ADS122C04: Idle Tones

Hi Glen,

You bring up the nasty words idle tones.  This is something that IC designers try extremely hard to eliminate or reduce as much as possible.  The core used for the ADS122C04 has been around for quite a few years and this is the first anyone has brought up idle tones for this device family that I am aware of and I've been supporting these devices since the beginning.  The biggest reason is most likely in how the device is expected to be used.

The clues as to why your situation is different is "the front end is super quiet and we use all the bits".  I would also assume that your reference is very quiet too.  Most sensors that where intended to be used with the ADS122C04 such as thermocouples, RTDs, bridge sensors, etc. are inherently noisier than the ADC.  Noise, similar to dither, adds enough break in the modulator output to prevent repeating patterns from occurring.  If there are tones present they are most likely below the noise floor of the system most often used.

So we don't expect to see tones with the ADS122C04, but as you said, that doesn't mean they don't exist.  All I can say is that we haven't seen them.  The ADS122C04 is not a 'normal' delta-sigma device and the actual design is not something I can go into even with an NDA.  I can say that the behavior will change slightly with data rate, reference voltage and analog supply voltage.

Usually tones are most likely to show up near 0V, but it is hard to predict something you haven't seen.  It is also possible that the input chopper is leaking through the digital filter and folding back.  The chopping frequency should be FS/4 where FS is the modulator sampling rate of 256kHz in normal mode.

It would be helpful to know the device configuration settings, the reference voltage and supply voltages you are using.  And it would be good to know if you are averaging results to lower the noise.  I would also like to know what SCL frequency you are using for communication and the pullup values used for I2C.  I would like to try and replicate if possible.  

Best regards,

Bob B

Hi Bob,

LOL I understand about idle tones being nasty words.
Yes, our 24 bit systems produce around 138dB SNR based on peak to peak noise divided by full scale.  (peak to peak is all that matter...maybe the next sentence or two will explain why)  And yes, I know, 24 bits is 144 dB ideal.  Our stuff outperforms most demo boards.
However, I do need to say up front, the ADS122C04 is NOT directly acquiring our chromatography signal, as I do not think it would ever get to 138dB SNR p-p.
But the following explanation is still relevant to maybe how and why it matters and why we look for it or how we know to look for idle tones.

In chromatography, filtering is a no-no because our signal must retain its time domain Gaussian shape, and any kind of filter will mess that up.
Thus we go to great lengths to have a rock solid front end.  Then to make matters more interesting, our users (Chemists) are quite used to looking at a scaled output in volts over time.  Think of it this way, our device is a time domain spectrum analyzer, where the spectrum is a chemical composition.
Thus peak to peak noise at a very low SNR can then be mistaken as our gaussian signal.  And yes, we definitely work down in that range too.  Those same chemists often come to me with information that always includes their measurement of peak to peak noise over a 2 second period.
We can have a very large peak right next to a tiny peak that sits in the noise.  All those peaks represent a chemical composition that is smeared across the time domain, but not necessarily neatly separated.  There are over 150 distinct compounds in gasoline.  Our instrument is designed to pull all these compounds into the time domain over seconds and minutes and process the shapes, measuring the area of each shape to associate with concentration.  Start smearing the shape with a filter and the waters get muddy.  I guess you could say that the bandwidth of our signal = the bandwidth of the noise.  That, of course, is not the same as SNR which is based on amplitude.

Since we look for Gaussian shapes and we do not filter them, and we do this at a very low frequency over a relatively long period of time, with a signal that often moves slowly through zero (or another quadrant) you can imagine how an idle tone looks to us.  It looks similar to a Gaussian peak, except it really looks like a growing and then fading series of positive and negative peaks.  It stands out pretty well, but when you toss in another 150 peaks, it wreaks havoc.
However, they are hard to just go and find.  We were one of the first users of the ADS1210 in 1996, and that part spit out idles tones if you shorted the input.  I remember the designer of the part was amazed that anyone cared (he actually said he never expected anyone to use the part at such a low level that they would see the tones), but we did because our dynamic range is over 130dB and those bits mattered.
We patched it by putting in a software invokable small analog offset and when the software saw we were within so many millivolts of true zero, it would put in the step offset.  At least the step was easy to see it was not signal, vs a sinusoid that comes in and goes out over, essentially, a sinusoidal envelope.
We also tried the competing ADI part which had idle tones at quadrants (1/4 range) but its reduced input voltage range dropped us too far into the noise floor and we could not achieve better than 120 dB SNR

Later generations of parts from TI and ADI "hid" this better.  We thought we were good with another one of your parts, went all the way through extensive testing, went to production for a year or so, and then one day a chemist knocked on my door and showed me a chromatogram with what was undoubtedly idle tones.  It was not at "zero" but up out in a quadrant with just the right conditions.  At that point I worked with Mark Berarducci in Tucson to figure out what was going on.  At this point, perhaps you want to take this thread private.

Suffice to say, we are a bit touchy about idle tones, and no longer believe they don't exist; rather they are just hard to find and re-create.
We accept they exist...and if the manufacturer is up front with us, we can deal with them accordingly.  After all, they are, well, a theoretical certanty.

As I said, we are still looking at data, changing the dimensions of the variables to figure out what to tickle to make it worse, etc. before I could ever be able to tell you how to re-create it.  This is the first time we have had to involve temperature, so I cannot say if temperature causes it, or if we happen to be creating the right conditions at this temperature, with our entire system.  That, currently is 10 degrees C.

As I started out to say, this application is not directly in the chromatography, but it controls pressure, which can very directly affect the chromatography.
An idle tone in the pressure system will modulate the flow of the separation process and then show up in the chromatography.

It may NOT be idle tones of course, but it cannot be discounted.  I have a very thorough technician who is excellent at wiggling conditions and documenting it well, so he is off "doing some wiggling" right now to see if we can pare it down.
We have a sensor simulator that is made up of bulk metal foil resistors with a 0.2 ppm/C drift :-) and he was going to try to bolt that onto things so we could entirely remove some of the analog front end.  That may take him the rest of the week to get the recipe just right.  It is not you can just put a pot on that and adjust it till you find the conditions :-)

Please reach out by email or take this thread private.  I have shared all I dare to share on a semi-public forum and you probably feel the same.

Thanks

Glen

 Glen E. Schmidt
R&D Hardware Manager-Process Chromatography

Siemens Industry, Inc.
RC-US DI PA AP R&D
500 US Highway 60
Bartlesville, OK  74003
T: (918) 662-7763
M: (918) 440-3298
glen.schmidt@siemens.com
www.usa.siemens.com/processanalytics

Thread taken offline...