Low Cost DIY Differential Probe

[Dave_C]

Hi!

It's my first time posting here and I'm still relatively new ( finished learning electronics in trade school, about to start in college )
I'm not really sure if this topic should go here or in the 'Projects, Designs, and Technical Stuff' section so feel free to let me know if I should take it somewhere else.
I've been thinking about rolling my own differential probe for troubleshooting switch mode power supplies. Currently I only have schematic and I'm planning to build it on a proto - board. I'm not going to bother to design a full PCB for it.

My main requirements are:
- Minimal cost ( re-using components from scrapped projects )
- DC coupled
- At least 400V input common mode range
- 500kHz bandwidth or more
- Multiple attenuation ranges
- Ability to drive 50 Ohm Coax at full swing
- Powered by single external power supply or lab power supply
- At least 1M input impedance

DC accuracy is not a big concern here, but still I won't use 5% carbon resistors in a voltage divider 
I decided to use as many parts as possible from my own stock.
Long story short it's using 2 voltage dividers with instrumentation amplifier driving the output.

Dividers consist of five 220k resistors in series, with 10k and an array of nine 10k resistors in parallel (equivalent resistance of exactly 1111,(1) Ohms. Input signal is tapped off at both sides of the 10k resistor. All resistors are 1% 50ppm/K rated, 220k are also rated for higher voltage (350V). Overall Input impedance is 1,1111... MegOhm minus capacitor leakage.
Capacitors C1 - C10 are 8,2pF 500V NP0 dielectric, then a combination of caps to form 180,4pF parallel to 10k and another combination equivalent to 1,623nF. Trimmer caps are used to precisely set the capacitances to improve frequency response.
Diodes D1-D4 serve as a crude input protection. 1N4148 have relatively stable junction capacitances when biased with a reverse voltage higher than 2V ( In this case they will be biased with at least 10V at all times ). Ferrite beads FB1 and FB2 take the edge off a high frequency pulses that would couple through divider capacitors (don't have any data on them ).

Range switch is used to change the input divider ratio and INA gain to get 4 different ranges:
X1, X10, X100, X1000
All of the ranges will survive high voltage on the input ( up to 1750V - 5 resistors with 350V withstanding voltage ) but high common mode or differential voltages might saturate the INA - more info on the attached shcematic.

Instrumentation Amplifier uses classic 3 Op-Amp configuration with IC2 gain set permamently to 2 ( to compensate for output termination ) IC1 is a jellybean TL082 because of it's FET inputs and okay bandwidth. I haven't really settled on the choices for IC2 ( I have a LT1028 - kind of overkill except high bias currents or CA3140 - CMOS input with external offset trimming )
Gain of IC1 is set by the range switch selecting the coupling resistor ( at X1000 there's no coupling, at X100 theres an array of resistors equivalent to 2222,(2), at X10 and X1 it's 201,5). IC2 drives the output buffer consisting of Q1-Q2 and Q3-Q4 transistor pairs. Both pairs are paralleled to be able to withstand shorts at the output at full drive. D6 and D7 provide bias voltage and R38 is the 'current source'  . Ferrite beads at the bases prevent the output stage from oscillating ( I accounted for  lead inductances from my sloppy wiring ) whole buffer stage is under IC2's feedback loop to further improve accuracy and linearity.

Output of the INA is terminated with two 100 Ohm resistors connected in parallel to make 50 Ohm.

Supply rails are provided by a single floating power supply through C24-L1-C25 common mode filter. R61 and R62 force the supply to be symmetrical in reference to oscilloscope ground. C26-C29 filter the rails. Value of 330 Ohm is arbitrary, lowering the resistance decreases the loss of symmetry when output has a DC component, but higher resistance decreases the power dissipation. I'm not sure if I can get away with that, but the design needs to fit in a small case. Let me know if I should add some active components in there.
D5 is crude overvoltage/ESD/reverse polarity protection.

Soon I should get to actually building the circuit to see how it works.

One more thing:
What if I could use Thevenin resistance of input dividers as a gain resistor in a single op-amp difference amplifier?
This would greatly simplify the design and make it easier to fit in a case that I want to use.

Feel free to share your suggestions and ideas 
I'll keep you posted on the progress 

[JS]

  Nice project, I have to come around one of this sooner than later...

  Two things, you want to trim the CMRR in some way, just trusting the 1% resistor or even matching them just won't cut it for the relation between CM and NM you are aiming for. That's pretty hard to do when you have two divider ranges to use, as trimming one would affect the other unless you switch the trimming device, not just the place to trim it.

  Next, your design will be limited by the TL062, I don't know what you are going to use here but the important part is to use low noise fast opamps, if you were to keep the 1k input alone, and ignore the 10k there are plenty of BJT input opamps that should work better in its place. The second opamp is not as near as critical as the first, as it has less gain and there are quite a bit of amplification before so using an equivalent one won't impact much in noise or BW, the choice of the first stage is critical.

JS

[Dave_C]

Thanks for the reply 

I was thinking about inserting a 10-turn pot into the divider, but I'm not really sure how will this affect frequency compensation (I have a limited amount of trimmer caps)

As for opamps I just stumbled across a NoS INA103 from Burr-Brown (TI Datasheet: http://www.ti.com/lit/ds/symlink/ina103.pdf)
I think It's going to be a lot better than any discrete solution that I can build right now, but I'm concerned about it's high bias current ( 12uA ).
I'm tempted to save that chip for later and just roll with TL082 or NE5532 and do it as cheap as possible.
I think I'm gonna ignore that 10k (since it's used only for X1 range - not really useful here) because it will simplify the range switch and go for the NE5532 since it's basically the best BJT input opamp I have right now.

[JS]

  That amplifier is designed for audio, as mic preamp, so for an input impedance of 600Ω usually, if you do the math for optimal impedance based on noise at 1kHz the number is 500Ω. The high bias doesn't matter much on such application, where the input impedance is about 2kΩ and AC coupled, in and out so you can do what you want with the offset. In any case note that the input bias is over the range of temp, at a reasonable ambient would be closer to the typ. The NE5532 also has a high input bias current, not as high as the INA102 but not far away.

  There are better instrumentation amplifiers for this application, some even fets input and much faster, you just need to look further.

  Are you worried about the calibration? usually they have a trimpot as balance, calibrated first with DC for best CMRR. Then two trim caps, tie one input to ground, send a HF signal to the other, calibrate the trim cap as probe compensation, for the other side, use HF CMRR. As you are going for low frequency and not such high impedance you can potentially get away without the caps... place them in the PCB and check if you need it (look for PCB or dielectric hook IIRC, there was something about it in another diff probe topic)

  Then, also you should trim for offset, and if you care for precise gain, you should trim that too but there you might get away with 1% resistors or just selecting them, depending on your expectations for gain accuracy. In some cases there are trims for offset and gain for each range the probe has, so you set it once and forget, otherwise each time you change the range you would need to calibrate again.

JS

[Dave_C]

I've modified the input to include CMRR trimming ( hope I did it correctly )
As for the INA - I don't really have much to work with here 
Fastest opamp that I have in quantity is NE5532. Really fast JFET opamps/inamps are expensive and hard to get where I live.
I could build a discrete JFET input stages for both input stage amps using BF256s or a current source to compensate for bias current. Either of these options will inevitably complicate my circuit and it might not fit into the case.
I can also build the INA with TL082 as the input stage and see what results can I get with it.

[Wolfgang]

... just a hint:

- compute the possible range that a variation (or inaccuracy) of your divider resistors will create.
- I try to get some 0.1% metal film resistors. They are still available and not unaccepatbly expensive.
- Also consider tempco.

Otherwise, interesting project. 500kHz with low-cost amps is a bit sporty, however. Did you simulate this and like the results ?

Much luck !

[Dave_C]

I did not simulate anything   
I sort of prefer to test this in the real world and see what comes out - not a real fan of simulation tools when I have the components on hand. Just have to get around to build it.
As for bandwidth 500kHz might not be possible with my current opamps. I'm considering various options including building a discrete amplifier for the job.
I don't have many 0.1% tolerance resistors right now and DC accuracy is not my primary requierment.
The design is supposed to be the cheapest sort of useful differential scope probe out there
Probe will be used in my lab only (10 degrees C max difference throughout the year) so I'm not too worried about tempco either.

[Cerebus]

Immediate thought: there's no output current limiting. Unless you're much more orderly and meticulous than I am there is an almost 100% guarantee that you'll short circuit the output at some point as you fiddle/fight to get the right test leads in the right place while your mind is on debugging some problem that led to the test you're setting up. Ditto output protection for when it ends up getting -50V DC stuffed into it by some idiot who can't tell the difference between the connection to the scope and the power supply.

[Dave_C]

I don't expect it to survive negative DC at the output ( It will have a BNC connector so It will be hard to mess it up )
Current limiting will require more physical space and it can survive continuous short to grount - good enough for me.
Besides, output would need overvoltage protection if we wanted to make it completely 50V proof not just current limiting 

[Wolfgang]

... as long as you are having fun and a lot of spare time, who am I to stand in your way ? 

[JS]

  Current limiting is easy enough, just two transistors could do it. You could also change R38 with a current source, also not overly complicated or space consuming, a jfet and a resistor could do.

  For the CMRR trim, I don't think you need that much cap trimmers for the compensation, just two should do, not tapping the cap networks to the trimmer ends, just one network by side. You are dealing with pretty low frequency to worry too much about compensation.

  Then, for the resistors choice, I'd go with 1%MF as I have and here parts are also hard to get and expensive. If the CMRR trim is reachable from the outside you can tweak it every use, and just rely the trim caps are reasonable fine. Note that an INA wouldn't improve CMRR, as the input divider would be the limiting factor. And even then, you can't get it perfect as source impedance unbalance will make it change in every measurement.
  Let's say you have your CMRR perfectly calibrated, now you connect the probe to a real world application, where the output impedance is different in both sides, like if you are probing a 10k resistor to measure a low current floating around, now the CMRR is just 40dB, no matter how nicely are the resistor and caps matched on your circuit. Things get worse in HF as the impedance gets lower by the caps in the divider. That's why a higher input impedance is good, other than loading the circuit and generating some error. In this case the error due to the loading will be 1%, which could be totally acceptable but if you are trying to measure 1µA variations in that 10k resistor in a signal that has 10V CM now the CM noise is 10 times greater than your signal variations...

  Note that for measuring small NM signals on top of big CM noise isolation probes are better, as their CM input impedance is really high, and ridiculously high at DC, being just the leakage across the different components on the isolation barrier from the floating side to ground. The capacitance tends to be a bit high so HF CM input impedance is not as great tough, still much better than a differential probe.

JS