It is not only that the open loop is something which phase should be checked (as fscarpa58 pointed out), but also the term phase margin is usually discussed as something related to the open loop behavior (for determining of the minimum gain with regard to the stability).
To check the open loop phase, see this page:
http://www.ecircuitcenter.com/Circuits/opfeedback1/opfeedback1.htm
To estimate the stability of the closed loop amp: load the output with something you expect in the real world and (1) feed the input with square wave - it shouldn’t overshoot and ring and (2) check the AC small signal amplitude response - you shouldn’t have a peaks.
Btw, from the look of the Danny’s phase plot, i.e. from the fact the phase goes up after 130kHz, my guess is the phase in the observed range is in fact made by the capacitor across the feedback resistor.
Pedja
To check the open loop phase, see this page:
http://www.ecircuitcenter.com/Circuits/opfeedback1/opfeedback1.htm
To estimate the stability of the closed loop amp: load the output with something you expect in the real world and (1) feed the input with square wave - it shouldn’t overshoot and ring and (2) check the AC small signal amplitude response - you shouldn’t have a peaks.
Btw, from the look of the Danny’s phase plot, i.e. from the fact the phase goes up after 130kHz, my guess is the phase in the observed range is in fact made by the capacitor across the feedback resistor.
Pedja
Thanks Pedja. I should read my textbook more closely. My graphs are wrong for analysis, they are closed loop and measured from Vout not the feedback point. I'll go resim these tonight when I get home and observe it.
I have no caps in my feedback network. I'm not sure why the phase seems to go up a little after 130Khz. I haven't added any compensation caps either.
Thanks for the help everyone.
--
Danny
I have no caps in my feedback network. I'm not sure why the phase seems to go up a little after 130Khz. I haven't added any compensation caps either.
Thanks for the help everyone.
--
Danny
Hi,
To be exact it is the phase and amplitude response of the loop gain that is important. If open loop gain is described by the function A and feedback loop response is described by the function B; (B should be written as greek beta) then loop gain is described by A x B.
The stability criterion is that phase response of the loop gain should change less then 180 degrees when loop gain is 0dB.
What this means in practice is that even if the open loop response change phase 180 degrees at some point where open loop gain is higher then 0dB the amplifier can still be stable if the loop gain fulfil the above requirement. Typical examples are some OPamps that are stable only for gain above a certain value, (I remember LF 357 that is only stable for gain > 5) that is because the open loop phase response changes more then 180 degrees for high frequencies but the amplifier is stable for a certain gain as the loop gain fulfil the above criteria.
Summary:
It is important to look at loop gain AB, if phase response of loop gain is changing less then 180 degrees for AB = 0dB the amplifier with B feedback is stable.
Even if the amplifier in open loop change phase more then 180 degrees before the gain reaches 0dB the amplifier with feedback can be stable for a certain value of B
If the amplifier in open loop never changes phase more then 180 degrees the amplifier is unity gain stable i.e. stable for any value of gain with feedback.
BTW traditionally it is assumed that an amplifier with 30 deg phase margin is stable but such an amplifier will have overshoot, to minimise overshoot it is adviceable to try to achieve at least 60 degrees which will give ~ 9% overshoot for a system with 2 poles, phase margin > ~75 degrees give ~0% overshoot.
Regards Hans
To be exact it is the phase and amplitude response of the loop gain that is important. If open loop gain is described by the function A and feedback loop response is described by the function B; (B should be written as greek beta) then loop gain is described by A x B.
The stability criterion is that phase response of the loop gain should change less then 180 degrees when loop gain is 0dB.
What this means in practice is that even if the open loop response change phase 180 degrees at some point where open loop gain is higher then 0dB the amplifier can still be stable if the loop gain fulfil the above requirement. Typical examples are some OPamps that are stable only for gain above a certain value, (I remember LF 357 that is only stable for gain > 5) that is because the open loop phase response changes more then 180 degrees for high frequencies but the amplifier is stable for a certain gain as the loop gain fulfil the above criteria.
Summary:
It is important to look at loop gain AB, if phase response of loop gain is changing less then 180 degrees for AB = 0dB the amplifier with B feedback is stable.
Even if the amplifier in open loop change phase more then 180 degrees before the gain reaches 0dB the amplifier with feedback can be stable for a certain value of B
If the amplifier in open loop never changes phase more then 180 degrees the amplifier is unity gain stable i.e. stable for any value of gain with feedback.
BTW traditionally it is assumed that an amplifier with 30 deg phase margin is stable but such an amplifier will have overshoot, to minimise overshoot it is adviceable to try to achieve at least 60 degrees which will give ~ 9% overshoot for a system with 2 poles, phase margin > ~75 degrees give ~0% overshoot.
Regards Hans
Well, this is the open loop response with the gain measured from Vout (since Vfb is of course very attenuated) and the phase at Vfb. There isn't much difference between my closed loop and open loop charts.
Ok... so this should be something like 95 deg phase margin right and stable right?
--
Danny
Ok... so this should be something like 95 deg phase margin right and stable right?
--
Danny
Maybe I misread your post, but you have to measure also the
phase at the output. That is, you break up the connection from
output to the feedback network, connect the input signal to
the feedback network (where the output would otherwise be
connected), sweep the input and then plot the amplitude and
phase of the output signal wrt. the input signal.
One problem is that when you break up the feedback loop
you also lose the DC feedback so you may have to to add a
small DC offset to the signal (or somewhere else) to get rid
of extreme DC offsets at the output.
phase at the output. That is, you break up the connection from
output to the feedback network, connect the input signal to
the feedback network (where the output would otherwise be
connected), sweep the input and then plot the amplitude and
phase of the output signal wrt. the input signal.
One problem is that when you break up the feedback loop
you also lose the DC feedback so you may have to to add a
small DC offset to the signal (or somewhere else) to get rid
of extreme DC offsets at the output.
Christer-
I think you're saying that I should just sweep a signal over my feedback network. Since my feedback network is purely resistive, it would not do for determining my oscillations (it would have no phase change). A majority of the phase change is going to come from the amplifier either by stray capacitance or compensation caps. I want to know if this will cause oscilations when feedback is applied. The proper method to observe this is shown in some of the previous posts. Monitor the phase of the feedback take-off point under open loop when the gain goes to 0dB on the output.
W/rt breaking up the feedback loop, in order to compensate for the feedback network, I had to add some equivelent resistance to my diff-pair. Therefore, I also added this same resistance to my feedback input for these tests so that the diff pair would be biased correctly.
--
danny
I think you're saying that I should just sweep a signal over my feedback network. Since my feedback network is purely resistive, it would not do for determining my oscillations (it would have no phase change). A majority of the phase change is going to come from the amplifier either by stray capacitance or compensation caps. I want to know if this will cause oscilations when feedback is applied. The proper method to observe this is shown in some of the previous posts. Monitor the phase of the feedback take-off point under open loop when the gain goes to 0dB on the output.
W/rt breaking up the feedback loop, in order to compensate for the feedback network, I had to add some equivelent resistance to my diff-pair. Therefore, I also added this same resistance to my feedback input for these tests so that the diff pair would be biased correctly.
--
danny
I think you misunderstood me. You should input the signal to
the feedack network and measure at the amplifier output.
This way you measure the so called loop gain and its phase
behaviour. The loop gain is the feedback gain (which is usually
less than one) multiplied by the amplifier gain. See the Intersil app. note that jackinnij linked to earlier in this thread and which
gives a rather good explanation on why this is what you should
measure.
You could in principle skip the feedback network and just
measure through the amp, but then you have to scale the
gain plot according to the feedback.
The Intesil app. note 9415.3 that jackinnij recommended is
good reading on this, although it lacks a good figure on how
to measure, which you can find in fig. 4 of their app. note 9420
(which is on current feedback, but that doesn't matter for
this purpose).
the feedack network and measure at the amplifier output.
This way you measure the so called loop gain and its phase
behaviour. The loop gain is the feedback gain (which is usually
less than one) multiplied by the amplifier gain. See the Intersil app. note that jackinnij linked to earlier in this thread and which
gives a rather good explanation on why this is what you should
measure.
You could in principle skip the feedback network and just
measure through the amp, but then you have to scale the
gain plot according to the feedback.
The Intesil app. note 9415.3 that jackinnij recommended is
good reading on this, although it lacks a good figure on how
to measure, which you can find in fig. 4 of their app. note 9420
(which is on current feedback, but that doesn't matter for
this purpose).
You are right Christer, in many cases opening of the loop will significantly alter the DC points. But this can be solved by putting the signal source at one and the resistor to ground at the other input; you should find the necessary impedances that will provide acceptable DC points. Or better, keep the zero impedance of the source, and find what input should have a higher impedance for zero offset and put and tune the resistor at that input (for most NPN input amps this means the source should go to the inverting and resistor to the non-inverting input – pic below shows this).
Results I had this way relate well to those I got using the method suggested by Intersil. Usage of it in SPICE is described here:
http://www.spectrum-soft.com/news/spring97/loopgain.shtm
I should add that I am actually a bit confused with the fact that I got the strange results above a few MHz (phase shift) with lower values feedback resistors using this method.
Pedja
Results I had this way relate well to those I got using the method suggested by Intersil. Usage of it in SPICE is described here:
http://www.spectrum-soft.com/news/spring97/loopgain.shtm
I should add that I am actually a bit confused with the fact that I got the strange results above a few MHz (phase shift) with lower values feedback resistors using this method.
Pedja
Attachments
Pedja said:
Another good resource for this approach (Middlebrook technique) is Jim Thompson's site here: http://www.analog-innovations.com/LoopGain.zip. He's got the full derivation of the technique and some utilities for using it. Unfortunately the utilities he provides are PSpice-specific. For users of LTSpice, there's two files in the examples\educational folder that are useful. The audioamp.asc file demonstrates the simplified Middlebrook technique, which is only approximate but easy to use. The loopgain.asc file shows the full Middlebrook technique with both voltage and current source stimulus. One confusing aspect of the examples using the full technique is that they are done by drawing the circuit twice, which just isn't practical for non-trivial circuits. But by using the subcircuit approach, you can draw your circuit just once and create a symbol for it that has only two pins, corresponding to the two nodes where the loop is opened. Then two instances of the subcircuit can be imported into a project for computing the loop gain. It was a bit confusing at first to figure it out, but after looking at the loopgain.asc file and the LTSpice documentation on "Hierarchy" (under "Schematic Capture") it wasn't too bad.
What I mentioned about it actually is related to the higher load and not necessarily to the feedback R values and both strange phase and gain could appear. And sometimes it could come down even below the MHz range. And I still got such strange results with the loads that shouldn’t be a problem for the certain amps.Pedja said:
The “problem” exists in the SwCAD as well as in the MicroCap. If you run V(x)/V
AC analysis of the loopgain.asc file in the SwCAD (that Andy mentioned above) you will see exactly this. Vary the load (or feedback values) and both the phase and gain will be altered. Is this a real word event (hardly, I’d say), a pitfall of SPICE or problem with the measurement method? Measuring the amp in the real open loop with the resistor at the other input for the offset trimming, I got the results almost undependable of the load. And yet, just as it should, gain usually starts to fail at some point (going with higher loads).
Danny, to answer your question, assuming you did not have an excessive DC offset while you were measuring the gain, that amp should work. It is sill not bad to run the transient analysis of the closed loop amp with square signal at the input and certain capacitance at the output. Btw, your first post with closed loop behavior could signify proper working amplifier, at least considering shown range, because you don’t have really steep phase shifts. But of course, only the real amp could tell the whole truth.
Pedja
Pedja said:
The “problem” exists in the SwCAD as well as in the MicroCap. If you run V(x)/V
Calculating Gv = V(x)/V
in the loopgain.asc isn't the same as calculating the loop gain. It's an approximation to the more complicated Middlebrook expression that's referred to in the comments of that file, which involves both Gi and Gv. If you plot the full Middlebrook expression per the comments in the example, it should give the expected results with variations of load. The approximate expression V(x)/V becomes exact only when the current in the voltage source V2 of the example is zero. Since it's in a low-impedance part of the circuit, it's a bad approximation in this case. Maybe a better place would be in series with the inverting input right at the op-amp.So it's a problem with the measurement method I guess.
Christer said:
Hi,
You can breakup the (closed) loop without loosing DC feedback by inserting somewhere a large inductor (something of 1 kH) and feeding the AC generator through a large capacitor (something > 1F).
You can actually simulate the closed loop gain also, provided the closed loop gain is sufficient high enough (>20 dB). The phase plot at around the 0 dB gain crossing is then nearly the same as the open loop plot, accurate enough for stability analysis.
Cheers
Jan,
I think Pjotr were referring to simulations, not real-world
measurements, just like the original poster and most of us.
I get a feeling some talk about real measurements withouth
realizing others are talking about simulations, since the methods
will/may have to differ.
Anyway, I get no headache from finding a 1kH inductor in LTSpice,
I can even find one when I have a headache.
Outside the simulator, that's harder.
Pjotr,
Thanks, your L and C idea seems the best and simplest way
to do it. It is one of those things that is obvious in retrospect,
when somebody has suggested it. I often use the technique
of changing components to absurd values in Spice to "get rid"
of them or to achieve other effects that can be done more simply
than when restricted to real-world components. It is usually
a good way to change a circuit. It is much quicker to change
the value of one or two components than to insert/delete
components in the schematic. In your example one simply has
to change the L and C values to near-zero to get the closed loop
configuration, well and change at least the ordinary signal
source too, of course. Anyway, seems like much less work
to do it this way.
I didn't get the second paragraph in your post. Probably it is
just me missing something, but could it be a typo and one of
the "closed loop" should actually be "open loop"??
I think Pjotr were referring to simulations, not real-world
measurements, just like the original poster and most of us.
I get a feeling some talk about real measurements withouth
realizing others are talking about simulations, since the methods
will/may have to differ.
Anyway, I get no headache from finding a 1kH inductor in LTSpice,
I can even find one when I have a headache.
Outside the simulator, that's harder.
Pjotr,
Thanks, your L and C idea seems the best and simplest way
to do it. It is one of those things that is obvious in retrospect,
when somebody has suggested it. I often use the technique
of changing components to absurd values in Spice to "get rid"
of them or to achieve other effects that can be done more simply
than when restricted to real-world components. It is usually
a good way to change a circuit. It is much quicker to change
the value of one or two components than to insert/delete
components in the schematic. In your example one simply has
to change the L and C values to near-zero to get the closed loop
configuration, well and change at least the ordinary signal
source too, of course. Anyway, seems like much less work
to do it this way.
I didn't get the second paragraph in your post. Probably it is
just me missing something, but could it be a typo and one of
the "closed loop" should actually be "open loop"??
Pedja said:<snip>
The “problem” exists in the SwCAD as well as in the MicroCap. If you run V(x)/V
That's why they sell network analyzers.
It's also the reason that the largest group of Zoloft subscribers are analog engineers.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.