# High resolution ADC vs amplifiers

I have started a design for work where I want to interface some sensors, (strain gauge, bridge, thermocouple, low voltage stuff) to a national instruments sbRIO card. This card has analog inputs built in as well as DIO. The sbRIO can measure down to +- 1v and 16 bits, but in my experience thats not quite good enough for thermocouples and strain gauges where you're looking at <100 mV. We were going to already be making a "mezzanine" card with some other interface circuitry so I was going to add on some circuitry that could handle these lower voltages.

A while ago I had found a 32 bit ADC with SPI interface and I've been looking for an excuse to play with one and thought this might be a good fit. (https://www.protocentral.com/analog-adc-boards/1005-protocentral-ads1262-32-bit-precision-adc-breakout-board-0642078949630.html). It has a built in gain amplifier, and a few other bells and whistles.

My question is for any hardware designers out there is this. Would I be better suited to using amplifiers for each individual channel rather than using this unit of an ADC? ie using thermocouple amplifiers and bridge amplifiers where appropriate? I appreciate any insight you might be able to give me. Thanks for your time!

• "not quite good enough for thermocouples"... How would you do the cold junction compensation? Nov 9 '18 at 19:49
• what is the resolution requirement? Nov 10 '18 at 1:41
• How stable must your system be? Can you provide power clean enough to satisfy expensive opamps? Are you willing to PAY for expensive 5 PPM Vishay resistors? Or perform a calibration each time? Nov 10 '18 at 5:08
• I find it hard to believe 16 bits is not enough for those types of measurements. Considering most automation systems, lab DACs, chart recorders etc. are 16 bits. if you want to go beyond 16 bits you need to also start looking at noise coming from everywhere else in your system, otherwise you'll never even see the additional bits of signal. Nov 11 '18 at 18:01
• Labjack has some notes on using high resolution ADC with thermocouples. You do have to deal with cold-junction compensation one way or another. Nov 12 '18 at 5:18

This isn't quite an answer, but rather an anecdote.

High-bit ADC are quite the nifty thing. Great resolution, along with high dynamic range, take away many signal-chain concerns.

I built a system for biopotentials with a 32-bit chip. Signal quality was excellent, as all my calculations told me they would be, with only some minimal amplification and anti-alias filtering. That said, my data was riding on what seemed to be an *enormous" square wave that I didn't notice during my prototyping. It had me quite baffled for a while.

Working backward, though, I figured out that the magnitude of the square wave was truly tiny.

Eventually, I had the box where this thing lived open, and I noticed serendipitously that when the programmer on microcontroller dev board that I was using wasn't USB-enumerated, that an LED flashed perfectly in time to my mystery square wave. That was making something sag, in the microvolt range, that was just huge in my 32-bit signal. It wasn't present during prototyping, because my on-board programmer was enumerated! Those bastards!!!!! The problem was resolved by removing the current-limiting resistor on the LED.

Why was this frustrating? Well, for the first time in my life, I didn't amplify enough for me to actually see the signals I was working with on an oscilloscope!!! I didn't do it, because I didn't have to.

I suppose the point is that selecting a 32-bit ADC created a funny opacity in my signal chain that I had to learn about the hard way. This was much like my early experiences with microcontrollers, where you can't just peek inside and know what's happening.

Long story short, high-bit ADCs are a valuable tool that makes analog design a breeze. That said, they're a tool, like any other, and the learning curve can be a challenge. Fortunately, in my case, I managed to ID my issue. I can tell you, I was under some real time pressure, working under subcontract to a medical device company. I was under pretty substantial stress for a few days, until I found my problem. There's a time and a place to start using new tools, and a time and a place for the tried and true.

• @ Scott Seidman Thus the Power Supply Rejection was poor? Wondering how the "Sag" was affecting the conversion. Or was the Electric Field of the LED cathode, where the square-wave existed, coupling into the high-impedance bio-potential wiring? Nov 13 '18 at 14:46
• @analogsystemsrf -- absolutely all very good questions, and all very difficult to answer, as the square wave was calculated to be 11$\mu$V on the input of my on-chip PGA! In the next iteration, I changed the side of the isolation barrier that the dev board was on, and left the ADC on the side with the low-noise instrumentation amp. By the time I was done, I had my noise level <1$\mu$V rms. Nov 13 '18 at 15:28
• ... all design considerations that never come up until you start using high^2 res ADC's. My signals are generally orders of magnitude above the square wave magnitude, which would just simply ordinarily disappear into the noise. Nov 13 '18 at 15:32
• @ Scott Seidman Was the "square wave" square, or did it have lots of droop, which indicates a high-pass-filter effect, where the aggressor energy comes thru a metal-air-metal path? One way to model paths thru the air is to use a parallel-plate equation, where C = 8.9pF/meter * Area/Distance, and then scale down by 1/distance^3 because the underlying PLANES will capture most of the Efield flux. The displacement current is converted back to voltage, because the charge has to exit the node thru any/all available resistances (impedances) to get back home. How far away was the LED/resistor/driver? Nov 18 '18 at 3:50

32-bit ADC is misleading. Even at it's highest gain, the noise peak is roughly 60nV. A 5V 24bit ADC is 5/2^24 or 29nV per a bit. So the bottom 9 bits of the 32 bit ADC will be noisy. There are less noisy delta sigma ADC's on the market.

Would I be better suited to using amplifiers for each individual channel rather than using this unit of an ADC? ie using thermocouple amplifiers and bridge amplifiers where appropriate?

Depends on what your objective is, if it's lowest noise, an ADC with a mux will always be noisier than a standalone ADC, because the transistors from the MUX are noise sources.

As far as your amplifier question, again it depends on what the requirements for the project is. But there will be better control over how much noise is in your circuit if you use analog amplifiers, it will also cost more. The ADC also has many digital filters, so instead of using analog sensors and calculating the bandwidth you can change it with software.

Years ago I performed silicon evaluation of a 22 bit ADC. I expected to learn, to be surprised, to be puzzled. I was.

1) your hand or face or body emits heat, and silicon junctions CLOSER to the heat source will be warmer; two nearby diodes would drift apart by 500 microvolts, and you'll experience about 60 seconds of settling time to the new offset voltage; given 0.1 meter of copper has 114 seconds of thermal time constant, we can expect heat flows to be a constant problem; I'd designed those 2 diodes onto the Eval PCB, to examine the heating by my face; one diode partially shaded the other diode, to ensure a heat flux difference.

Why are heat flows a problem? The movement of 1 watt thru a square of copper foil, from edge to edge, will produce 70 degree Centigrade temperature gradient. Yet the joining of dis-similar metals produces 5 to 40 microVolts per degree Centigrade, and PCBs have lots of such metallic transitions. The thermal mismatch of differential paths (Vin+, Vin-) becomes your challenge.

2) dielectric absorption of capacitors showed up; input filtering using RC lowpass, to explore the ADC's noise floor, showed 2 or 3 minutes of settling; when shorted briefly then opened up, nearly a millivolt of stored charge would slowly appear

3) the resistance of 1 ounce/foot^2 copper foil is 0.000500 ohms per square, for any size square; 1milliAmp thru a square will generate 500 NanoVolts of error; plan on using Finite_element modeling to design your PCBS at the 32 bit level. [edit the NanoVolts was firstly microVolts]

4) 1 amp of 60Hz pure sinusoid (no spikes) at 1 meter from 10cm by 1cm loop, will induce this voltage onto your PCB

Vinduce = 2e-7 * Area/Distance * dI/dT

Vinduce = 2e-7 *10cm*1cm/1meter * 377

Vinduce = 2e-7 * 1e-3 * 377

Vinduce = 1e-10 * 754 = 75 nanoVolts

Why? because thin copper foil will not shield against 60Hertz magnetic fields. At 60,000 Hertz, just barely. At 60,000,000 Hertz, quite well. But not at 60Hz.

5) those "quiet" digital interface pins, with either a 1 or a 0 level, are still buzzing with 200 or 500 milliVoltsPP of MCU rail noise; how close can you let a digital interface trace get to the 32-bit signals, given the MCU trash has pseudo-random (program dependent) patterns, and cannot be trusted to "average out" ?

6) some useful values for switched-cap noise

1000 picoFarad ............ 2 microVolts RMS

100,000 picoFarad ........ 200 nanoVolts RMS

10,000,000 picoFarad ..... 20 nanoVolts RMS

1Billion picoFarad ............. 2 nanoVolts RMS

using the formula: VnoiseRMS = sqrt( K*T/C)

What is use of this table? to achieve 2 nanoVolt noise levels, the equivalent energy of charging 1Billion picoFarad (0.001 Farad) must be provided from the signal source or from buffers or from amplifiers.

You are missing a very important consideration on any design of this sort: firmware/software/drivers.

Using an existing DAQ card provides you with all of that and allows you to concentrate your resources on the problem itself via high-level abstractions and not on the technical details of the interfacing.

Besides, I really doubt you can get your analog noise to a level in which 32 bits or 24 bits would make any difference.