Easy DIY 5.5 Digit DVM + Volt Ref./Cal. (LTC2400+LTC6655 / SPI uC / Arduino)

[Kleinstein]

It is possible to add a temperature sensor and compensate for temperature effects. This can be more than just the reference alone, but also amplifier gain or a divider that can be temperature dependent. The problem is however that the references in a plastic case have more than just a simple temperature effect, but usually also some humidity effect and possibly an effect of the past temperature profile / hysteresis. A NTC can help, but it is not a magic solution for stability. It also needs some time and effort to measurent the correction function.

A 5.5. digit meter is usually no enough to really check a good bandgap reference - this is more something where one wants a good 6 digit meter with a reference that is considerably better than the DUT.

[Andreas]

Is it a practice to monitor voltage reference temperature and compensate for the temperature variation known ppm in the MCU or later on the PC?

It gives a improvement of ~factor 10.
But it needs some pre-requisites:
The reference must be repeatable (no hysteresis (<1ppm)) over the temperature range of interest (e.g. 10-40 deg C)
To get no additional noise from temperature quantisation the temperature steps need to have a good resolution (e.g. 0.1 deg C with a optimized NTC and 10 Bit ADC).
Further you have to select references with < 1ppm/K in the interesting temperature range to keep compensation steps below 0.1 ppm.

The best results I have with AD586LQ references (buried zener in hermetically CERDIP8 package).
Usually I get 2-3 devices from a batch of 10 which meet my requirements.

The other possibility would be to put a reference in a oven. (but this prohibits battery operation).

with best regards

Andreas

[jorgemef]

I purchased 5 AD586LQ on Aliexpress I was considering to use instead of ADR4050. I measure ~35uVPP noise. They seem popcorning if I am not wrong. Pin8 has 3.5V so they seem genuine but popcorning. Measurements where performed in breadboard with 1uF poliester cap in pin8. At least they have good TCR.
Alternatively I will use LM399H like circuit shown in attachement as still have a pair of 5K NOMC from when I built the voltage reference.

[Andreas]

I measure ~35uVPP noise.

noise without bandwidth is useless.
so how did you measure it (how many NPLC + measurement rate of your K2000?)
Did you put a 100 nF capacitor across the output? (could help if there are any EMI sources, but usually the dents go in negative direction).

I hope you did not use a switchmode supply for the references; otherwise I fear these parts are defective.

35uVpp is way too much for a DMM measurement.
By the way: the 1uF capacitor at the NR pin helps only to reduce wideband noise. (above 10 Hz).
Below 10 Hz it is nearly useless.

with best regards

Andreas

[Andreas]

Hello,

as comparison a quick + dirty measurement of a AD586LQ of my ADC28 with K2000 (10 NPLC ~600ms per measurement) battery supplied.
~7uVpp of AD586 + K2000 noise.

with best regards

Andreas

[jorgemef]

Hello,
I was measuring with 1 NPLC 1 sample per second.

Was not aware that SMPS could damage it. Is this applicable even if LDO used after it or some filtering? So far I used 5S battery with LDO after it but was planning to go SMPS with LDO and filtering to reduce number of batteries required.

Added the 100nF on the output plus sampled at 10NPLC with 0.6s sampling period. Is more comparable with yours. Not sure what factor it may add by being a breadboard (not soldered).

Cheers,
Jorge