Twice in the last couple of years I have needed a probe to diagnose problems with a non-isolated power supply that connected directly with the 125VAC mains voltage here in the USA. I have been able to work around it by using less than satisfying methods. My cheap oscilloscope (a Rigol DS1102E) is good enough for most of my purposes, but the math function to subtract Channel 1 from Channel 2 really sucks. And even though it is a cheap scope, I'd rather not have to buy a new (more expensive) scope because I connected the scope ground incorrectly.

Needing a tool on a yearly basis hardly qualifies as necessary, but I had some down time waiting for components and PCBs for another project and began to consider this as an interesting project.

I'd been following Paul Versteeg's progress to make a differential probe on his blog. I recommend that you read it -- he covers most of the basics and has some good references of other approaches to make this kind of probe. There are also several persons within Hackaday that have completed projects successfully, or are still ongoing.

My expectation for success is rather low. I don't have the necessary expertise to be confident that there will be a finite amount of revisions prior to meeting my initial goals. But that never stopped me before.

The Specifications:

  • Input impedance: 20MegΩ// 3pF.
  • Differential Gain = 1/10 V/V. Any lower than this and the DS1102E could not resolve a 1V signal with any clarity. (Many of the cheaper differential probes are switchable between 50X and 500X attenuation. This would be good if the signals in which you are interested are very large, but that's not the case so far with my needs. The better probes ($400 and up) provide 10X/100X attenuation.
  • Common Mode Range > ±340V (240VAC produces a sine wave 679V peak-peak)
  • Differential Voltage Range > ±20V
  • Bandwidth > 100MHz (Dependent upon signal amplitude) This will be a stretch for my current skill set.
  • Noise: TBD (Wider bandwidth = more noise)

The Schematic:

The input attenuator is 200X. It is very similar to what Paul Versteeg implemented. The series R and C network is necessary to avoid exceeding the voltage rating on any of the components when subjected to high voltages. The resistors attenuate the low frequencies and the capacitors attenuate the higher frequencies. Both networks require a trim to reduce common mode rejection at low frequency, and gain and common mode rejection at high frequencies. 1 percent tolerance components help to keep the trim ranges small. Spark gaps, SG1 and SG2, and clamp diodes, D1 and D2, limit damage from ESD or inadvertently exceeding input voltage specifications.

The first gain stage has a differential gain of 20 and a common mode gain equal to unity. This stage recovers some of the attenuation that was necessary to provide a relatively large common mode range. These amplifiers need a gain-bandwidth (GBW) greater than 2GHz in order to meet the 100MHz system bandwidth requirement. 

The second stage is a unity-gain differential to single-ended converter. Its GBW should be around 200MHz. R9 matches the 50Ω impedance of the expected coax cable to the scope, even though the scope input is 1MegΩ//15pF. My cheap scope doesn't have a 50Ω input.

The circuit that occupies the bottom develops the power supplies. The input power requirement is 5V/50mA. I expect that it will be provided by a simple 5VDC wall adapter. Ferrite beads and capacitors should clean it up nicely. U3 splits the ±supply rails to generate a GND rail at 2.25V above VEE. I set the GND rail lower because the LTC6269-10 and LTC6268 opamps have a common mode input limit of VCC-0.5V, so the lower GND should allow for symmetric clipping of the input.

It's a good start, but the PCB layout is probably more important in obtaining the performance goals and that part of the design is just beginning. I'll fill in the blanks as I go.

Initial Simulation Results:

Here's the LTspice...

Read more »