EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: c4757p on July 14, 2013, 02:31:41 am
-
Simple question: would something like this HCPL-7840 isolation amplifier (http://www.mouser.com/ProductDetail/Avago-Technologies/HCPL-7840-300E/?qs=s%252bgfxR24RQ4TiD3JOvQTtA==) be a reasonable thing to base a low-cost, low-bandwidth isolated probe on? Obviously with external attenuators, input protection, etc.
Alternatively, if I wanted higher bandwidth than 100 kHz (really, I'd like a MHz or two, but can't find any affordable isolation amps with such a bandwidth), how about rolling my own with a reasonably quick ADC feeding a digital isolator, and a quick DAC on the other side? Can anybody recommend an ADC and DAC for this purpose? I don't have much experience with ADC/DAC and really suck at navigating catalogs for them. I don't need precision, 8-bit would be just fine, considering that's what DSOs usually have anyway...
Edit: What about:
ADS7949 8-bit, 2Msps ADC (http://www.digikey.com/product-detail/en/ADS7949SRTER/296-27833-1-ND/2409512) and AD5450 8-bit, 2.7Msps DAC (http://www.digikey.com/product-detail/en/AD5450YUJZ-REEL7/AD5450YUJZ-REEL7CT-ND/2063129)? Perhaps a little CPLD or something to manage shuffling the data between them, unless (I haven't looked and I doubt it) it really is as simple as feeding SDO into SDI and pumping SCLK...
-
If you look at TI's high speed ADC's they are differential. The recommended Op Amps in front of this are also differential. And in the examples shown, some have transformer coupling.
C
-
Wouldn't transformer coupling eliminate DC response? Or is this transformer coupling on the output? Maybe my brain just farted, but I'm not really sure what you're talking about.
Maybe the question wasn't clear enough - I don't just want a differential probe, I want an isolated one, up to at least a couple hundred volts isolation. I'd just do all analog if I wanted low voltage differential. I want something to blast the signals across an isolation barrier.
I know the ADC input itself is differential, that's normal.
-
Wouldn't it be possible to split DC/LF and HF with a C-R filter and use separate isolators for both and then combine them again?
The voltage across the capacitor would be referenced to "ground" with a difference amplifier and transmitted through a classical VFC->digital isolator->FVC setup. The voltage across the resistor would be transmitted pure analogue through isolation capacitors, on the other side you need another difference amplifier to recover that voltage and then you can add them again?
Regardless how you do it I think this type of probe is going to present a fair amount of capacitive loading (not for the MHz range, but for the 100+ MHz range) if you just float the non isolated part of the probe with whatever the ground clip is connected to ... the stray capacitance of the "ground" plane of the non isolated part of your probe is put into the circuit after all.
-
When you switch a scope to ac coupling you break the dc path to just one side of signal path.
The direct connected differential input op amp would allow DC to very high frequency. With the transformer couple output you lose some low freq response just like a scope on ac. input. But here you would not have to worry about the problem of the second half of the scope's signal input being ground tied.
So transformer output gives some low frequency to Ghz.
To handle the dc to low frequency part you could use one of the dc to low freq optical parts.
A transformer with both windings having a Split center tap. Connect each of the two center connections by two low speed optical analog connections.
You have a differential DC to low frequency optical connected center section with an outside low frequency to high frequency section.
There may be some impedance problems with this in the frequency overlap region interacting with op amp's impedance. If input to above is linear then the output should be linear.
C
-
@Marco
That would be possible, I will look into it. One problem I can see is that I'd like to avoid having too much capacitance between the two, and I think getting the passbands to sufficiently overlap with a very small coupling capacitor would be difficult. I'll definitely try it out, though.
I do want to keep capacitance down, but not "RF" down. I'm just looking to attach this to low frequency/near-DC "grounds" floating at a higher voltage than true ground (the ground rail of the primary side of a SMPS, for instance), not to use it as a true differential probe and put it on things with high frequency content, like the switching nodes of said SMPS. I will definitely try to keep the capacitance between the two circuits relatively low, though, for instance in the power isolation transformer.
I can accept roughly the capacitive loading of the common lead of a bench DMM.
@C
My 2:00 AM eyes and brain are having trouble making it through your post, I'll come back to this in the morning. I would prefer a transformer to a capacitor. Thanks!
-
A transformer with both windings having a Split center tap. Connect each of the two center connections by two low speed optical analog connections.
You have a differential DC to low frequency optical connected center section with an outside low frequency to high frequency section.
Ah, you use the magnetics as the signal adder ... but I don't understand why you need the split transformer, wouldn't this simply work as well.
(http://i278.photobucket.com/albums/kk105/_MfA_/isolation_zps7d9fd185.png) (http://s278.photobucket.com/user/_MfA_/media/isolation_zps7d9fd185.png.html)
-
I'm working on something similar. You'll need to add a crossover filter and the 'DC' part needs enough bandwidth so it rolls off far beyond the crossover frequency. Same goes for the transformer coupling.
-
Is there anything to the ADC->DAC idea? Seems like it might be easier to get everything right with relatively inexpensive off-the-shelf parts.
-
I'm working on something similar. You'll need to add a crossover filter and the 'DC' part needs enough bandwidth so it rolls off far beyond the crossover frequency.
You can always use a higher order filter.
-
But then you'll get a lot of pass-band ripple and its sensitive to component variations. Ideally the high pass and low pass section of the filter should be made using the same component values so its likely you end up with components from the same batch and hence mostly the same error.
-
if you want the best possable you should use all differential signal paths and a balanced design every thing equal and opposite.
C
-
But then you'll get a lot of pass-band ripple and its sensitive to component variations. Ideally the high pass and low pass section of the filter should be made using the same component values so its likely you end up with components from the same batch and hence mostly the same error.
AFAICS you don't need seperate low pass and high pass section at all ... you can just use a high pass filter and then use a difference amplifier to extract the low passed voltage between the input and the output of the filter, as shown in the above schematic. This will guarantee an exact splitting of the signal, pass band ripple is then not really a problem. The transformer frequency response might be though, but hey ... you can always split it three ways and use two transformers to make it easier to cover a wider frequency range :) (I think capacitive coupling has the advantage in getting a flat response over a wide range.)
-
if you want the best possable you should use all differential signal paths and a balanced design every thing equal and opposite.
C
That would complicate the design considerably though, you need to use a high impedance divider to create a virtual ground for the circuit and then you basically have to double the circuit complexity ... for instance there is no single DC/LF signal from the differential point of view, there are two relative to the virtual ground from the divider.
Just using one of the inputs as virtual ground simplifies things ... and at that point going differential for the magnetics doesn't make much sense any more, there's already so much asymetry in the way the inputs are treated.
That said, using a virtual ground from a high impedance divider would get rid of the capacitive loading issue ... even then I'd just put a difference amplifier in front and then handle everything afterwards as in my schematic though, I doubt doubling up on everything will gain you much.
-
First you should look at the good differential input & output chips.
These chips are very easy to use while curring many problems.
Second you should not push poor single side design things in to the differential design.
It is very simple, single side design puts the return path in with noise.
Two resistors and a wire can make a big difference just by not putting that return signal in to the noise that exists in common paths.
To sense a voltage, instead of two resistors and one signal path, you use three resistors and two signal paths.
With differential, you are not measuring one side to ground and the other side to ground, you are measuring between the two points. You do not need or want a ground, you just need the two signals paths in the common mode range of the amps. To push the limits you add a third wire so that the output of one amp and the input to the following amp can work together to keep two signals in common mode range.
Look at modern scopes. the single ended signal from the probe will as soon as possible shift from single side to differential.
Look at everything happening. PCIe, SATA, SAS, USB, RS486, CAN, The before if it existed was single ended at some point, now the speeds are only possible with differential. With SAS even the before was differential and there were big problems with time on wire differences of the many parallel paths
That would complicate the design considerably though, you need to use a high impedance divider to create a virtual ground for the circuit and then you basically have to double the circuit complexity ... for instance there is no single DC/LF signal from the differential point of view, there are two relative to the virtual ground from the divider.
In simple terms this is single ended thinking. When you take a reading with Your Dmm. it is reading the differential of the two leads. I repeat you do not want or need a ground with differential.
For differential, A voltage divider is three resistors instead of two.
in place of a filter that is a cap- inductor-cap with the two leads of the caps connected( a three wire connection) you insert a second inductor connecting the two caps to change from single ended to differential( a four wire connection). Not two separate single ended filters, but one built to work in a differential mode. What is happening in one inductor should be the opposite of what is happening in the other inductor so you could replace the two inductors with a common mode choke.
C
-
You don't absolutely need the virtual ground from a divider ... but if you don't have it the circuit becomes yet more complex, the virtual ground allows you to share some parts of the circuits, which is to say the power supplies. If you don't use a virtual ground from a divider AND you want to keep everything differential then you need to duplicate the supplies as well.
All very unlikely to be worth it IMO.
PS. when you take a reading with your DMM it's measuring the difference, but it's treating one of the inputs as a virtual ground for the circuit. One input is special, changing them around can change the measurement beyond just the polarity (the stray capacitance to earth isn't equal between the inputs).
-
Is there anything to the ADC->DAC idea? Seems like it might be easier to get everything right with relatively inexpensive off-the-shelf parts.
Look, I don't mean to be pushy, but this isn't something I'd like to spend a long time on. There are lots of ways I could go about this as an all-analog circuit, but as far as I can tell that just complicates things. Good isolation gets trickier, matching between DC/LF and HF gets trickier, and parts get perhaps more expensive. Does anybody see a reason why I can't just use the digital method? One of those cool digital isolator chips that use an RF coupling to do >100Mbps signal transfer, and an electrostatically shielded transformer (or hell, a battery) for power? That would make something small enough to fit in a slightly chunky handheld probe (not sure if I want to do that, but it would be possible if I did) with no (or few) adjustments to make. I'm only looking for 1MHz or so of bandwidth, which is certainly not "just throw it on a breadboard" territory, but not really hard either. I'm aware that I have digital noise to worry about at that point, and data rates that high do need some attention paid to them, but I can handle that.
-
Sure it can work, but would you really need a CPLD? I think some of the faster MCUs should be able to handle the serial signals and be the glue between the ADC/DAC.
Of course if you're simply comfortable with CPLDs and have all the tools already I guess it would make sense.
-
Nah, I don't need a CPLD, it was just a quick guess. I've got a tube of cheap Altera MAX 3000As for "I want this finished yesterday" glue logic. Looking at the data formats, it looks like I could do with a bit of discrete logic (possibly two 74HC00 or similar) and a slow AVR or PIC to handle initializing everything. I'd rather not use a MCU I've never used before and have to learn the toolchain and everything, and I don't have any experience with ones that fast.
-
You know, you have just reinvented the Tekscope there..........
If you have digitised the signal, isolated it and passed it past an isolation barrier you might as well just use a FPGA and USB to connect it to a computer like a Picoscope but actually usable. otherwide converting back to analogue to feed a scope will make it more usable.After all you only need an 8bit converter on both sides, a PGA on the input but nothing on the output aside from an anti aliasing filter.
-
If I run it to a computer I have to implement the triggering system and all the software myself, which I really have no interest in doing. If I just blast the signal into a scope I can use its trigger and interface.
otherwide converting back to analogue to feed a scope will make it more usable.
That was already the plan! Did you mean "less", or am I just misunderstanding you?
-
Lets not get too excited about differential unless noise and crosstalk are a real problem.
edit: What is needed is a differential probe. There are many diagrams out there to build your own but they can also be bought on Ebay for a reasonable amount of money.
-
Lets not get too excited about differential unless noise and crosstalk are a real problem.
Considering the architecture of an isolated probe, I really don't see any point in a "true" differential system. It has its own isolated ground and only measures one signal, why not just ground it to the signal's own reference? Differential is great for many things, but kind of pointless here.
All the fast ADCs are at least "pseudo-differential" anyway.
-
edit: What is needed is a differential probe. There are many diagrams out there to build your own but they can also be bought on Ebay for a reasonable amount of money.
Not high voltage ones, they're quite expensive. I do have a low voltage one. Besides, it's just a little hobby project, I'd rather like to design this myself.
-
edit: What is needed is a differential probe. There are many diagrams out there to build your own but they can also be bought on Ebay for a reasonable amount of money.
To handle high common mode voltages they need high ratio dividers ... by using galvanic isolation you can do a low voltage differential measurement at high common mode voltages with small ratio (low noise) dividers.
It's also quite amazing the amount of money people dare ask for what is in essence two high ratio dividers and an instrumentation opamp in a box ... high voltage differential probes are ridiculously expensive.
-
.
You don't absolutely need the virtual ground from a divider ... but if you don't have it the circuit becomes yet more complex, the virtual ground allows you to share some parts of the circuits, which is to say the power supplies. If you don't use a virtual ground from a divider AND you want to keep everything differential then you need to duplicate the supplies as well.
Still pushing single ended thinking I see.
You do not need to "duplicate the supplies" for each op amp.
If you look at the proper design of single output op amp circuits the two inputs need some balance even for single ended use. With a voltage based differential input & output op amps you are just adding a second matching feed back path to the second side. The changes are small as most is in the chips.
You really need to look at some good differential design practice in stead of pushing your single ended thinking.
At one time you had to build the differential input & output from single ended chips. Today all that mess is put in a chip as a package. The big problems are fixed by having it all in one chip.
It can even become easer to do balanced design.
c4757p
With a floating supply for power you will be limited by the by the isolation you select.
If you build with Differential input & output chips like TI and others have, you can easily go for speed or high accuracy that will be hard to do with single output op amps and it will be easer.
With a digital output across the isolation, the digital speed of the isolation.
If you are feeding the output to a scope then digital will need a DAC, again Differential input & output chips will make it easer.
Just copy some of the examples that TI and others have.
If you do not use digital and use transformer or caps then you loose some low frequency response. but the circuit will be cheaper.
At the low end there are some slow speed analog isolation blocks that include DC.
onsidering the architecture of an isolated probe, I really don't see any point in a "true" differential system. It has its own isolated ground and only measures one signal, why not just ground it to the signal's own reference? Differential is great for many things, but kind of pointless here.
With a 10 Meg input you will have 10 Meg in one side for single ended and all the strays will be connected to one side. Like your hand to probe strays
With a 10 Meg differential input you have 5 Meg to any strays and less effect on the measured points.
Remember that both test points can be changing so differential will effect what you are trying to measure less.
C
-
You really need to look at some good differential design practice in stead of pushing your single ended thinking.
Normal differential design practice doesn't reference the supplies to the inputs, they simply reference the supplies to true ground and present an identical input capacitance to ground to both inputs.
By having nothing to reference the supplies to except for the inputs we have a few choices :
- reference the supplies to one input, easy but destroys symmetry and introduces uneven capacitive loading of inputs (this is what DMMs do). At this point you might as well build the rest of the circuit single ended as well.
- reference the supplies to the centre of a divider between the inputs
- build everything symmetrically, including the supplies.
The supplies need to be referenced to something, and something close to the common mode voltage of the inputs, what other option do you see?
-
I'm OK with asymmetric loading. The only thing the "ground" terminal of this probe will be connected to is something that's already "ground-ish" and can handle a bit of abuse. I may even intentionally put a few pF between the grounds to keep the power supply from radiating noise when the circuit it's connected to isn't solidly ground referenced.
Bickering about differential circuits aside, I think I'm going to go with my current design. Obviously the analog frontend is still yet to be done, so there's still room for this sort of thing there.
Obviously there's no point in differential anything on the output. It's just driving the signal down coax into a scope. Unless you want me to build a new scope as well, with differential inputs, that is.
-
I've designed similar signal isolators before using a couple of PICs, really quick and dirty sort of stuff, although admittedly at much lower bandwidth. The 10-bit internal ADC result from one PIC gets fired over an opto-isolator out of the UART into the RXD pin of another PIC. Then straight into the ECCP module in PWM mode to act as a cheap-as-chips DAC. Not hugely accurate, but surprisingly effective. Using SPI instead can increase speed at the expense of another opto.
Obviously, you couldn't get this thing to go all the way to 1MHz bandwidth, but with external converters and a fast enough micro/FPGA on either end as you suggest, I don't see it as being a problem.
Or, use an isolation amplifier: http://www.analog.com/en/specialty-amplifiers/isolation-amplifiers/products/index.html (http://www.analog.com/en/specialty-amplifiers/isolation-amplifiers/products/index.html)
-
Or, use an isolation amplifier: http://www.analog.com/en/specialty-amplifiers/isolation-amplifiers/products/index.html (http://www.analog.com/en/specialty-amplifiers/isolation-amplifiers/products/index.html)
Ha... yeah, that's what I said at the beginning. I decided I wanted to extend the bandwidth, though, and I haven't even seen one of those above 200 kHz before, and if they do exist they're probably very expensive. (I'd love to be proved wrong, though!) Then I still have to build the frontend and power supply, and at that point I'd probably have paid for a lot more bandwidth than I got.
-
Here you go... I reckon you could pick one up for 10 bucks or something... http://www.thinksrs.com/products/SIM984.htm (http://www.thinksrs.com/products/SIM984.htm)
-
Look at the output stage of a op amp
you are changing the resistance of the output pin to the two supply leads.
In simple terms you have a pot with the wiper as output.
With a differential output you have two of these,
As one pot is changed one way the other is changed a matching amount the other.
The output will self balance at the center of the supply.
With the output at center the input will also be at center.
c4757p
save your self some design time and look at TI or one of the others with full differential chips and look at the examples for the the analog path
Will be easy and quick. The examples will also cover going to or from single ended
C
-
I want to design it myself. I don't gain experience designing electronics by buying them!
-
I want to design it myself. I don't gain experience designing electronics by buying them!
You will be designing it, you will just be building with better op amps for better lower noise circuits.
The same ideas on picking single output op amps are needed on the differential outputs.
C
-
Completely agree!
-
I want to design it myself. I don't gain experience designing electronics by buying them!
I always find it way more exciting to build something I can't buy OR come up with a solution which is way cheaper O0
-
I wasn't talking about differential output amps, I meant prepackaged isolation amps. An isolated probe isn't something I have a pressing need for - I'll gain more knowledge from building one than use from buying one.