Phase-lead compensation for uncompensated op-amps

After a search yielded no relevant information, I thought I would pose this question.

Let's say I have a basic non-inverting op-amp stage, and use an uncompensated op-amp for the gain element (think uncompensated DOA or perhaps a partially compensated op-amp like a NE5534), and run it near unity gain (6dB).  If I insert a phase-lead element (e.g. cap) into the feedback loop I can trim away gain at high frequencies (e.g. above the audio band) that perhaps I don't need.

Can a phase-lead capacitor take the place of internal op-amp compensation?  It seems to me that if I place the pole well below the poles and zeroes of the internal op-amp elements, I am in fact replicating the dominant pole compensation used in most op-amp designs to make the system appear to contain a single pole.

My LTSpice simulations show a test system containing a simple 2520-type DOA with no internal Miller compensation, wired as a non-inverting gain stage, which is clearly unstable at 6dB overall gain up in the 10MHz range:  about -23 degrees of phase margin.  After adding a 470pF cap into the feedback path, the closed loop gain falls to unity at about 30kHz, and at that point there is about 43 degrees of phase margin which looks to be pretty stable.  I did not however test with capacitive loads.

It's seems to me that with some careful tailoring of the phase-lead cap value, I can get better slew rate into relatively tough loads by omitting the Miller compensation and bringing the closed loop-gain down to just above where I need my band-pass region of operation.  It seems that with Miller compensation, the closed loop gain starts falling off VERY early (e.g. in the 10Hz range) which means the gain tends to unflatten over the audio band.

Is there some other aspect to internal compensation that I'm missing here?

No.  The Phase lead across your feedback resistor INCREASES Loop Gain which is what determines stability.  This is especially true with the large value you used which makes your amp unity gain at HF where the evil stability stuff happens.

To assess stability with GM & PM, you need to look at Loop Gain NOT Closed Loop gain.

Look at LTspice4/examples/Educational/LoopGain2.asc and learn how to use it.

If you do your Closed Loop sim with your 470p up to 100MHz, you'll see your peaking.

Matador said:

After a search yielded no relevant information, I thought I would pose this question.

Let's say I have a basic non-inverting op-amp stage, and use an uncompensated op-amp for the gain element (think uncompensated DOA or perhaps a partially compensated op-amp like a NE5534), and run it near unity gain (6dB).

Click to expand...

I think these are called de-compensated when not unity gain stable.

If I insert a phase-lead element (e.g. cap) into the feedback loop I can trim away gain at high frequencies (e.g. above the audio band) that perhaps I don't need.

Click to expand...

Do you mean in parallel with feedback R?  Reducing closed loop gain at HF for de-compensated (under, un-compensated) opamps generally reduces stability.  To increase stability you want the NF to to be less than unity by the HF where internal delay makes 180' phase shift.

Can a phase-lead capacitor take the place of internal op-amp compensation?  It seems to me that if I place the pole well below the poles and zeroes of the internal op-amp elements, I am in fact replicating the dominant pole compensation used in most op-amp designs to make the system appear to contain a single pole.

Click to expand...

don't follow you there.

One technique to stabilize un-de-compenasted opamps for lower gain is to add an RC from the - input to ground, this makes the noise gain look stable at HF while not adding noise gain at lower audio frequencies

JR.

My LTSpice simulations show a test system containing a simple 2520-type DOA with no internal Miller compensation, wired as a non-inverting gain stage, which is clearly unstable at 6dB overall gain up in the 10MHz range:  about -23 degrees of phase margin.  After adding a 470pF cap into the feedback path, the closed loop gain falls to unity at about 30kHz, and at that point there is about 43 degrees of phase margin which looks to be pretty stable.  I did not however test with capacitive loads.

It's seems to me that with some careful tailoring of the phase-lead cap value, I can get better slew rate into relatively tough loads by omitting the Miller compensation and bringing the closed loop-gain down to just above where I need my band-pass region of operation.  It seems that with Miller compensation, the closed loop gain starts falling off VERY early (e.g. in the 10Hz range) which means the gain tends to unflatten over the audio band.

Is there some other aspect to internal compensation that I'm missing here?

Click to expand...

Sorry I wasn't clear:  I am asking if compensation can be applied in the external feedback network in as opposed to relying on whatever compromises were made by another designer.

I am trying yo ask generally: but the real engineering answer may be (as always) "It depends."

The criteria for stability involve internal and external factors, so clearly the same opamp can be stabilized by raising the bridge or lowering the water. There can be different trade-offs between different stability approaches.

It seems the external compensation approach of increasing the noise gain (separately from the forward signal gain) leaves the internal compensation pole free to deliver higher slew rate. The opamp slew rate is limited by input stage current density or max current out, and stability pole capacitance value. So tweaking noise gain externally to stabilize should result in a faster slewing opamp than increasing the compensation cap value. Albeit with the cost/penalty of higher noise gain.  Putting a pole in this noise gain tweak (RC instead of R from - input to ground) can shift the noise increase to only be above the audio band. Note: for non-inverting topology you must use RC in noise gain compensation to keep closed loop gain correct.

I have also seen external compensation tweaks that involve an R or RC between + and - input pins which doesn't affect closed loop gain while increasing stability (when driven from relatively low impedance). This may also mess with input impedance.

If slew rate is a problem this may be worthwhile, but understand nothing is free. These days there are modern off the shelf opamps that are generally much better than we need for audio so such tweaking is not a huge concern. Back in the early 70's opamps were often slow enough that we needed to explore such techniques. (There's more but IMO only of academic interest at this point for original designs).

JR

I found a good article from TI which covers most of this:

http://www.ti.com/lit/an/sloa020a/sloa020a.pdf

I believe John's method is called "lead-lag compensation" in the paper.  I'll do some experiments with this.

Matador said:

I found a good article from TI which covers most of this:

http://www.ti.com/lit/an/sloa020a/sloa020a.pdf

I believe John's method is called "lead-lag compensation" in the paper.  I'll do some experiments with this.

Click to expand...

Yup they show that one.  My RC between + and - opamp inputs is exactly what you get when you use lead-lag compensation with a simple inverting opamp that has a grounded + input. Placing the RC between the two inputs for a non-inverting topology makes subtle differences. This may be easier to analyze and visualize with just a resistor between them, while we don't want the DC voltage errors and other problems associated with DC connection there.

That TI papers mentions another obscure issue. Package and input pin stray capacitance. Sometimes adding a small cap across an opamp NF network that is not unity gain stable can be helpful, while generally this would drop closed loop gain below the stable minimum gain, a gain divider is formed between the feedback cap and stray capacitance to ground, just like the feedback resistors set gain with a divider. So perversely adding a few pF there can sometime help stability, while adding more capacitance would hurt stability.

There were other esoteric compensation schemes, but like i mentioned not very useful in the context of modern silicon.

JR