I see casual comments like "optimize the open loop gain" before adding feedback. OK, fine. Cordell shows a crude method that adds too much distortion in the first place.
So for a generic amplified, LTP input. how the heck can you do this? Using LTSpice. Can I just feed a 180 signal to the right side?
So for a generic amplified, LTP input. how the heck can you do this? Using LTSpice. Can I just feed a 180 signal to the right side?
You can increase the the noise gain, which lowers the effective nfb and increases observed distortion level.
Add a resistor between the two inputs, roughly 1/10 to 1/100 of the value of the small nfb resistor to ground.
Of course. Try.
I once made a amplifier LTP input, - open loop 120 kHz. Easy to close with 15dB feedback (a design based on Jean Hiraga/Kaneda)
In practice, there are small lead/lag phase effects that should be looked at live (=simulation might not work)
This Hiraga guy used small lag caps, not just in the feedback. I used trimmers and I could set the square wave with a monimal overshoot. I used his selected components . . .
However, later, not repeatable with other components In the same circuit.
Just to say, closing the loop is complicated.
I once made a amplifier LTP input, - open loop 120 kHz. Easy to close with 15dB feedback (a design based on Jean Hiraga/Kaneda)
In practice, there are small lead/lag phase effects that should be looked at live (=simulation might not work)
This Hiraga guy used small lag caps, not just in the feedback. I used trimmers and I could set the square wave with a monimal overshoot. I used his selected components . . .
However, later, not repeatable with other components In the same circuit.
Just to say, closing the loop is complicated.
I am trying to optimize the basic circuit before the output feedback. You know, optimize current in the IPS, VAS. How much emitter degen, effect of different CCS etc. Things masked by global feedback. Not trying to design a no-feedback amp. But I am looking to see if I can improve it enough to reduce the gain and feedback keeping the result reasonable.
Getting somewhere with a 1H in the global feedback line and feeding it from both sides of the voltage source. I may have to add a load to the right side of the LTP to balance the base current the left side drives.
Getting somewhere with a 1H in the global feedback line and feeding it from both sides of the voltage source. I may have to add a load to the right side of the LTP to balance the base current the left side drives.
Dug out the old Cordell book. He suggests a GH in Spice. Re-reading. One can glean a tip every time.
before you close the loop you should have the DC's right.
yes, local feedback will help in reducing the gain. A low collector/drain resistance is a way that is overlooked - most often one adds emitter/source degeneration. A common resistance can be used also as part of the collector/drain resistance: it then keeps the DC same level, but amplification can be severely reduced without problems. . . .
yes, local feedback will help in reducing the gain. A low collector/drain resistance is a way that is overlooked - most often one adds emitter/source degeneration. A common resistance can be used also as part of the collector/drain resistance: it then keeps the DC same level, but amplification can be severely reduced without problems. . . .
Not sure I follow your argument.
Are you simulating a complete power amp with LTP input stage?
NFB can be "eliminated" using a large inductor (1MH or 1GH) in series with the feedback resistor.
That will give you an indication of your OLG except at low frequencies of course when the inductor impedance is not so high.
If you have a high open loop gain as a result, you will need to attenuate the input signal by about the same so that you do not see large distortion due to overloading.
You may well be able to define an open loop gain with controlled phase which allows NFB to be applied, but I would urge caution that you should not rely on this. What I have found to be important is the transient effects in the LTP and other stages (slew induced if you like).
That often requires more attention to avoid transient distortions than attending to an open loop condition. In other words, designing for stability with a closed loop system is often actually more difficult than it appears from getting OLG/phase right.
Are you simulating a complete power amp with LTP input stage?
NFB can be "eliminated" using a large inductor (1MH or 1GH) in series with the feedback resistor.
That will give you an indication of your OLG except at low frequencies of course when the inductor impedance is not so high.
If you have a high open loop gain as a result, you will need to attenuate the input signal by about the same so that you do not see large distortion due to overloading.
You may well be able to define an open loop gain with controlled phase which allows NFB to be applied, but I would urge caution that you should not rely on this. What I have found to be important is the transient effects in the LTP and other stages (slew induced if you like).
That often requires more attention to avoid transient distortions than attending to an open loop condition. In other words, designing for stability with a closed loop system is often actually more difficult than it appears from getting OLG/phase right.
Yes, simulating a complete amp. I found I could only go 1H as more than that would cause instability. ( Miller compensation)
Probably easier to simulate without the output stage. Can't do the input as the VAS has a large effect on it. Gleaned a few tips from Gloner.
Probably easier to simulate without the output stage. Can't do the input as the VAS has a large effect on it. Gleaned a few tips from Gloner.
Don't use an inductor. Insert a voltage source at the inverting input, then plot vf/vm. There are YouTube tutorials on how to do this.
Edit: When doing this, run the simulation with the non-inverting input grounded.
Edit: When doing this, run the simulation with the non-inverting input grounded.
Last edited:
getting the DC right up front: short out R6. Then adjust for the DC. Then the feedback does not have to go awry because of minute DC differences. Little differences will create different settling points for the output. Little differences have great effect - in harmonics and other aspects. Such as stability.
I'm surprised you had a problem with an inductor. It is a standard technique. Could you post your circuit as the diagram hpasternack showed only seems to have a gain of 3x. Im guessing there is either more gain in the output stage or it is perhaps just a headphone amp. in his post.
The point about a large inductor is that if it represents the open loop gain, there should be no instability, by definition.
So you may have some odd feedback path. But you say you have a Miller compensation which should be very stable.
If you do not want to use an inductor (and I've never found it to be a problem, and nor do I find it a problem using it in a FET input, Blameless design) then really you need to be using a Tian probe, not just a single voltage probe.
That will give you the loop gain from which you can obtain the "true" open loop gain, although in practice you need to assess stability with both AC and transient simulations.
Are you including the feedback capacitor in series with the inductor or are you placing the inductor only in series with the feedback resistor? If there is an LC circuit there then that could well be unstable. However, any feedback capacitor from the output is also part of the network, and will generate a single slope roll off in addition to any other roll-off components. Although I see a potential for oscillation, even with a 1G inductor, the AC sim I ran did not show a problem.
The point about a large inductor is that if it represents the open loop gain, there should be no instability, by definition.
So you may have some odd feedback path. But you say you have a Miller compensation which should be very stable.
If you do not want to use an inductor (and I've never found it to be a problem, and nor do I find it a problem using it in a FET input, Blameless design) then really you need to be using a Tian probe, not just a single voltage probe.
That will give you the loop gain from which you can obtain the "true" open loop gain, although in practice you need to assess stability with both AC and transient simulations.
Are you including the feedback capacitor in series with the inductor or are you placing the inductor only in series with the feedback resistor? If there is an LC circuit there then that could well be unstable. However, any feedback capacitor from the output is also part of the network, and will generate a single slope roll off in addition to any other roll-off components. Although I see a potential for oscillation, even with a 1G inductor, the AC sim I ran did not show a problem.
I regularly use a 1G L in series with the feedback resistor on the output side to plot open loop gain. I use a voltage source between the output and the feedback resistor to plot loop gain.
These are indeed standard techniques for ascertaining these (different) quantities.
These are indeed standard techniques for ascertaining these (different) quantities.
Yes, it's a headphone amp. I just used a model I had as an example. Also, yes, this gets you the loop gain, not the open loop gain. It's a smart idea. I just followed the instructions I found on YouTube. I don't really care what the open-loop gain is, only the transfer function around the whole loop.
By the way, this method produces an unexpected rising response at very high frequencies. I believe it's because it accounts for current in or out of the inverting input, which you normally don't think about in basic op-amp models.
By the way, this method produces an unexpected rising response at very high frequencies. I believe it's because it accounts for current in or out of the inverting input, which you normally don't think about in basic op-amp models.
Yes, the gain you mentioned (Vf/Vm) is a loop gain, not the OLG. The OLG can be obtained by multiplying by 1/alpha (alpha being the attenuation of the feedback components).
It's still a cut-down version of a Tian probe. The "real" loop gain needs a voltage and a current injection and some extraction to calculate it. There are LTSPice versions of the Tian probe available. Frank Wiedmann gives a good account. The Tian probe will account for the impedances into the feedback node - that's its main point.
Rising output is often due to feed forward: signals transferring directly through device capacitances.
If you are getting a similar result with a single voltage drive to a Tian probe then it may be because the FET input is almost ideally high impedance.
It's still a cut-down version of a Tian probe. The "real" loop gain needs a voltage and a current injection and some extraction to calculate it. There are LTSPice versions of the Tian probe available. Frank Wiedmann gives a good account. The Tian probe will account for the impedances into the feedback node - that's its main point.
Rising output is often due to feed forward: signals transferring directly through device capacitances.
If you are getting a similar result with a single voltage drive to a Tian probe then it may be because the FET input is almost ideally high impedance.
Thanks. I started with an inductor to bad effect, then found this LTSpice video:
It does not mention the full Tian probe. I assume they didn't want to scare people away. The method described in the video is adequate for my purposes. It seems to work as expected up to around 10 MHz with the circuits I've been modeling.
I determined the HF anomaly is due to input transistor capacitances, as you say. I'm familiar with how to set up the Tian probe in SPICE, but have been busy and haven't gotten around to trying it yet. IMHO, the basic voltage probe is worth trying as a first step.
It does not mention the full Tian probe. I assume they didn't want to scare people away. The method described in the video is adequate for my purposes. It seems to work as expected up to around 10 MHz with the circuits I've been modeling.
I determined the HF anomaly is due to input transistor capacitances, as you say. I'm familiar with how to set up the Tian probe in SPICE, but have been busy and haven't gotten around to trying it yet. IMHO, the basic voltage probe is worth trying as a first step.
This is my "tuned" Hafler DH-120. Converted to Exicon outputs, Miller compensation and a few small tweaks.
My goal was to try and linearize the stages before global feed back as it tends to mask your changes. It sounds quite nice, but I was just looking to see what little changes I might make short of the FET cascode input mod.
IF I am reading the Bode plots correctly, it models pretty stable. It has been running for years in the real world so at least that part is probably OK.
Note, this model used the Cordell library and Exicon supplied library. Repaired is as it is running now. Tuned is what I am playing with. It also has a higher current Hexfred rectifier and 8 x 6600 main filter bans, removed the Dynaquad etc. The input pair was hand matched.
I have not succeeded with the Titan probe and am not at all sure on the .noise. I am getting very strange anomalies in some of the .four displays I don't quite understand. I had not picked this up in about 7 years.
My goal was to try and linearize the stages before global feed back as it tends to mask your changes. It sounds quite nice, but I was just looking to see what little changes I might make short of the FET cascode input mod.
IF I am reading the Bode plots correctly, it models pretty stable. It has been running for years in the real world so at least that part is probably OK.
Note, this model used the Cordell library and Exicon supplied library. Repaired is as it is running now. Tuned is what I am playing with. It also has a higher current Hexfred rectifier and 8 x 6600 main filter bans, removed the Dynaquad etc. The input pair was hand matched.
I have not succeeded with the Titan probe and am not at all sure on the .noise. I am getting very strange anomalies in some of the .four displays I don't quite understand. I had not picked this up in about 7 years.
I've now investigated the single voltage injection referred to in post 15. My quick calculations indicate that what it measures is the loop gain based on the "full" open loop gain. There are two open loop gains in an amplifier (this may surprise some). One is the "intrinsic" open loop gain which is extracted from the Vout=A(Vinp-Vinm) but the other is due to an attenuated signal which arises from the feedback resistor impedances and the amplifier input impedance.
In some cases, where the impedances are high, as I hinted at earlier, these OLG's may be very similar. In other cases, they may be quite different. The Tian probe will determine the loop gain for the actual configuration, and should be used in preference to the single voltage injection.
A prime example is the case of a single transistor input stage, such as used in the original JLH10 amplifier. The "full" open loop gain calculated with a zero feedback grounding resistor can be 1300. (Example: input stage operates at ~270uA, gm is 10mA/V approx. A high gain driver transistor (eg 2N3019) may be 240 but is shunted by the base resistors down to about 200. Output transistors could have a gain of 45 but they are doubled for free (the same base current swing in the driver pushes and pulls the two output devices) but the shunt load of the bootstrap resistors reduces this by about 10% so we get a net current gain of about 80. All together that equates to .01x200x80x8=1280. The closed loop gain is 12.8 (simulated or calculated using these figures) and the loop gain is 100, which is exactly what is calculated with the single voltage injection. But it isn't right.
The actual OLG depends on the input stage gain (10mA/V) modified by the feedback grounding resistor, for which 220 ohms takes it to about 3mA/V. That makes the "effective" OLG only 400. Therefore the real loop gain is about 32. Using a Tian probe, the simulated LG is indeed 32.
The "real" OLG is actually correctly indicated using a 1G inductor in series with the feedback resistor, FWIW.
As regards the circuit the OP showed, I have not seen a problem using a 1G L with a very similar design (not exact, but does use cascode VAS and ECF10x20 output devices, after tracking down the SPICE models for these). There are several areas of concern I would suggest could be improved.
In some cases, where the impedances are high, as I hinted at earlier, these OLG's may be very similar. In other cases, they may be quite different. The Tian probe will determine the loop gain for the actual configuration, and should be used in preference to the single voltage injection.
A prime example is the case of a single transistor input stage, such as used in the original JLH10 amplifier. The "full" open loop gain calculated with a zero feedback grounding resistor can be 1300. (Example: input stage operates at ~270uA, gm is 10mA/V approx. A high gain driver transistor (eg 2N3019) may be 240 but is shunted by the base resistors down to about 200. Output transistors could have a gain of 45 but they are doubled for free (the same base current swing in the driver pushes and pulls the two output devices) but the shunt load of the bootstrap resistors reduces this by about 10% so we get a net current gain of about 80. All together that equates to .01x200x80x8=1280. The closed loop gain is 12.8 (simulated or calculated using these figures) and the loop gain is 100, which is exactly what is calculated with the single voltage injection. But it isn't right.
The actual OLG depends on the input stage gain (10mA/V) modified by the feedback grounding resistor, for which 220 ohms takes it to about 3mA/V. That makes the "effective" OLG only 400. Therefore the real loop gain is about 32. Using a Tian probe, the simulated LG is indeed 32.
The "real" OLG is actually correctly indicated using a 1G inductor in series with the feedback resistor, FWIW.
As regards the circuit the OP showed, I have not seen a problem using a 1G L with a very similar design (not exact, but does use cascode VAS and ECF10x20 output devices, after tracking down the SPICE models for these). There are several areas of concern I would suggest could be improved.