PA92 OPAMP phase compensation

I'm having trouble to understand the phase compensation on the PA92 opamp. First I'll explain quickly the circuit I'm building.

1.  The circuit

I'm using a 200V battery and would like to create a virtual Gnd for my circuit. The PA92 OPAMP is the ideal component for that, because it can operate on high-voltage supplies (up to 500V). Below you can see the circuit I've built:

simulate this circuit – Schematic created using CircuitLab

R1 and R2 create a reference voltage aimed to be 0V. Capacitors C1 and C2 provide some stability for the reference.

D1 and D2 are 3V6 zener diodes that keep the inputs close to each other. A differential input voltage of more than 20V can damage the OPAMP.

You can find the datasheet of the OPAMP here: https://www.apexanalog.com/resources/products/pa92u.pdf

2.  Phase compensation feature

This OPAMP has a "phase compensation" feature. You have to connect pin 4 to pin 5 through an RC circuit. The choice of the resistor and capacitor values define the phase compensation of the OPAMP. The datasheet explains briefly:

The PA92 is externally compensated and performance can be tailored to the application. Use the graphs of small signal response and power response as a guide. The compensation capacitor CC must be rated at 500V working voltage. An NPO capacitor is recommended. The compensation network CCRC must be mounted closely to the amplifier pins 4 and 5 to avoid spurious oscillation.

These are the "small signal response" and "power response" graphs:

The datasheet also provides this table:

3.  The problem explained

I have to admit that I don't fully grasp what this phase compensation actually does. To make the component behave as a general everyday OPAMP, I thought I had to choose the last line from the table (10pF, 0Ω). I couldn't be more mistaken. The output oscillates at about 60V peak-to-peak, as you can see on my oscilloscope screenshot:

> YELLOW: V+ input
> BLUE:   V- input
> PURPLE: Vout

> Horizontal: 400ns per division, 10 divisions in total
> Vertical:   10V per division, 8 divisions in total


Note: For this test, I used two batteries in series of 100V each, so I have a real Gnd right in the middle. These oscilloscope screenshots are measurements respective to that real Gnd.

I then changed the CC and RC values to 100pF and 100Ω respectively. This corresponds to the second line in the table. I now get this output:

> YELLOW: V+ input
> BLUE:   V- input
> PURPLE: Vout

> Horizontal: 1us per division, 10 divisions in total
> Vertical:   1V per division, 8 divisions in total


There is still an oscillation, but it's only 5.5V peak-to-peak (note the different scale!).

Finally, I would like to do a test with CC and RC equal to 150pF and 100Ω. I'll order 150pF capacitors for that. Hopefully, there will be no more oscillation.

4.  My questions

I have basically two questions. A theoretical one, and a practical one:

1. What does this "phase compensation" feature on the PA92 actually do?

2. What is this "Gain" from the small table in the datasheet? Open-loop or closed loop? At what frequency?

1. What practical advice can you give to get rid of these oscillations? I'm really puzzled, I'd never expect this on such a simple voltage follower...

5.  Notes

• The circuit I've drawn above is really all I've put on the board. For now, the only load is the 100kΩ resistor.

• @laptop2d says that the "phase compensation" feature "allows the OPAMP to respond faster at high frequencies". In short, can I look at this feature as a way to tune the "rise time" and "fall time" of Vout? So I could tune it from like 5V/µs to 50V/µs?

• Q: Why do you need this? Are you sure you need a virtual ground in circuit with floating battery source? Aug 30, 2018 at 15:46
• Yes, I need that. Aug 30, 2018 at 16:55
• The PA92 is unlike the jellybean opamps commonly used, this amplifier is not unity gain stable and requires compensation when employed as a unity gain buffer Aug 31, 2018 at 1:34
• Hi @sstobbe. Applying the R=100 and C=150p to those special pins, would it make the opamp "unity gain stable"? Or do I need other fixes? Aug 31, 2018 at 8:54
• Hi @K.Mulier Yes, applying the RC values you describe will yield a unity gain stable amplifier. You may also run into other stability issues with a capacitive load on your virtual ground amplifier. Aug 31, 2018 at 14:14

What does this "phase compensation" feature on the PA92 actually do?

In short, it allows the op amp to respond faster at high frequencies, in other words, high gain at higher frequencies. Since your application operates mainly at DC, I would set the compensation 150pf, because you don't need the extra gain\bandwidth at high frequencies.

What practical advice can you give to get rid of these oscillations? I'm really puzzled, I'd never expect this on such a simple voltage follower...

This isn't a simple voltage follower because you have 20k in the feedback loop and the diodes. The diodes add a small amount of capacitance, there may also be small amounts of inductance from wiring the circuit up. The feedback loop has a resonance point from one of these things, and it's at 375kHz.

Since I'm too lazy to simulate this, the 20k resistors are not necessary, because the opamp inputs are high impedance, and the only other connection is the 150k resistors, which will give you a worst case shunting current of 1mA through the diodes if the 20k resistors aren't there.

If that doesn't work then try changing the op amp from a voltage follower to an inverting op amp configuration and across the feedback resistor, add a capacitor to limit the bandwidth (thus changing it to a low pass filter) to lower than 300kHz.

EDIT (from @K.Mulier)
Eventually I got it working with RC = 100Ω and CC = 147pF. For the other PA92 OPAMPs (which are used in more complex circuitry), I needed to solder RC = 100Ω and CC = 447pF to the compensation pins. For more details, see the other post.

• Hi @laptop2d. Thank you for your answer. I already tried removing the 20k resistor, but it only got worse. The oscillations are still present, and the DC-component of Vout goes quite close to the negative rail. Very weird. Tomorrow I will try the "phase compensation" feature with a 150pF capacitor. So if I understand your answer well, this "phase compensation" just makes the OPAMP slower/faster at high frequencies: so a slower/faster rise time? Usually expressed in volts per microsecond... Aug 30, 2018 at 19:49
• By the way, the PA92 is very expensive, if you were to do this again, you could use voltage regulators for a positive and negative rail and an op amp with a much lower voltage range that is cheaper. Aug 30, 2018 at 20:35
• I needed some PA92 OPAMPS for other parts of the circuit. While using them anyway, I added just one more to create the virtual Gnd (instead of doing it with other components). This saved me some time in the design. And it's just a prototype, so that's okay. Anyhow, is it correct to look at this compensation capacitor as a way to lower the "rise time" (which is about 50V per us)? Is it that simple? Or is there more complicated stuff involved? Aug 30, 2018 at 20:58
• The comp pin functions kind of like a low pass filter. You need to band-limit your op amp below 375kHz. Aug 30, 2018 at 21:40

I mailed the question directly to Apex Microtechnology and got an answer from Mr. Eric J. Boere (Apex Field Application Engineer, responsible for European support). With his permission, I share his answer with you:

I have checked out the link provided in your e-mail and all answers in there are kind of right, especially the one by @sstobbe; "The PA92 is unlike the jellybean opamps commonly used, this amplifier is not unity gain stable and requires compensation when employed as a unity gain buffer"

If you look at the small signal response graph, you will see that instability is spelled by the open loop gain sloping down at -60dB per decade (octave) by the time it’s 1 (0dB)

A unity gain -stable opamp, would have showed a small signal frequency response sloping down at -20dB/decade all the way down to and below 0dB, after which there could be other pole/zero clusters to bend the graph to -40db/decade or even beyond that.

In order to stabilize the opamp, the external phase compensation can be used. The phase compensation component(s) RC and CC slow the opamp down. Just picture yourself a capacitor slowing down the response of transistor Q3 in the intermediate stage of the equivalent schematic of PA92:

To get rid of the oscillations, use the phase compensation components recommended by the datasheet:

So, for your unity gain PA92, Cc should be 150pF, and Rc be 100Ω. Please adhere to the recommendation to use a Cc that is rated for the full supply voltage, in your case +100 – (-100) = 200V (if you check out the equivalent schematic above you’ll understand why; this capacitor can ‘see’ almost the full supply voltage as the differential voltage between gate and drain can change that much while the output is driven from (almost!) rail to rail). It would be even better to use a higher rated part, for instance a 500V capacitor. By the way: there would be nobody holding you back from using 220pF + 100Ω to ensure the part doesn’t oscillate.

I highly recommend you to read our Application Note #01 as it can save you a lot of time using our power opamps. Just to call out a few things to pay attention to:

• Power supply bypassing (not doing so can cause the part to oscillate, too!)
• Output protection by means of (ultra-)fast reverse recovery diodes (if you have reactive loads)
• Input protection by means of (small signal) diodes, rather than Zener diodes (because the latter exhibit higher capacitance)
• Heat sinking (Iq=14mA, VSS=200V, Pq=2.8W; the part is dissipating 2.8W doing nothing!)
• Etc.

Using this tool can also save you a lot of time!

Mr. Eric J. Boere sent me another mail to answer my question about the table in the datasheet:

Thank you Eric for helping me out!

2.  Practical fix

2.1  TEST 1: RC = 100Ω and CC = 147pF

I have soldered RC = 100Ω and CC = 147pF onto the compensation pins of the PA92. This corresponds to the first line of the table from the datasheet:

The oscillations are gone! At least on the PA92 OPAMP used as mere voltage follower. The other two PA92 OPAMPs used in the more complicated gyrator circuit (simulating an inductance, known as the "Antoniou circuit") are still oscillating:

2.2  TEST2: RC = 100Ω and CC = 247pF

I soldered an extra capacitor of 100pF, raising CC up to 247pF. I still got oscillations in the Antoniou circuit (about 10V peak-to-peak), although the DC-components of the signals are behaving as expected:

2.3  TEST3: RC = 100Ω and CC = 347pF

I soldered again an extra capacitor, raising CC to 347pF. The oscillations in the Antoniou circuit are weakened to 5V peak-to-peak (compared to 10V peak-to-peak before).

2.4  TEST4: RC = 100Ω and CC = 447pF

After adding yet again 100pF (CC is now 447pF), all oscillations in the Antoniou circuit are gone:

Note: The three oscilloscope screenshots I pasted here from the Antoniou circuit, cannot be compared indiscriminately to each other. They all have different scales and are measured under different input conditions. Unfortunately, I didn't make consistent scope screenshots accross all 4 tests ( C=147pF, C=247pF, C=347pF, C=447pF). These scope screenshots merely illustrate the fact that oscillations die out with greater capacitance applied on the compensation pin of the PA92 OPAMP.

2.5 Test conclusions

The conclusions I make from these test results are:

• The PA92 OPAMP used as simple voltage follower works fine with RC = 100Ω and CC = 147pF.

• The PA92 OPAMPs used in the Antoniou circuit (gyrator to simulate an inductance) need RC = 100Ω and CC = 447pF to eliminate oscillations.