I'd say the 10 ns rise time I used in the sim to show stability is very fast. This is always a good Q'n'D (Quick and Dirty) test for stability. Keep in mind that a simulation is always an ideal situation, as demonstrated with the comments about decoupling the supply rails with an SMD->DIP adapter. And about a lowpass filter at the input.
Correct me if I'm wrong, but.....a 10 nSec rise time would occur with a signal with Period = 40 nSec, thus a frequency of 25MHz. What has that to do with audio?
This thread is about stability, and checking the small-signal step response is one of many ways to check the small-signal stability. Unfortunately the step used in post #10 was too large to be regarded a small-signal step for bipolar op-amps.
The relation with audio is that an amplifier that oscillates unintendedly usually also can't handle normal audio signals well.
The relation with audio is that an amplifier that oscillates unintendedly usually also can't handle normal audio signals well.
I routine decouple opamps with a cap between Vcc and Vee. I do not want to decouple to ground as that can work both ways - introduce gnd junk back into the circuit.
I normally use SOIC to DIL adapters with a single 1206 1u/50V decoupling cap on the bottom side. Never had any oscillation issues from it.
Jan
When the output switching occurs causing noise instead of power supply sagging it takes power from capacitor instead.
Analogue isnt usually as bad as digital as digital has hard switching.
The analogue is just switching slower.
Analogue isnt usually as bad as digital as digital has hard switching.
The analogue is just switching slower.
In what sense? I meant coupling it onto the supply rails through the decoupling cap.
It can also interfere with a carefully laid out grounding scheme.
Jan
There certainly is a lot of good information. I appreciate everyone's contributions.
The circuit was tested with the inputs shorted. When it oscillated I did connect a 4.7K load to the output. The only consequence was that the circuit drew a few more mA current from the supplies.
I rarely use a bypass configuration with two caps grounded. I only use that configuration in a circuit that delivers an appreciable amount of power (like a headphone amp or power amp). Then I always run a clean ground and dirty ground. However, if it would fix this issue then I would do it. I'm in the camp of keeping my grounds as clean as possible.
I did some evaluation of the 2134 a couple of years ago and I found that if you put a resistor (47-100K) in parallel with a small capacitor (22-47 pF) in parallel and connected them between the inverting input and output, and then put a small resistor (1-2.2K) in series with the noninverting input, then it performs fine as a unity gain buffer. Obviously this is not always a desirable solution. Furthermore, it seems to conflict with other people's experience.
The circuit was tested with the inputs shorted. When it oscillated I did connect a 4.7K load to the output. The only consequence was that the circuit drew a few more mA current from the supplies.
I rarely use a bypass configuration with two caps grounded. I only use that configuration in a circuit that delivers an appreciable amount of power (like a headphone amp or power amp). Then I always run a clean ground and dirty ground. However, if it would fix this issue then I would do it. I'm in the camp of keeping my grounds as clean as possible.
I did some evaluation of the 2134 a couple of years ago and I found that if you put a resistor (47-100K) in parallel with a small capacitor (22-47 pF) in parallel and connected them between the inverting input and output, and then put a small resistor (1-2.2K) in series with the noninverting input, then it performs fine as a unity gain buffer. Obviously this is not always a desirable solution. Furthermore, it seems to conflict with other people's experience.
With rail-to-rail bypass, the "junk" goes to the other rail. If the op amp has no ground references internally or in the feedback, that is OK. The delta between rails is maintained, and that is enough to prevent oscillation, in most cases.
However, power supply spikes become common-mode spikes, so supply disturbances are more likely to show up on the output with rail-to-rail decoupling. I prefer coupling from rails to ground, but if that's not feasible, or the local ground is sensitive to noise, then rail-to-rail coupling is generally OK.
I would feed a realistic rise time signal into any sims. 10 ns rise times are great to stress circuit for overshoot etc, but you will never encounter these on any music signal from any source. 2-5us is practical. You will often see overshoot in a sim (or with a fast rise/fall time generator in a practical circuit) that you will never get in practice).
Also, make sure you include an output isolating resistor between the opamp and the cable or next circuit you are driving. In you sim, add c. 100pF per cable metre load on the output. On a fast opamp operating at unity gain, this is often enough to push it into oscillation. The series isolating resistor (47 to 100 Ohms) normally cures this. Place this resistor close to the opamp as possible.
Mooly did some work on this a few years ago where he swapped some opamps out and looked at the output - one of them oscillated.
I've had bad oscillation despite following the above rules because of inadequate decoupling as well.
Also, make sure you include an output isolating resistor between the opamp and the cable or next circuit you are driving. In you sim, add c. 100pF per cable metre load on the output. On a fast opamp operating at unity gain, this is often enough to push it into oscillation. The series isolating resistor (47 to 100 Ohms) normally cures this. Place this resistor close to the opamp as possible.
Mooly did some work on this a few years ago where he swapped some opamps out and looked at the output - one of them oscillated.
I've had bad oscillation despite following the above rules because of inadequate decoupling as well.
Last edited:
Good points, yes. I often wonder if the main benefit from rail-to-rail decoupling is just providing a local 'short' to improve stability as you mention, rather than actually smoothing the rails, what with high performance regulators and opamp huge PSRR.
I normally put the decoupling cap at the back side of the PCB, between the opamp pins. Keep leads short, that's important too.
Jan
That's irrelevant, as the simulation is just meant to check whether there are any pole pairs close to the imaginary axis - the step not meant to be a model of a musical transient. In my opinion, 10 ns is rather long, I would rather use 100 ps or so (or any other value << 1/(2 pi GBP)).
If you don't have a ground plane layer, it's probably the best way. With sockets, what else can you do?
You know, the rail to rail chips are popular for low voltage applications. The inputs can typically go to the rails too on these chips.
A side benefit is that these chips are much more immune to latching up. And it seems like the good old 5532 is resistant to latch up. I've had zero problems using it as a unity gain buffer at line level.
Remember, in a typical inverting configuration, both inputs are at ground (inverting at virtual ground) and so the voltage (theoretically) always stays at zero. And also, the beta of a unity gain inverting buffer is 2; the beta of a noninverting buffer is 1.
That's how I always do it.
For example, suppose one of the circuits were at the verge of oscillating at 30 MHz. When you apply a small step with 100 ps rise time, you will see it ring severely at 30 MHz and know there is trouble.
When you apply a small step with 2 us rise time, you just see the 2 us-rise-time ramp and have to zoom in to the start or end of it to notice that there is anything ringing. If you use the wrong integration method and insufficient accuracy settings, the simulator may even step over it without showing any ringing at all.
When using a SMD-to-DIP adapter the PCB is layed out for DIP. So, in 99% of cases there are already 2 decoupling caps (often 2 x 10 µF electrolytic caps) that now are a bit too far away from the SMD IC power supply pins. One could use that GND point for the new decoupling caps and remove the old ones that are too far away. If the device originally does not have 2 decoupling caps per opamp it probably is not even worth it to try out better opamps.
Last edited:
Fast opamps often need 100nF ceramic decoupling from rail-to-rail typically to stabilize them at HF. This is different to bulk decoupling of the rails to ground which is about removing sources of noise in the audio band.
Attachments
I was not clear. I meant adapter PCBs to adapt SOIC to DIP sockets.
In my head, I knew what I meant, but I didn't write it. The SOIC PCB would have a V+ to V- bypass cap. There is no ground available on the adapter PCB.
In my head, I knew what I meant, but I didn't write it. The SOIC PCB would have a V+ to V- bypass cap. There is no ground available on the adapter PCB.
But if you feed these ultra fast rise time signal into any amp, you will provoke severe overshoot unless you have an input filter. So ideally you would need to do it without the filter. How do you know that a real world audio type opamp would even behave properly with a signal like that? What about stray capacitances? Does the sim consider these effects?
Surely a simple loop gain/phase plot would tell you if you were going to have stability problems? If yes, what is the point of looking for pole pairs near the imaginary axis by feeding in a 10 ps rise time signal that does not take into account board parasitics etc?