TL072D ultrasonic oscillation

[shadewind]

Hello, first post here

I've been designing two different circuits around the TL072D and in both of them, I've been experiencing HF oscillation (about 500 kHz) on the output with about 6-7 Vpp. This is, of course, unacceptable and even seems to lead to audible distortion in one of them.

These are pictures of the output stage section of both designs:
http://cl.ly/image/0C2l1u3Z3l1R
http://cl.ly/image/072Q1m051G0I

As can be seen, both are unity gain configuration though one is inverting and the other is not. They both share the 10 µF output capacitor, though.

Layout and decoupling does not seem to be the problem (0.1 µF cap in 0603 package right next to the IC which is an SOIC package all of which sits on a continuous ground plane with both the ground on the IC and cap stitched directly to it through vias) but I've heard that the TL072 is quite prone to oscillation at unity gain.

I'm quite new to analog design but I've seen that people often add low value caps in the feedback path to correct the phase margin (or is it simply for attenuation?). Is this the best solution?

[R.G.]

As can be seen, both are unity gain configuration though one is inverting and the other is not. They both share the 10 µF output capacitor, though.

Not the same, actual, single 10uF cap, as in "outputs tied together"? Each has its own separate 10uF cap, right?

[shadewind]

As can be seen, both are unity gain configuration though one is inverting and the other is not. They both share the 10 µF output capacitor, though.

Not the same, actual, single 10uF cap, as in "outputs tied together"? Each has its own separate 10uF cap, right?

Oh, yes of course. I meant "share" as in "both designs have identical output caps". They are two separate designs

[R.G.]

Layout and decoupling does not seem to be the problem (0.1 µF cap in 0603 package right next to the IC which is an SOIC package all of which sits on a continuous ground plane with both the ground on the IC and cap stitched directly to it through vias)

This appears to be a section of a commercial design. Is there logic in the same box or on the circuit board? Is there noise on the reference voltage? Does this happen at more than one location - as in, are you testing while sitting near a radio transmitting tower?

Also, ground planes are not necessarily the best approach for analog/audio frequency designs. RF needs them, but this is by no means a sure thing at lower frequencies.

Does that high-impedance node at the + input wander around all over the PCB (ground plane or not) near noise sources or near the outputs of the opamp it references?

but I've heard that the TL072 is quite prone to oscillation at unity gain.

I use TL07x parts at unity gain all the time, in volume. They are specified for unity gain. However, we are now well past the era when one could trust an electronic part to be what the label says it is. Are you certain that these are real TL07x parts? Sorry - I have to check.  And any high bandwidth, high gain part will oscillate if you give it half a chance. There are periodic moans about the TL07x series I see, but if they were really that bad, they would not be used in such huge quantities. *I* wouldn't use them in even my tiny quantities if they were that bad.

All opamps are most prone to oscillation at unity gain. This is the gain with the most feedback available to power oscillation. But if you have bona fide TL07x parts and they really will not do unity gain, return them and get another brand. Be sure you can demonstrate this independent of your layout. If all of them oscillate in a socket jumpered for unity gain from a battery, you have a case.

If the same part does not oscillate when lashed up with wires and sockets, but does in your layout, then I'd say you have to examine the layout again. Doing all the conventional-wisdom processes in a layout does not ensure success, it only avoids the problems which have been found by other people before.

I'm quite new to analog design but I've seen that people often add low value caps in the feedback path to correct the phase margin (or is it simply for attenuation?). Is this the best solution?

Adding a cap across the feedback resistor is not something that necessarily corrects phase margin. What it does is to increase feedback with increasing frequency above the rolloff of the cap and any parallel resistances. This may cut gain below unity before the phase hits oscillation, but it's not a slam-dunk, especially as the nodes you really want to use a phase compensation cap on are buried inside the IC package. That's why opamps used to bring out compensation pins in the bad old days.

If your opamp is not unity gain stable, using a cap from output to input can sometimes cause oscillation. As someone who's new to analog design, you have a long - but very interesting and fun - path ahead of you. If you've been extruded through some kind of linear controls/feedback training, go read the books by Walt Jung and Jerald Graeme on opamp applications.

[acromarty]

Sometimes opamps don't like driving capacitve loads directly. This can be the cause of high frequency oscillation.
You may find that adding a low value resistor (15 or 22 ohms) in series with the opamp output helps.

From the TI TLC072 data sheet:


Depends on what is to the right of your output coupling capacitor. If it's low impedance then this might be relevant. If it's high impedance then ignore me ;-)

[shadewind]

Depends on what is to the right of your output coupling capacitor. If it's low impedance then this might be relevant. If it's high impedance then ignore me ;-)

I'm assuming that almost all modern amps are high impedance loads.

This appears to be a section of a commercial design.

A commercial design? Why?

Is there logic in the same box or on the circuit board? Is there noise on the reference voltage? Does this happen at more than one location - as in, are you testing while sitting near a radio transmitting tower?

There is no logic in the same box or on the board though I am testing this in open air. There are no nearby radio transmitting towers nearby though I haven't tested this anywhere else except on my bench.

Are you certain that these are real TL07x parts?

I ordered them from Farnell, this exact part: http://se.farnell.com/stmicroelectronics/tl072idt/ic-op-amp-ln-j-fet-dual-8-soic/dp/1750153

Adding a cap across the feedback resistor is not something that necessarily corrects phase margin. What it does is to increase feedback with increasing frequency above the rolloff of the cap and any parallel resistances. This may cut gain below unity before the phase hits oscillation, but it's not a slam-dunk, especially as the nodes you really want to use a phase compensation cap on are buried inside the IC package. That's why opamps used to bring out compensation pins in the bad old days.

I knew the feedback cap would reduce gain but from what I've read, the main reason is for phase compensation. But then again, I have not read much.

I've examined the problem some more and come up with some interesting observations. With nothing connected to the output, the opamp is stable and does not oscillate. As soon as I connect a cable with nothing connected to the other end, the oscillations start. If I then connect the other end to the amp, the result depends upon the length of the cable. The long cable still produces oscillations but the shorter cable does not when connected to the amp. Just loading the output jack with a 100 k resistor does not produce this problem. Moving the cable around does not change the oscillation at all in any way.

When using the shorter cable, I can make the oscillations die away in a couple of seconds by touching either across the contacts on the output jack or the contacts across the other plug of the cable. With the longer cable, I can do the same but only if I first wet my fingers.

[R.G.]

I'm assuming that almost all modern amps are high impedance loads.

Didn't know the output went to an amp.  Modern guitar amps are high impedance inputs, with the capacitance and inductance of the cable in front of them. This doesn't matter at audio frequencies, but may be a big deal for RF.

A commercial design? Why?

Appearance of the schematic.

I ordered them from Farnell, this exact part: http://se.farnell.com/stmicroelectronics/tl072idt/ic-op-amp-ln-j-fet-dual-8-soic/dp/1750153

OK. Had to check. Counterfeit ICs are endemic now.

I've examined the problem some more and come up with some interesting observations. With nothing connected to the output, the opamp is stable and does not oscillate. As soon as I connect a cable with nothing connected to the other end, the oscillations start. If I then connect the other end to the amp, the result depends upon the length of the cable. The long cable still produces oscillations but the shorter cable does not when connected to the amp. Just loading the output jack with a 100 k resistor does not produce this problem. Moving the cable around does not change the oscillation at all in any way.

Ah. You have found the issue of oscillating followers. Capacitive loading of feedback-style followers frequently causes oscillation. Happens in transistor and FET followers too for some kinds of loads.  At high frequencies, the output of a follower can look inductive. The cable capacitance tunes and forces the follower into oscillation. This is the reason all modern solid state power amps include the Zobel network across the output and a damped series inductor.

In your case, the solution is probably to put a load resistor on the board itself. The "best" version might be a 0.001uF + 10K to ground, as a mini-Zobel, but a simple 10K to ground after the 10uF cap probably works as well, since the opamp will happily drive a 10K load in parallel with a cable/amplifier.

When using the shorter cable, I can make the oscillations die away in a couple of seconds by touching either across the contacts on the output jack or the contacts across the other plug of the cable. With the longer cable, I can do the same but only if I first wet my fingers.

This makes you the resistive load.

Try the 10K resistor across the output. If this still gives you problems, put a series resistor in the output, 100 to 1K in value. This series-damps the cable-capacitance/pseudo inductance of the output.

[R O Tiree]

As it says in acromarty's posted picture, "...for capacitative loads greater than 10pF, it is recommended that a resistor be placed in series (Rnull)..." or ringing and oscillations will result.

A long guitar cable could easily exceed 10pF capacitance... try inserting a small resistor in between the output pin of each opamp and your off-board connections (P4 and P7). Anything between 22 and 330 ohms will do, I reckon.

[amptramp]

You may need a resistance in series with the output capacitor.  Typically anything from 100 to 1000 ohms is used to isolate the cable capacitance from the output stage.  If you are going into a high impedance amplifier input, this will have very little effect on gain but a great effect on stability.  Think of the output of the op amp as having a somewhat variable resistance.  The cable capacitance to ground creates a low-pass filter which reduces the amount of feedback at high frequencies.

I also always use a capacitor across the feedback resistor to eliminate the effect of the capacitance to ground at the inverting input.  This way, the feedback signal sees a resistive divider (as intended) but also a capacitive divider that prevents inverting terminal capacitance to ground from raising the gain at higher frequencies.  If the input resistor and its capacitance to ground and the feedback resistor and its parallel capacitance have the same product (R * C = time constant), then there will be no change in gain with frequency.