Diode voltage drop stability/accuracy

[Infraviolet]

Thanks for the app note, I had a go at simulating Fig 15, yet found that for any given input amplitude of signal the output is not temperature stable. That is to say, if both diodes have their drops changed together the output voltage still changes. Shouldn't this circuit layout cancel out such changes when both diodes change by the same amount?

Yes, a multi-channel voltmeter might be a good way to hink of it, the trouble is I can only measure voltage once the peak (envelope) level has been taken, I can't measure the peaks as they happen due to their "high"* 3MHz speed.

*very slow by real high-speed standards, but pretty fast compared to previous things I've done

[Kim Christensen]

The 2nd diode in Fig 15 isn't for temperature compensation. It turns the single diode halfwave rectifier circuit, into a halfwave voltage doubler rectifier circuit. In other words, Fig 15 produces double the DC voltage for a given AC input voltage when compared to Fig 2.
I posted that ApNote to show that there are diodes available that do not require a DC bias voltage to detect small AC voltages. Note the very low reverse breakdown of these diodes which is a trade off for their very low forward voltage drop at low currents. Sometimes known as a "zero bias Schottky diode".

[Infraviolet]

Is there any configuration which can cancel temperature effects of the diode drop changing?

If I can't find a diode or transistor array with enough separate units in it (often they have common on one pin or another, which makes most of them unsuitable here), and resort to separate diodes or transistors physically very close together on a PCB, to what extent could I likely expect different channels to vary from one another due to temperature changes? Would being close together and of the same manufacturing batch be enough to have all of them have the same change in drop voltage with temperature changes, even if one channel were say detecting peaks from a 2Vpp input signal while another was trying to handle a 50mVpp input? What levels of deviation in voltage drop between them could I expect per celsius change in the ambint environment? I could get away with not cancelling temperature effects if I could ensure they were the same for all channels.

[Kim Christensen]

Depends on what you are doing with the peak detected DC voltages. If you were feeding them into a MCU ADC, you could have a single temperature sensor to monitor the ambient PCB temp, and then do the subtraction in code.
You could also do the math, on the peak detected voltage, with differential Op-Amp circuits by subtracting the voltage change from a "diode temperature reference" located on the same PCB.

[Infraviolet]

My thoughts were to have a dummy channel with no signal feeding in, and measure than one by ADC as well as all the real channels. Would that work though, or would I be getting substantially different variation on each channel (especially depending whether the channel was at that time detecting large or small peaks?) from environmental effects?

Would it be reliant on having all diodes/transistors in a monolithic array (somewhat trickier to find than expected, most have one pin or another common which isn't suitable here), or would close together on the PCB with discrete SMD parts be enough?

Also, I found an idea for temperature compesnation in Fig 5 at https://www.analog.com/en/technical-articles/measurement-control-rf-power-parti.html
I guessed that R1 and R2 ought to be equal, and replaced the RF specific 68ohm bias with a high resistance divider. It seemed in simulation be have a lot less effect on the output when the diode drop varied, but not fully cancelled, however only when R1 and R2 were equal. Trying for R1 as a quarter of R2, so as to improve the voltage of the output, lead to the diode drop change cancellation effect mostly ceasing to exist. Is this principle any use?

[Kim Christensen]

My thoughts were to have a dummy channel with no signal feeding in, and measure than one by ADC as well as all the real channels. Would that work though, or would I be getting substantially different variation on each channel (especially depending whether the channel was at that time detecting large or small peaks?) from environmental effects?

You'd still have to feed a known DC voltage into the dummy channel. Then subtract the known voltage from the read dummy channel voltage for your correction factor.
If you had a reasonably accurate DAC, you could even feed a 0-3V ramp into the dummy channel and build a correction factor table and then occasionally refresh it to compensate for temperature changes.

Would it be reliant on having all diodes/transistors in a monolithic array (somewhat trickier to find than expected, most have one pin or another common which isn't suitable here), or would close together on the PCB with discrete SMD parts be enough?

You did spec a tolerance of +/-20mV between channels, so as long as the diodes were reasonably matched and the temperature gradient between the parts wasn't too high, it should be doable. You'd have to prototype and test.

Also, I found an idea for temperature compesnation in Fig 5 at https://www.analog.com/en/technical-articles/measurement-control-rf-power-parti.html
I guessed that R1 and R2 ought to be equal, and replaced the RF specific 68ohm bias with a high resistance divider. It seemed in simulation be have a lot less effect on the output when the diode drop varied, but not fully cancelled, however only when R1 and R2 were equal. Trying for R1 as a quarter of R2, so as to improve the voltage of the output, lead to the diode drop change cancellation effect mostly ceasing to exist. Is this principle any use?

You'd definitely have to drive those diode detectors from a relatively low impedance source, otherwise loading will also introduce errors. Yes, R1 != R2 wouldn't make sense.

[Infraviolet]

"You'd still have to feed a known DC voltage into the dummy channel. Then subtract the known voltage from the read dummy channel voltage for your correction factor."

Is either of these an appropriate setup then? The dummy and real outputs both feeding in to ADCs, or possibly through a noninverting op ampconfig first, given they are close to DC so easy for op amps to keep up with. The dummy channel is biased like the real but does not have capacitive coupling to an input AC waveform, a real channel does have coupling. I show the idea in two images for two different types of peak detector biasing arrangment.

In both examples the peak holding cap is 1nF, and the resistor parallel to it is 100K, in the horizontally stacked image the resistor pair R20,R21 are close to 50K each so they add to around 100K.

In the vertically stacked image I find the R12,R13 pair to work best when fairly low (2.2K), so long as the input signal can cope.

In the horizontally stacked I find R19 around 3K9 and R18 at 1K gives a seemingly sensible biasing level. This horizontally stacked version is supposedly able to somewhat compensate for diode drop variation so long as R21 and R20 are the same size, but only has voltage outputs about half those of the vertically stacked sort for any given input amplitude.

P.S. I'm starting to think I'm in a situation where what really matters is somehow getting out the ratio of signal amplitudes on multiple real channels, does this make life easier? Is it the case that the dummy then only has to provide a "zero amplitude" reference for the other channels? Would the ratios of the peak detector outputs (each with the value of the dummy subtracted from it first) then be a good match to the ratios of the amplitudes of the incoming sine waves?

[Kim Christensen]

P.S. I'm starting to think I'm in a situation where what really matters is somehow getting out the ratio of signal amplitudes on multiple real channels, does this make life easier? Is it the case that the dummy then only has to provide a "zero amplitude" reference for the other channels? Would the ratios of the peak detector outputs (each with the value of the dummy subtracted from it first) then be a good match to the ratios of the amplitudes of the incoming sine waves?

So, for example, with 2V into chan A and 1V into chan B...
What you really care about is whether channel A reads 2x channel B but don't really care if:
Chan A reads 1.8V
Chan B reads 0.9V

or

Chan A reads 2.2V
Chan B reads 1.1V

etc...

[Infraviolet]

I think so, though I need to check further. But this ratio needs to hold true for far greater than 2:1 ratios. If channel 1 had a 2.5V peak to peak and channel 2 had a 140mVpp signal I'd need* the outputs** to be accurately in a ratio of 17.85 to 1. I'd also still ned accuracy if channel 1 was at 2.1V peak to peak sine wave and chan 2 was at 2.15V,  would need outputs in a 1 to 1.075 ratio. And I'll have more than two channels, so ratios between all of them need to be maintained when converting from sine wave amplitudes to DC levels.

Put in terms of ratios I'll need a while to rethink what this will mean in terms of the accuracy required, though I can say that the +/-20mV matter still applies in as much as saying that I'm expecting noise and error levels on the original sine wave inputs of somewhere a little below this, so +/-20mV is the most accurate those initial signals can really ever be.

Thanks

*I'll have to double check a few more things further, but I increasingly think it will be only the ratio which matters
**relative to either some truly fixed voltage or to a dummy channel which would be as subject to temperature and manufactruing variations as the real channels

[Terry Bites]

Don't use a diode to make a peak detector. Use a fet.
Use a zero threshold mosfet . see https://www.aldinc.com/ps_epadmosfet_zero.php
They'll even throw in a spice model. Register for the full sp

Harvest energy from (all) your signals to charge a low leakage cap and power a nanopower opamp from it.
Lots of sources for 500nA Iq opamps. Maybe use those epads again here. Check out harvesting ics.

Oh, and LEDs and regular pn junctions have equal and oposite tempcos- well about the same.

[David Hess]

Oh, and LEDs and regular pn junctions have equal and oposite tempcos- well about the same.

That depends on the LED type.  Old red LEDs have a temperature coefficient which closely matches a silicon PN junction, so they can be used to compensate for the Vbe of a transistor.

[mawyatt]

Don't use a diode to make a peak detector. Use a fet.
Use a zero threshold mosfet . see https://www.aldinc.com/ps_epadmosfet_zero.php
They'll even throw in a spice model. Register for the full sp

Interesting devices, thanks!!

Wonder if these are floating gate or native devices, Recall some folks at Intersel disclosing a floating gate voltage reference at the ISSCC way back, they utilized a clever means of "programming" the floating gate voltage (charge) by means of some matching Fowler-Nordheim tunnel diodes, maybe this technique is utilized to "program" zero threshold devices which would need to be floating gate types?

Anyway, here's something wrt native transistors (zero threshold devices) used in energy harvesting.

https://pdfs.semanticscholar.org/c9ad/4efe83afbbb18ffb43743513a964352ebbde.pdf

Best,

[Infraviolet]

Terry Bites, what circuit layout would that MOSFET solution use? Is it somehow able to feed much slower signals to an op amp* that the actual input sinewave is? You're saying that with a MOSFET I can add a bit more charge to the cap wth each incoming signal then measure the value that cap settles to after tens or hundreds of incoming peaks of the same amplitude?

*you mention nanopower op amps, so that's a pretty strong indication that a pretty slow one can work for this application

LEDs versus typical pn junctions, does this mean an LED of the right voltage drop (say a red one) in series with a typical diode (ay 1n4148) acts, even at pretty minimal signal current (too low for the LED to visibly light), like one big diode with a larger voltage drop (V_dropled+V_dropdiode) but which is very stable over temperature and anything else which might vary?

P.S. I'm not in a micropower situation as regards my overall circuit, I've tens of mA to play with from my 5V Vcc rail, a hundred even if I absolutely had to, I guess this negates a use for enrgy harvesting, unless that principle is howthe charge is getting transferred to the cap you describe.

EDIT: another thought, a two transistor differential amplifier is already temperature compensated as both transistors vary the same with temperature? This applies too if you hold one input at fixed voltage and just amplify the difference between an incoming signal and the fixed voltage? Is there a way to modify one of these so it serves as a peak detector outputting an analogue DC level proportional to the incoming sine wave's amplitude, rather than outputting a gained up sine wave.

[David Hess]

LEDs versus typical pn junctions, does this mean an LED of the right voltage drop (say a red one) in series with a typical diode (ay 1n4148) acts, even at pretty minimal signal current (too low for the LED to visibly light), like one big diode with a larger voltage drop (V_dropled+V_dropdiode) but which is very stable over temperature and anything else which might vary?

It means that the difference in the forward voltage drops between a silicon PN junction and red LED is stable over temperature because they both have a roughly -2 millivolt per degree C temperature coefficient.

[Infraviolet]

Difference between, so you'd have to set them up so they were subtracted from one-another rather than being added in series?

[David Hess]

Difference between, so you'd have to set them up so they were subtracted from one-another rather than being added in series?

That is right, but it is not difficult to do.  For instance the LED could control the base voltage of a transistor, so now the emitter voltage is constant, about 2 volts, despite changes in Vbe over temperature.  Put a resistor between the emitter and common and you have a temperature compensated constant current source.

[Infraviolet]

How would that constant current source then get me a temperature compensated peak detector? Do you mean running an LED from the incoming sine signal, then in to a transistor's base, putting the collector to the high voltage rail and using then having the emitter run to the parallel cap and res of a peak detector?

Wouldn't the big voltage drop from an LED plus the B-to-E of a transistor also mean the peak detector would have very poor linearity and sensivity for the smalelr amplitude input signals in my plausible range?

Also, I keep wondering about whether a two transistor differential amplifier can be modified to form a peak detector. My thoughts are that if it could then the use of two transistors in that manner would compensate temperature effects? I can't find any search results about the idea though.

Thanks

[David Hess]

How would that constant current source then get me a temperature compensated peak detector? Do you mean running an LED from the incoming sine signal, then in to a transistor's base, putting the collector to the high voltage rail and using then having the emitter run to the parallel cap and res of a peak detector?

It does not, unless your peak detector requires a temperature compensated current source.

Wouldn't the big voltage drop from an LED plus the B-to-E of a transistor also mean the peak detector would have very poor linearity and sensivity for the smalelr amplitude input signals in my plausible range?

There is an alternative peak detector design which replaces the diodes with bipolar transistors.  The advantage is that the collector supplies the emitter current instead of the base.  See figure 97 on page 44 of Linear Technology application note 47.  The transistor Vbe voltages still need to be matched.

Also, I keep wondering about whether a two transistor differential amplifier can be modified to form a peak detector. My thoughts are that if it could then the use of two transistors in that manner would compensate temperature effects? I can't find any search results about the idea though.

An operational transconductance amplifier can be used to make a closed loop peak detector without the slew rate limitations of an operational amplifier.

[Infraviolet]

Thanks for the app note idea. It looks like the following works as a peak detector, with pretty decent linearity with input voltage as compared to a simple diode based detector.

R28 and R29 give a low quiescent biasing point to the incoming signal, R30 affects the quiescent point of the output and is set large enough to prevent "clipping"* on the largest plausible input waveform, C16 and R31 set the decay rate (1n with 100K works nicely for my signal).

To temperature compensate it and protect against manufacturing variation all I have to do is have both the NPNs be in the same physical package? As long as both NPNs for each channel's copy of this are within a dual IC, that will be enough, wouldn't need to have the transistors of allchannels share an IC I hope? If its that simple then I think this could be the solution to my problems, guess I would no longer need a dummy channel to compare to it either with this method in use?

Thank you indeed

P.S. completely forgot t attach image, sorry, Also is this a circuit which ought to have emitter degeneration resistors added, or is this topology immune from needing them to improve stability? r_e_prime on this is, in theory, about 130 ohms on the left hand transistor and 1.3K on the right hand.

[David Hess]

R30 could be replaced with a current source for better performance.

The transistor Vbe voltages still need to be matched for best accuracy.  Usually just having them in close proximity is good enough for temperature tracking.

[Infraviolet]

"The transistor Vbe voltages still need to be matched for best accuracy"
Matching, in this context, is mostly thermal. Two transistors in close proximity, or ideally in the same IC, even if it isn't truly monolithic inside, are all that is required, not selecting specially manufactured matched transistors? And no need for emitter degeneration resistors here either? Am I right in understanding the only property of a transistor which has a strong effect on this circuit and is variable is Vbe, and with this topology the two transistors cancel any changes they both undergo.

P.S. are there any keywords I can use online to find more information about the circuit type I show in reply #68. The app note didn't give it a name as such, and while I think it is related to a current mirror circuit, those don't seem to take input signals on to NPN's emitters.

[TimFox]

The actual V_BE of an individual transistor of a given part number is a (presumably weak) function of the actual geometry, which is going to be different during manufacture from unit to unit.
Monolithic dual transistors are inherently matched, since the two geometries are formed together and should match each other well.
Measuring V_BE of individual transistors (at a given collector voltage and current) from a batch received from Mouser (probably from the same production lot) is straightforward.
Thermal matching thereafter is a matter of design:  you can epoxy TO-92 packages together on their flat side, for example.

[wizard69]

Cn I quickly check how stable the 0.7V drop across a typical small signal diode is, and how accurate between diodes of the same model.

If this is production size devices then you have to consider difference between batches and manufactures.

If one is in a situation where several separate analog signals are all passed through diodes* then can one rely (to within what percentage accuracy 1%, 5%, 10% ,20%...) on all the signals being dropped by the same amount?
Thanks

That all depends upon what you mean by the same amount, the imple answer is NO.  I'm not sure why you are running signals through diodes but this sounds like a device demanding calibration.

*for my situation I can definitely assume small currents, so no signficant heating of the diodes from this, and all physically close together so all the diodes would be getting the same rise or fall in ambient temperature if the environment around the PCB gets hotter or colder

Like I said above not sure what you are doing here but this sounds like an instrument that would need calibration and characterization.   However that depends a lot on what sort of error you can tolerate or correct.

[Infraviolet]

I think its probably got to the point where the only thing to really do is some temperature testing. I don't have much in the way of equipment to get things to precise temperatures, let alone to keep them there when in the breadboard, but...

I'll take a particular pair of transistors, see how they peform in this application at room temperature, then let them heat sitting on a 3d printer bed for a while and try again, then pop them in the fridge for a few hours and another try. That gives a far larger temperature range than I'll actually need the finished peak detector to work for, so it will show an extreme limit of how bad temperature variation can be. Maybe it won't be so bad afterall, I might be over-worrying about these sorts of effects simply becaue I haven't done anything analogue and precise with discrete transistors before, it just I see temperature and manufacturing variation effects mentioned in the context of discretes where you never hear them warned of for op amps or digital chips. If it is enough to be troublesome, calibration is the way to go I guess, I can think of an easy way to make it automatic* and provide a dummy channel for reference which can give datapoints for a signal of known size and for a DC state.

*I'm not mass producing these, but I am making a sort of module which can be included in a variety of future projects and I like to ensure any I build can be quickly put in to use and not need huge amounts of probing for individual calibration, but I think there might be a nice way to make it self-calibrating.

[Infraviolet]

The dual transistor peak detector design (see image in my reply #68) inspired by Linear App Note 47, Fig 97, doesn't seem to be anything like as linear* in the real world as it was in simulation. Any suggestions as to why? In sim, with the values shown, it gave a a straight line for inputs amplitudes above about 0.1V, which outperformed any form of diode of B-E junction based peak detector I tried. On breadboard it is behaving much worse than any diode detector, for larger input amplitudes the peak detector DC voltage falls far below that expected from a linear equation, and farther below even than from the diode of B-E junction detectors.

Why is this working so much worse in the real world than with simulations?

*in this context I mean how closely the output DC voltage follows a linear equation of the form Vo=A*Vpp_in+Offset

Thanks