Ultra low-noise, short-term stable references

[Echo88]

I havent yet tested my idea that thermal bridges (e.g. Vishay THJP) may be able to reduce the thermal gradients across components like your current limiter or relays. If youre interested to test it i can send you some for free.

[dietert1]

Thanks for the thermal bridge hint.
Yes, i will definitely try other geometries. This kind of current limiter (600V, some mA, low thermal EMF) can be a very useful part. It could replace the input protection resistors in vintage multimeters like the Advantest R6581T or HP 3457A, thus reducing noise. I would also use it to protect the multiplexer inputs of ADS1256 or ADS1263 setups.
By the way, similar circuits are used in power electronics, e.g. https://www.electronicdesign.com/technologies/power/article/21808402/a-better-way-to-build-current-limiters-and-circuit-breakers.

Image shows a first 10 V LiFePo4 set with its thermistor embedded in ETP copper. The copper part was pretty difficult but probably serves the purpose. Maybe the second one will be nice. Meanwhile leakage currents dropped to 2 uA.

Regards, Dieter

[Echo88]

While waiting for the new pcbs i thought about the mentioned paper https://www.wellesu.com/10.1142/s021947750700388x and tried to see whether it could potentially provide an interesting path.
However, some questions came up reading it:
Taking the square root of the used Power spectral density gives the usual Noise Spectral density, right? V²/Hz -> nV/sqrt(Hz)
They only mention/calculate the white noise contribution of the used voltage reference, current setting resistor and the dynamic diode resistances.
Yet their calculations seem to correspond to their measured curves in Figure 7, while they are working clearly inside the 1/f noise region. Cant make sense of that.
How did they arrive at the mentioned 8 x 10^-17 V²/Hz spec for the used AD586?

Attached is a general first idea how it could be implemented.
ADR1399 7.05V *2 gain stage provides the necessary >10V reference voltage.
The transistor stage Q1/Q4 provides the current to the diodes, without loading the OPA189.
The original SSM2220 are obsolete, but THAT320 arrays or similar should also be suitable.
Then again i just opted to go for 12x ZTX951 transistors (lowest noise according to AoE3, about 1€/piece) attached to a temp controlled aluminium core pcb, which would sit inside the already temp controlled chamber mentioned a few posts earlier.
The 12x diode connected transistors provide the wanted 10Vout by adjusting the flowing current through them, the heater stage keeps the 10V stable by minimizing the diode TC.
The paper only mentioned the equivalent diode resistance and doesnt go into depth about 1/f noise contribution of diode connected transistors, so i just thought its best to use low noise (low rbb) transistors for this job.

[Kleinstein]

The heated diodes system is somewhat related to the idea we had here. In the article the temperaure control works as a kind of slow thermal low pass filter. I am not so sure if exactly this way is a good idea, as it may not scale so well to lower frequencies and it is easy to get temperature variations from the outside to add low frequency noise.  On the upside regulation of a thermal system is well known.  The case of Li cap (battery) and regulator for the voltage may actually be a bit on the slow side - especially for testing.

For the initial tests and learning one may think about a more classic super cap or even a large normal electrolytic capacitor. This won't work perfect, but could test of the correction loop works.

[dietert1]

The diode array comes with a TC of about -2 mV/K / 600 mV = -3300 ppm/K, while the LiFePo4 batteries i tested have around 12 ppm/K, a 278x advantage. In the paper they are looking above 0.1 Hz. Certainly ambient temperature variations are easier to suppress at short time scales.
I think the LiFePo4 averager can work at time scales of hours or days. And yes, the cells are terribly slow for development and tests. While i glued the cells into the copper block they already lost some charge. When i connected them again, i made a minor mistake that changed the state of the cells even more. One definitely needs an automatic controller to charge and balance the cells as fast as possible and make them finally track the zener reference. Otherwise the idea is useless.
The schematic is a possible starting point. It still needs some level shifters. As Kleinstein wrote above, resistor R13 won't be necessary, so optional or switchable. I have been using it to measure leakage current.

Regards, Dieter

Edit: The schematic isn't more than a sketch. The amplifier can't feed the 10R charger resistor. And there should be a hardware protection when reaching 3x 3.6V = 11 V total voltage for whatever reason. And the nullmeter needs its own battery tap in order to measure battery voltage independent of charger current and during output off.

[Kleinstein]

I think the bancing part should have a way to turn it off all the way (e.g. via the FETs that are there for the protection anyway). Just switching the OP-amps to nominal zero current may not be good enough.  Once balanced and not using much charge / discharge it should be OK for the nex years, possibly the life of the cells.

The diode array has the high TC, but this high TC is also the way to do the adjustment / regulation. So they kind of need at least some TC to get regulation. The speed of the regulation part sets the lower limit for the useful frequency. This means that slow regulaton would be wanted to also get to lower frequencies. The analog regulator used with the diode stack works OK for relatively fast regulation, but gets more tricky with even lower speed. One kind of gets the RC leakage problem at the regulator instead of directly using it for the voltage.  At some frequency there should be a cross over where digital regulation can be better. Digital has no issue with integrator drift from leaking / bias.

[dietert1]

Meanwhile i made another current limiter with 2x BSP135, this time with a 150R resistor to get about 5 instead of 8 mA. Total resistance is about 200R. Also i turned off an external hard disk that had been running about 10 cm away from the thermal EMF test location. This time thermal EMF came out to be 0.5 nV for the low thermal short, 2.0 nV for the PTC 20R and 5.2 nV for the new BSP135 limiter. Its geometry is "folded" to make it more compact. Both mosfets are glued to a small piece of fiber isolation sheet. The JFET limiter is in a short piece of aluminum tube.

Regards, Dieter

[branadic]

If you were located in Germany, I could have sent you a brass tube with glass feedthroughs and some parafine to make a hermetically sealed, oil-filled version of it, which should reduce thermal gradients between all components.

-branadic-

[Echo88]

Yeah, i missed the giant TC of the diode stack when posting and only recognized it later.
They dont mention the achieved oven stability or even the big TC problem at all, shame.
With about -3000ppm/K i dont think theres a single heater solution that would get it sufficiently close to useable, without an additional PTC component (which gives the standard Z-diode + diode ref topology).
I have two ideas apart from using Z-diodes for PTC compensation that ill test to see if they have merit.

[Kleinstein]

The diode stack solution would likely want a 2nd oven layer, to keep out external temperature variations and so that the critical inner regulator does not have to work that hard. The oven stability is not such an issue, as the regulation is from the target voltage, not actually the temperature.
It is still an interesting article to compare. One could be free to replace the diode stack with a chemical battery - though I don't think the batteries would like temperature variations that much.

[dietert1]

Meanwhile i got the setup in the image, with 2x AD587 references, 2x chargers similar to my sketch above and 2x copper-embedded LiFePo4 10 V battery packs, everything in a card box. The setup is wired to a K2700 with a 7706 MUX in order to log six charger currents and the two thermistors inside the copper.
Charger circuit is based on OPA4140, MUX4053 and NOMCT 8x 5K resistor array. For battery protection there are BS170 mosfets with Rdson about 1 or 2 Ohm. They limit the gain of constant voltage charging to about 0.67 uA for each uV of deviation. Separate mosfets for force/sense could speed that up somewhat as the battery ESR is less than 80 mOhm. Until now the batteries exhibit a time constant of about a day near the desired state, so it may take a week to descend to the desired precision.
Until now i can see the expected noise hierarchy of the three batteries in each pack. That means the noise in the charger currents is already dominated by the AD587 noise.
Next step is the constant humidity, temperature and pressure oven. Later one may want separate ovens for the two filters.

Regards, Dieter

[dietert1]

Here i have a log of charger operation for battery pack A. During the night, when room heaters are off and temperature is decreasing slowly, charger currents are in the low uA range. Heater turn on in the morning causes discharge due to positive battery voltage TC.
Meanwhile there is a design for the oven(s). I printed two brackets to put the battery packs inside a Weidmüller K41 enclosure that i have been using for hermetic oven tests. One or two 9x7 cm boards will also fit inside. That enclosure will run at 22 °C, using TEC temperature control.
Each battery pack received a small heater to use the copper as an inner oven, running at 23 °C, 1 °C above outer oven temperature. See Kleinsteins remark above about double ovenizing the diode array. Yesterday i made a linear temperature control for oven B. I got a log of battery pack temperature, still exposed to ambient temperature. Heater power is 15 V times the measured current, so about 45 mW on average. Without the heater, the temperature curve would look like the one of pack A, so the heater adds about 1 or 2 °C. Once the battery pack goes inside the outer oven, temperature variations will be much smaller. One may need a digital controller though, as the large integrator cap won't fit inside the outer oven.
While working on the temperature controller i screwed up the charging state of battery pack B. So i don't yet have a charger log. Some more work until things can go inside the hermetic enclosure...

Regards, Dieter

[dietert1]

Meanwhile i tuned the temperature control loop. The PID is now much faster, with the integrator cap reduced from 100 uF to 2,2 uF. The OPA140 now runs at a gain of 86 dB, retaining about 40 dB for negative feedback.
To my surprise the loop is still stable - probably due to the fact that there is about 1 cm in ETP copper from heater to thermistor. Temperature noise came down to about 100 uKpp, still exposed to ambient temperature change. During the night the controller runs into it's 6 mA current limit, when ambient temperature cools down more than 2.5 K below "oven" temperature.
This is to demonstrate the importance of PID controller tuning. A stable loop doesn't mean the controller is well designed.
The analog PID test wasn't a waste of time, although: 2x OPA140 at 15 V consume about 54 mW, while a ADS1256 + some PWM will be more like 40 mW and the ADS1256 does eight channels.

Regards, Dieter

[dietert1]

Meanwhile i made two prototypes of the outer oven. The hermetic enclosure shown above (Weidmüller K41) resides inside a box with 30 mm PUR sheets for thermal isolation. As the enclosure has two parts it got two TEC12706, one at the top, one at the bottom. Those are chinese TEC cooler kits including heat sinks and fans. The setup has four I2C sensors: A BME680 and two SHT45 inside the oven and another BME 680 for ambient monitoring. The two SHT45 sit next to the Peltier elements. The whole setup runs at TEC currents of about 100 to 200 mA total at an oven temperature of 22 °C.
Image shows the TEC driver. It implements two independent PID controllers based on the SHT45 readings and supports heating and cooling. The diagram shows an 11 hour log with an ambient temperature change of about 1.7 K. Oven temperature deviations are about 1 milliKelvin (standard deviation of 1 minute averages). The internal BME680 sensor not used for temperature control exhibits about 4 mK deviations.

One disadvantage of using an enclosure with rubber seal is that after closing the lid one gets a little higher pressure inside, due to the movement of the lid. One needs to choose a day with average or a little less than average air pressure.

Regards, Dieter

[dietert1]

@iMo: When i tried to use those bluepill modules from ebay, the USB CDC link did not work with the drivers provided by STM. I knew it works with STM32L433 MCUs. They have the same pinout as the F103, except one of the boot mode pins. I also modded the USB re-enumeration pull-up to be controlled by a GPIO, in order to get automatic re-enumeration during boot. That helps with debugging. The 48-pin MCU of the bluepill has the clock battery pin and i get a time tagged log data stream via USB to be recorded with a general purpose terminal program (no special application required).

By the way those ribbon cables i am using with epoxy and some additional aluminum parts to make a hermetic wire feed-through aren't air tight as their wire strands let air pass. One needs to seal all strands with epoxy to get constant pressure inside the enclosure. Those ribbon cables are copper and provide low EMF connections.

Regards, Dieter

[iMo]

@dietert1: ..thanks, I removed my post on the BP not to distract you .. I did the same with stm32F303CB (found several in my junkbox) and the upload of binaries and communication via usb works fine as well (and it is the CC silicon so twice the flash for free).

[miro123]

Hi Dieter.
 have one question. As far as I understood you are using SHT45 input to control temperature. You measurements shows deviation in mK and sub-mK range.

How do you achieve it since SHT45 resolution is 10mK?

I have build small oven in the past and I have used two NTCs in bridge configuration for resolution and PT1000 for accuracy. The more resolution I have the more easy control was. I my case I have dual control loops. Inner controlled the Peltier plate temperature and outer the DUT

[dietert1]

I tried using those digital sensors as i wanted to have all analog circuitry inside the oven. I have those nice glas NTCs inside the battery copper block used as internal ovens, using them with an analog PID. To use them for the outer oven, i need to put an ADC inside the oven, like the ADS1256 + ADG736 circuit i showed elsewhere.
The SHT45s worked better than expected. The image shows a random sample of the raw log data. For the PID i am using a median of three filter to suppress some of the sensor noise. The oven has a lot of thermal mass, so it doesn't matter that the noise filter introduces a timing jitter of 2 seconds. The D gain of the controller is null.
The diagram i showed above had 1 minute averages of those raw data. I think averaging is justified as the sensor noise dithers enough.
The device will have more temperature sensors, anyway, for example the heater currents of the two inner ovens. The device also has two LTZ1000 references to discipline the LiFePo4 batteries. Their temperature control part can be used as well. The control board has some space left, so one could even implement the two LTZ temperature controllers in the standard fashion. First i am trying to just tune the LTZs for zero TC near the 22 °C outer oven temperature.

Regards, Dieter