EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: TERRA Operative on January 17, 2020, 03:34:33 am
-
I have a few bits of old equipment that specify the use of a null detector in their testing and calibration, so I thought it would be fun to build one to play with (null detectors don't seem to pop up on the auction sites here in Japan too often).
I think I might give the Conrad Hoffman one from part 2 of the Mini Metrology Lab article in Electronics Now.
http://conradhoffman.com/mini_metro_lab.html (http://conradhoffman.com/mini_metro_lab.html)
Attached below is the schematic too.
I have two questions to ask the experts here before I start though.
With the advancement in electronics, are there any improvements that can be made to this circuit with different parts etc since this was designed?
And is it possible to modify this circuit to use an external voltage reference?
-
I see no need to update the components. Might be fun (and a bit of a challenge) to incorporate a adc/display section to make the thing stand alone. Or hack in one of the many panel meter/voltage monitor packages that are available today.
-
You mention a voltage reference. There is none in the circuit. The resistors set the zero and gain. However these resistors must be very low noise with so much sensitivity.
-
The rail splitter is often referred to as a voltage reference. I see no benefit to moving it external, and some challenges. The chopper amp already has substantial supply rejection so there is little effect from minor changes in the supply voltages over time. The absolute values are totally unimportant as the the offset is adjusted out. An external supply would introduce the possibilities of ground errors and noise picked up in the wiring.
-
It seems that offset adjustment is referenced to supply rails; I think more error will be caused by normal battery discharge than by the rail splitter ::)
LTC2054HV has similar performance at significantly lower supply current.
And the TL-whatever splitter is kinda pricey. It would likely be cheaper to get a dual opamp and use one channel plus a resistive divider to generate the virtual ground.
-
Either that or find a battery holder that takes 6 x AA cells and wire a ground tap to the center point.
-
The divider at the input is relatively low resistance, this could be use higher resistance values.
The capacitors at the input are rather large and should be PP type to avoid possible DA related unexpected input currents / slow settling. I don't think one really needs that much capacitance with modern AZ OPs.
With a reduced supply (e.g. 3-5 V), there would be more OPs to chose from - it is still a compromise between input bias and noise.
With low noise and thus higher bias types (e.g. AD8628) one could think about compensating some of the bias current in one way or the other.
If one decides to combine the amplifier with a ADC µC and LCD, filtering is better done digital than analog as one can choose a FIR type filter that settles faster. Range switching may than use CMOS switches and for a 100 mV or 1 V range and one could get away without the divider and switch at the input, just switching the amplifier gain.
The Offset of the low bias AZ OPs is usually rather small (e.g. < 5 µV) so referencing the adjustment to the supply is not that bad. The rail splitter should not be an issue at all, as the PSRR is very good at low frequency.
-
True, regarding noise etc with an outboard voltage source. I like the idea of a bunch of AA batteries. Would provide more capacity than a 9V battery so should remain at a stable voltage for longer too.
For an inbuilt display, I found a 6-digit 5-digit version on ebay which I used as a display on a PSU, and hooking up my 6.5 digit meter, it seemed pretty darn spot on, so one of those could do the job maybe, unless I just decide to have binding posts to hook cables to 6.5 digit as it's a known calibrated device.
Looking at the LTC2054, it doesn't seem to incorporate the capacitors internally for the chopper amp. I guess I'd need to add those somewhere if I was to use this part?
I guess this would be the same deal if I used other opamps too?
-
Most AZ OPs don't need external capacitors. The few old types I know that do need external caps are some ICL7650 / ICL7652, LTC1050, LTC1052 and a similar Ti/microchip part. Modern AZ OP all include the caps.
An idea if the Zero detector is to have it battery powered so that it has very good isolation from ground. So the display unit should also be battery powered. There is no need for high accuracy / resolution there. So some 3.5 - 4 digits are enough. A small LCD panel meter would be an option, though without much digital filtering.
-
I worked with similar circuit for many months. Used LTC1050 and TL061 as the rail splitter, 9V battery (1.2mA).
Input resistor 100k (or 10k?), then 2x1n4148. Gain set to 100x (100/10k, or 1k/100k) thus I got 1uV resolution on my 200.0mV meter (also floating). 1000x gain was a bit noisy in my setup.
Some foil 0.47uF in the fb, 47ohm output protection resistor (and 100nF across DMM inp). No offset comp trim (it did a few uV output error w/o it).
The references LM399 (14V out) or REF01 or LT1021_10V nulled by a 10k 10t WW bournes and 10ohm fine tune trim.
Laborious (!!) to null because of rather large 10k 10t wire wound pot "step", however.
I measured "thermal drifts" of other 5 and 10V references.
It worked fine, the 1n4148 at the input messed a bit at even diffused in-room daily light so I had to cover them (not sensitive to extremely strong LED flash light, btw).
As that null meter and the DMM were floating I connected ie. the voltage reference (or better to say the output voltage off the nulling pot) to its virtual GND (the input GND of the null meter), and the measured 5/10V voltage to its input.
PS: I would recommend to have those 2 input diodes wired there, and also put some 100-330ohm resistor into the LTC1050 +Vcc rail. One of my 1050 smoked as I was messing without the diodes and most probably caused a latchup inside the 1050. The small 9V battery (NiMH acc) was powerful enough to fry the chip then (I burned my finger).
-
Wow, it's been a really long time since that circuit was designed, so an opamp update seems wise. The rail splitter is just an equal value divider and a buffer to get the impedance down. You could make your own with an opamp easily enough. Back when I did it, the rail splitter was inexpensive, maybe a dollar or less. Today it's about 3, so you can certainly do it cheaper. OTOH, you only need one and it saves work. I took the rail splitter just because I didn't want to mess with a bunch of batteries and their holder.
-
So, I'm laying out the PCB now and I went with a bunch of AAA batteries to power it (along with an LTC2054HVMP unless there is a better suggestion?)
One question. Is it worth using 8x AAA batteries, split into 2 groups of 4 for +/- 6V dual voltage, then have a couple of vregs to provide a stable +/- 4.5V, rather than run straight off 2 groups of 3 batteries at +/-4.5V?
Would that improve performance at all?
-
Many modern AZ OPs are made for up to some 5 / 5.5 V supply. So there is little need for a much higher supply.
If one wants the maximum useful range from such an OP one could drive the other side the opposite direction in stead of a fixed virtual ground.
The LTC2054HV is an option, but even if it can stand a higher supply, using only 5 V would result in less bias. It is a balance between input current and noise. The higher bias ones could profit from bias compensation. A few good ones (with decreasing noise, increasing bias) are:
LTC2054
max4238
AD8551
LT1052 / ICL7652 / TLC2652 (external caps)
AD8628 (supposedly low current noise)
LTC2057
ADA4528 (high bias and current noise)
-
I'll take a look at those suggestions on the weekend after I've had more than 5 hours sleep a night... :D Hard to digest data sheets in my current sleep deprived state.
How about the idea of using vregs for the 4.5V supply instead of taking the voltage straight from the batteries? Any benefit to that?
-
A regulated supply is a small benefit, but not very much. The bigger point is probably to use a supply that is not right at the edge. For the LTC2054HV they give a typical bias of 3 pA at +-5 V and 1 pA at 5 V. So the lower supply can help quite a bit. This also saves battery power. Unless one really need high input Z up to a few more volts, I would not use a higher supply than 5 V. With a driven negative side one could go to some +-4.5 V even with only a 5 V supply.
The 1N4148 / 1N914 diodes are also not ideal, as they have relatively high leakage and a light sensitive glass case. I would prefer something like BAV199. For better protection (e.g. from ESD events) there should also be an extra resistor (even if just 1 K) between the diodes and the OP : the external diodes limit the voltage to maybe 1 V, but it may still need some more resistance to limit the current flowing through chip internal "diodes".
Unless one wants / need the filtering, there is no need for the huge caps at the input. If analog filtering is needed this would be better in the feedback path, as the cross over frequency would not depend on the source impedance. Some input filtering is still needed to keep out/in EMI (both directions), avoid aliasing and keep to fast signals away from the OP.
-
Armchair opinion, others probably know far more. It's generally thought that batteries are quieter than regulators, but it depends on the type of battery and its state of discharge, plus the current draw. I've done some very quiet regulators, but it takes a few more parts. Hopefully the opamps isolate you from a lot of that, so I'd probably go without regulators just to avoid the quiescent current. IMO, it's all down at a level where you need to physically compare and measure circuits.
-
When adjusting for DC null, e.g. Wheatstone bridge, I find it much easier to view a moving-needle meter rather than a digital meter. Your circuit can use either (or both) for the indication.
For AC nulling, an analog cro works well as the indicator. My favorite Wayne-Kerr admittance bridge uses 4 cascaded magic-eye indicators (2 duals) to make a progressively more sensitive indication of the AC voltage.
-
[...]
For AC nulling, an analog cro works well as the indicator.
[...]
I've found a digital scope in combination with a x100 amplifier works extremely well - turn on averaging, and the noise goes away. I'm able to resolve down to 10^-7 this way, nearly lock-in amplifier territory! :scared:
-
I posted this question on another topic but it was ancient and no activity for months. If one were to only want to match resistors using the wheatstone bridge method would there be any drawbacks to just using an ad620 INA? https://www.analog.com/media/en/technical-documentation/data-sheets/AD620.pdf (https://www.analog.com/media/en/technical-documentation/data-sheets/AD620.pdf)
It seems far easier to just switch in 2-3 resistors to set the gain. I also was curious about the diodes in the original schematic. are they there for protection and voltage/current limiting. https://www.eevblog.com/forum/beginners/how-to-increase-an-analog-galvanometer-sensitivity/ (https://www.eevblog.com/forum/beginners/how-to-increase-an-analog-galvanometer-sensitivity/) shows a galvanometer gain circuit using transistors as protection. Is there a difference between original circuit at this one? Also After reading many posts on building the mini metrology lab as well as picoammeter circuits I have now confused myself on what we are supposed to measuring on the wheatstone bridge, I or V or either one depending on method used? I have mocked up and waiting to order a pcb using the original design minus the mA range due to only being able to find 4way rotary switches at decent price. My goal is to dual purpose the basic ad620 +/-9V supply with center tap reference to measure both wheatstone bridge balance at possibl higher voltage than INA supply voltage as well as null voltage references to each other. I prefer to use the analog panel meter in a scrap Fluke 931B voltmeter which measures a little over 3k across the inputs. While trying to null 2 of my references I had a lose connection on the input and it caused the needle to hard peg to one side. I connected the INA output directly to one side of the panel meter and center tap reference to the other. Should I have put some current limiting resistance in series on each side of the panel or once again was I measuring voltage. Any help is greatly appreciated.
-
The diodes at the input are used for protection. One can use transistor junctions instead of the diodes shown. The transistors can make reasonable good low leakage diodes and can be cheaper / better to get than low leakage diodes.
The AD620 is low offset, but still higher drift than the usual AZ OPs. So there may be a little more error. The gain range is also limited, so one may still need a divider at the input for the higher voltages. So the gain switching is not that much easier. For just resistor matching there can be a small advantage that the AD620 has a differential input and could thus uses the same supply as the bridge drive - the null meter would need a separate supply (but could be lower voltage).
-
Im slowly understanding but looking at one of the circuits Cerebus sketched out at https://www.eevblog.com/forum/beginners/how-to-increase-an-analog-galvanometer-sensitivity/ (https://www.eevblog.com/forum/beginners/how-to-increase-an-analog-galvanometer-sensitivity/)
[attach=1] I wanted to modify that circuit to fit the ad620 using jellybean npn as protection but don't understand why the emitter on both transistors isn't connected. Also im not understanding why the null meter would need its own power supply. Is it because of the output drive of the ad620? I cant find any info other than input current for the device so im assuming so but it seemed to work fine the way I had it wired driving the meter directly to the point of being able to see my voltage references drift when heating one of them.
-
Broadband noise is irrelevant in this application but some modern chopper stabilized operational amplifiers have lower flicker noise, or at least say they do.
I would replace the switching diodes with low leakage diodes like 2N3904 base-emitter junctions. I would put two in series for each one also.
The input filter capacitors could be bootstrapped to reduce leakage and dielectric absorption effects. This would also allow for a longer time constant if that was considered desirable. The bootstrap drive can be provided by a separate CMOS or JFET voltage follower. Low noise voltage references use this trick to lower drift contributed by their high impedance input filter.
I would replace the dedicated rail splitter with an operational amplifier based one; I do not like using such specialized parts if I can avoid it.
-
I imagine that it should be possible to use jellybean lm358 as a rail splitter and output drive to the panel meter at the same time as well. I know there would be awful tc drift and would have to work out a way to null offsets but would that be as easy as setting the ad620 to mid to upper gain limit and shorting inputs? The AD620 safely withstands an input current of ±60 mA for several hours at room temperature. This is true for all gains and power on and off, which is useful if the signal source and amplifier are powered separately. For longer time periods, the input current should not exceed 6 mA. Am I right to be thinking that it should see 4.5V at the input and shorting it with 4.6k resistor will limit the current to just at 1mA. Is this the proper way to null the panel meter
-
What are you going to do with the null meter? If you are going to use it to null two low impedance sources, such as voltage references or a low resistance bridge for a 25 ohm SPRT, then you don't need a particulary low input bias opamp - low noise is more important.
It doesn't even need to be auto-zeroing as you can do that manually using an input shorting switch providing the amp's drift isn't excessive and the thermal environment is reasonably stable. Alternatively you could servo the offset using an autozero opamp getting the advantage of zero-drift and the low noise of the main amp.
If it is a general purpose null meter there is nothing to stop you having more than one input amplifier, optimized for different source impedances, manually switching in the required amp.
-
The exact purpose will be to null buffered or un-buffered Ref102 10V references to each other as well as nulling multiple hamon dividers to one another. Secondary purpose was to be able to both null a wheatstone bridge which would also allow me to see T.C. of specific resistor in the bridge. After going through my parts drawer and finding a handful of AD620 https://www.analog.com/media/en/technical-documentation/data-sheets/AD620.pdf (https://www.analog.com/media/en/technical-documentation/data-sheets/AD620.pdf) and ordering 5x LMC662 http://www.ti.com/lit/ds/snosc51c/snosc51c.pdf (http://www.ti.com/lit/ds/snosc51c/snosc51c.pdf) while I already have 5x LTC1050 https://www.analog.com/media/en/technical-documentation/data-sheets/1050fb.pdf (https://www.analog.com/media/en/technical-documentation/data-sheets/1050fb.pdf) I just breadboarded the AD620 and was under the impression that it would be a possible substitute with far less parts than the mini-metrology lab null detector circuit http://conradhoffman.com/MML%20files/2_null_p2.jpg (http://conradhoffman.com/MML%20files/2_null_p2.jpg) for my specific wants. To sum it up I want to measure tens of micro-ohms difference if possible as well as sub micro-volt.
-
The comparison of references is a relatively easy job. Even without a buffer they are low impedance and still relatively noisy.
A Hamon type divider is usually more higher impedance, so that one wants a high impedance version with low bias current.
For the resistor bridge it really depends on the resistor values: PT100 is low impedance, but also low voltage. With resistors in the MOhms rage one would need really low bias, though not all the way to the LMC662.
With the bridges there may be the option to reverse the driving voltage to the bridge, so that one may not absolutely need an AZ OP - though it can still help, as they can have low low frequency noise.
-
I may be hung up in the idea that the null point has infinite impedance. You seem to think the lmc662 is overkill for low bias current. I have it on hand so why not use it if its the best option of what I have I figure. You also mention resistor bridge in MOhm range, why would this not work for lower value resistor bridge given I drive the bridge with very low voltage/current?
-
The idea of infinite impedance at zero is just wrong and confusing. With the old analog meters one zero current at zero, but still the relatively low impedance of the meter. Still having zero current is good and for some of the electronic meters the current is also low near zero, but not necessary exactly zero.
The tendency is that OPs ( amplifiers with lower noise have more bias and also current noise). So the best suited amplifier depends on the source impedance. The lower current / higher noise amplifier can also be used at low impedance sources, but these may need better resolution.
So the LMC662 is not a good choice for something like PT100 or PT10 where one wants sub µV resolution. Especially at low frequency simple CMOS OPs can be quite noisy. At the other end the ADA4522 is not a good OP for high resistance bridges, because of too much bias and also current noise. It is however a good choice for something like PT100.
Depending on the Application one could different amplifiers. A relatively good choice would be
MCP6021 / LMC662 für > 500 M
LTC2054 for high impedance (e.g. 10 M)
AD8628 for something around 50K-5M
LTC2057 at some 5-50K
ADA4528 below 5 K
The last 2 OPs may need a compensation of the bias.
-
Time to finish this project.
Parts and PCB's are ordered. :)
I went with two groups of 4x AAA batteries for dual rail, and an LTC2054HVMPS opamp.
I designed it to fit into a standard enclosure I can get here in Japan and am using a black PCB for the front panel (too lazy to make a front panel myself. :) )
Hopefully it works. :D
-
That looks good! What design software are you using for this?
-
Diptrace for the PCB layout, Rhinoceros 3D for the physical design and packaging.
-
Unless one really needs a high output voltage, one should get away with a lower supply, like 4 V total and than the normal LTC2054 OP. Higher voltage usually leads to more bias current and more self heating. There are regulators that can work with less minimum current than the LM317L and have less drop out, so one could get away with less supply current and fewer batteries. A low power way would be 4xAA, a 5 V LDO and a virtual ground instead of the true +- supply.
The trimmer for the zero adjustment looks odd in the circuit: with both sides to ground there would be very little effect.
The 1N914/1N4148 diodes are not the best choice for input protection, as they have a relatively high leakage / parallel resistance. A better choice would be transistor junctions or a special low leakage diode like BAV199 (SOT23 SMD, but easy to solder, especially in the back to back circuit).
The front part of the circuit is still missing, so it may be there. The OPs output should no directly go to the output, as OPs don't like capacitive loading. With an AZ OP capacitance at the output can also have a small effect on the offset.
So better have some minimal filtering, like 1 K from the OP and than some 1-10 nF to ground.
The capacitors at the OP are quite large. Depending on the use one may want the meter to react faster.
Some of the resistors in the FB path look quite large and they may add noise or offset.
-
Ah, that trimpot is to the 4.5 volt rails, a mistake on an older revision of the drawing. :)
Can you draw your suggestions so I can see exactly what you mean?
Linked below is the original schematic:
-
At the time, the LT1050 was chosen so the user wouldn't have to chase zero all the time. Input current is a bit high and there are lots of better choices if the source impedance is a concern. The real trick here is battery operation and the use of a DVM, so everything is floating. The input network was designed so few real world overloads would damage the thing, because, well, things happen!
-
[...] The input network was designed so few real world overloads would damage the thing, because, well, things happen!
...and we probably all have the lingering smell of smoke in the lab area to prove it! :D
-
I would stay closer to the original circuits with the following changes:
Most modern AZ OPs are for 5 V supply, so one should have a 5 v regulator and thus +-2.5 v after the virtual ground.
There are low drop regulators like MCP1700 that get away with very little current and something like 4xAA to start.
The virtual ground can use a normal divider and OP, like MCP6001 or LMV321 - this OP is not critical.
For the protection I would change the diodes to something like BAV199 (dual low leakage diode). The 1N914/1N4148 is still OK with the relatively low impedance divider, but not ideal. The 1N4148 in glass case can be light sensitive and produce extra bias from light.
I would suggest a higher impedance divider, more like 1 M. It would be a little noisy when open circuit, but this should not be a big problem. With 10 M the filter caps would make the reaction a bit slow, as the /10 setting would have a considerably longer time constant.
The 2 filter caps at the input can likely be smaller, like 100 nF. Ideally they should be PP or C0G type. Most DMMs already do quite some filtering.
Most modern AZ Ops are rail to rail and this get enough output range even with only 5 V supply.
Have an extra 100 Ohms resistor at the output to avoid capacitive loading.
-
These forum topics might also be interesting:
https://www.eevblog.com/forum/metrology/avm-2000/ (https://www.eevblog.com/forum/metrology/avm-2000/)
https://www.eevblog.com/forum/metrology/null-voltmeter/ (https://www.eevblog.com/forum/metrology/null-voltmeter/)
-
Ok, so looks like I need some help.... |O :-/O
I built this thing, and I'm not getting the expected outputs at all. I'm sure there's some boneheaded thing I did wrong somewhere. :-BROKE
Below is the full schematic, which seeeemmss to be correct-ish to the original schematic barring my modifications, and also a picture of the completed assembly.
(The diodes are doubled up, as I didn't have BAV199 in my stock of parts, so I could choose what to populate on the PCB during assembly)
The outputs I am getting using the voltage divider on page 4 of the instructions are as follows:
(Hooked to my Fluke 731B DC Standard for the 10V reference)
http://conradhoffman.com/MML%20files/2_null_p4.jpg (http://conradhoffman.com/MML%20files/2_null_p4.jpg)
100mV - 66.7mV
10mV - 627.5mV
1mV - 124.7mV
100uV - 93mV
10uV - 147mV
Any ideas how to get this thing reading right before I give up and scrap it for parts?
-
Maybe I should try putting a 9.09k resistor for R3, rather than a 90.9K...... :D
[EDIT]
Ok, closer now:
| Range | Output |
| 10uV | 92.6mV |
| 100uV | 97.9mV |
| 1mV | 96.6mV |
| 10mV | 99.2mV |
| 100mV | 91.8mV |
I wonder if using trimpots for R1-R4 would allow for finer calibration?
Also, the output seems a little noisy, any ideas for stabilizing it a bit?
[EDIT]
Looks like the resistor divider I made from cheap Chinese resistors was shittier that I expected.
When using my Advantest TR6142 to produce the required voltage for each corresponding range, I get the following:
(closer to 100mV output the better)
| Range | Input Voltage | Output |
| 10uV | 9.79uV | 121.75mV |
| 100uV | 100.11uV | 104.47mV |
| 1mV | 1.00001mV | 99.68mV |
| 10mV | 10.00078mV | 100.34mV |
| 100mV | 100.0016mV | 99.05mV |
-
A null meter is usually not about very accurate scale factors, more about an accurate null (offset). So normally low tolerance (like 0.1% or 0.5%) resistors for R1-R4 can be good enough. There are not that expensive any more, often cheaper than an extra trimmer. One may have one for the amplifier part, though it would also be possible to use low tolerance resistors there too.
The values all look a little low. Part of the problem could be just loading the divider by the chain of R1-R4.
Otherwise it would be more like a trimmer / adjustment at the amplifier.
-
Well it worked well enough to make a reasonably accurate measurement:
I'll tweak it and play with it at my leisure until I'm happy, then figure out what to do with it after that. :D
https://youtu.be/zcgIFeHUpH8