0

I'm designing a precision source meter and I wanted to put a couple ideas out to the community. One thing I will note is that this device has 32 channels, so I need to be very cognizant about both component cost, and component size.

Current sense circuit You can see the current sense circuit below. The dynamic range of this circuit is pretty big. It needs to measure up to 125 mA and down 100 nA. The load voltage is +/- 10 V. I was thinking about using an instrumentation amplifier with a digital potentiometer as the gain resistor. I set the reference pin to 2.5 V to give the output a 2.5 V offset. There's also an output buffer to clamp the ADC input to 0 to 5 V. Regarding the potentiometer, I would then select three resistor values to give me the gains that I desire, and then calibrate them in software.

To calibrate this I will basically store gain and offset values in the firmware to satisfy V = Gain*V_adc+offset. I understand that these are temperature dependent, and depending on how much time I can dedicate to this, I will add a polynomial factor in as well.

My concern about this design is error. Is this idea propose at any greater risk of injecting noise due to the digital signal? What about any other noise sources. I understand that temperature variation will always be a factor, is this particularly at risk of this?

I'm interested in your thoughts. Do you have any other ideas? I'll post the voltage sense circuit tomorrow.

Thank you!

FYI, I know I need caps on the power rails for the op amps and digipot. I made this schematic quickly and didn't put them in, but they'll be there in the prototype.

enter image description here

AMacDonald
  • 69
  • 5
  • Consider switching gain by selecting shunt resistor value, much less punishing on in-amp DC-offsets specification. It's easy to have make-before-break switching to allow constant through current, without the selection switch resistance appearing in the I to V equation. – Neil_UK Jul 24 '20 at 03:31
  • How would you accomplish this? Relays are too big and expensive. Transistors are small and cheap, but a single FET or BJT won't work due to the bipolar nature of the circuit. What I need is something like an AND gate, but digital and analog make poor bedfellows. – AMacDonald Jul 24 '20 at 04:31
  • Depends on what you are going for in terms of accuracy. High quality relays are how this is normally solved, although if you can afford the series resistance, you can put the shunts in series and use individual op-amp stages into separate ADC channels. – Dean Franks Jul 24 '20 at 05:00
  • Is 100nA the resolution or do you actually expect to read a 100nA current to any level of accuracy? That's 330nV, which is rather on the small side. – Spehro Pefhany Jul 24 '20 at 05:27
  • Are you using a uC to process the ADC measurements? – mr_js Jul 24 '20 at 05:58
  • I need 100 nA accuracy. I'm using a 29 bit ADC, so I've got tonnes of precision, but it's very easy to drowned out the signal with noise. I can't use separate op-amp stages because of the cost and size that would incur over the 32 channels. I am using a microcontroller for signal processing. – AMacDonald Jul 24 '20 at 14:58

1 Answers1

0

schematic

simulate this circuit – Schematic created using CircuitLab

Something like this would improve your dynamic range significantly.

As Dean said in the comments, this sort of switching is normally done with relays. However, there are many devices available with varying on resistance, leakage and voltage compliance available that could replace them. Look on Digikey, take your time and look thoroughly. You can buy opto isolated FET output switches if you look. You might want to look at ADG1411 (or ADGS1412 maybe better, higher current and SPI interface) type devices to see whether they might be good enough for your application, to integrate several switches into one package. Look at other manufacturers' devices as well. The ADGs are quad, logic level drive, +/- 15 V analog supply, 100 mA or 200mA per path (schematic shows two switches in parallel for the high current range with the 100 mA switch, but you could have 4 ranges with the 200 mA type), leakage in the sub and low nA range.

I've shown the low leakage BAS116 across the switches as path protection, should all switches be open at once. However, if you keep the voltage low enough, then most silicon diodes will do. Don't use schottky, much higher leakage than silicon. Interestingly, every time I've measured the leakage on silicon diodes in the 100 to 200 mV drop range, the big beefy 1N400x has been lower leakage than a 1N4148.

Neil_UK
  • 158,152
  • 3
  • 173
  • 387
  • I looked at using an analog switch. The problem with this is that it's very expensive. At $12 for two channels, that would add nearly $200 to the cost of the project, not to mention the additional board space. Also, the non-linear temperature response concerns me. – AMacDonald Jul 24 '20 at 21:04
  • The advantage of using a digipot would be that I can always set it to the same settings every time, which would then set the gain of the instrumentation amplifier. There would be some temperature variation, but it would be linear (and it's not huge with that chip). The downside of this is that it caps the maximum gain at about 1000 due to the wiper resistance. How do you think that this is punishing on the op-amp? – AMacDonald Jul 24 '20 at 21:09
  • @AMacDonald It depends what specification you want, I said 'punishing its offset specification'. 100 nA in 3.3 ohms is 0.33 uV. The INA2126 has an offset voltage tempco of up to 3 uV per C. While you can calibrate out the 250 uV offset voltage, you can't calibrate out the tempco, so with only 0.2 C temperature change, you wouldn't be able to tell the difference between +100 nA and -100 nA. If you're happy with that temperature sensitivity, then go right ahead with a fixed shunt resistor. I'm not sure I'd say I was measuring down to 100 nA unless I had 100x better performance at 100nA. – Neil_UK Jul 25 '20 at 08:57
  • Thanks for the clarification. Certainly agree with you that the voltage across the sense resistor would need to be greater. I think I found a chip that might solve many of my problems. Analog's ADA4254 and TI's PGA280 are virtually identical. They're fancy, digital, two channel op-amps, programmable gain op-amps. Moreover, the two channels are digitally selectable and there are GPIOs as well. – AMacDonald Jul 25 '20 at 19:14
  • I think I've come up with a solution: have two shunts in parallel, one 8 O and one 3 kO. The 8 O shunt will be connected to two parallel switches (for current purposes) in an ADG1414. Each chip can be used for 4 channels, which would mitigate the cost. The 8 O resistor would be used for measurements between 100 uA and 125 mA, and the 3 kO resistor would be used for measurements between 300 uA and 100 nA – AMacDonald Jul 25 '20 at 20:24
  • I like the ADA4254 op amp. One channel on the op-amp could be for current measurement and the other channel could be for the voltage measurement for the same channel. – AMacDonald Jul 25 '20 at 20:25
  • What's the real accuracy you need? 1%, 10%? The 1000:1 ranges really tax the accuracy at the bottom end of the range. Caution, the ADG1414 has an on resistance variation with channel voltage of up to 2 ohms, so 1 for 2 in parallel. That variable 1 in series with the 8 ohm therefore means 10% accuracy as voltage coefficient on the high current range. Think carefully about two anti-parallel diodes as the sensing element for a moment. 125mA to 100nA in one range, no range changes, 10% accuracy, and dirt cheap amplifiers and ADC can be used, but tempco calibration required. Just a thought. – Neil_UK Jul 26 '20 at 09:07
  • I've just re-read your OP, a 29 bit ADC doesn't exist, at least not one that will read to 29 bits. I don't know what your application is, but it sounds like resistive tomography. The diode solution would enable you to build 32 channels at <10% of the cost of high precision ADCs, in-amps, and analog switches, and try out your application at low cost and risk. As it's log reading, it will give you constant relative accuracy. Every time I do a cheap'n'quick'n'dirty experiment, I learn something unexpected that I need to know. Then if you need higher precision, build something better. – Neil_UK Jul 26 '20 at 09:12
  • The ADC I'm using is an LTC2449. It's classified as a 24 bit ADC, although there are an additional 5 bits on the bottom that can be used. We're actually building a mulit-channel source meter for PV characterization. – AMacDonald Jul 26 '20 at 18:16
  • Your anti-parallel diode suggestion intrigues me, but I'm not sure what you mean by it. Tempco calibration is not optimal, but definitely possible if it will solve some of the other problems. As for accuracy at the bottom end of the range, 100 nA +/- 10% would be acceptable. – AMacDonald Jul 26 '20 at 18:29
  • Also, I was going to calibrate each channel to compensate for any variation in resistance between channels of the switch, and the total resistance would be 8 ohms. – AMacDonald Jul 26 '20 at 18:30
  • Neil, I'm actually quite interested in your proposal about using diodes. On a glance, I would say that the IV curve for diodes is less than ideal for measuring currents, but I suppose that if you're measuring small currents, greater than the current at the breakdown voltage, and less than the current at the forward voltage it could work. Is this what you meant? – AMacDonald Jul 31 '20 at 02:31
  • @AMacDonald The question is where to put the variable gain, with respect to the unstable voltage offsets in the amplifier, and where you want your measurements to be accurate, and where you can afford for them to be less so. Using a linear shunt like a resistor gives linear input to output, but with large *relative* inaccuracy near zero, to the extent that small +ve could read as small -ve and vice versa. Using anon-linear device like a diode means +ve and -ve are always correctly resolved, but scale errors could be larger. Which is more important to your application? – Neil_UK Aug 02 '20 at 17:08
  • @AMacDonald If you need unambiguous resolution of +ve and -ve, then a diode will help. A 1N4148 I measured a while ago had an almost perfect log characteristic from 10 nA to 10 mA, from 100 mV drop to 700 mV drop, at 100 mV per factor of 10 in forward current. It's quite temperature sensitive, so a second co-mounted diode is recommended for temperature correction. A quad of high current diodes in a bridge package is ideal, two for sense, two for compensation. See [this](https://electronics.stackexchange.com/questions/340330) answer of mine where I illustrate a similar idea. – Neil_UK Aug 02 '20 at 17:12
  • @AMacDonald The fun thing about using a diode as the shunt element, and targeting a modest accuracy, say +/- 10% (the temperature issue means you can't do much better than that anyway) at any current reading, means that the amplifier noise, offsets and drift become easily achievable. Taking my 1N4148 (earlier comment) at 100 mV per factor of 10 in current for example, then 10% is about 4 mV. While the very cheapest amplifiers have comparable offsets, you don't need to pay much for <1 mV offsets. That's 10% accuracy at 1 mA, or 10% accuracy at 100 nA, both give you 4 mV at the opamp! – Neil_UK Aug 02 '20 at 17:24
  • I've actually found a digitally programmable amp that looks like it will fit our application well, ADA4254. A 3k resistor for low currents in parallel with an 8 O resistor attached to an analog switch should give us sufficient wiggle room, even at the modest gain of only 128 V/V – AMacDonald Aug 02 '20 at 19:02