LTC2400 ADC offset err

[Oznog]

I have a 24-bit LTC2400 I'm trying to use for measuring current on a 50mV @ 200 amp shunt. I'd like to measure low current leakage (under 1 amp) out of my deep cycle battery as well as the high currents.

I have it up and working. The problem is, I'm seeing a DC offset of 1.3mV when Vin is connected to ground at the ADC's pin (not ground somewhere far from the ADC). This constitutes an error of 5.2 amps so we're totally not in the ballpark here.

The LTC2400 lists the max offset error as 2ppm of Vref, which should mean 0.01mV. I've got 130 times that. There is no temp differential here so I'm ruling out a thermoelectric Seebeck effect. What's the problem here??

 

[Nigel Goodwin]

I've never used such a device, but 1% tolerance is generally considered very good (and better than most circuits) - but 1% is less than 7 bit resolution. 24 bit is something like 0.00001%, it seems totally unrealistic for your circuit to approach anywhere near that degree of accuracy.

 

[Oznog]

I've never used such a device, but 1% tolerance is generally considered very good (and better than most circuits) - but 1% is less than 7 bit resolution. 24 bit is something like 0.00001%, it seems totally unrealistic for your circuit to approach anywhere near that degree of accuracy.

On the contrary, even the worst ADC out there would never give you a 1% err for offset. Even PIC is 1.5 lsb offset and 1 lsb gain error. So the offset error's around 0.1%, and that's a pretty "basic" ADC, about as low an accuracy as you're ever going to find. They don't isolate the digital and analog grounds, which makes high accuracy impossible for high current consumption devices, and they don't spend a lot of attention with the accuracy issues for obvious cost reasons.

Anyways, a 24 bit's obviously higher accuracy, otherwise the additional bits are meaningless. Actually, the LTC2400 gives 28 valid bits, but the 4 LSBs are below the noise floor and thus only useful if you're averaging a lot of readings. I found it fascinating that they honestly only advertised bits which you'd be able to use. In any case, the spec's clearly guaranteeing 10uV offset error and I'm seeing something different.

 

[Nigel Goodwin]

On the contrary, even the worst ADC out there would never give you a 1% err for offset. Even PIC is 1.5 lsb offset and 1 lsb gain error. So the offset error's around 0.1%, and that's a pretty "basic" ADC, about as low an accuracy as you're ever going to find. They don't isolate the digital and analog grounds, which makes high accuracy impossible for high current consumption devices, and they don't spend a lot of attention with the accuracy issues for obvious cost reasons.

I was refering to the rest of the components in your circuit, 1% resistors are about as good as you get, and everything else has massively higher tolerances.

Anyways, a 24 bit's obviously higher accuracy, otherwise the additional bits are meaningless. Actually, the LTC2400 gives 28 valid bits, but the 4 LSBs are below the noise floor and thus only useful if you're averaging a lot of readings. I found it fascinating that they honestly only advertised bits which you'd be able to use. In any case, the spec's clearly guaranteeing 10uV offset error and I'm seeing something different.

I think the 24 bit accuracy is pretty meaningless, as I've already suggested - where do you find suitable support components?.

From what I've seen about these sorts of devices in the past, I seem to remember that layout is absolutely critical - do you have a manufacturers proven PCB layout you're using?.

BTW, what are you trying to do with it?.

 

[Roff]

Oznog, I'm also baffled. Your basic question has nothing to do with the rest of your components. As I understand it, you simply connected the input pin to the GND pin, at the IC,, and got an output code which indicated 1.3mv on the input. This is, as you said, way outside the spec, which is 2ppm of Vref. You implied that Vref is 5v, which should yield a maximum offset of 10uV. My calculations show that the maximum code with zero input should therefore be +/- 34 LSB's, or 22H. Your reading of 1.3mv would be 110AH. Am I interpreting the spec and your reading correctly?
Regarding the precision of your other components, It sounds like Vref and the current sense resistor are all you have that figure into your calculations. Both of them contribute to scaling errors, but not to offset errors. When you are trying to measure from milliamps to 100's of amps, I can see why offset would be your primary concern.

 

[Oznog]

I was refering to the rest of the components in your circuit, 1% resistors are about as good as you get, and everything else has massively higher tolerances.

I think the 24 bit accuracy is pretty meaningless, as I've already suggested - where do you find suitable support components?.

From what I've seen about these sorts of devices in the past, I seem to remember that layout is absolutely critical - do you have a manufacturers proven PCB layout you're using?.

BTW, what are you trying to do with it?

Of course utilizing a divider with 1% resistors would destroy the gain accuracy, though it should not affect the offset. However, there is no voltage divider present for the reasons you give.

I found the number for LTC tech support and got ahold of Mark Thoren. It turns out the device will have a predictable offset error from the lowpass RC filter I added to the input, it's in the spec sheet (but it's a very long spec sheet). Both the input resistor and capacitance combined to a calculated 1.5mV offset err. He pointed out that the ADC input stage inherently has a 12 Hz lowpass on it, so the filter's not necessary.

In any case, the intolerance of input resistance/cap is a killer. I was planning to do a high side shunt and pump the charge down to ground reference. It was questionable before, but it's out of the question now. They've got an app note about making a new ground 5V below the 12V rail so the ADC can measure the high side shunt directly, then use an optoisolator to get the digital signals to the real ground.

I looked at that possibility before, but was turned off by the fact that all the optos I found were dead, dead slow. Mark also suggested several high speed optos that would go 10x or more faster than the ones I'd found. Score!

 

[Roff]

The problem is, I'm seeing a DC offset of 1.3mV when Vin is connected to ground at the ADC's pin

Well, Gonzo, you sure sent me on a wild goose chase. I took your quote literally. Oh, well. At least I learned something about a pretty nice part.