-
The LTZ1000 reference circuit needs a divider of some 1:12 to 1:15 to set the oven temperature. This are usually 2 of the most critical resistors. It is mainly about long term drift, not so much the instant TC, but a good TC is also wanted.
The required stability is somewhere in the 5 ppm range resulting in some 0.1 ppm drift contribution for the final reference voltage.
There is no need for a really fine adjustment, though a little adjustment range may be nice. Also linearity is not important.
There are a few possibly gotchas. So It is probably keep the discussion separate here and not the the long LTZ1000 thread.
The general setup would likely be something like:
1) A buffer for the 7 V ref voltage. Likely a relatively large buffer cap at the output to filter charge injection from the switch.
2) A SPDT switch to choose between 0 and 7 V
3) A resistor of some 100K and low pass filter.
4) A buffer for the output. Likely an AZ OP, as the filter is high impedance and drift matters.
5) Some filtering (against AZ OP artifacts) and resistance to get an output impedance comparable to the usual 1K:13 K divider.
The PWM signal would likely come from a small µC.
In addition one would likely need means to turn off the heater for the initial time until the DAC voltage has stabilized after power on.
Overall the demand is not so extreme, so chances are good it would work. It is more the question how much effort is needed.
Chances are one would first test the PWM DAC without the reference and only than if everything works use it would a LTZ1000.
There are other uses for a simple PWM DAC, like getting a stable 5 V , 2.5 V or 1 V ref. output. The 7 V to 10 V step is a little different, as the PWM DAC would be in the feedback. Still some point may apply. There was a thread on a 7 to 10 V step with PWM DAC:
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg2092108/#msg2092108
One point open to discussion is whether it is worth compensating part of the ripple with an added inverted AC coupled signal. Not sure how much it helps. -
Now, that's a novel idea, I like it.
For the filtering, I have a suggestion. I made a similar PWM divider a few years ago for a project with the LTZ1000, for background calibration of 5V input ADCs.
The Datron 4912 uses a PWM divider, and has a quite unique filter topology. The Opamps marked as U207, U208, U209 are doing some out of DC path filtering. I cannot recall the name of this topology from the top of my head. So I used this method to filter the PWMed 7V reference, it worked surprisingly well. The JFETs on the input might even be unnecessary, or use a JFET input opamp. I've used regular NE5532, maybe 5534, and couldn't see too much reason for the JFETs. C210 and C216 shouldn't be ceramic or electrolitic cap, Film cap, I used polypropylene. Microphonics, and leakage current. Generally, the more capacitors you replace with film, the lower noise you can reach, at least in theory. Somewhere there will be diminishing returns. Depending on how fast your PWM is, you might not even need 7 pole filtering, but it's not expensive in terms of components to include it.
My circuit was compared to a 3458A for a few months and it worked quite well. I think it can be improved, but this was for work, and management's position was: " OK, done, next project" -
I have had similar ideas for many years. Last year I even drew a schematic diagram, but for some reason, I stopped at the PCB layout stage. Maybe once I finish the task at hand, I will continue to try to finish it.
-
I have had similar ideas for many years. Last year I even drew a schematic diagram, but for some reason, I stopped at the PCB layout stage. Maybe once I finish the task at hand, I will continue to try to finish it.
The circuit still has a slightly weak point: The ADG5236 is a rather high resistance switch (some 200 Ohms). The low charge injection help in not loading the ouput very much, but this also give more nonlinearity, as the difference in switch resistance for the positive and negative side give a small nonlinear contribution to the output ratio.
My current favorite for the switch would be more like DG419L (some 10 Ohms, though 25 pC or so of charge injection).
I have not understood the ouput part with the LTC1043 - but that is more like specific for that circuit, no longer part of the PWM DAC.
The OP(s) for the filter can contribute to the noise and as the circuit is more like high impedance the current noise of am NE5532 or similar would be a problem. Moden JFET OPs would be OK, no need for highest DC performance like the OPA140. The problem is more that one may need a negative supply or a extra filtered virtual ground.
For the reference use there is likely no need for fast settling - so the 3 rd oder filter should be good enough. -
Given the PWM runs at a constant (low jitter) freqency, I would look into using a sample and hold circuit instead of a continous time filter. Only a a little filtering would be needed after the S/H...
Cheers
Andreas -
You can think of my schematic as just a concept. All the components I use are those I can find 'in stock' at home. The LTC1043 is used here as a subtractor to try to cancel out any 'common mode' interference between two PWM-DACs such as temperature drift of analog switches, etc.
-
Considering that the transition time of analog switches variety with temperature, even using a low jitter clock source with a constant frequency cannot guarantee the stability of the PWM output. This is why my design is differential.
-
No idea about the target specs, but I think this principle should work, too:
https://www.eevblog.com/forum/projects/fun-circuit-to-play-with/
Is "PWM DAC" a dedicated term for a specific DAC topology, or is it just the ad hoc name for a PWM voltage source followed by an integrator? -
The PWM DAC is usually the principle of PWM to set the voltage from the on / off ratio and than a low pass filter. There are different options for the low pass filter and buffer.
In the link to the LTZ1000 thread, there is a special version to get half the voltage. This special case allows to use the 2 resistors and this way get away with less filtering. The same principle is however more or less limited to a few special rations like 1/2 , 1/3, 1/4 ,as more an more resistors would be needed. It is a nice circuit in that there are "matched" resistors inlcuded, but this is mainly to reduce the ripply amplitude. The output is the average over the two cases. So the 2 resistors are swapped an averaged. So the resistor error would only enter to a higher order. However the circuit gets large with 14 resistors / ouputs.
I had a quite look at the ripple reducetion with the AC coupled inverted signal. It acts like adding another filter order. So the capacitor is not really more effective than as part of the filter. It may still have some merits in not relying on the OPs qualtity / GBW. So it is not a great improvement (as claimed in the EE article), but also not totally stupid.
The GBW and loop gain limit the ultimate attenuation for the usual filter with the capacitive coupled OP (e.g. like the circuit shown by Gamalot). So a fast JFET OP there makes absolute sense. Somthing like 120-140 dB ripple rejection seems reasonable - so starting with some 1 V this would be residual ripple in the 0.1 - 1 µV range. This should be just acceptable. -
The circuit from https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg2092108/#msg2092108 is using two SPST switches.
The circuit from https://www.eevblog.com/forum/projects/fun-circuit-to-play-with/ is using DPDT switches, thus making the circuit symmetric. Same idea can be used to implement a digitally adjustable voltage divider with the same R1==R2 at all times, by varying only the "PWM" (the on/off timing, which is easy to control digitally).
The advantage of using DPDT will be its symmetry, when compared with the SPST.
Another advantage (when compared with classical fixed resistive dividers) is that it should be easier to get a single well matched pair (over temperature) of two resistors of the same value, R1==R2, than it would be to get a precise and stable divider where one resistor is many times bigger than the other. -
The divide by two FF voltage divider can be easily modified for other voltage ratios (ripple is higher tho) but as mentioned limited to a practical few integer ratios. Long ago prior to the divide by two technique we developed a method to create a voltage division ratio of M/N, where M and N are arbitrary long digital words as required, which could be consider PWM and much later patented (4958159). During this time period of the 70s and early 80s we employed MOS & CMOS for many analog, or analog type functions and having an in-house fabs help fuel these endeavors. The 1st CMOS capacitive ratio DAC and ADC came out of these labs, as well as other interesting technology like the 1st camera Auto-Focus technology and some unique and interesting CCD devices.
CMOS is such an interesting and useful technology, with the many available ICs not much need to develop your own functions unless you are doing full custom IC design. However, back in the day none of this was available and you had to roll your own which had the side benefit of having to learn the semiconductor technology involved, rather than just putting some components around a COTS part
Best, -
One can consider the divide by 2 circuit as a special case of a 2 phase PWM DAC. With a 50:50 PWM ratio the phases cancel out quite good and thus less need for filtering. One could still use a similar circuit also for other PWM ratios (so no simple flipflop but finer stepped PWM with 180 deg phase shift). One would than need more filtering, but still have a slight advantage in starting with less ripple.
The resistor ratio only needs to be accurate to reduce the ripple. For the final output the more critical point is the R_out of the Flipflop for the 2 sides. 2 resistors means one can start with higher resistor values.
For use inside a chip, a classical charge pump type divider may be more suitable and also lower power. -
For a first test I though about copying the Datron 4910. (Stability(±1°C) 1Year 2ppm | TC 0.12ppm/°C | Noise 0.01-2Hz 0.1ppm). The spec of the 4910 are for the complete reference, hence I expect the PWM divider to be even better. For the Opamps in the 7 Pole filter I thought about LTC6240. As Input offset Voltage shouldnt matter in a AC coupled filter? For the switch maybe the ADG1419?
In order not to get further offtopic in main LTZ1000 thread I will reply to Dr Frank here:QuoteI think that such questions should always be started from the (engineering) requirements and from design goals, not backwards from an anticipated solution... as this would be better and professionell engineering practice. Consequently, all the replies to your three questions so far did not give any concrete answer.
Datron 4910 specs make a PWM divider seem a possible solution. I was not aware of any reasonable solution for Resistors. Vishay b.m.f. resistor are at least 13€ for 1kOhm and 13kOhm isnt available. Parts for a PWM Divider could be easily sourced and <20€ seems possible. My idea about tempreature adjustability is to be able to initially pick the perfect tempreature.
So, to answer your questions concretely: I think that a PWM would be a very bad approach in all of your categories.
At first stability: what stability requirements do you mean, like short term (noise), temperature, long term drift, or susceptibility to E.M.C. , pressure, humidity, vibration?
These 2 resistors theoretically contribute only by 1/75 .. 1/100 of their own stability figures, so the LTZ chip itself has paramount influence on mostly all these parameters; that has been measured / demonstrated in this lengthy thread many times. Practically, the residual T.C. of up to 0.3ppm/K, despite e.g. T.C. matching of the resistors, is not yet understood here, and is obviously not relevant, as the overall T.C. can easily be trimmed, down to <0.02ppm/K.
As the oven regulation and in turn the reference voltage is very sensitive to external noise, keyword voltage dips, for a PWM solution you'd need a very elaborate filtering, which would probably yield high cost and in the end worse noise figures than the passive divider solution.
Concerning cost, these Vishay hermetically sealed, oil filled, luxury resistors are not really necessary; much cheaper ones will do the job as well. These only have to fulfill certain minimum technical requirements regarding e.g. T.C. and timely drift.
Please, first make yourself aware of the BOM cost of these cheaper solutions, they are in the ballpark of < 20€ for two b.m.f. resistors, or even much less for other technologies, like e.g. TiN has presented on the discussion about the clone reference of the 34470A.
Concerning oven set point change, that question makes no sense at all to me.
Instead, what would be the purpose/application, or your design goal for that feature?
If you would start from the requirement "timely stability", you would know, that the lower the temperature, the better this parameter.
50°C oven temperature gives about -0.8ppm/year, each 10°C more will double that number. Therefore, simply chose a fixed, lowest possible oven temperature.
Changing the oven temperature will change the reference voltage by about +50ppm/K, which would violate the overall stability requirement as such.
If your feature goal would be to have an easy and cheap(er) possibility to make thermal cycling like in the Pickering patent, well that's also already done in the Fluke 7000 devices, so no need to re-invent the wheel by another complicated solution.
From all these aspects, I can not imagine at all, what would be the benefit of a PWM solution, other than over-engineering an already solved problem
, just like this other voodoo stuff here, as PCB meander cuts, fancy resistors, ceramic and multi layer PCBs and so on.
Frank -
Just a quick comment on that: once understood how to decrease the initial unovenized 50 ppm/K of the zener itself, the requirements for the oven temperature setting resistors becomes less critical. The solution costs <9€ plus shipping and is demonstrated in W/F/M7000.
-branadic- -
The ADG1419 has rather high capacitance , charge injection and low resistance. So it would need quite some filtering for the input side.
The switch is kind of a compromise between one resistance (can nonlinearity and a little drift) and charge injection / capacitance (needs more filtering and may cause more EMI). The relevant part is the difference in switch resistance between H and L position relative to the resistor for the filtering. The difference is expected to drift would something like some 5000 ppm/K and thus should be smaller than the resistor by a factor of 10000 or so. So with 100 K at the filter the difference should be smaller than some 10 Ohms.
So I don't see a need to have 2 Ohm total resistance and thus a difference in the 0.2 Ohms range. So I would consider a slightly higher resistance switch.
The LTC6240 looks OK for the filter and one could also use the dual version. DC precision is not needed, but to some extend the low frequency noise from the filter transition region (e.g. 10 Hz range) can matter effect the result. I don't hink one would really need a 7 pole filter, the 5 pole version and maybe even the 3 pole one could be sufficient with a little larger caps. For the reference a little longer settling should be acceptable.
Without a negative supply one would also need a filtered "ground" for the OPs. This may not need an extra buffer, as it is only going to the OPs inputs.
The butter for the input side would not be so critical with DC drift, as it is relative to 7 V. The more tricky part is the buffer for the output, as it could be relative to the smaller / divided votlage.
Reducing the unheated TC would of cause be a much easier solution, if it does not come with an increased sensitivity of other resistors. The methods I know so far tend to add errors from added resistors. So it only shifts the problem to other resitors. -
Quote
Reducing the unheated TC would of cause be a much easier solution, if it does not come with an increased sensitivity of other resistors. The methods I know so far tend to add errors from added resistors. So it only shifts the problem to other resitors.
This is based on a misconception due to the lag of actually building and testing an LTZ reference, I guess. There is no shift in problems at all and I can say that for the reason that I've actually build it already and have the example of how it works in front of me, even with my two left F7000 references next to it. But keep going on raising the effort on something that's not worth it.
-branadic- -
The spec of the 4910 are for the complete reference, hence I expect the PWM divider to be even better.
Datron 4910 specs make a PWM divider seem a possible solution.
Hello,
when looking at the Datron a bit closer then you will see that the 7 to 10V transfer is mainly done by a 3.01 K and a 8.25 K resistor.
The PWM divider (and so the noise + stability of it) has only less than 8% influence on the output voltage to correct the stray of the output voltage tolerance of the LTZ1000.
With best regards
Andreas
-
This is true for the 7 to 10V conversion, but not for the 10 to 1V conversion. "In addition the Datron 4910 has two separate PWM dividers which divide down the average 10V output to provide 1.018V and 1V outputs" Users Handbook 1-7
1V is much closer to 0.5V hence I listed the 1.xV outputs specs above. -
The required stability is somewhere in the 5 ppm range resulting in some 0.1 ppm drift contribution for the final reference voltage.
Can you elaborate how you calculated this? If the TC of the unheated Zener is about 50ppm, then a 5ppm change (regardless of over time or temperature) on the oven-temperature setting resistor will (assuming temperature changes linear with the value of the resistor) yield a change of 5ppm*50ppm = 0.00025ppm, won't it? -
The attenuation factor for the R3/R4 ration is around 70 to maybe 100. So 5 ppm change in the set point voltage would cause something like 0.07 ppm change in the referene voltage. This should be just smaller than the natural drift that is possible for the chip itself. So the 5 ppm are a kind of good enough limit. No real need to get much better. For a PWM divider this range may be within reach - at least that is my hope. An the data from the Datron 4910 seem to support this.
A slightly higher dirft (e.g. with resistors) could be still OK if one does not have extreme requirement. So for resistors I would consider something like 20 ppm as good, for a lower cost version also 100 or maybe 200 ppm. For the TC this would be something like 10-20 ppm/K. At least the TC part is relatively easy to reach with not really expensive resstors (e.g. 10 ppm/K thin film). In addition there can be an additional adjustment for the residual TC via R9 ( ~400K). So no need to build the PWM divider just for the TC.
The tricky part is however the long term drift and possible humidity effects, which are hard to predict as the datasheets rarely give good data on this.
For the lower end just a good resistor divider is OK. The PWM divider it more an possible alternative to those who want more than the S102.
Some magic to improve on the attenuation factor would of cause be much easier. The more classic (shown in the datasheet) way is a resisor in series to the zener, but this makes the resisor R1 more critical, so one does not really gain that much. A resistor (e.g. 10 K range) parallel to the zener and than a slightly low value for R1 may work, but I have not seen data or calculations on this. If it works - great, but it looks to simple. It kind of has to be simple to be stable.
A PWM divider may also be useful for other things (like a 1 V or 2.5 V output), so even if not useful for the set point it is not totally lost effort. A 50% divider (e.g. for a 3.5 V refrenence to a SD ADC chip) can probably use a simpler circuit, like the flip-flop idea, that starts with much lower ripple.
edit:
A crude estimate shows, that the parallel resistor idea is likely not that practical: With some R_Z = 5-10 Ohms differential resistance for the zener, the change in curren with temperature contributes some R_Z/120 * -2 mV/K to the TC. That would be some -100 to-200 µV/K. To get a significant improvement on the overall TC one would need about twice that effect and thus about half the value for R1 and thus quite some extra current (e.g. 5 mA) from the parallel resistor. This would than make R1 and the extra resistor quite critical. So it could be similar to the series resistor way, just with more power consumption and thus less attractive. -
This thread made me curious about LTZ1000 TC so I gathered some data.
Uncorrected TC of LTZ1000 = 33-37 PPM/C.
Corrected TC with 17 ohm resistor = around 0 PPM/C.
See attachment for charts. Top line in each series is the heat up (using internal heater), bottom line is the cool down.
My next step is probably to combine this TC correction with the usual heater regulation circuit, then put 10k/etc in parallel with 17 ohm resistor and get a measure of the sensitivity of reference output due to this TC correction resistor.
-
The sensitivity to R1 is relatively easy to calculate, as the current is quite constant.
The tricky part it that the sensitity to R2 also changes and gets larger. As a crude estimate about 3 times, as R1 is about twice the differential resistance if the zener.