Precision synchronous demodulator

[Spehro Pefhany]

What's a good approach for use at 1MHz, give or take 2:1?

Things like the AD630 are slow, and my usualy go-to approach (Gilbert
Cell balanced demodulator with LO >> Vt) has input referred drift and
offset typically in the ~1uV/K and a couple mV. I'd like both to be
better by at least an order of magnitude, and preferably with some
gain like the Gilbert cell things, which have 15-20dB of gain.

Analog switches have negligible offset- how much trouble will all that
charge injection (~4pC for a good one) cause at 1-2MHz?


Best regards,
Spehro Pefhany

 

[John K]

What's a good approach for use at 1MHz, give or take 2:1?

Things like the AD630 are slow, and my usualy go-to approach (Gilbert
Cell balanced demodulator with LO >> Vt) has input referred drift and
offset typically in the ~1uV/K and a couple mV. I'd like both to be
better by at least an order of magnitude, and preferably with some
gain like the Gilbert cell things, which have 15-20dB of gain.

Analog switches have negligible offset- how much trouble will all that
charge injection (~4pC for a good one) cause at 1-2MHz?


Best regards,
Spehro Pefhany

How about a AD8551? 1uV offset and 5nV/C. $1.84 at Arrow:

http://octopart.com/partsearch#search/requestData&q=AD8551

Get the signal up out of the noise then use any analog switch to suit.

JK

 

[John K]

How about a AD8551? 1uV offset and 5nV/C. $1.84 at Arrow:

http://octopart.com/partsearch#search/requestData&q=AD8551

Get the signal up out of the noise then use any analog switch to suit.

JK

Oops - thought you were interested in DC. Same principle though. If you
are demodulating a weak 1MHz signal, use a low noise amplifier to boost
it, then chop.

OTOH, the Tayloe demodulator might be interesting. Using a LT1115
preamplifier at the output, it can get down to 17.3 nV or -142.1 dbm for
3dB SNR in 1KHz bandwidth:

http://www.norcalqrp.org/files/tayloe_mixer_x3a.pdf

I don't know how long until his patents expire, but you might ask. Since
he has promoted it so widely in ham use, he might be willing to let you
use it for a nominal fee.

Another alternative is the H-Mode mixer, invented by Colin Horrabin,
G3SBI. Martein Bakker, PA3AKE, has done extensive research on it. The
site is

http://martein.home.xs4all.nl/pa3ake/hmode/

JK

 

[Spehro Pefhany]

How about a AD8551? 1uV offset and 5nV/C. $1.84 at Arrow:

http://octopart.com/partsearch#search/requestData&q=AD8551

Get the signal up out of the noise then use any analog switch to suit.

JK

Hi, John:-

The demodulator (and following LPF) will eliminate low-frequency noise
(including DC offset). We'll be operating well above the 1/f corner on
the power spectral density plot of the amplifiers, so amplifier TCVos
and Vos really don't matter until you run out of headroom.

The purpose of the demodulation is to accurately recover a very small
signal buried deep in noise. Think old-school analog lock-in amplifier
(no modern ADC->DSP/FPGA digital signal processing) with a relatively
large 'dynamic reserve'. Before demodulation we can only amplify
signal+noise, and you can't let that clip or you lose the signal (and
it shouldn't distort too much or accuracy is lost).


Best regards,
Spehro Pefhany

 

[Adrian Tuddenham]

What's a good approach for use at 1MHz, give or take 2:1?

Things like the AD630 are slow, and my usualy go-to approach (Gilbert
Cell balanced demodulator with LO >> Vt) has input referred drift and
offset typically in the ~1uV/K and a couple mV. I'd like both to be
better by at least an order of magnitude, and preferably with some
gain like the Gilbert cell things, which have 15-20dB of gain.

Analog switches have negligible offset- how much trouble will all that
charge injection (~4pC for a good one) cause at 1-2MHz?

Has anyone tried using a fully-balanced system to null the charge
injection?

 

[Jasen Betts]

Has anyone tried using a fully-balanced system to null the charge
injection?

You mean like a diode ring modulator?

 

[John K]

Hi, John:-

The demodulator (and following LPF) will eliminate low-frequency noise
(including DC offset). We'll be operating well above the 1/f corner on
the power spectral density plot of the amplifiers, so amplifier TCVos
and Vos really don't matter until you run out of headroom.

The purpose of the demodulation is to accurately recover a very small
signal buried deep in noise. Think old-school analog lock-in amplifier
(no modern ADC->DSP/FPGA digital signal processing) with a relatively
large 'dynamic reserve'. Before demodulation we can only amplify
signal+noise, and you can't let that clip or you lose the signal (and
it shouldn't distort too much or accuracy is lost).

Best regards,
Spehro Pefhany

Yes, I guessed you were interested in detecting a weak signal, so I
appended some info on the Tayloe and H-Mode mixers. These have pretty
good MDS (minimum discernible signal) and excellent dynamic range,
perhaps 120dB or so. They are limited by the noise floor and large signal
handling of the following amplifier stages. Charge injection does not
seem to be a problem with these approaches.

Your application is quite similar to sensitive receivers, except you
probably don't have strong adjacent signals to cause intermodulation. But
you will be sensitive to phase noise from the local oscillator which may
also limit the performance.

Another thing that may help is to use a transformer at the input to boost
the signal before detection. You are apparently interested in a narrow
band of frequencies, so a tuned input might even work.

This is one of the most interesting design challenges - detecting a weak
signal that can have a wide dynamic range.

JK

 

[Spehro Pefhany]

Yes, I guessed you were interested in detecting a weak signal, so I
appended some info on the Tayloe and H-Mode mixers. These have pretty
good MDS (minimum discernible signal) and excellent dynamic range,
perhaps 120dB or so. They are limited by the noise floor and large signal
handling of the following amplifier stages. Charge injection does not
seem to be a problem with these approaches.

Yes, it's interesting to see that circuit-- thanks very much for the
link. I was thinking of something along those lines with a SPDT switch
and a zero-drift instrumentation amplifier at the output.

Your application is quite similar to sensitive receivers, except you
probably don't have strong adjacent signals to cause intermodulation. But
you will be sensitive to phase noise from the local oscillator which may
also limit the performance.

Crystal oscillator, so I don't think so.

Another thing that may help is to use a transformer at the input to boost
the signal before detection. You are apparently interested in a narrow
band of frequencies, so a tuned input might even work.

Got all of that, and more. The input BPF can't be too narrow bw, for
obvious reasons.

This is one of the most interesting design challenges - detecting a weak
signal that can have a wide dynamic range.

JK

"May you live in interesting times" is allegely a Chinese curse, but
apparently there's no evidence of that source, according to Wikip*dia.
They do list this one "May you come to the attention of important
people", which is suitably ominous.

 

[John K]

Crystal oscillator, so I don't think so.

A cheap crystal oscillator can have terrible phase noise at 1MHz.
Low phase noise is an art. A good crystal can cost an arm and a leg.

You need to know the amplitude of the noise in nanovolts per root Hz at the
operating frequency, then compare that to the phase noise from the crystal
at the same frequency. If the crystal noise is equal to or higher, you may
need to find a better oscillator.

JK

 

[Bill Sloman]

Yes, I guessed you were interested in detecting a weak signal, so I
appended some info on the Tayloe and H-Mode mixers. These have pretty
good MDS (minimum discernible signal) and excellent dynamic range,
perhaps 120dB or so. They are limited by the noise floor and large signal
handling of the following amplifier stages. Charge injection does not
seem to be a problem with these approaches.

Your application is quite similar to sensitive receivers, except you
probably don't have strong adjacent signals to cause intermodulation. But
you will be sensitive to phase noise from the local oscillator which may
also limit the performance.

Another thing that may help is to use a transformer at the input to boost
the signal before detection. You are apparently interested in a narrow
band of frequencies, so a tuned input might even work.

Tuned inputs can be a problem before a phase-sensitive detector. If the phase-shift through the input stage changes. so does the output from the phasesensitive detector.

Larsen N T 1968 Rev. Sci. Instrum. 39 1–12 used a band-pass filter beforehis phase sensitive detector, but added an all-pass network to cancel any phase shifts through the filter. National Bureau of Standards employees canbe like that.

http://ieeexplore.ieee.org/xpl/logi...re.ieee.org/xpls/abs_all.jsp?arnumber=5040488

 

[Adrian Tuddenham]

You mean like a diode ring modulator?

I was thinking of doing something with CMOS where the signal is balanced
and any charge injection on one leg is precisely balanced by charge
injected into the other leg. The shorter transition time of CMOS
switching should give lower noise than a diode ring modulator and there
is no chance of unbalaced D.C. injection from the control waveform into
the signal.

My particular application could be in modulating audio with a non
sinusiodal waveform by pulse-width modulation, but the charge-balancing
principle should work equally well at R.F.

 

[Spehro Pefhany]

Can you bandpass filter some first? That would directly take burden off the
synchronous detector.

Yes, I have a 2nd order BPF. Q is not very high. An all-pass such as
Bill suggested might make sense in this case.

There are cmos switches rated for below 1 pC injection. And they have a
common-mode voltage sweet spot, where injection crosses through zero. You can
tweak the power supplies so's to operate there.

That's an interesting technique.

There should be some clever dual-path sync demod architecture that cancels most
charge injection offset errors. Build two identical detectors and feed them
antiphase signals and take the difference, something like that.

I'm thinking the glitches will only cause major troubles if there is
nonlinearity.

I guess I could make a fully-differential output amplifier (or use an
ADC driver chip) to keep the signals closely antiphase. Or (horrors)
use a little RF tranformer with a grounded centertap.

Phemts have absurdly low capacitances and especially g-d capacitance. There may
be something there.

Aren't they leaky? That could screw up the DC performance.

 

[Phil Hobbs]

Has anyone tried using a fully-balanced system to null the charge
injection?

Sure, all the time. The late lamented Si8601 quad MOSFET was amazing
for that.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

 

[George Herold]

What's a good approach for use at 1MHz, give or take 2:1?

Things like the AD630 are slow, and my usualy go-to approach (Gilbert
Cell balanced demodulator with LO >> Vt) has input referred drift and
offset typically in the ~1uV/K and a couple mV. I'd like both to be
better by at least an order of magnitude, and preferably with some
gain like the Gilbert cell things, which have 15-20dB of gain.

Fun, I don't have any suggestions though. :^(
You'd think someone would make a faster AD630.. but I couldn't find one on digikey. (maybe I don't have the right search term.)
For the Gilbert cell I assume you're talking about an analog multiplier.
I've used the AD734.. good to ~10MHz, you could trim out the offset with a (gack) pot., but the drift remains.

Analog switches have negligible offset- how much trouble will all that
charge injection (~4pC for a good one) cause at 1-2MHz?

So rolling your own switched gain stage. I assume there is a preamp so you can keep the impedance down... still it looks like ~uA's of current.

Jamming on JL's dual path approach, maybe have one path that does the charge injection offset, (grounded input) that later gets subtracted from the signal.
(Or a single path, but every once in a while ground the input and measure the offset? Kinda a double lockin)

Any idea how they do it in fast lockins? (SRS or Zurich)

George H.

 

[Spehro Pefhany]

You've probably already considered this, but the best low-offset
multipliers I've found all live in digital hardware, and work on a data
stream that's been run through an ADC.

Take an ADC that has good linearity, sample your signal at the absolute
fastest that the ADC will go, decimate it if necessary in digital-land,
then do the demodulation as a multiplication-and-sum in an FPGA, DSP, or
ordinary processor.

At your speeds, you're probably at the dividing line between a really
hard-working DSP chip and an FPGA that's loafing along. I'd go with the
FPGA if I could find the talent to do the implementation; if I couldn't
then I'd flip a coin between trying to cram it into a DSP, or using my
own inexpert FPGA skills to make it work in that realm.

Thanks, Tim:-

Yup, have that concept working elsewhere in more than one design
(using FPGAs). This one is a quickie design- analog + some supervisory
digital.

 

[Lasse Langwadt Christensen]

Den mandag den 23. september 2013 12.05.47 UTC+2 skrev Spehro Pefhany:

What's a good approach for use at 1MHz, give or take 2:1?



Things like the AD630 are slow, and my usualy go-to approach (Gilbert

Cell balanced demodulator with LO >> Vt) has input referred drift and

offset typically in the ~1uV/K and a couple mV. I'd like both to be

better by at least an order of magnitude, and preferably with some

gain like the Gilbert cell things, which have 15-20dB of gain.



Analog switches have negligible offset- how much trouble will all that

charge injection (~4pC for a good one) cause at 1-2MHz?

I wonder how well a multiplying dac would work?

-Lasse

 

[Phil Hobbs]

Den mandag den 23. september 2013 12.05.47 UTC+2 skrev Spehro Pefhany:

I wonder how well a multiplying dac would work?

-Lasse

Multiplying performance up in the megahertz is generally very disappointing.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

 

[miso]

I was thinking of doing something with CMOS where the signal is balanced

and any charge injection on one leg is precisely balanced by charge
injected into the other leg. The shorter transition time of CMOS
switching should give lower noise than a diode ring modulator and there
is no chance of unbalaced D.C. injection from the control waveform into
the signal.

My particular application could be in modulating audio with a non
sinusiodal waveform by pulse-width modulation, but the charge-balancing
principle should work equally well at R.F.

There are papers on such schemes using CMOS switches for cancellation.
Also schemes regarding controlling the DV/DT of the clocks. You would
probably want to go fully differential on the clock as well if you want
to be anal about this.

 

[Lasse Langwadt Christensen]

Den mandag den 23. september 2013 21.07.52 UTC+2 skrev Phil Hobbs:

Multiplying performance up in the megahertz is generally very disappointing.

I just saw analog have several rated for ~10MHz multiplying BW and parallel in out so they should be real easy to drive. But I never tried them so I'm not sure what that means in the real world

-Lasse

 

[Spehro Pefhany]

The effective bandpass of your system will be tiny. You need the get the phase
close at the demod frequency, but that doesn't need an all-pass. It's customary
to trim the digital clock phase to match the analog paths. Cos is flat on top,
so small phase errors have a tiny effect on gain.

Of course you have to let the sidebands through the BPF.

The usual hazard is DC offset, and offset vs temperature.
Oh, beware of charge-injection spikes getting into opamps, into their inputs or
their outputs. That can cause bizarre problems.

That's the kind of thing I'm worried about. Op-amps like to act as
detectors all on their own sometimes.

They don't have to be very closely antiphase. A modest amplitude or phase error
will just make a small gain change.

I don't think so.. but please tell me if I missed something. If I have
a 1mV signal and 1V of offset (LF noise), the average will be:

1mV + 1V - (-1mV + 1V) = 2mV signal

If the 2nd one is * 0.99, then I'll have

1mV + 1V - (-0.99mV + 0.99V) = 1.99mV signal + 10mV error

Well, yeah, microamps maybe.

Kind of a lot when you're looking for a microvolt. Are they good for
anything below 100's of MHz? Apparently they still work nicely when
it's a bit chilly (eg. sub-4K). How about a cheapish scope probe for
probing crystal oscillator nodes?