HPM7177 ADC from CERN

[Kleinstein]

Thin film resistors can be noisy. So even if they are low TC the noise can be a problem. The TDP1603 series only specifies <-30 dB of excess noise. This would be too much for a low noise voltmeter.  There are a few thin film resistor arrays with lower noise specs, but not that many. Going by the data sheets the LT5400 arrays may be an option in a few cases, if suitable values are available.

Even if the resistors inside the BMF arrays are individual substrates, they are very likely from the same batch and thus still very similar.  The thermal coupling is not as good, but this is compensated by a much better absolute TC.

With temperature control the TC is not that important any more. Though it could still effect the linearity by local self-heating.

[jaromir]

Perhaps you'd like to repeat after me: "I'm sorry but I was talking complete b***ocks. Please accept my sincere apologies for attempting to mislead readers here that I had any sort of inside imformation. I didn't, and was simply conjecturing without disclosing that fact. I apologise to anyone that I mislead as a result".

I do have inside info - and this topic started as a result; but apparently it was partially wrong, I apologize to anyone offended. LTZs are selected, indeed. ADCs are not. Errare humanum est.
Fortunately Castorp explained in more detail what they are doing and I'm grateful for that. Instead of personal attacks I'd prefer keeping it civil and perhaps focus on a topic.

[jaromir]

Castorp: I'm intrigued about opamps in analog frontend. Did you try any other comparable autozero opamps? What about AZ switching noise? It can be problematic, depending on input impedance. Perhaps classic (as not AZ) opamps would be OK since the whole unit is temperature regulated?
I'm sure your choice was optimal for this project, but I'm curious about thinking and reasons behind it.

[Theboel]

Hi all,

Sorry If my question not appropriate in this thread.
Castorp can you described a little about Your temp management in rack and in field.
is that also CERN design etc etc

[Andreas]

Another question to the cirquit:

Do you really use a MMBFJ202 for T2?
According to the data sheet it will deliver between 0.9 and 4.5 mA at zero Gate voltage which is too low for the 4-5 mA Zener + the voltage dividers (temperature setpoint + 5V divider).

I am asking because I used the BF245C/BF545C up to now. (12-25 mA zero gate voltage current).
But I have recognized that it is obsolete now and I am looking for a replacement.

With best regards

Andreas

[edavid]

Do you really use a MMBFJ202 for T2?
According to the data sheet it will deliver between 0.9 and 4.5 mA at zero Gate voltage which is too low for the 4-5 mA Zener + the voltage dividers (temperature setpoint + 5V divider).

I am asking because I used the BF245C/BF545C up to now. (12-25 mA zero gate voltage current).
But I have recognized that it is obsolete now and I am looking for a replacement.

MMBF4392 and MMBF4393 are still in production.

[Castorp]

Thin film resistors can be noisy. So even if they are low TC the noise can be a problem. The TDP1603 series only specifies <-30 dB of excess noise. This would be too much for a low noise voltmeter.  There are a few thin film resistor arrays with lower noise specs, but not that many. Going by the data sheets the LT5400 arrays may be an option in a few cases, if suitable values are available.

Even if the resistors inside the BMF arrays are individual substrates, they are very likely from the same batch and thus still very similar.  The thermal coupling is not as good, but this is compensated by a much better absolute TC.

With temperature control the TC is not that important any more. Though it could still effect the linearity by local self-heating.

Indeed, NI of -30 dB would be catastrophic. Once I tried to measure the excess noise of Vishay bulk metal foil resistors and concluded that it was below the detection capabilities of my setup. That was done with a Wheatsone bridge biased with 40 V.

There is a bit of self-heating effect in the array. I see it as a slight thermal tail when I step the input voltage from 0 to 10 V. It's on the order of 0.1 ppm with time constant in the tens of seconds. I know it's thermal, because I can see the reaction of the temperature controller correlated with it. My conclusion was that it's a small effect that could be ignored.

Castorp: I'm intrigued about opamps in analog frontend. Did you try any other comparable autozero opamps? What about AZ switching noise? It can be problematic, depending on input impedance. Perhaps classic (as not AZ) opamps would be OK since the whole unit is temperature regulated?
I'm sure your choice was optimal for this project, but I'm curious about thinking and reasons behind it.

Initially I used ADA4522-1. So far I haven't seen any hint of switching noise with those modern auto-zero amps. It could be due to the fact that we always have low source impedance. Or it's just way out of the signal band, or both.

I wanted something with lots of open-loop gain and high CMRR, so that it makes a good buffer where I don't have to worry about linearity. Also, I checked that they don't exhibit thermal tails. That's something I discovered with another design, where the buffers were built of AD8512. The tail was on the order of 1-2 ppm, time const. 20-30s, for a 10 V step.

From top of my head, I measured the SAR ADCs sampling at 1 MSPS. When it comes to broadband white noise, their behaviour is pretty straightforward. There's a fixed rms noise value, so you get the lowest density when you run at the highest sampling rate. In other words, you spread this noise over a wider bandwidth. I guess that's the kT/C noise of the sampling capacitor spread over a very wide bandwidth, which aliases multiple times in the Nyquist band. I just checked the datasheet of LTC2378-20 - it states a -3 dB analog bandwidth of 34 MHz. That's way over the maximum sampling rate.

hmmm ok i was able to curve fit the noise plot of 7177, the plot i could assume is done at 10k SPS. the total rms noise 1Hz~10kHz will find 1.8uV rms, this fits the datasheet (the 1/f noise contribution below 1 Hz appears to be very small). but starting with 2380, the 100nV white noise is high, the rms noise will have to be >10uV, this suggest the 2380 is likely running at around 500k SPS (1M SPS ~ 16uV rms). above 125k SPS the 2380 has extra 2~3dbm of noise power (~165). whereas for 7177, the noise power between 1k~10k is nearly the same (~164), but 1k SPS is still a tad lower as reflected by the datasheet. "it looks like the dots are connecting" isnt it?

this kind of scratches an itchy curiosity regarding how much 1/f noise would impact the overall noise content of the ADC, esp below 1Hz. i think the 1/f noise corner need to be quite high to force 2380 into an "uncompetitive" seat. like the 7176-2 ?

so i could still assume, the 2380 should still beat the 7177 when using 2k~ 16k SPS range ... maybe 

but this is all paper numbers. there is also a chance i curve fitted wrongly ! i would say my fitting is very low precision.
i highly recommend some of you out there to try curve fit and see for yourself, maybe an actual expert will find something i didnt "notice".

I'll double-check. I did those measurements 2 years ago, which almost feels like a previous life... because I have a small kid 

Sorry If my question not appropriate in this thread.
Castorp can you described a little about Your temp management in rack and in field.
is that also CERN design etc etc

It's a totally appropriate question.

Those cabinets are not CERN design - they are commercial. They keep the temperature at 23 degrees C. The temperature down in the LHC tunnel doesn't vary with time of day or seasons, but there's a lot of equipment that dissipates lots of power, so locally it could vary by 5 degrees or more. It's correlated with the machine cycle - the power converters output max current when the beams reach their nominal energy.

Those LTZ1000s certainly have a good life there  First of all, they are powered all the time. And then, there's the on-chip heater and its loop, the Peltier-stabilized module, and then all that's within the large cabinets. All in all, 3 nested control loops.

Another question to the cirquit:

Do you really use a MMBFJ202 for T2?
According to the data sheet it will deliver between 0.9 and 4.5 mA at zero Gate voltage which is too low for the 4-5 mA Zener + the voltage dividers (temperature setpoint + 5V divider).

I am asking because I used the BF245C/BF545C up to now. (12-25 mA zero gate voltage current).
But I have recognized that it is obsolete now and I am looking for a replacement.

Yes, that's what we use. I guess it operates with a bit of -Vgs to source those milliamps - maybe a volt or so.

Unfortunately, lots of discrete JFETs are becoming obsolete these days...

[Kleinstein]

With a reasonable low impedance source and a slow circuit the AZ OPs are probably OK. I can imagine that it could still help to have some filtering at the input, like pi type low pass with something like  100pF , 1 K (and/or ferrite), 100pF. Some of the AZ OPs may react to high frequency components emitted from the OP that come back from a not properly terminated line.

However there could be a slight problem if used not as a single OP, but in combination with a very fast OP, like the ADA4522 + AD8065 combination for the reference drivers. The fast OP may interact with the fast spikes from the AZ OP.

[edavid]

Another question to the cirquit:

Do you really use a MMBFJ202 for T2?
According to the data sheet it will deliver between 0.9 and 4.5 mA at zero Gate voltage which is too low for the 4-5 mA Zener + the voltage dividers (temperature setpoint + 5V divider).

I am asking because I used the BF245C/BF545C up to now. (12-25 mA zero gate voltage current).
But I have recognized that it is obsolete now and I am looking for a replacement.

Yes, that's what we use. I guess it operates with a bit of -Vgs to source those milliamps - maybe a volt or so.

Unfortunately, lots of discrete JFETs are becoming obsolete these days...

-V_GS would reduce the drain current below I_DSS, which is already too low 

[Andreas]

-V_GS would reduce the drain current below I_DSS, which is already too low 

Hello,

Yes that is what I fear. In this case the gate source diode will be forward biased to get enough current through the zener.

MMBF4392 and MMBF4393 are still in production.

Thanks for that.

I think the MMBF4392 will be a good option for my LTZ1000 circuit.
Unfortunately I have a 2nd use of the BF545C/BF245C as 3-5V up to 3 mA voltage regulator with zero idle current:
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/?action=dlattach;attach=402884

Here the MMBF does not fit with the pinch off voltage.
In the mean time I have found the MMBFJ310 which seems to be suitable for both applications.

with best regards

Andreas

[Castorp]

Gentlemen, you're right. There's something fishy here. I'll investigate, as soon as I get back to the lab. Thanks for spotting it, and thanks for the suggested replacements!

I believe this transistor is already a BOM change for an obsolete part. However, I don't know when it was done.

[MegaVolt]

except TiN's measurements with ADS1263

Tell me in what topic was it?

[jaromir]

except TiN's measurements with ADS1263

Tell me in what topic was it?

https://xdevs.com/review/ti_ads1262_p1/

[Castorp]

However there could be a slight problem if used not as a single OP, but in combination with a very fast OP, like the ADA4522 + AD8065 combination for the reference drivers. The fast OP may interact with the fast spikes from the AZ OP.

So far I haven't seen evidence of that. Actually, that composite amplifier configuration was suggested to me by engineers from Analog Devices. The fast amp was originally something else that was hard to get.

Anyway - I've probed these circuits a lot with various wideband active probes. I've also looked with a 10 MHz FFT analyser (89441A). I didn't see anything funny going on. Then again - it could have been due to circuit conditions (source/load impedances, etc.)

...

@3roomlab - I know what you mean. I also got cross-eyed with those datasheets. And that was not even done for fun, it was work 

At some point I also considered testing ADS1262, but right now I can't remember why it didn't make the short list. The information is probably buried somewhere in my lab notebooks. Too bad I don't have the time or patience to document all these things better.

By the way, we use ADS1281 in yet another system - a radiation-tolerant controller. They actually tested the part and qualified it as rad-tolerant:
https://www.researchgate.net/publication/291205919_Development_of_radiation_tolerant_components_for_the_Quench_Protection_System_at_CERN

We read the bare modulator bitstream, which is available on some pins, and then do the filtering externally. The older Sigma-Delta digitizer I've referred to several times in this thread also outputs bare, undecimated bitstream, which goes over an optic fibre to the controller where the FPGA does the filtering and decimation.

[branadic]

By the way, we use ADS1281 in yet another system - a radiation-tolerant controller. They actually tested the part and qualified it as rad-tolerant:
https://www.researchgate.net/publication/291205919_Development_of_radiation_tolerant_components_for_the_Quench_Protection_System_at_CERN

Interesting, ADS1256 was tested for radiation hardness before and succeeded as well. TI seems to have a decent fabrication process.

-branadic-

[iMo]

I've never been in CERN but spent several hours walking inside DESY tunnel touching the great stuff there (the beam was off) .
While asking them on issues with early experiments I was told by the guys there the synchrotron radiation was so strong the optical fibres (running nearby the tube) got grey cataract, thus they had to replace them by copper. Even radhard, the location of the boxes inside the machine is therefore critical, I guess

[Castorp]

From top of my head, I measured the SAR ADCs sampling at 1 MSPS. When it comes to broadband white noise, their behaviour is pretty straightforward. There's a fixed rms noise value, so you get the lowest density when you run at the highest sampling rate. In other words, you spread this noise over a wider bandwidth. I guess that's the kT/C noise of the sampling capacitor spread over a very wide bandwidth, which aliases multiple times in the Nyquist band. I just checked the datasheet of LTC2378-20 - it states a -3 dB analog bandwidth of 34 MHz. That's way over the maximum sampling rate.

hmmm ok i was able to curve fit the noise plot of 7177, the plot i could assume is done at 10k SPS. the total rms noise 1Hz~10kHz will find 1.8uV rms, this fits the datasheet (the 1/f noise contribution below 1 Hz appears to be very small). but starting with 2380, the 100nV white noise is high, the rms noise will have to be >10uV, this suggest the 2380 is likely running at around 500k SPS (1M SPS ~ 16uV rms). above 125k SPS the 2380 has extra 2~3dbm of noise power (~165). whereas for 7177, the noise power between 1k~10k is nearly the same (~164), but 1k SPS is still a tad lower as reflected by the datasheet. "it looks like the dots are connecting" isnt it?

this kind of scratches an itchy curiosity regarding how much 1/f noise would impact the overall noise content of the ADC, esp below 1Hz. i think the 1/f noise corner need to be quite high to force 2380 into an "uncompetitive" seat. like the 7176-2 ?

so i could still assume, the 2380 should still beat the 7177 when using 2k~ 16k SPS range ... maybe 

but this is all paper numbers. there is also a chance i curve fitted wrongly ! i would say my fitting is very low precision.
i highly recommend some of you out there to try curve fit and see for yourself, maybe an actual expert will find something i didnt "notice".

**sorry guys, im like having a ADC diarrhoea and sample rate virus infection ... the bitrates are coming out of my armpits ...

I found my old notes. Indeed, LTC2380-24 was running at 500 kSamples/s. I made a brief note about some timing issue in the digital interface, but I'm not sure if that's the explanation for the sampling rate.

So, at 500 kS/s, the RMS noise was 32 uV. 500 kS/s means 250 KHz Nyquist band, so the noise density is 64 nV/sqrtHz. When I compared it with the other tested ADCs, I multiplied it by 1.25 to account for the lower full-scale range (I used the "digital gain compression" feature). That leads to the 80 nV/sqrtHz visible on the plot.

Now, I knew at this point that operating at higher sampling rate would give better white noise floor. In theory it can go up to 2 MSa/s, so twice lower white noise floor. However, I noted that operating at lower sampling rates (250 kS/s and 125 kS/s) yielded the same 1/f noise. And since it was 3-4 times higher than in AD7177-2, I decided not to do any further tests and abandon the part.

Interesting, ADS1256 was tested for radiation hardness before and succeeded as well. TI seems to have a decent fabrication process.

Yes, there are possibly many ICs that fall in the gray area of "radiation tolerance". That's good enough for many of our purposes. Rad-hard parts are a different story - they are specially packaged and much more expensive.

I've never been in CERN but spent several hours walking inside DESY tunnel touching the great stuff there (the beam was off) .
While asking them on issues with early experiments I was told by the guys there the synchrotron radiation was so strong the optical fibres (running nearby the tube) got grey cataract, thus they had to replace them by copper. Even radhard, the location of the boxes inside the machine is therefore critical, I guess

For us synchrotron radiation is not the issue, it's losses of high-energy protons that hit the walls and activate all sorts of materials. The LHC is pretty good in this sense - there aren't so many losses. So these rad-tolerant controllers can be placed very close to the small corrector magnets, right under the beamline. They do experience single-event upsets sometimes, but it's tolerable.

There are very radioactive regions in the smaller accelerators, for instance in the PS (Proton Synchrotron) and PS-Booster. Imagine dose rates of 1 Gy/minute. A person standing there would be dead in 5 minutes. It's no place for electronic devices either.

[Castorp]

3roomlab - just to double-check - by FS you mean the unipolar full scale? The ADC AIN tab in your table?

In principle, you should get the same noise floor estimate regardless of the digital filter type. That is - if you know precisely the "effective bandwidth" of the filter.

In my tests, the VREF was definitely not noisy. It came from a 10 V travelling standard, divided down to 5 V using a 10K SMN array. Then buffered with a decent low-noise op amp. Plus - those measurements were done at zero differential voltage, with physically shorted inputs, biased at common-mode voltage of 2.5 V (again derived from the clean 10 V). Supply voltages, etc., were also good.

…

Andreas - you were right. I probed the voltages around T2 - the junction is forward-biased. Thanks again for spotting this!

[splin]

Perhaps you'd like to repeat after me: "I'm sorry but I was talking complete b***ocks. Please accept my sincere apologies for attempting to mislead readers here that I had any sort of inside imformation. I didn't, and was simply conjecturing without disclosing that fact. I apologise to anyone that I mislead as a result".

I do have inside info - and this topic started as a result; but apparently it was partially wrong, I apologize to anyone offended. LTZs are selected, indeed. ADCs are not. Errare humanum est.
Fortunately Castorp explained in more detail what they are doing and I'm grateful for that. Instead of personal attacks I'd prefer keeping it civil and perhaps focus on a topic.

Sorry if I was a bit harsh, given your statement was actually in good faith. It is extremely frustrating and annoying though to take on board someone's confidently stated information only to later find out it was wrong.  Life's too short to fact check every interesting data point found on the net.

[splin]

I think the MMBF4392 will be a good option for my LTZ1000 circuit.
Unfortunately I have a 2nd use of the BF545C/BF245C as 3-5V up to 3 mA voltage regulator with zero idle current:
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/?action=dlattach;attach=402884

Here the MMBF does not fit with the pinch off voltage.
In the mean time I have found the MMBFJ310 which seems to be suitable for both applications.

I don't understand - the 4392 has a max cutoff of -5V, the J310 is -6.5V. The 4393 is only -3V but IDSS min is only 5mA. Should be fine for most LTZ1000 applications. What am I missing?

[Andreas]

What am I missing?

For the LTZ1000 all is ok. 

But I am using the BF545C often as cheap 4V (+/-1V) voltage regulator with zero quiescent current. Here I need a corresponding (higher) pinch off voltage. See picture on the left side (T1). (A alternative would be a XC6216 but with 3-4uA quiescent current which is rather high against 10uA average consumption of the processor)

Sorry had initially linked the wrong schematics:

https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/?action=dlattach;attach=402886;image

https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg1449488/#msg1449488

with best regards

Andreas

[Castorp]

I think something is still off.

Do you have a column somewhere with the noise bandwidth? For a given filter type it's a fixed fraction of the ODR.

Assuming white noise, the density is: rms noise/sqrt(bandwidth) [V/sqrtHz]. Or (rms noise)^2/bandwidth [V^2/Hz]. That's one step away from having a scaling relative to the FS.

About the dynamic range - is it for a full-scale sine wave? That's what they usually give in datasheets. It's the theoretical maximum SNR.

[splin]

3roomlab, you've got a few errors in your table.  The first column is the input referred noise level which is generally independent of reference voltage and input voltage range. Yet for the LTC2508-32 your numbers are all half of the datasheet values.  I guess you have adjusted the values for some reason but you shouldn't be doing it here.  Similarly, the LTC2380-24 also look wrong, but I can't see where the numbers come from just now.  The 1.5MSPS value however should be twice your value, based on the transition noise spec of 55.7LSBs, rms. 33uVrms = 55.7/2^24 * 10Vpp input range.

The second major problem, as hinted at by Castorp,  is that you are not comparing like for like - you have picked the lowest noise value for a given ODR (where there is a choice of filter, such as the 2508-32). But ODR by itself doesn't mean anything - you can take any ADC output and increase the sampling rate through interpolation but it won't improve it's noise performance.

To compare ADCs fairly,  you have at least two options: normalise to the fully settled sampling rate, or to the input bandwidth (but the difference will be small). The AD7175 and AD7177-2 output samples are fully settled at or below 10KSPS for the Sinec1 + 5 filter whereas the ADS127 has a fully settled output data rate of 1/2 to 1/5  the ODR for the low latency filter,  depending on the OSR.

For the LTC2500-32 the averaging filter, unlike the SSINC filter values you selected, are all fully settled at the ODR so can be compared directly to the AD7177-2 numbers. The tables also show the filter bandwidths so you can normalise the noise to bandwidth easily.

[Castorp]

3roomlab, you've got a few errors in your table.  The first column is the input referred noise level which is generally independent of reference voltage and input voltage range. Yet for the LTC2508-32 your numbers are all half of the datasheet values.  I guess you have adjusted the values for some reason but you shouldn't be doing it here.  Similarly, the LTC2380-24 also look wrong, but I can't see where the numbers come from just now.  The 1.5MSPS value however should be twice your value, based on the transition noise spec of 55.7LSBs, rms. 33uVrms = 55.7/2^24 * 10Vpp input range.

That's pretty close to what I measured.

I remembered another detail. My setup for LTC2380-24 and for LTC2378-20 was completely identical, down to the test PCB. These two parts are pin-compatible, so it was not even a different executions of the board - it was the same board with different ADC chips populated. The rest was just a tweak in the FPGA that read out the digital interface to get 24 bits per conversion instead of 20. I did conclude that the 24-chip is better in terms of 1/f noise, but not that much better. And it chilled my enthusiasm for the "32-bit" SAR.

splin, I agree with the rest of your explanations and reasoning.

[splin]

I remembered another detail. My setup for LTC2380-24 and for LTC2378-20 was completely identical, down to the test PCB. These two parts are pin-compatible, so it was not even a different executions of the board - it was the same board with different ADC chips populated. The rest was just a tweak in the FPGA that read out the digital interface to get 24 bits per conversion instead of 20. I did conclude that the 24-chip is better in terms of 1/f noise, but not that much better. And it chilled my enthusiasm for the "32-bit" SAR.

Yes,  I was going to point out that Linear sell loads of variants of just a few core ADCs but with varying extras, principally built in filtering, for higher prices. But I got bored editing on a tablet. 

The LTC2500-32 seems to be an LTC2378-20 at heart - identical noise and linearity specs. I can't imagine how/if they managed to get any worthwhile sales volumes of the expensive variants given most designs are going to include an MCU capable of all the filtering required with much more flexibility (or at most with the addition of a $.50 MCU). The 2500-32 for example, unlike the 2378-20, doesn't support 1MSPS which can be very useful.

I can possibly see some niche markets, where the product developers want to sell on the basis of a state of the art, high-performance 32 bit ADC, but the price differential would, I expect,  exclude most volume applications.