LM399 based 10 V reference

[fmaimon]

Zeners are current driven objects. The input cuurent noise is in the femto ampere range for an ad706 and i. The picoamp range fro the op177. Furthermore the ad706 is dc stable. This opamp is built for dc operation. The input bias current is also stable over temperature, that of an op177 is not.

This drift in n put bias current willl shift the operating point of the zener ever so slightly.

Could you elaborate a bit more? It is obvious from the datasheets that the AD706 has less drift than the OPA177 (and the OP177, which seems to be exactly the same). Codeboy's calculations shows exactly that with the 177 being about 4 times worse than the 706. What is not obvious is what makes an opamp DC stable? How can I check if an opamp is DC stable and/or built for dc operation by reading the datasheet? AD706's datasheet mentions "High DC Precision" in it's features, but I tend to ignore this "features" and try to find what I want from the specs.

[fmaimon]

BTW: Resistors can be bought over here http://www.rhopointcomponents.com/components/resistors/precision-through-hole.html

Nice find! I've just quoted some resistors. Let's check their prices.

[oilburner]

Thank you Branadic, I got this book and it already payed for itself...  I was looking for a MOSFET that could be used for a linear voltage to
current converter and saw the circuit on page 227 that used an ON Semi  MTP3055V.  The original MOSFET I picked was a fast switcher
type that was not as stable with temperature.

Greg
=============
BTW: Current Sources & Voltage References by Linden T. Harrison is very nice "evening-reading".

[branadic]

Thank you Branadic, I got this book and it already payed for itself...

Pretty good stuff isn't it? I still read serveral chapters from time to time.

[Andreas]

As I said in the other post, I don't really care of the accuracy. I will trim it elsewhere in the power supply. I just want it stable.

You should consider to stabilize the heater voltage of the LM399 instead of using a (unstabilized?) 18V supply. I have observed 0.5 - 1.5ppm/V output voltage change between 10-16V heater voltage.

Further the LM399 changes output voltage with orientation. Upside down or legs sideways will change output voltage by some ppm's. Current consumption of the heater is a minimum with upside down orientation if you want to save battery power.

With best regards

Andreas

[quantumvolt]

I have a LM399 I will start to play with. In addition to the "staged" version schematic (Linear regulator 15 V, Current source (Resistor or IC),  LM399 Ref., Adjustable Gain Amplifier/Buffer) there are two circuits that uses the reference voltage for current/self bias.

The two circuits discussed in this tread so far seem to be the self bias version and the staged version. What do you think about the "combined voltage/current regulator" version vs. the "self bias" version (compared to the simpler staged version)?

(Edit for pictures)


"Combined"



"Self Bias"



"Staged"

 

[Andreas]

The two circuits discussed in this tread so far seem to be the self bias version and the staged version. What do you think about the "combined voltage/current regulator" version vs. the "self bias" version (compared to the simpler staged version)?

Hello,

For the first: you cannot compare circuits which have a higher output voltage than the reference (10V) with a circuit for a lower output voltage (1.018V).

Further you will have to calculate the error budget of the different circuits by yourself. And perhaps publish the results here.

So from my side just some thoughts to the different circuits:

For the combined solution:
+ regulated heater supply (otherwise 0.5 - 1.5 ppm/V drift for input voltage change)
- the load/temperature dependent adjust current influences the zener current
  fortunately the zener itself has a very low impedance of about 0.5 Ohms
  but a 10uA change (e.g. 25-75degrees chip temperature) will result in 5uV (0.7ppm) at the zener.
- the LTC1049 has a relative large noise (2uVpp) against the 7:1 divided noise of the zener (4.2uVpp/7 = 0.6uVpp)
- the LTC1049 can only deliver around 0.5-1 mA at the output. For a buffered reference typically 10 mA are usual.
I would use a LTC2057 instead of the LTC1049.
If you have capacitive loads you will have to do further measures at the output to get it stable.

Self Bias:
- unregulated heater supply (will save battery power but give additional unstability)
+ zener current has lower noise than with other solutions
* the zener current stability depends only on the voltage divider + offset drift and the 5K resistor drift.

Staged:
- unregulated heater supply
- unregulated zener current
- unstable trimming scheme
for this circuit I would use a low noise 14-15V voltage regulator.
up to 20V input voltage a LTC1763 could be used.

With best regards

Andreas

[quantumvolt]

Thank you.

I understand that noise and drift in the output will be different for 6.95 V divided down / buffered to 1.018 V and 6.95 V amplified up to 10 V. Also the quality of the resistors and ICs will determine the end result. But the resistor arrangement around the op amp on the output as either a voltage divider + voltage follower or a non inverting DC amplifier does not change the operating modus of the circuit.

I was more interested in having opinions on the principles for the 3 solutions. The staged one is "simple" with no "backward interdependence", and the combined solution has "complexity" by making the linear regulator dependent on reference voltage and then letting it feed both heater, reference and op amp.

If no one thinks that any principle is much better then the other, I think I prefer the staged version. Very easy to rig up and set in long time measurement, and then test out refinements by bettering the DC supply / battery, use 18 V regulator for the heater and then step down to 15 V for the reference, use a dedicated current source in stead of a resistor, get better low tempco resistors for the amplifier, try different op amps, put the whole board in a box that is heated to some 40-50 degrees, etc. The simple design makes it easier to see which experimental change produces which effect where ... Imo.

[Andreas]

But the resistor arrangement around the op amp on the output as either a voltage divider + voltage follower or a non inverting DC amplifier does not change the operating modus of the circuit.

I was more interested in having opinions on the principles for the 3 solutions.

The resistor arrangement has the most influence on the stability of the output voltage. I recommend highest quality precision resistors (either precision wire wound or Vishay metal foil) for this. And yes: they will cost more than the reference itself.

If you look at the schematics of precision DMMs like the HP 34401A (document 34401-90013.pdf) you will always find the self bias circuit to generate the +10V reference voltage for the ADC.
Of course they use a hermetically tight (long term stable) hybrid resistor string for the voltage divider.

With best regards

Andreas

[quantumvolt]

Thanks again. I guess this is the Agilent 34401A reference schematic:




That makes my choice. I have bought a LTC2400 ADC cheap from fleabay and want to try different chip references at 5 V but also make a 10 V based on the LM399. I hope to make a single range voltmeter with around 0.01% / 1mV accuracy for 10 V measurement.

I will try to use OP177 (have one already) and hand-measured-tempco-selected-matched standard metal film resistors (I have some bands of 1000 pcs. to pick from ). If that does not work, I have some Vishay 0.01% 2.5 ppm/degree that can be used in series/parallel coarse value and then a trim. If necessary I can put a a temperature sensor and a correction table in the microprocessor that will display the measured (corrected) value. All I need is a voltmeter that emulates my 34401A to 4 1/2 digits or better.

If everything else fails, I will buy an AD706 or similar and some decent resistors with the right values.

Is the heater running on 30 V, or have I misunderstood something?

(From the manual: "The instrument precision voltage reference is U403. Resistor R409
provides a stable bias current for the reference zener diode. R408 and
CR404 provide a bias to assure that the reference zener biases to +7 V
during power up.")

[quantumvolt]

To answer my own question: Yes, the LM399 heater resistor (pin 3 and 4 in U403) in the schematic over does run at 30 V. In this datasheet is another example http://cds.linear.com/docs/en/datasheet/1001fb.pdf. With heater voltages from 12-30 V and power (U^2)/R this means that the chip operates on a wide power/temperature range.

[Andreas]

Hello quantumvolt,

If you have a negative voltage supply, you should use it for the heater. According to my measurements
the power supply rejection ratio of the heater voltage is better the higher the voltage is.
And since heater power is a constant, a higher voltage gives a lower supply current in steady state.

0.01 % longterm stability is a challenge with normal film resistors. They will drift already during soldering up to around  0.1 - 1%. I use capacitive dividers (LTC1043) to divide the 10V input voltage to the input range 0..5V of the LTC2400.
With a good 5V voltage reference and a temperature compensation you can reach below 20ppm/year stability after some pre-ageing.

I think that the OP177 is a better choice for the circuit than the AD706. The important key features are open loop gain, offset voltage drift over temperature and time.
Of course newer chopper OPs like the LTC2057 could get better results.

With best regards

Andreas

[quantumvolt]

Very good. I have +- 15 V. I did not know that the heater was not "purely resistive".

Due to the odd values of my bands of 1% metal film surplus resistors, the "precision resistors" will be made up from 10-15 resistors in parallel/series where I begin with the large values matching positive and negative tempcos and add smaller and smaller corrections before ending up in a small trim added as "parallel center tap". When I build this "composite resistor" I can check value and tempco after soldering all the resistors together on very small PCBs. The final soldering heat will be on leads added to the PCB's afterwards. I'll do it for fun - if it doesn't work, I'll get a decent divider.

Thanks again. I'll go for the self bias circuit principle with +- 15 V heater / OP177 amplifier, and will be prepared that I may have to get a better ratio divider for amplification to 10 V as a general reference and/or division from 6.95 V to 5 V in order to reference the ADC directly.

[branadic]

I did not know that the heater was not "purely resistive".

Have a look at this:

http://i007.radikal.ru/1202/ad/4d39ad65f42f.jpg

BTW: I was asked for the marking code of the crystal heater. Attached is an image of the board, all parts can be identified by this. But I'm the opinion that it makes no sense to duplicate this board as the circuit is assembled on a ceramic substrate and you won't use FR4.

[babysitter]

About 12EUR in germany, this heater. Nice little beast.

[branadic]

I think a very good design consist of a cascaded LM317 arrangement feeding the heater and a temperature compensated constant current source for the zener (e.g. LM334). Together with a chopper amp such as the named LTC2057 and some pi filter to limit the whiteband noise the design could be completed. I would agree that its worth setting the ouput voltage to 10V nominal and divde them with LTC1043 to 5V or other values.

[Andreas]

I think a very good design consist of a cascaded LM317 arrangement feeding the heater and a temperature compensated constant current source for the zener (e.g. LM334). Together with a chopper amp such as the named LTC2057 and some pi filter to limit the whiteband noise the design could be completed. I would agree that its worth setting the ouput voltage to 10V nominal and divde them with LTC1043 to 5V or other values.

What do you mean with cascaded arrangement? (schematics).

I did the LM334 solution for my first 2 LM399 References. It is possible but very time consuming to get the tempco of the current source to below 30ppm. I started with the datasheet suggestion of 10:1 resistors (137/1370 Ohms) and ended in a tempco of 305 ppm/K (3600 ppm over a 12 degree span). After some changes to the resistors I ended with 145+1200 Ohms for the one reference and 147+1200 Ohms for the other with a tempco of -30ppm/K.
So its not easy to compensate the +3333ppm/K tempco of the LM334 with a -2mV/K tempco of a diode.

So in the meantime I'm convinced that using good 15ppm/K metal film resistors (RC55Y or RN73) together with the self biased design will reach the same/better result with less effort.

With best regards

Andreas

[branadic]

What do you mean with cascaded arrangement? (schematics).

"LM317+cascaded"

http://www.acoustica.org.uk/t/3pin_reg_notes4.html

Just one of many possible pages. This will improve PSRR, you maybe remember this thread:

http://www.mikrocontroller.net/topic/295213#3148897

[Andreas]

What do you mean with cascaded arrangement? (schematics).

"LM317+cascaded"

http://www.acoustica.org.uk/t/3pin_reg_notes4.html

Ok that makes sense when using mains line powered sensitive equipment.
Of course a good designed ADC or meter will add additional mains line frequency reduction by choosing appropriate integration time.

With best regards

Andreas

[branadic]

Of course a good designed ADC or meter will add additional mains line frequency reduction by choosing appropriate integration time.

Sure, but you would do at least everything you can as long it's worth but cheap before increasing integration time.

[alm]

You're missing the point. By choosing the integration time as an exact multiple of the mains frequency, the ADC can have a much better mains ripple rejection. For 60 Hz, a 16.7ms period will work much better than a 20ms period, for example, even though the integration time is shorter. This is why old meters would have a jumper/front panel setting, or would even have different crystals, for 50 Hz and 60 Hz. Modern meters tend to use a PLL from the mains frequency to produce the sampling clock.

[Andreas]

This is why old meters would have a jumper/front panel setting, or would even have different crystals, for 50 Hz and 60 Hz. Modern meters tend to use a PLL from the mains frequency to produce the sampling clock.

Why not choose a multiple of 100 ms as integration time?
This would give multiple of 5 PLCs on 50 Hz and multiple of 6 PLCs on 60 Hz.

Ok the LTC2400 has to be configured by PIN to 50/60 Hz suppresion (> 110 dB) at Pin 8.
When connecting to VCC the integration time seems to be 80 ms.
(can be measured as spikes on the input of the LTC2400)
The first 80 ms of the 160 ms measurement time are obviously some kind of self calibration mentioned in the data sheet.

With best regards

Andreas.

[alm]

Why not choose a multiple of 100 ms as integration time?
This would give multiple of 5 PLCs on 50 Hz and multiple of 6 PLCs on 60 Hz.

That would work. 10 samples per second is a fairly slow sampling rate, though, compared to the 50/60 S/s that are possible with 1 PLC.

Ok the LTC2400 has to be configured by PIN to 50/60 Hz suppresion (> 110 dB) at Pin 8.
When connecting to VCC the integration time seems to be 80 ms.
(can be measured as spikes on the input of the LTC2400)

This pin indeed controls the internal conversion clock to be a multiple of 16.67ms/20ms.

[branadic]

My reference is still running, stable at 10.00183 - 10.00184V depending on the labs humidity. I still don't know if this is due to the 34401A or my reference. Attached the actual assembly with the crystal heater on top of the 5ppm/K smd resistors.
But what is hidden to the eye comes visible using an IR-cam.

[Andreas]

Good pictures.

By the way: Is the crystal heater not thermally isolated from environment?
Or did you just remove the isolation for the camera?

With best regards

Andreas