Negative Voltage Reference for Bipolar ADC

I am considering using ADS1258 - a bipolar ADC. It needs an external reference across REFP(+) and REFN(-). How do I generate precise -2.5V assuming I have +12V and -12V analog voltage rails? I can generate an approximate -2.5V using something like LM337 but I need something more accurate as that voltage is going to be used as a reference. A more precise way is to use a low input offset voltage opamp to invert a positive reference but that would still add some error.

What is the standard way to get good quality bipolar reference voltage in a circuit? • LM385 is a commonly available micropower voltage reference. It has the Vref connected to cathode (instead of to anode as in tl431). It may be easier to connect to the negative rail. – Indraneel Aug 15 '18 at 11:33

You are over-complicating this. The power rails for the circuit in your question are +2.5 volts and -2.5 volts and it uses a REF3125 to produce a REFP voltage that is precisely 2.5 volts above REFN. This makes REFP 0 volts but if REFN drifts to -2.6 volts, REFP must also track to -0.1 volts.

The internal reference voltage needed by the ADC is REFP-REFN hence, providing REFN is within the constraints dictated by the chip, it doesn't need to be tied to -2.5 volts. For instance if your power rails are 0 volts and +5 volts then you tie REFN to 0 volts and use a regular reference chip to produce (say) +2.5 volts.

In other words REFN doesn't need to be precisely set or even fairly constant but REFP needs to be a fixed value above REFN.

What is the standard way to get good quality bipolar reference voltage in a circuit?

You don't need one - a common/semi_rough -2.5 volt rail (that also feeds the op-amps) is fine for REFN. You need REFP to be precisely set above REFN - that's all.

• If REFN is not precisely set then the ground referenced readings would be wrong. For example, if REFN is -2.4V instead of -2.5V and I give 0V to ADC input, the ADC would measure it as -0.1V. So as I understand REFN does need to be set precisely. – Dojo Aug 15 '18 at 11:54
• Read the data sheet: Full-Scale Input Voltage (VIN= ADCINP–ADCINN) = ±1.06 VREF where VREF = VREFP–VREFN. It acquires the difference voltage between VREFP and VREFN by charging a capacitor and disengaging that cap from the VREF input signals and reapplying it to the successive-approximation part of the ADC hence VREFN does not need to be an accurate value. Look at fig 34. – Andy aka Aug 15 '18 at 12:00
• The signals are not referenced to VREFN. They are referenced to circuit ground. According to what you are saying if I apply ground referenced 0V to ADC input then the ADC would read 0V irrespective of whether VREFN is -2.4V or -2.5V with respect to ground. That would be absurd or there is some problem in my understanding. – Dojo Aug 15 '18 at 12:16
• That would be absurd or there is some problem in my understanding. It's the 2nd one - get your head around it. After all, why would they state VREF = VREFP-VREFN? And then go on to state that max VIN = ±1.06 VREF. Think about it and think what happens when you apply a voltage to a capacitor and then disconnect that capacitor and re-apply it to another disparate circuit - does the capacitor "remember" the voltage applied to both legs or does it remember the difference. – Andy aka Aug 15 '18 at 13:04

If the only purpose of the reference is to feed the ADC just use whatever crude (but it should be low noise) -2.5V source you like, Andy is right, a TL431 with a resistor will likely work just fine.

If you need a ground-referenced precision reference for other purposes in your circuit you can add a precision op-amp and a precision resistor network (eg. 10K+10K). That will give you double the precision reference voltage, so if you wanted 2.5V total you'd need a 1.25V reference. That's the best way if you are using a really top- drawer reference (eg. ovenized or extreme low drive).

Alternatively, you could simply use two shunt references in series, suitable for cheap low accuracy (eg. 0.1%) references such as LM4040, where you need the resolution of that converter but don't care much about accuracy or stability.

• I have trouble understanding how a crude -2.5V would do. I want to measure signal with respect to ground and not with respect to the crude -2.5V rail. So if the crude -2.5V is actually -2.5V + x mV would the ADC not measure voltages that are also offset by x mV? – Dojo Aug 15 '18 at 12:31
• In other words, the ADC has no Ground reference (other than in the digital section which it won't use for measurement) so how will it precisely measure ground referenced voltages unless the reference it uses is in turn precisely referenced to ground? – Dojo Aug 15 '18 at 12:42
• @Andy is correct - the voltage REFP-REFN is what has to be stable. The reference input is differential and the ADC input is differential (ADCINP-ADCINN). Unless you need the reference to be relative to analog ground you don't have to care. – Spehro Pefhany Aug 15 '18 at 12:44
• This ADC can be used in single ended mode also. Do I still not need a precises -2.5V for REFN? – Dojo Aug 15 '18 at 12:47
• You only need REFP-REFN to be precise (and appropriate for the range) and you need both REFP and REFN need to be within the 5V or so analog power supply (within 100mV). Those are the requirements. If you're driving it with an op-amp or two op-amps you need the op-amps to be able to drive at those voltages so you can't pick +/-2.5V with a +/-2.5V supply, but +/-1.25V might be okay with the correct op-amp. The example circuit drives the REFP near 0V so it's easy for the op-amp. – Spehro Pefhany Aug 15 '18 at 13:00