too much phase shift from there.
LVDT mechanical displacement sensor can be read out with a lock-in. Similar DMS can use a carrier frequency amplifier, which is a lockin pluse oscillartor for the reference.
A LCR bridge could use a lock in amplifier - though this really needs the phase shift part.
I have done photo-acoustic detection of light: so use a microphone to detect the pressure rise from modulated light to heat up some gas (e.g. water wapor to absorb 950 nm) or solid surface.

[Picuino]

Thank you for your advice.

I was also thinking about making a LIA with a digital processor, but I need first to have a reference analog LIA to check that the digital one works well as I program it.

I will make the preamp with 2 separate stages. One stage with 4 options (x1, x2, x5 and x10) and another stage that can have 2 options (x1 and x10) so I can choose many options between x1 and x100 with simpler knobs and it won't lower the bandwidth too much. If I need more amplification, it can be done with an external preamp.

Another advice I am missing is about the PCB routing, which is not shown in the tutorial and I think it can be important in such a sensitive instrument. How do I connect the grounds? Is a ground plane enough or do I have to wire all the lines to a single point with separate routes?

[Kleinstein]

With very high gain and precision in mind it is usually better to have a single ground point (star ground). A ground plane is more thing to avoid EMI and good for a digital circuit.
The point with a lock-in amplifier is to keep the input and reference part separate, so avoid the ref. side to cause any input signal with the same frequency. In the plan there is not yet much for the ref. side. For quite some applications one would want a phase adjustment there.
One would want at least some protection.

If one needs fine steps with the gain depends on what is used to read / display the output: with an analog meter one may need 1, 2,5,10 steps for the gain. With a digital read out with sufficient resolution steps of 1 , 10, 100, 1000 are enough. Only a gain range from 1 to 100 may be a bit small. 
Depending on the precision of the amplifier at the input one may want a 2nd AC coupling step before the phase sensitive detection.
The display / output side also determines how much gain is useful at the output (DC) side. With a rel. low resolution ADC or an analog movement one can use quite some DC gain there. Chances are the AD630 output can be stable to something like 10 µV. So the output side should ideally be able to resole to that level. If a good dmm is used the range settings there can provider some of the total gain.

[RoGeorge]

In case you don't have the time to build a dedicated LIA, this oscilloscope trick might be a good-enough replacement:
https://www.eevblog.com/forum/projects/oscilloscope-with-trace-averaging-as-a-lock-in-amplifier-(rigol-ds1054z)/

[Picuino]

I don't really have much time right now. I am preparing the project for this summer when I will have enough time to assemble and test it.

I want to make a real LIA, which is fast enough to decode sounds in real time (10kHz minimum) from a higher frequency signal.

I don't care if it has a lot of quality, just enough for basic experiments to work. That's why I'm considering making the LIA with cheaper, simpler powered components (amplifiers). The problem is that I can't replace the AD630 with a simpler power supply. Sometimes I consider doing a digital LIA directly, but then I would not have any standard to compare.

[Kleinstein]

One can replace the AD630 with a CMOS swich (e.g. DG419) a few resistors and OP-amp.

For simple testing one should include some simple generator that also generate a quadrature signal. Many of the experents may wand a ref signal from a generator and it easier to start with 2 or 4 x the frequency and than make it an accurate 50:50 signal and get a quadrature signal than to generate this from a PLL. A simple µC to generatore the ref. signal and a source drive signal could be a good idea, as it allows to get at least some phase shift rather easy.

Parts of the design depend on how one looks at the output / result. E.g. when using an ADC there anyway, one would not really need the very long time constants at the output. The resolution at the output recording also determines how many gain steps and outout gain is useful. HIgh resolution there can substiture some gain steps.

[RoGeorge]

A LIA only works if you have a strong and clean reference signal, and in sync with the small and noisy signal to be measure, and only works at a constant frequency, or with very slow variations.  A LIA does not decode an unknown signal, it only averages the signal by a pattern given by the reference signal.

In regards to the max frequency, a classic LIA usually goes up to 100kHz or so, but they are very sensitive and very low noise in their analog input stage.  In contrast with that, the oscilloscope method is less sensitive, more noisy, but can work at frequencies as high as the oscilloscope can display, so virtually hundreds of MHz.  Which one to use depends of the measurement that needs to be done.

for basic experiments

For what kind of experiments do you need the LIA, what do you plan to measure with it?

[Picuino]

I want start measuring several experiments:
 * Milliohms in pcb traces
 * Coupling between different twisted pairs of cables.
 * Voice with a laser in a crystal. (Microphone)

[Terry Bites]

I agree with Kleinstein, Analog switches are great for low cost mod demod.  The switch method gives the same result as mutiplying the input with a squarewave. Plenty of examples in AD and LT app notes.

[shabaz]

This won't meet the OP's needs, but it might be of slight interest to beginners: I created an educational project. I was hoping people could create new experiments with it to improve on my simple ones, but there's no traction (at least, I'm not aware of anyone trying it out).

The project is called Wave Miner (Github link) and has 10 starter experiments so far.

It uses a DSP board from AliExpress and a simple breakout board to BNC connectors (the photo shows it attached to a Raspberry Pi for programming the DSP; the Pi can be detached afterward).

The Gerber files for that breakout board are at that Github link. From the 10 starter experiments, experiment #9 is a very crude lock-in amplifier.

Here is a video demo of the LIA  (before I constructed the breakout board).


The experiments are very basic and for fun/learning rather than serious use, and the DSP used is extremely limited, but on the plus side, very little construction is needed (the breakout board is entirely through-hole construction for simplicity).

[Picuino]

One can replace the AD630 with a CMOS swich (e.g. DG419) a few resistors and OP-amp.

For simple testing one should include some simple generator that also generate a quadrature signal. Many of the experents may wand a ref signal from a generator and it easier to start with 2 or 4 x the frequency and than make it an accurate 50:50 signal and get a quadrature signal than to generate this from a PLL. A simple µC to generatore the ref. signal and a source drive signal could be a good idea, as it allows to get at least some phase shift rather easy.

Parts of the design depend on how one looks at the output / result. E.g. when using an ADC there anyway, one would not really need the very long time constants at the output. The resolution at the output recording also determines how many gain steps and outout gain is useful. HIgh resolution there can substiture some gain steps.

So, if I generate a square signal from a microcontroller and with the same microcontroller take samples with the ADC synchronized with the square signal, I could make a simple Lock-In Amplifier without multiplying the analog signal. I could do that with a simple microcontroller as a proof of concept.

The microcontroller should internally perform the operation of adding or subtracting the value read by the ADC depending on the state of the output signal.

[Picuino]

I am trying to set up an STM32 board (NUCLEO-L412KB) for testing the digital LIA. If I don't succeed, I already have other boards running with 16-bit PICs. They are slower, but I think they will also work without problems and can be powered at 5V (an advantage).
Ultimately I can buy an Arduino UNO R4 and do the project with it. That way it would be much more replicable by others who want to take advantage of the project to assemble it themselves.

[Picuino]

I also have a microchip MCP6G02 circuit (dual programmable R-R amplifier x1, x10 and x50) to make the input stage.

It has much worse quality than the AD620, but it is much easier to start testing. It has a wide range of programmable gains from the microcontroller.

https://ww1.microchip.com/downloads/aemDocuments/documents/OTH/ProductDocuments/DataSheets/22004b.pdf

[Kleinstein]

The MCP6G02 is not really a good choice for an input amplifier. The noise is quite high - so if at all only for later stages.  In addition it only is for a 5 V supply and thus has a limited dynamic range.

The amplifier part for the input would be similar for an analog an digital solution.

[RoGeorge]

So, if I generate a square signal from a microcontroller and with the same microcontroller take samples with the ADC synchronized with the square signal, I could make a simple Lock-In Amplifier without multiplying the analog signal.

It is not possible to take samples without multiplying.  Sampling is multiplying, just that the multiplying factors only take the values 0 and 1, and yes, the setup you proposed will work. 

The classic LIA uses analog multipliers because back then there was no other option.  Cheap and fast ADC were not yet available, resolution was lame, there were no fast enough CPUs and no DSPs to compete with analog.  Even today, it is hard to achieve digitally the same specs as a good analog LIA.

Where a dedicated LIA shines is in the analog performance:  expensive ones are very carefully designed, their performance is close to the theoretical limits, with very low noise, very high dynamic range, very good separation between signals and between channels, they often have a low jitter PLL to generate accurate quadrature signals, etc.

[Kleinstein]

The theory for lock in amplifiers often assumes an analog multiplication with a sine wave. However the actual implementation often uses the +-1 square wave multiplication (like the AD630 or with CMOS switches). A problem is that the analog multiplication (e.g. AD633) tends to be drifty and not very stable.  One can limit the response to the sine part with an extra bandpass filter before the multiplier.
Working around the multiplier drift is tricky, e.g. with an additional feedback of a compensating square wave.

It is already quite some time (starting around 2000) that good digital lock-in amplifiers are superior to analog ones in most aspects, especially at not so high a frequency.
One can get really low noise, close to the limits with both versions. Even with a primarily analog LI it makes sense to do at least the final integration / low pass filtering at least partially digital. An analog low pass filter has an anoying long settling time, while digital averaging or a FIR low pass fitler is easy and settling faster.

For a digital lockin, the way that DSOs implement the averaging mode is a good starting point. Boxcar averaging looks different from classic lockin function, but one can still start with the waveform averaging. The the lockin does the multipication with the ref. waveform (e.g. a sine or square wave) first and than averages / integrates. The boxcar method first averages the data point by point in the time domain and to get the lockin result one multiplies only at the end after averaging.  From the math side the result is the same - one can swap the squence. The boxcar method needs more memory and fewer multiplications. In addition one could provide additional information from the waveform or fast look at different harmonics or afterwards apply a different test function besides the classic sine and square.

The ADC for a digital lockin amplifier should be relatively good, so that one has a reasonable dynamic range. The µC internal ADCs of often 12 bit or less would limit the dynamic range and thus the ability to detect really small signals. So this is more something for a fast 16-24 bit ADC, maybe audio type ADCs.

[mawyatt]

There is no known better signal distinguishing factor that Bi-Phase Modulation (multiplying by +- 1 or BPSK). Any other type of basic waveform modulation is inferior to BPSK when it comes to signal recovery from noise within coherent systems, and why it's chosen for deep space communications such a Voyager.

With this in mind it should be no surprise that noise buried signal recovery utilizing Synchronous Sampling or Demodulation by means of Bi-Phase demodulation is the best form of signal recovery and why we utilized such in many of our applications dated back over half a century ago.

Honestly, would be surprised if a Lock-In amplifier based upon Sine-Wave Demodulation would outperform one based upon Bi-Phase Demodulation. If so this might point to an inferior Bi-Phase Demodulation design rather than the fundamentally superior sine-wave approach.

Intuitively if one considers a mixing or multiplication process, flipping the waveform in polarity (+- 1 multiplication) has less of the multiplying waveform unwanted artifacts than a sinusoid multiplicand such as close in phase noise.

For example it's well known that for a low noise heterodyne systems with the usual means of a diode based mixer for down conversion, when the diode mixer is heavily driven produces the lowest noise result, and the heavily driven approaching a Bi-Phase multiplication by the LO. This not only improves the mixer conversion loss but also reduces the amplitude uncertainty or noise from the driving signal (LO) further improving the result.

So one might suspect the same for utilizing a pure Bi-Phase vs. Sinusoid multiplicand for Synchronous Sampling or Demodulation as in a Lock-In amplifier application.

Anyway, just some random thoughts on such.

Best,     

[RoGeorge]

I was wandering if using a poly-phase mixer could improve a LIA, because of the reduced noise factor specific to poly phase mixers. 

Should be easy to double the frequency of the reference signal (in order to divide by 2 again, while producing precise quadrature I/Q of the same frequency as the reference) then to use the I/Q to drive the switches of the poly-phase mixer.


Side note now that the quadrature of the reference signal was mentioned, the schematic of the kit in the first post is not exactly a LIA, in my opinion that schematic is rather a synchronous detector.  What I'm trying to say is, a LIA (for me) has to have 2 mixers, one for each of the I/Q components of the quadrature reference, in order to extract both the amplitude and the phase of the fain signal.

It's 2:30 AM here and didn't check against the datasheets, but the schematic of the kit seems to have only a single analog multiplier, and no I/Q splitting of the reference signal.  Though, sometimes the architecture with 2 mixers is called a two-phase lock-in amplifier.  Well, I've always thought a LIA should be able to tell both the phase and the amplitude of the small signal.  If it doesn't have quadrature reference and 2 multipliers, I think of it as a synchronous detector (and I don't call it a LIA).  But I'm not sure if I use the proper naming.

[mawyatt]

The PPM will naturally create I and Q outputs. The 8 phase version seems to be the "sweet spot" wrt to performance vs complexity.

Best,

[Kleinstein]

A basic lock in amplifier has 1 phase only, but there are better models to have 2 phases in quadrature. In some applications this can help quite a bit, while for others (e.g. optics) it is not really helping (only faster settling to check the quadrature part). So far I have not seen an analog build lock in that uses more than 2 phases.
The extra effort for an extra quadrature channel is not that high, if one has a PLL for the reference section anyway. A PLL for the reference section make absolute sense to get stable phase shifts, especially the 90 deg. steps. Much of the effort is in the input amplifier, filters and ref. side PLL / phase shift, not so much the actual phase sensitive detector.

It depends on the application if a sine demodulation makes sense or the simpler +-1 case is better.  In some cases if one looks more at phase shifts or frequency dependent effects one does not want the contribution from the harmonics. Its not about noise, but avoiding side effects / systematic errors that complicate the interpretation of the data.
The point is fitlering out the harmonics, not how the actual demodulator is build.
When it comes to noise the best demodulation function corresponds to the signal waveform, or at least the stable part of it.
With optics experiments one may well have a more square modulation can also use the power in the harmonics, tough if there is jitter (e.g. from a mechanical chopper) it could help to exclude some of the transition region.
For a digital implementation one has anyway the option to use the boxcar like averaging first and than use a suitable demodulation waveform, that may be neither sine not square suitable for the experiment. One could also still save the full (or somewhat filtered) averaged waveform and decide even later which way to look at the data.
Especially digital (but also a few analog ones) lock-in amplifiers also allow to look at the harmonics separately (sepcially 2 x the frequency, 3 x the frequency can have artifacts from the demodulator).

[Picuino]

The MCP6G02 is not really a good choice for an input amplifier. The noise is quite high - so if at all only for later stages.  In addition it only is for a 5 V supply and thus has a limited dynamic range.

The amplifier part for the input would be similar for an analog an digital solution.

Ok. I have ordered an AD8422 amplifier that can be used with 5V single supply. I am going to use it along with a microcontroller to make a milliohm meter as a starter project.

https://www.analog.com/media/en/technical-documentation/data-sheets/ad8422.pdf

[RoGeorge]

Also from Analog Devices:  AN-306, Synchronous System Measures \$\mu\$\$\Omega\$s
https://www.analog.com/media/en/technical-documentation/application-notes/AN-306.pdf

[zrq]

I have exactly the same thought.
But sine wave may still have advantages for systems with nonlinearity, the modern DSP lockins usually support demodulation at an arbitrary harmonic frequency, although myself only used second harmonic for peak locking which should also work with a square wave modulation.
BTW sometimes in physics experiments we also do boxcar averaging for low duty cycle signals.

For the original post, working with sound cards is probably a good idea for low frequency experiments. Inexpensive sound cards can have >120 dB dynamic range, probably even more if you don't care about harmonics and spurs for lock-in detection, which is in priniple comparable with todays' best lock-in amplifers from SRS or ZI, not to mention the older analog ones. Although not sure if channel cross-talk can be limiting. I was thinking about playing with my Focusrite but couldn't get the time...

Also although I have a very bad feeling towards red pitaya, they also works as a DSP LIA for higher frequency. Koheron also sells a interesting Zynq based board which my colleagues built laser phase noise characterizations systems around.

[gnuarm]

I have done photo-acoustic detection of light: so use a microphone to detect the pressure rise from modulated light to heat up some gas (e.g. water wapor to absorb 950 nm) or solid surface.

So, what were you using this for?

Many years ago, I did a bit of work with photo-acoustic signal collection.  We were using it to obtain spectra from samples that could not be put into solution so easily.  We used a rotating disc with slots cut evenly, to produce a modulated light beam.  Then the wavelength of the light was varied to produce a scan.  The AC component of the signal gave us the amplitude of the absorbance of the sample. 

Is this anything like what you were doing?  Back in '75, this was pretty virgin territory.