-
When debugging circuits, I've often found it would be useful to have an active differential probe. The vast majority of the circuits I work on are low voltage (<15v), which means the high division ratios provided by commercial probes (eg x50 and x500 for the low-cost Micsig DP10013) are not required, and attenuate the signal much more than is ideal.
As such, I've decided to design my own probe. I am aiming for the following specifications:- 10x attenuation ratio
- Around ±30v input limits
- >100MHz bandwidth (for use primarily with a DS1054Z)
- Low cost (which means a 2-layer PCB)
- Low circuit loading (around 5Meg input impedance)
I have decided to use a standard instrumentation amplifier design, consisting of three discrete opamps (two buffers, one summing). For the opamps, the ADA4857 seems to be a good fit for my requirements, with 850MHz -3dB unitity gain bandwidth, a 2.8kV/us slew rate, and fairly low input bias current.
http://www.analog.com/media/en/technical-documentation/data-sheets/ADA4857-1_4857-2.pdf
Basic instrumentation amplifier arrangement
I have put together a simulation (see attached), which seems to indicates the performance should be pretty good, with a very flat frequency response, and a -3dB bandwidth for the entire probe of around 300MHz. Before I move on from simulation to designing the final schematic, it would be great to get some feedback on my design from the forum. For the actual design, the 3.3Meg resistors will probably be slit into multiple (to allow for a higher voltage variant to be made if required), and some of the tuning capacitors replaced by variable ones.
In order to avoid reflections from unterminated coax, I have designed it to be used with a 50R feedthru terminator. Obviously this means the probe itself only needs 5x attenuation (as the 50R series resistor and 50R terminator add another 2x).
Gain lineup:- 0.216 gain in the first divider
- 1.133 gain in the buffers (adjustable via R2)
- 0.846 gain in the summing amplifier
- 0.500 gain in the series 50R resistor/coax termination
The final stage seems very sensitive to small amounts of capacitance (C1, C10). I decreased the size of R4,R5,R6,R7 to help with this, but I wonder if a few pF of parasitic capacitance from the PCB might cause issues here. Does the magnitude of the resistors I have selected (hundreds of ohms) seem appropriate?
The AC simulation response looks great until you get to very small input signal (eg the 10mV curve attached). I presume this is due to the probe's noise floor, as even with microvolts of input this 'hump' is much the same. If this is the case, that's fine (obviously an active probe will introduce some noise) - hopefully I haven't missed anything.
I'd be interested to hear any thoughts or suggestions on the design!
-
The ADA4857 is not really suitable for a high impedance source like the divider. It has way to much bias an current noise: At 1 M input impedance and 1 pA/Sqrt(Hz) this would be noise in the µV /sqrt(Hz) range thus rather high. It is not as bad as it looks because the higher frequency part is much lower impedance due to the capacitance, but still quite some LF noise.
For such a high impedance BJT based amplifiers usually don't work well. The more obvious choice are JFET based ones.
With 100 MHz BW this sounds like a difficult task. -
Thanks Kleinstein, that's helpful.
Initially when I was looking at opamps, I considered the ADA4817 - that's a JFET amplifier with 20pA bias and 2.5fA/Sqrt(Hz) current noise. The slew rate is lower at 870v/us, but that should just be sufficient (an output of 4v pkpk @100MHz).
http://www.analog.com/media/en/technical-documentation/data-sheets/ada4817-1_4817-2.pdf
I've switched for that in my simulation model - however, the AC response still has some weird stuff going on. The 10mV curve still has a lot of high frequency gain, and the bandwidth seems to decrease slightly as the signal amplitude decreases (see the attached response).
Note that this is all due to the final summing opamp - looking at the test point 'VBUF' after the buffer stage shows an excellent flat response at all amplitudes (to around 700MHz).
Is there other important parameter I'm missing for the selection of the final opamp, or is my simulation (without parasitics) not trustable at these high frequencies?
-
This graph also shows how the capacitance in the feedback paths (C1 and C10) changes the response for the 100mV curve in the previous post - it looks much more sensitive than I would like.
Not being very experienced with this high speed circuitry, I'm not sure how high the parasitic capacitance will be - ideally I'd like to be closer to that 0pF curve...
-
You'll want to examine the phase response, too.
-
Oops, didn't see you wanted -/+ 30V .... that makes this less than relevant, still.
The Chipwhisperer probe is still available for 57$. It uses an AD8129 which has an internal architecture using transconductance stages, much more suitable to differential amplifiers at high frequency. Just need to watch out that you can't use the zero'ing pot, because it screws up the low frequency response. -
Many JFET OPs have the problem that input capacitance can depend on the common mode voltage. This may not be reflected correct in all models. In addition one may overlook the resulting voltage dependent frequency response.
The ADA4817 looks good due to the really low input capacitance.
One could consider discrete JFETs too, as the input OPs are mainly used as buffers. -
To drive capacitive loads (some pice of coax) you might want to add a resistor at the output of the driver opamp inside the feedback loop, you still need to have some capacitive feedback before the resistor. This leads to poor HF response, you might want to add some extra gain to compensate up to the desired frequency, some capacitance in parallel with R4 and R5 in your schematic, like the input dividers. Congrats! now your circuit is unstable! Add small resistors in series to those caps, or even some tiny inductance that kicks in once your desired BW is over.
I drawn an schem to explain myself better.
I'm also planning on a custom differential probe but for low level low speed stuff. Trying to avoid all normal mode attenuation while maintaining CM attenuation, I'm being bitten by the simulator with this! But this is low noise, audioish BW which what I'm familiar with, so I think something will come out. Worst case I do very low noise stage and live with the attenuation. I'm not aiming for more than 1MHz but at least X0.1 probe with about 30V CM input. So shouldn't be so hard. This are for measuring audio circuits, power supplies (floating current sense while measuring output at the same time) and that kind of stuff.
JS
-
Some active probes do use bipolar inputs but they use much lower impedance input dividers. Since the source impedance unbalances the input impedance limiting common mode rejection of a differential probe, good common mode rejection performance requires the highest input impedance possible. If anybody made a bipolar input differential probe, then it was probably intended for power supply and digital signal measurement where the impedances are very low.
Active probes limit the input voltage range of the FETs to like +/-1 volt to limit capacitance modulation and other effects although cascodes can be used up to at least 150 MHz in discrete designs to ameliorate this allowing at least +/-10 volts. If you want +/-30 volts, then input dividers are going to be needed.
Maintaining AC common mode rejection at high frequencies is difficult and high impedance input dividers make it even more difficult because the printed circuit board substrate becomes part of the circuit; search for circuit board "hook". Dual devices like transistors and operational amplifiers are a problem because of coupling between the sides so dual devices are not necessarily an advantage.
When you layout the board, consider including the test jig for adjusting the high frequency performance using a precision 50 ohm source.
-
I have been collecting notes on differential probe design for a while and your post encouraged me to finalize some of my notes last night which might be helpful.
Study the design of the Tektronix P6046 which was the industry standard low voltage differential probe for decades. Notice that it uses transconductance amplifier stages throughout instead of shunt feedback amplifiers which are more problematical.
Linear Technology (LT1193/LT1194) and Analog Devices (AD8129/AD8130) (1) make some transconductance difference amplifiers which approximately duplicate how the differential to single ended conversion stage in the P6046 works and these will be much better than any instrumentation amplifier. They have good high frequency performance and do not require AC common mode adjustment. They also support gain and offset adjustment at the output instead of an earlier stage.
I think you could get pretty good performance with a pair of JFET buffers driving an AD8129/AD8130. Something to watch out for is the drive capability of the AD8129/AD8130; 35 milliamps into a double terminated 100 ohm load is 3.5 volts but if the AD8129/AD8130 supply voltage is 24 volts, that is 350 milliwatts. Add that to the 240 milliwatts of quiescent power, and 590 milliwatts into 120C/W is right at the power limit. Adding a separate output amplifier or buffer running on a lower supply voltage might be worth doing. If the AD8129/AD8130 supply voltage is reduced because its full common mode input voltage range is not required which is likely, then power will not be an issue.
(1) I had a hell of a time finding these parts in Analog Device's selection guides. Some of them are under difference amplifiers, some are under operational amplifiers, and some are completely missing.
-
This thread got me thinking. Hypothetically speaking what would be the difference between an isolated battery powered metering system that sends data via radio with single ended inputs and another with differential inputs?
Cause in my opinion the extra complexity of a isolated link + battery power makes the complexities of a balanced differential input useless??
Rather than precision analog over RF link, why not use just an isolated transformer ? -
Rather than precision analog over RF link, why not use just an isolated transformer ?
The conundrum is the single ended vs differential system in a totally isolated system. The rf box is just an example that gives perfect isolation, doesn't really matter if its transformer, analog opto, digital opto.... -
This thread got me thinking. Hypothetically speaking what would be the difference between an isolated battery powered metering system that sends data via radio with single ended inputs and another with differential inputs?
Cause in my opinion the extra complexity of a isolated link + battery power makes the complexities of a balanced differential input useless??
From a practical use standpoint, there are a couple of big differences.
The isolated single ended probe is equivalent to an oscilloscope which has isolated inputs like a Tektronix TPS2000B or an old Tektronix A6902 isolation amplifier. The input impedances for a single ended probe are asymmetrical with the ground lead having 10s to 100s of picofarads of input capacitance to ground. This makes the series inductance in the ground lead connection limit high frequency performance just like with a passive oscilloscope probe and the ground side must be driven from a low impedance source. The low input capacitance of a differential probe is much more tolerant of probe lead length and different source impedances.
The single ended isolated probe potentially has a better high frequency common mode rejection but only where a low impedance ground and coaxial connection are available.
-
Rather than precision analog over RF link, why not use just an isolated transformer ?
Because that only works for some definitions of "isolated". Transformers can be galvanically isolated and still have all sorts of coupling. Check out the capacitive coupling between primary and secondary, for starters.
For even more education, research the term "isolation capacitance" and set aside some time.
I too have been bitten by the "I can design a DC-coupled isolated differential probe" bug and I have some ideas. I don't know if they're novel ideas, or if they'll work, or what bandwidth I'll get, and someday when I have some free time
I'm going to experiment with them. But I'm tempering my expectations because I've had long and painful experience with trying to achieve isolation and it can be a serious PITA.
True isolation is a deep, dark rabbit hole. -
True isolation is a deep, dark rabbit hole.
Too late for me, already deep in the hole with this two rabbits, blame David Hess.
-
The isolated single ended probe is equivalent to an oscilloscope which has isolated inputs like a Tektronix TPS2000B or an old Tektronix A6902 isolation amplifier. The input impedances for a single ended probe are asymmetrical with the ground lead having 10s to 100s of picofarads of input capacitance to ground.
agreed, so basically the objective would be to minimize that capacitance, by not having for example a huge metallic enclosure connected to ground. Eventually there will be a small mismatch in input capacitances. Also the "probes" can be as small as a spring loaded terminal pin. So for frequencies of some 10s of Mhz this could be simpler solution and even behaving nicely with high and fast common mode voltages as you pointed out. -
From a practical use standpoint, there are a couple of big differences.
What about adding a common mode choke balun to the single ended input? At low frequencies, it appears as a direct connection and the stray capacitance is much less significant, while at high frequencies, the balun does the differential to single ended conversion.
The isolated single ended probe is equivalent to an oscilloscope which has isolated inputs like a Tektronix TPS2000B or an old Tektronix A6902 isolation amplifier. The input impedances for a single ended probe are asymmetrical with the ground lead having 10s to 100s of picofarads of input capacitance to ground. This makes the series inductance in the ground lead connection limit high frequency performance just like with a passive oscilloscope probe and the ground side must be driven from a low impedance source. The low input capacitance of a differential probe is much more tolerant of probe lead length and different source impedances.
The single ended isolated probe potentially has a better high frequency common mode rejection but only where a low impedance ground and coaxial connection are available. -
agreed, so basically the objective would be to minimize that capacitance, by not having for example a huge metallic enclosure connected to ground. Eventually there will be a small mismatch in input capacitances.
The difference in capacitance is an order of magnitude or more and the common mode capacitance varies with the phase of the moon, which cats are sleeping on the workbench, etc. The math is fun.What about adding a common mode choke balun to the single ended input? At low frequencies, it appears as a direct connection and the stray capacitance is much less significant, while at high frequencies, the balun does the differential to single ended conversion.
Sometimes this is done with standard passive probes by winding the cable through a ferrite core but I have never found it to be very effective. I don't think you can get a high enough self resonate frequency compared to the low frequency cutoff for this to work well.
An isolated single ended probe works great in low impedance applications where you have a low impedance neutral or ground. DSOs with isolated inputs are a real product niche. But I am dubious about using them to measure things like ground bounce unless you use two probes in which case two non-isolated probes could be used or a non-isolated differential probe like we have discussed here.