Enprovement of a DIY fiber optic isolated voltage probe.

[FriedMule]

Hi I have found a galvanic isolated probe that look like it's a great start but maybe a bit noisy, max 70MHz and bw of 35MHz.
I was thinking if it would be possible to reduce noise, get higher max MHz and maybe a wider bandwidth, by replacing some of the "old" chips.


The author wrote about the possibility to improve the design, but to me do it look like the project is abandoned.

Full article with original schematic here: https://hackaday.io/project/12231-fiber-optic-isolated-voltage-probe#menu-description

Here are some of the most important exemption from the authors notes:

Ideally, the probe will have 50-80MHz bandwidth, without horrible distortion and with less noise that an oscilloscope.
Goal, an isolated oscilloscope probe with 10x attenuation, +-50 volt input voltage range, less than 2mVrms noise, and more than 30MHz bandwidth.

The core of the probe are the IF-E91D and IF-D91 plastic optical fiber transmitter max 70Mhz.

Managed to get 9-10nS rise and fall times. Works out to an approximate bandwidth of 35MHz. Roll-off past 35MHz is first order as best I can tell with a noise source and scope FFT. Might be able to compensate with the post-amplifier and get more bandwidth.

Had to remove trimmer C5 due to the unbalanced capacitive dividers.

First problem is that the OPA 847 is oscillating a bit near 100MHz. The output is also rather noisy. So since the OPA847 has more bandwidth than required due to the 35MHz front end, Boosting it to 15K was enough to quiet the oscillation, but I went all the way to 39K to help with noise. Also dropped the LT1819 gain because of the increased sensitivity of the OPA847 stage... (from 10k)

Noise is now about 15mVrms at the output or 150mVrms at the input.... The noise level is a lot higher than I'd like and limits the useful signal amplitude range of the probe.

 I did attempt to extend the bandwidth of the probe by adding a 22pf capacitor in parallel with R31. While this did extend the bandwidth, it also resulted in a blunt peek at 60-70MHz. Didn't keep the capacitor because the distortion was just too obvious.

I get 636uVrms of noise. (3900MHz gain bandwidth, 0.85nV/Hz voltage noise, 2.5pa/Hz current noise, 30Kohm Rf, Cf around 0.2pf, Cin around 7pf) That's consistent with my 7-10mV peak to peak noise I'm seeing after the 3-6x gain from the LT1819.

A noise-free op-amp would drop my noise floor about 5x. Looks like it's worth spending time to develop a better amplifier

Found a few links online I want to keep track of. First is paired plastic optical fiber. This would allow a V2 probe to use differential optical signalling to eliminate offset drift and allow gain compensation. POF pair http://www.afwoptics.com.au/index.php?route=product/product&path=119&product_id=495

END OF NOTES

I have tried to find some alternative to some of the chips but have a hard time to select one that is more usable: https://www.analog.com/en/parametricsearch/11091#/

[Kleinstein]

The OPA847 is the wrong type of OP for the photo-detector. It just has way to much current noise. The noise matching source impedance is more like .9 nV / 2.5 pA = 360 Ohms. A really fast TIA is a tricky thing, but this OP is more like the wrong one. The more obvious choice would be either a fast FET based OP (e.g. OPA659 or similar) or a discrete FET (JFET or maybe dual gate MOSFET) based front end.
There may be ready made TIAs for the job. Usually the FB resistor is smaller than 30 K for a BW.

[FriedMule]

The OPA847 is the wrong type of OP for the photo-detector. It just has way to much current noise. The noise matching source impedance is more like .9 nV / 2.5 pA = 360 Ohms. A really fast TIA is a tricky thing, but this OP is more like the wrong one. The more obvious choice would be either a fast FET based OP (e.g. OPA659 or similar) or a discrete FET (JFET or maybe dual gate MOSFET) based front end.
There may be ready made TIAs for the job. Usually the FB resistor is smaller than 30 K for a BW.

Thanks a lot! I did fight a lot to find something like that chip, but as you say, it just seemed wrong. But I did think it were me who were wrong.
Am I wrong in also assuming that the LM317 is a sort of a bad choice here, I am thinking on it's mediocre specification?
The LT1819 do also confuse me a bit, is that really the "best" chip, would a faster and more silent chip not be of advantage here?
Sorry for the many questions but I am still a totally noob! :-)

[StillTrying]

I've looked at that hackaday a few times over the years, I don't think these IF-E91D and IF-D91 ideas are worth much, but I'd still like to test the bare devices.

You wouldn't need all that circuitry to test their analogue bandwidth and linearity, which would probably be the first step, I don't think analogue light bandwidth can get anywhere near 70MHz or even 20MHz.

There's some analogue light experiments by JS and me here:
www.eevblog.com/forum/beginners/floating-probe!-for-$2-50/

Another mention of the IF-E91D that goes nowhere!
www.eevblog.com/forum/testgear/cheap-diy-fiber-optic-isolated-voltage-probe-with-bandwidth-up-to-70-mhz/

[Marco]

Lets first guesstimate if you need a transimpedance amplifier, lets say you simply use a 330 Ohm resistor to ground, that shouldn't hurt bandwidth. Lets say you use modest modulation depth to keep things linear and you're left with 10 uW at the receiver, so that gives 6 uA, so that gives 2 mV. Peak to peak resistor noise is 100 uV ... well okay, you need the transimpedance amplifier.

[FriedMule]

I've looked at that hackaday a few times over the years, I don't think these IF-E91D and IF-D91 ideas are worth much, but I'd still like to test the bare devices.

You wouldn't need all that circuitry to test their analogue bandwidth and linearity, which would probably be the first step, I don't think analogue light bandwidth can get anywhere near 70MHz or even 20MHz.

There's some analogue light experiments by JS and me here:
www.eevblog.com/forum/beginners/floating-probe!-for-$2-50/

Another mention of the IF-E91D that goes nowhere!
www.eevblog.com/forum/testgear/cheap-diy-fiber-optic-isolated-voltage-probe-with-bandwidth-up-to-70-mhz/

LOL WOW yes that shows a lot of promises - for failure:-)
Hmm a month ago did I think of a combination of a Probe -> ADC -> Fiber -> DAC -> Scope connection. But came away from it when I read that analog signal were possible to send via fiber optic.

[FriedMule]

Lets first guesstimate if you need a transimpedance amplifier, lets say you simply use a 330 Ohm resistor to ground, that shouldn't hurt bandwidth. Lets say you use modest modulation depth to keep things linear and you're left with 10 uW at the receiver, so that gives 6 uA, so that gives 2 mV. Peak to peak resistor noise is 100 uV ... well okay, you need the transimpedance amplifier.

Hmm yes, hmm that do solve a lot, but what do you say about other commenting that the project being dead before start?

[StillTrying]

Are you going to get a pair to test, they're discontinued of course.

We know what data sheets are like for listing 'applications', if they were good for analogue even <200kHz it would be on the data sheet.

"so that gives 6 uA"

Are there any TIA getting 35 to 70MHz of analogue bandwidth from 6 +/- 3 uA, it would need to have 50kR of gain as well.

[Marco]

A really fast TIA is a tricky thing

The big problem is that opamps are a lousy fit and integrated solutions are all so highly integrated that it becomes useless for a hobbyist (ROSAs) or boutique stuff the manufacturers sell to labs, generally needing to be wirebonded.

So what's left is going back a couple decades in technology and doing it discretely ... would this work for a low noise moderately broadband transimpedance amplifier?

[StillTrying]

Does it work in the simulation. The output impedance looks very high to me and I can't work out the gain.

[Marco]

Does it work in the simulation.

For what it's worth yes, but I don't have a good mental model of either the internals of simulators or what parasitics I have to model (that 10K resistor better be a small one though). So it don't mean much.

The output impedance looks very high to me and I can't work out the gain.

The current source current mostly goes through R1, so it has 10k transimpedance gain. You'd have a buffer on the output, but that's trivial and I have to be careful in that simulator to not go over the analog node limit for the free version.

This is a form of regulated cascode.

PS. the biasing is a bit ad-hoc because of the large variations in threshold voltages for the JFET it might need more or less diodes on the source. Not sure how to solve that elegantly, especially while keeping everything DC coupled. Also the relatively high supply voltage is inelegant, but with Q3 bias current going through R1 it's hard to avoid ... there's no good PNP transistors left for high side current sources compliant at higher frequencies in the discrete realm. The exact transistors shouldn't be important, as long as they have high enough ft (no idea what high enough is).

[Marco]

On second thought ... ignore all that and just use LTC6560. I had looked around for faster analogue TIAs in the past so I thought there wasn't much available, but it seems at these slower speeds there are decent ICs at reasonable cost. The performance is far better than the X GHz opamps you can abuse as TIAs and it's going to be much easier to work with (also easier than doing it discrete obviously).

[StillTrying]

I thought it was a version of LTspice, I might try to simulate it tomorrow, it's tomorrow now, but it'll have to wait.

"ft (no idea what high enough is)."

You're not telling what the bandwidth is! For a probe to be useful I think 20MHz to 50MHz, and the dynamic range will be bad, which just leaves getting the LED to output light linearly at 20MHz to 50MHz, good luck with that.

In my simple 1 and 2 transistor ~TIAs there's quite a difference in speeds between a normal 300MHz TR and a 600-1000MHz RF TR.

The Fiber Optic Photodiode IF-D91 data sheet suggests analogue video up to 23MHz but I can't find any examples of it.

"On second thought ... ignore all that and just use LTC6560."

OK.

[StillTrying]

The inputs are all 0 to 10uA 4MHz square waves.
The outputs of the 1st 3 are around 95mVpp, the LTC6560 one is 370mVpp.
The input impedances seen by the 10uA inputs are about 40R, 50R, 60R, and 240R for the LTC6560.

[FriedMule]

Very sory for first replying now!!

I hav now used the last half of a year to try to find a DIY isolated probe, not high voltage, differential or anything needed, just an isolation so I can measure willi nilli without screwing the measurements or blow my scope up.

I have found several articles where a electronic engineer is using pages to explain how it works, what he tested the unit to and what may be needed to improve the design. But when I run it by your smart guy's, you do all say, this wont work or it is not doing as described in the article or it's in fact not isolated. And I have to start all over again, I have even tried several solution my self, but with my near zero knowledge, has it not a chance to work.
I am not interested in paying for a commercial solution, it's fare to expensive or is not usable to my purpose.

Isn't there some sort of sollution where I can just isolate the ground lead from the scope, am I the only one who just need to measure two pleases where the ground are not common or who have can't spend $1000 to trow out on that?

[StillTrying]

If it was easy to make an analogue opto isolated probe we'd all be making them.

If <200kHz is good enough, it becomes a bit more possible.

[Marco]

The point of the circuit I posted was to be able to use a jfet at the noise determining input transistor and it really needs a bloody fast BJT to make it work in simulation (like BFP720FESD). A single transistor transimpedance amplifier can work with a BJT, but current noise is higher and the noise corner is at 100 kHz instead of 1 kHz.

[Marco]

But when I run it by your smart guy's, you do all say, this wont work

I don't see why the LTC6560 and that LED/photodiode pair wouldn't work ... you'd still have to experiment, but I wouldn't be unoptimistic. You'd still need to speed up the LED driver of course, but just using a single fast BJT with a resistor on the base and emitter and the LED on the collector with a signal generator will be enough to experiment.

[FriedMule]

I have read all your answers and so on, I have to admit I may have misread some of your reply's but I'll try to rephrase my question, from all the great new knowledge you have given me:-)

My goal is to be able to measure low voltage signals (i.e. noise in a circuit) and "larger" signals (i.e. output from a step down transformer)
I would like to be able to use the probe to detect fault, different errors and so on, I do not need to test main voltage but on the other hand, would I not mind to be able to.
About frequency, DC or almost DC to what ever is realistic possible, so if 100MHz is max, I have to accept that, but my scope goes to 350MHz

In short I would like to be able to measure as much as possible, but it do also have to be realistic.:-)

[StillTrying]

"My goal is to be able to measure low voltage signals (i.e. noise in a circuit) and "larger" signals (i.e. output from a step down transformer)"

Even making a just DC analogue light isolated probe with inputs of +/- 10mV to +/- 10V would be a bit of a nightmare.

Turning a +/- 1V input into 5mA -15ma through the LED would be the easy bit, but by the the time it's turned into light, turned back into current in the PD and then back into a +/- 1V to the DVM or scope, it will be drifting all over the place.

There's the IL300 with 2 internal PDs so that one can be used for feedback but <200kHz.
www.vishay.com/docs/83708/appnote50.pdf

Someone else will be along in a minute to say making a 100MHz analogue optically isolated probe is quite easy.

[Kleinstein]

Even making a just DC analogue light isolated probe with inputs of +/- 10mV to +/- 10V would be a bit of a nightmare.

Getting the large dynamic rage optically makes things more complicated. The more practical way would likely include range switching before the optical transmission. With just the usual 8 bit  DSO resolution drift and linearity are not that difficult / impossible. DC drift could still be a weak point. So one may end up with a kind of AC coupled probe only.

With reduced demand on the dynamic range one might get away without a full TIA, but just a simple amplifier in base or gate configuration. A photo-diode can be reasonable fast with a 50 Ohms impedance.

[FriedMule]

As I understand it, is there the optos, like the IL600, as written, and transformers (not DC), digital isolation and fiber.

I did try to make some sort of combination, where the IL300 could take the DC and up to a few KHz, then a transformer up to maybe a few MHz and again a high speed transformer, up to some hundred MHz.
As I did understand it, could it be done, but not easily.

[GonzoTheGreat]

Would it be possible to use the an SFP+ module containing a 10Gbps transceiver for optical fiber, to modulate an analog signal on that fiber by tapping the SFI signal before the L2/SerDes/Ethernet encoder/decoder ? 
Nowadays, these optical transceiver modules for multimode OM3 fiber can be had for as low as $20. The ones for single mode fiber are more expensive (see here) but they are less likely to send data in parallel using multiple wavelengths.

In case of limiting transceivers which accept only SFI digital differential signals (e.g. a pre-detected digital serial data stream), perhaps some form of semi-digital modulation could be used to encode the analog signal, e.g. DSM or PWM because it is easy to convert a PWM signal back to analog by integration at the receiving end ?

See a sample datasheets of such optical transceiver modules:
https://eu.mouser.com/datasheet/2/1116/10G_LR_APSP31B33xDL10-1949253.pdf
https://www.mouser.com/catalog/specsheets/ELX10GDL0610710ELT_ka.pdf

Are there any GHz PWM modulator chips available off the shelf ? 
They would not have to drive the LED/Laser directly because the circuitry integrated into the SFP+ module already takes care of that.

[Kleinstein]

Chances are high the transceivers are very much integrated and not much chance to tap of the signal before limiting / nonlinear amplification.
There is no may to convert a high speed signal to PWM modulation - if at all it would be having the ADC before the fiber and than send digital data at some 10 Gbps which may allow up to 1 GSPS at 8 Bit.

[GonzoTheGreat]

Chances are high the transceivers are very much integrated and not much chance to tap of the signal before limiting / nonlinear amplification.

If you read the SFP+ module specification you will find out that some of them are linear (non-limiting)
The 2^nd datasheet, which I have linked to in my previous message, refers to such linear transceiver.

There is no may to convert a high speed signal to PWM modulation

Even if the limiting type of module is used, I don't see a reason why its ~10s of picoseconds timing accuracy wouldn't allow a VCDL to PW modulate 10s of MHz of analog bandwidth onto it.

Also, PWM is not the only modulation that can be used. I think a continuous-time DSM is a possibility, too: See:
https://nanoscalereslett.springeropen.com/articles/10.1186/s11671-020-3284-4

Also, I don't see a reason why FM or PM schemes couldn't be used to turn a 10-16Gbps digital link into a 100MHz analog link.

[mawyatt]

Interesting idea!!

If you could get into the WDM effects, would think you could use a linear Phase or FM modulator (Lithium Niobate comes to mind, or maybe "pull" the laser diode to modulate the optical frequency) to allow a reasonably good analog bandwidth. If you could use two WDM channels, one would be fixed with no modulation while the other is modulated, this opens up many possibilities for communicating the analog signal down the fiber.

Anyway, hope someone gives this a try, seems like a interesting approach whichever way it's done

Best,

[GonzoTheGreat]

The optical Tx modues in SFP+ transceivers are sealed.  I do not know how to linearly pull their optical wavelength millions times per second.
Take a look at this video. It depicts a 1Gbps (SFP) module (not the 10Gbps SFP+) that illustrates the limited access to the optical components:
https://youtu.be/GGuqtZOurZo

[Marco]

They make ROSA/TOSA for CATV which should be somewhat linear, but I never see them sold as components.

[StillTrying]

They make ROSA/TOSA for CATV which should be somewhat linear, but I never see them sold as components.

They look too expensive and complex, at least for me to play with, and as usual I couldn't see any analogue use on the data sheets of the ones I looked at, and they seemed to start at 5kHz rather than DC.

[Marco]

The ROSA/TOSA themselves are as simple as it gets (laser) diode and wirebonded die amplifier, with pigtails and optical connector.

They are not DC coupled, but all the bare die really fast transimpedance amps aren't either. They have to be analogue and fairly linear or they would screw up the tv channels. As I said, "for CATV", not ROSA/TOSA in general.

[GonzoTheGreat]

This $30 optical module is linear but AC coupled.
https://www.mouser.com/catalog/specsheets/ELX10GDL0610710ELT_ka.pdf

[StillTrying]

"This $30 optical module is linear"

It looks digital use only to me.

[Terry Bites]

As your transmit side is mains powered why not digitise and then send data over the isolated link? Digital isolators are fast and cheap. OK the cost of a fast ADC, chipageddon, micro controller faff.
TI and AD have photodiode design wizards, well worth a look.
You can also compensate for drift using a diferential configuration . Though not with an antuique fron the TL0 series!
Other toplogies include https://www.digikey.com/en/articles/design-transimpedance-amplifiers-for-precision-opto-sensing

[Marco]

AFAICS only the SFP+ receiver on a linear module is guaranteed to be "linear" (except it can still have AGC). I see some papers where they abuse transmitters for commodity modules sold for digital modulation with PAM, so the transmitter can be linear too I guess, but don't count on it.

Here's a laser diode with fiber connector on alibaba supposedly for CATV. The price on the page is certainly very interesting, but it's alibaba so who knows.

[Terry Bites]

Ah.
I assume your measurement side is a scope of some sort, earth referrenced.
I that case, powering the HV side from mains means that all youve realy got between the scope in the way of isolation is the transformer insulation in the PSU.
Worse still youve got interwinding capacitance pasing even more leakege curent.
So that kind defeats the object?

The mains powered parts need to be on the "earthy" side of your system. A psu failure may occur because of the high voltage difference between the iso side and the earthy side. That voltage will be seen across the transformer. Shazam!

[StillTrying]

When I was doing the photodiode and linear LED light and experiments I must have read 100s of web pages and pdfs and there was nothing much out there on it.

There seemed to be not many more than JS and me that's tried it.
Some experiments here www.eevblog.com/forum/beginners/floating-probe!-for-$2-50/msg1817129/#msg1817129

To get any linearity and a least some dynamic range out of the LED light the speed seems to be stuck at not much more than around 300kHz. It seemed to me I had to get the LED current very high (X10) to get the speed up, but then they could only be used pulsed for just a few us and duty around 1%, because at high currents the LED light droops after only a few us.
You'd think there was at least one video speed linear light transmitter and receiver out there but I didn't find it.

[Marco]

I wonder what the modulation bandwidth is of those ridiculously cheap red laser diodes is (the ones with just some PCB and metal sheet stuffed into a lens housing). They aren't exactly build for low capacitance, but it's probably low enough for a couple 100 MHz.