AD717[2567]

[jordens]

Quick analysis of pin- and software-compatible alternative ADCs.

tl;dr: The AD7175 is an interesting option over the AD7172 for fast loops (faster than 50 Hz data rate). Let's stick with the ad7172 for now.

• 2,5,6,7 are all pin-compatible and register compatible. This is stressend by the common ADI noOS driver.
• AD7175 is very similar to the 2, (almost) same base rms noise, but higher base sample rate (250 kS/s vs 31 kS/s, factor 8).
• AD7175 has 8 times higher clock/modulator frequency and about 8 times the current consumption.
• AD7176 is the AD7175 without buffers
• AD7177 has 32 bits but the same in-band noise as the AD7175. It's useful if the data rate is under 400 Hz compared to the 5.
• The 5's higher base sample rate (before the filters) at the same base quantization noise directly leads to less voltage noise (about sqrt(8)) at the same output sample rate.
• Once the filtered voltage noise drops below the filtered (24 bit) quantization, the later dominates. That's at 400 Hz and 60 Hz for the 5 and 2 respectively.
• Tempco is dominated by the 5ppm/K reference resistors. There are 2ppm/K resistors available from the same series. Going to <1 ppm/K reference resistors (through hole) and six wire measurement (remote excitation) or a <0.5 ppm/K current source and four wire measurement would start to make the ADC the dominant part. Both alternative setups are possible with the current design.
• The ADCs have 0.2 (AD7172) and 0.4 ppm/K (AD7175) typ. But looking at the actual distribution of those drifts they are indistinguishable to me. Unclear what the measurement accuracy is and the device drift might even be indistinguishable from zero.
• 10k||(5k+5k) Johnson noise in 500 Hz is 0.2 µV. 24 bit quantization is 0.17 µV rms. The two reference resistors additionally contribute 0.06 µV/V current noise (in one decade bandwidth, 1/f) each. We run them at 1.25 V.
• NTCs also have current noise. Maybe even 1 µV/V (20 µK) in 0.01-1 Hz depending on the design (Ryger et al. Rev. Sci. Instrum. 88, 024707 (2017). We run at 2.5 V.
• Thermocouple effects between nominally same materials (Cu-Cu) are 0.1-0.5 µV/C. Can be much worse on oxidized connections.
• 50 Hz should be the sweet spot for the AD7172 (maxes out 24 bit and 50 Hz notch with minimal latency), 400 Hz for the AD7175
• The current design has a 160 Hz RC filter (220n, 10k||(5k+5k)) before the ADC limiting the bandwidth, noise, aliasing (1 MHz modulator), rf pickup, but also contribute latency.
• A 6 by 1.38 mil Cu trace has 0.035 Ohm/cm and 4e-3/K tempco. Compared to the 5k <5 ppm/K reference resistor, the trace resistance should be less than 2 Ohm to not contribute.
• Thermal gradients between effective temperature of interest and sensor are typically larger than 1 mK (50 µV, 10 µHz-0.1 Hz, after settling, with a decently designed multi-layer actively shielded commercial ULE cavity).
• For faster loops with closely coupled TECs, thermal mass, and sensors, or other types of sensors and measurements, the lower noise of the AD7175 is superior.

Current Thermostat noise is 0.5 µV 2.5 LSB at 3.3 V ref and 27 S/s (0.45 µV datasheet), 5 µV/K (100 ppm) tempco.

[gkasprow]

For me is fine, it would be good to use it in both Thermostat designs.

[jordens]

Agreed. These were just my notes what would need to be done to get to faster and even more accurate/stable/low-noise control loops.