Here's something new.

I replaced the zero pot with a new one due to the bad connection. I let the unit run to check the operation and I got the graph shown here.
I determined that the pattern was due to my house heat cycling on and off. The detector's chassis was closed but it was still affected by the changing temperature of the room air. I'm going to have to insulate the circuitry. It must be the resistors being affected by the temperature. I never thought they'd be so sensitive.​

Rd values are 820 ohms, Rs is 1.46K ohms, same as the last run. The diff amp output was 3.739V, to begin with probably due to diffrent overall resistance of the new zero pot compared with the old. I readjusted the zero pot to bring the output closer to zero so the software would show the signal amplified for clarity.​

(Actually the zero pot wasn't exactly centered.)

 

Here are all the new stats with the new pot. I found that the drifting values would settle down in a minute or less so these are the steady state values, thus they are more accurate than the drifting values.
The battery is still on the antenna (0.926V)

Rd1 = 820 Ω
Rd2 = 822 Ω
Rs = 1.46K Ω
Zero = 1K, wiper centered (572, 572)
Vd1 = 24.23V
Vd2 = 21.01V
Vs = 14.93V
Ids1 = 3.87 mA
Ids2 = 6.07 mA
Vds1 = 9.02V
Vds2 = 5.82V
Rd1+zero = 1.357K Ω
Rd2+zero = 1.361K Ω
Rds1 = 2330.7 Ω calc
Rds2 = 958.8 Ω calc
pwr1 = 34.9mW
pwr2 = 35.3mW
Output = 3.131V

 

First of all, I think my suggestion of a 1k pot in the source circuit wasn't by any means ideal - I think I was too concerned about your parts bin (at least I know you've got 1k pots!). I'll try and do a better analysis of the circuit, but I think a 500 ohm or 250 ohm pot would be better (though you could probably use 1k pots if you perhaps were to put a 220 ohm or so resistor between each end and the wiper).

I must emphasise that there is a relationship between the diff amp outputs and 0V, because they're referred to the +15V line, which in turn is referred to 0V - so all 3 circuits are OK, including the one with dual supplies and no divider - in fact, this is very much the one I'm recommending.

You can describe an amplifier as having voltage gain or current gain - or power gain, which is a combination of the two. A lot of amplifiers have a voltage gain of 1, but a substantial current gain (they're called unity-gain buffers usually); equally, they have a large power gain (power is volts x amps).

As I say, I'll try to analyse the circuits in a bit more detail, but I think with the zero pot in the source circuit we'll be better of in terms of temperature effects. I'm quite surprised by the sensitivity to your heating though. Bets results will be had with those monolithic dual fets, but it will be interesting to tweak the circuit using your existing devices.

By the way, what is your parts bin like? Do you have an extensive range of fixed resistors, and what pots do you have (also small-signal diodes)? It's obviously a lot easier to pay around if you have easy access to a wide range of components, but I sense that you don't (which would mean I'll have to try to be more accurate in my circuit analysis and more precise in my suggestions).

 

I guess it's my lack of experience with these things, but with what I know, I just don't see how the circuit with dual supply and no divider can work. There's something I'm just not seeing.

I've seen unity gain amps mentioned and I always wondered what anyone would have to use a unity gain amp for. It seems like why bother with an amp if it's unity?

I'm going to look into the dual fets the other guy mentioned a while back.

I guess I have a decent assortment of resistors including a lot of 1M ohm low noise metal film resistors, much fewer caps from pF up to 22,000 uF (not sure if all the electrolytics are ok due to some minor polarity reversal), mostly zener diodes, the only small signal diodes are 1N4148, a variety of pots, most are fairly new. Also a large photoflash cap for a laser I never finished back in school, 2 neon sign transformers, a couple things like variable vacuum caps for microwaves (??), large magnets, high wattage blacklights, & various things picked up from companies going out of business. Some of my parts date back to the late 60's!!

 

Something's wrong. I tried out the unit after replacing the zero pot and all I get is noise. So far I can't find the source of the problem. The unit works a bit, it does detect my presence when I approach the antenna. Unfortunately, the indoor circuitry seems more sensitive. Sometimes I get spikes when I get up from my chair or turn the room light on or off. Everything is shielded and grounded. It normally doesn't do this, but I have seen it before. I don't know what causes it. The attachment shows the entire graph and a blow up portion for detail.

 

I suspected something had happened to the ground fet so I replaced it. Now I get the following graph. Still noisy plus some 4V spikes that I've seen before but I've never figured out how they could be real. If not real, then they might be due to the DVM or the data collection software that works with the DVM. They might represent data peaks are of a size and rapidity that the meter can't sample them fast enough. On the number list on the left, a peak like that causes the symbol "OF" which means "overflow".

Working with the outdoor part of the circuit is difficult. In the past when I got peaks like this I would fiddle with the bias settings and could get rid of them. That's about all I could do. I wouldn't go out and replace any fets outside unless everything else failed. Now I don't know, since I just replaced one and touched nothing else. I guess I'll replace it again.

 

I replaced the ground fet again. Now I have 40 V spikes. Nothing in the system is 40 V.

 

Regarding the previous posts concerning temperature effects, I've realised that the temperature cycling was applied to the resistors only (and the power supplies), and not the fets. I'd imagine the source of the problem is the pot. What sort of pot is it? One with a knob and 270 degrees rotation, or a multi-turn trimpot? I suspect the trimpot might actually be better here, but I'm not certain.

As for the instrumentation amp: these are designed to amplify only the differential voltage applied to the inputs, regardless of the absolute voltages - so 100mV differential can exist as -6.7/-6.8 V or +9.3/+9.2 V and you'll get identical outputs. That's because the amp has a very high common-mode rejection ratio. However, the voltages are not allowed beyond a certain range, and for a +/-15V supply, this happens to be -15V to +15V, and is called the common-mode range. You couldn't therefore apply voltages outside of this and still get sensible results - and that's why you're advised to keep the inputs within range of 0V.

Now for the spikiness: I think the next thing to try is to short the inputs together (and preferably connected to 0V), with the pot centred. We need to exclude the possibility of external influences, such as rf pickup. After all, you have an antenna up there, and antennas are used to receive electromagnetic waves.

If you still get spikes or untoward noise, then something else is happening - possibly it is oscillating. You really could do with putting a 'scope on the outputs, but I'm guessing you don't have access to a decent oscilloscope?

If it does turn out to be rf interference (it's possible you live close to an amateur radio enthusiast, for example), then I'm going to have to think about how we can solve this problem while maintaining the ability to monitor static charge.

Apart from this, and just reflecting on the nature of the problem to be solved, it occurs to me that this is all actually quite a difficult task: we're looking at a very high impedance source whose differential potential is somewhat undefined, whose common-mode potential can vary, we seem to need to be able to observe small changes in the differential potential over a long period, and in the presence of fairly large changes in ambient temperature. On top of that, we may have to take steps to reject interference from high-amplitude unknown sources!

I wonder if you have information regarding such as the range of static potentials we may experience, and the magnitude of the variations we're trying to detect? If the variations are very small compared to the background values, it's certainly going make life difficult!

 

Regarding a unity-gain buffer, as I say, these are used for current gain - in other words, when you have a high-impedance voltage source which needs to be presented as a low-impedance source.

 

The new 1K pots I just bought are 0.5W, with knob, 300* rotation. They don't say what the resistive material is but it's probably carbon. They were $2 each. There were other pots at $7 each which were 2W carbon, and that was about all they said. I do have a few 1K ohm, multi turn trimpots made of cermet. Rd1 and Rd2 are this type.

About the in-amp, where is the bias return path on the inputs? This is what I don't see. There needs to be a path for each input according to the datasheet.

I'll short the gates together again and see what happens. I can try to connect the shorted gates to a cable shield, but this is going to be earth ground, not the 0V ref line. But since the shields are earth ground, I don't need to connect to them because this is the same ground as the gates would already be connected to.

Correct, no scope.

I'll go through the info I've collected and see what I can find about the magnitude of signal variation.

 

I found the dual fets the other guy mentioned, there are several to choose from. If you can tell me some parameter values that you feel they should have I'll make a choice. Do you think we could make this job easier with these or would we be getting ahead of ourselves at this point?

 

There is a path, and it involves anything between the diff amp outputs and 0V, i.e. Rds, the fets etc. and the power supplies.

It's probably premature to go for the dual jfets at this stage. I had a look at the Digikey offerings, and no mention was made of whether or not they were monolithic (i.e. on the same chip), and no mention of matching - so they may just be two separate fets in the same package.

I would think that something like a 2N3958 would be suitable, but they're about £10 each in the UK. I'll continue looking.

By the way, a crude solution to the possible rf interference problem might be to put a large inductor in series with the antenna input. Something like the primary of a mains transformer (for safety's sake use one with a low-voltage secondary) or an old vacuum tube audio power output transformer primary. The whole thing would need to be isolated from anything at ground potential using a thick insulator (like a thick sheet of plastic or glass).

 

A couple months ago I wound 259 turns of wire on a large snap-on ferrite core from Radio Shack (rectangular core, roughly 1''x2''). I installed this into the antenna line (the FETs were indoors at that time so it was easier to make changes.) There was no improvement in the noise. I also tried an FM radio trap from Radio shack. There was a tiny improvement. I found a hack online that described a simple change that could be made to the Radio Shack FM trap which would vastly increase its trapping capability. I made the change but there was no further improvement. I
also tried the coils with a 10 ohm resistor in series and there was no noticable change here either. I ended up removing the extra stuff to keep things simple.

At one point it looked like I might have 60 Hz interference coming in through my house ground (could be a neighbor with a bad ground.) I made a couple 60Hz traps and put them in the ground lines but it didn't help. (I don't detect any 60Hz in the power supply output, or in the diff amp output, but when I operate the in-amp, I detect several volts of 60 Hz along with other unidentifiable high freq noise, in the otherwise DC output. Putting decoupling caps around the circuit and ferrite inductors on the in-amp output also did not reduce the noise.)

I also tried a fluorescent lamp ballast on the in-amp output with no effect. I've tried extremely low pass filters in various places from the antenna to the in-amp output with similar results.

I used to have a 10M ohm bleeder resistor on the antenna input made with regular resistors. When I got my low noise resistors, I made a 50M ohm low noise bleeder. This added more noise than the 10M.

I've often checked signal appearance with the fluorescent room lights on and off and I've never seen a connection. At most, with the "wrong" circuit, I might get spikes when something is turned on or off, but never continuous RF interference.

------------

I replaced the gnd fet several times and got the same results each time. Then I replaced the antenna fet, swapping it with one of the gnd fets I'd used. Now the antenna fet and gnd fet would seem to be similar since they both acted similarly in the gnd position. I'm running a test now. The graph shape is similar but is shifted more positive by about 12V. It's more noisy. So far, no big spikes.

All these fets I'm swapping have been used before, but they check out as being good. I'm careful handling them and I ground myself and the antenna when swapping them.

Attached is a typical graph from one of the gnd fets being tested.
I'm wondering why the graph suddenly turns upward and also why it turns horizontal and gets noisy?

 

Here are some graphs of atmospheric electric field from the literature and one of mine from my original circuit using 1 FET in a Wheatstone bridge. They are lined up by time. The vertical scales are different, mine is in mV, the others are in volts/meter. There is also one made during a thunderstorm for field comparison. Typical fields in fair weather are around 100 V/M.

You might find some of the info in the following paper helpful. It gives a schematic, some calculations, field intensities and sensor output values: http://arxiv.org/pdf/physics/0701296

 

I grounded the gates and ran a test. Afterwards I took the following measurements:

Rd1 = 820 Ω ​

​

Rd2 = 822 Ω ​

​

Rs = 1.46K Ω ​

​

Zero = 1K, wiper centered (572, 572), output not zeroed.​

​

Vd1 = 25.65V​

​

Vd2 = 25.23V​

​

Vs = 9.16V​

​

Ids1 = 3.04 mA equilibrated down from 5.5​

​

Ids2 = 3.30 mA equilibrated down from 5.5​

​

Vds1 = 1.047V​

​

Vds2 = 1.026V​

​

Rd1+zero = 1.392K Ω ​

​

Rd2+zero = 1.394K Ω​

​

Rds1 = 344.4 Ω calc​

​

Rds2 = 310.9 Ω calc​

​

pwr1 = 3.2mW​

​

pwr2 = 3.4mW​

​

Output = 0.401V ​

​

​

​

I think the blip on the right is from me turning on my microwave oven in the next room. Soon after this, there was a 1.3 V spike. This was the only interference in the almost 2 hr test.​

​

​

​

After this, I removed the ground connection from the antenna. So far it has run for 2 hrs with no spikes. It is noisy though. Too early to tell signal size. Output is about 7.5V.​

​

 

The figures suggest that Vg is about -5V; I wonder why there's such a discrepancy between external ground and 0V? You could try measuring this directly.

I don't understand Vds1=1.047V, Vds2=1.026V, Rds1=344.4 Ω calc and Rds2=310.9 Ω calc, but I don't think it matters anyway.

Regarding the inductor, you'll need as many Henries as possible - I'm not sure the components you've tried will have sufficiently high values of inductance. I think a mains or audio transformer would be getting closer, but the disadvantage here would be the fact that the cores would tend to be magnetic at low frequencies only. Maybe worth a try if you have any to hand though.

I can't help feeling that your 24 hour graph would be very similar to a graph of temperature for the period. It might be a good idea to get a crude feel for temperature sensitivity by using a hair dryer (with the gates both grounded).

Thanks for the link to the paper - I haven't read it fully yet, but it describes the use of an op-amp as a unity gain buffer. Interestingly, it has a "bootstrapped" input guard, which is something I had been toying with: my idea involved ditching the fets and just having a unity gain buffer with a bootstrapped screen on the cable attaching to the antenna (in other words, the op-amp would be driving the cable screen). The biggest problem with this could be that op-amps are notably intolerant of input signals which have high amplitude, high frequency noise, and that's why I hadn't mentioned it. But it still may be a viable alternative if, for example, we can't get round temperature sensitivity problems.

Also, I'm wondering why you hadn't tried to repeat Bennet and Harrison's experiment, using their antenna/circuit design?

 

At present, with no load I measured between the power supply 0V line and the grounded chassis and I get 14.41V, with the 0V line being positive. When I switch on the detector (fet circuit) I get an erratic reading. Switching to AC, I measure 75 mV which climbed to about 160 mV, then dropped back. If I switch off the detector (still measuring AC) I read 9 mV. I've measured this before and did not get results like this at all. It looks like something went bad in my power supply. First guess is a cap is leaking the AC under load. I can't say what is causing the high DC difference between 0V and earth, but I suppose it's related. There have been a couple times when I blew a fuse in the power supply, but as the schematic shows, I have them in the output lines and they are about 0.25 to 0.75 amps as I recall. I did this specifically so they would blow long before any other parts of the circuit got stressed since it can put out 1A. Just now I measured the DC outputs separately (relative to 0V) and got +15.00V and -14.79V under load, and +15.01V and -15.00V with no load. I'll look into this today.

As for Vds1 and Vds2, that's what I measured. I measure from the same points each time. There was a little time needed for some of the numbers to stop drifting and settle down but only Ids changed a lot, so I can't say there was a large drift to these when measured. They are what they are. I calculate Rds as Vds/Ids like this 1.047V/0.00304A = 344.4 ohms.
I just remeasured these (gates are not grounded together) and I got: Vds1 = 16.94V and Vds2 = 8.72V. Not that it matters at this point.

I used the largest inductors I could find. That's all I had.

You're right, the temperature variation over 24 hrs is similar to the shape of the atmospheric potential graph. When I got that graph last year I had felt sure I was seeing a potential signal as well as a temperature signal. One instance comes to mind where it was in the high 90's during the day and quite cool at night, 60's, maybe less, and I got the same size signal as when temperatures were not so extreme, like 90's in the day and about 80* at night. I know there is some contribution from temperature, that was why I moved the fet indoors in the first place.

I found the Bennet & Harrison paper only 4 months ago. I'd already been working on my circuits nearly 2 years then. I found online around 2 dozen different circuits of varying complexity for measuring electric fields. I tried a couple other simple circuits that were "supposed" to work easily, but barely worked at all. I have no way of knowing which of all these circuits would be good and which not so good and I don't want to build a lot to find out. My circuit started from a tiny thing consisting of a 9V battery, a FET, and an LED that I played with a bit. I got interested and gradually built it up to what I have now just to see what I could do. I was learning as I went. I'd never worked with FETs before, much less in-amps. I kept things as simple as possible so I could understand what was happening and be able to modify the circuit as necessary. At this point, I don't know enough about op amps and in amps to be able to make intelligent changes to the circuit if there was a problem. Working with discrete devices allows me to make changes more easily and see what each part is doing. I'm old enough that IC's still seem like "black boxes" to me. I don't really know what goes on inside, and even if I find out, I can't change it. I like tweaking things. This project has kept me busy for 2 years, and though I've spent WAY more money on it than I ever expected to, it has been relatively inexpensive considering how much it has kept me busy. I have a lot of time on my hands. Even if the circuit in the Bennet & Harrison paper would work perfectly, I wouldn't want to make it. My work has been more about keeping busy than about any interest in weather phenomena. Originally I had hoped to be able to detect approaching lightning so I could get a warning signal and turn off my computer if a close strike was imminent. I'm not sure I have ever picked up that kind of signal. There have been some signs that I might have, but they were small and lost in the background noise of a storm.

 

I found a broken connection on the 10uF 63V cap at lower right of the power supply schematic. It's the one between the neg rail and 0V, NOT the one in series with the diode between neg and zero & connected to the LM337.​

The DC output is ± 15V relative to 0V with no load. With the detector as load, the outputs are +14.95V and -14.98V.
The DC difference between 0V and the chassis at earth gnd is fluctuating, with load and is 14.41V DC without load with the 0V line being positive vs the chassis.
The AC difference is about 100 mV AC at 60 Hz with load and 8.1 mV AC without load.
I'm looking for the old measurements to check the 14.41 V measurement. It sure does not ring any bells.​

Here is the circuit and graph from 1.5 months ago when I had both fets indoors and only the antenna outside, so there was no temperature change like we were talking about. The signal strength is lower from the diff amp so I'm running it through the in amp at a gain of 1080x. I still got a signal which lined up with previous graphs. This had to be from detected atmospheric potential differences.​

 

Here are re-measured variables from above (gates not grounded together):

Rd1 = 820 Ω
Rd2 = 822 Ω
Rs = 1.46K Ω
Zero = 1K, wiper centered (572, 572), output not zeroed.
Vd1 = 24.78V <
Vd2 = 22.60V <
Vs = 12.90V <
Ids1 = 3.52 mA equilibrated down from 8.9 mA <
Ids2 = 5.29 mA equilibrated up from 1.3 mA <
Vds1 = 11.76V <
Vds2 = 9.83V <
Rd1+zero = 1.392K Ω
Rd2+zero = 1.394K Ω
Rds1 = 3340.9 Ω calc
Rds2 = 1858.2 Ω calc
pwr1 = 41.4 mW
pwr2 = 52.0 mW
Output = about 2-4V, unsteady

 

I think I get the picture now regarding your hobby! I suppose contributing to these forums is one of my own little time-occupiers...

Anyway, I'm not sure I fully understand your grounding situation. Which of these are connected to each other: chassis (i.e. "house ground"), 0V and external ground?

I think it's important for you to have a decent appreciation of how the diff amp works, and to have a decent appreciation of what the fets are doing. If you take a look here:

http://en.wikipedia.org/wiki/File:JFET_n-channel_en.svg

you'll see a graph of drain current versus drain voltage. Note at the left hand side drain current increases linearly as you increase voltage - just like a fairly low value resistor. Here, the fet behaves like a resistor whose value is determined by the gate voltage. However, on the right hand side, the drain current remains largely unaffected by changes in drain voltage. This is actually called the saturation region (note that I got the region names mixed up when I last used the term "saturation").

Now, if you're wanting a voltage amplifier, you'll appreciate that the gate controls drain current. If you want to maximise voltage gain, then you'll want to organise any change in current to result in as large a change in voltage as possible. You would do this by passing the current through as large a resistor as you can, and this would imply an appropriately high positive rail, in order that you maintain a reasonable drain voltage. If you let this voltage drop into the resistive "linear region", you'll see that a change in current produces far less change in voltage - in other words the gain is drastically reduced. Effectively, the fet resistance is shunting the high value load resistor.

That's why you shouldn't be calculating Rds values - they're not at all valid if you're in the saturation region (which is where you should be aiming to operate).

As for the source end of things, the reason for the negative supply is to allow the inputs to have a decent common-mode range, i.e. the gate voltages can go up and down in unison without too much effect on the working of the diff amp. The "tail" resistor is used to set the standing current, and you estimate this by assuming that the source voltage will be at approximately the same potential as the gates. So, for the gates at 0V, the tail resistor would be 15V/current. Once you know the tail current, you can calculate a sensible value of drain load resistor, bearing in mind that each fet will handle half the tail current (assuming matched fets).

By the way, it occurs to me that the reason the simulator was screwing up was that it also had the gate and drain connections switched. If this was the case, you could perhaps correct this and use the simulator to predict the diff amp behaviour (but bear in mind it will assume matched, middle-of-the-road fets).

I wonder if some of the "noise" and spikiness is due to the pot? I would imagine it's very sensitive to small mechanical inputs - what happens when you tap it, for instance?

It's a pity that you don't have a 'scope. I must say, I'd feel lost without one if electronics projects were hobby of mine - they're not, though, so I don't own one myself. If you come across a cheap, old (but working) analog 'scope, I'd recommend you get it!