The schematic of the Active summer I built to perform the Bode plot tests - its results are pretty close the current probe (with a single turn) around the transition region, the active summer works better down towards DC as to be expected.
An externally hosted image should be here but it was not working when we last tested it.
Last edited:
The V1 prototype PCB with the Active Summer PCB:-
https://dl.dropboxusercontent.com/u/86116171/IMG_6687.JPG
https://dl.dropboxusercontent.com/u/86116171/IMG_6687.JPG
Still battling with my Bode plot measurements - I will update the progress.
I turned my attention to Output impedance measurements as with the design they are a reflection of the OL Gain so helping to bring some sanity to my world.
These measurements are just the output stage without global feedback (the output stage has its own nested FB loop), so the final output impedance will be lower.
The Output Impedance plots are referenced to 1 Ohm so:-
0dB = 1R
-20dB = 100mR
-40dB = 10mR
-60dB = 1mR
-80db = 0.1mR
-100dB = 0.01mR (0.00001 ohm)
The first plot is the simulated output impedance:-
https://dl.dropboxusercontent.com/u/86116171/MDAC2 Single MOS OPS OPZ 100R 330pF Simulated.jpg
Measured output impedance with HP3577 setup:-
https://dl.dropboxusercontent.com/u/86116171/MDAC2 Single MOS OPS OPZ 100R 330pF 20Hz Measured.jpg
Atleast within a few dB (fractions of an ohm) the measured output impedance agrees with the simulated results.
0.00001 Ohm output impedance at 20Hz is not bad - shame that without Kelvin sensing cables the connectors and cable resistance will dominate the results in practice.
I turned my attention to Output impedance measurements as with the design they are a reflection of the OL Gain so helping to bring some sanity to my world.
These measurements are just the output stage without global feedback (the output stage has its own nested FB loop), so the final output impedance will be lower.
The Output Impedance plots are referenced to 1 Ohm so:-
0dB = 1R
-20dB = 100mR
-40dB = 10mR
-60dB = 1mR
-80db = 0.1mR
-100dB = 0.01mR (0.00001 ohm)
The first plot is the simulated output impedance:-
https://dl.dropboxusercontent.com/u/86116171/MDAC2 Single MOS OPS OPZ 100R 330pF Simulated.jpg
Measured output impedance with HP3577 setup:-
https://dl.dropboxusercontent.com/u/86116171/MDAC2 Single MOS OPS OPZ 100R 330pF 20Hz Measured.jpg
Atleast within a few dB (fractions of an ohm) the measured output impedance agrees with the simulated results.
0.00001 Ohm output impedance at 20Hz is not bad - shame that without Kelvin sensing cables the connectors and cable resistance will dominate the results in practice.
Last edited:
RNMarsh,
Not being able to find much practical information on the techniques of making Amplifier Gain and phase margin plots with a VNA, I've "thought myself" - there are still some issues I don't fully understand, once I've a better understanding of whats going on I'll post the information here.
My design posses particular measurement problems - which has not eased the "learning process"!
Maxim Integrated AN3245 follows up on the earlier HP App note on current transformer based injection techniques.
https://dl.dropboxusercontent.com/u...nt Transformer Injection based Bode plots.pdf
The important information to be aware of is that the Measurement node to the VNA must be lower circuit impedance then the VNA Reference measurement node otherwise the current transformer will not correctly couple the stimulus signal into the feedback node and this will produce erroneous measurements.
This is especially relevant above the unity gain transition point where the Amplifiers feedback will no longer be to reduce the output stage impedance, so in effect the amplifiers output impedance can be higher then the feedback node thus preventing the transformer to inject a correct stimulates voltage into the feedback node.
The current probe injection technique cannot be relied upon above the Unity Gain transition point, thus making Gain margin measurements very suspect. In fact due to this injection point impedance "reversal" the measured gain is seen going positive once again above the true unity gain transition point - making a mockery of any attempt to measure the Gain margin.
Below the Unity Gain transition point, the Phase Margin measurements seem fairly close to the simulated results, in fact the measured PM appears to measure slightly less then the real PM, possibly due to the "incomplete" coupling of the stimulus current due to impedance "mismatch losses" - what I mean to say that the amplifiers output node is not a perfect zero ohms, so some stimulus signal is "attenuated" at the feedback node.
A good Oman is that the unity gain transition frequency is almost spot on as predicted by the simulations - so its only above the UG transition point that the Current probe injection method falls short - sadly making GM measurements results practically useless unless the output stage node can be guaranteed to remain lower impedance then the feedback node.
Not being able to find much practical information on the techniques of making Amplifier Gain and phase margin plots with a VNA, I've "thought myself" - there are still some issues I don't fully understand, once I've a better understanding of whats going on I'll post the information here.
My design posses particular measurement problems - which has not eased the "learning process"!
Maxim Integrated AN3245 follows up on the earlier HP App note on current transformer based injection techniques.
https://dl.dropboxusercontent.com/u...nt Transformer Injection based Bode plots.pdf
The important information to be aware of is that the Measurement node to the VNA must be lower circuit impedance then the VNA Reference measurement node otherwise the current transformer will not correctly couple the stimulus signal into the feedback node and this will produce erroneous measurements.
This is especially relevant above the unity gain transition point where the Amplifiers feedback will no longer be to reduce the output stage impedance, so in effect the amplifiers output impedance can be higher then the feedback node thus preventing the transformer to inject a correct stimulates voltage into the feedback node.
The current probe injection technique cannot be relied upon above the Unity Gain transition point, thus making Gain margin measurements very suspect. In fact due to this injection point impedance "reversal" the measured gain is seen going positive once again above the true unity gain transition point - making a mockery of any attempt to measure the Gain margin.
Below the Unity Gain transition point, the Phase Margin measurements seem fairly close to the simulated results, in fact the measured PM appears to measure slightly less then the real PM, possibly due to the "incomplete" coupling of the stimulus current due to impedance "mismatch losses" - what I mean to say that the amplifiers output node is not a perfect zero ohms, so some stimulus signal is "attenuated" at the feedback node.
A good Oman is that the unity gain transition frequency is almost spot on as predicted by the simulations - so its only above the UG transition point that the Current probe injection method falls short - sadly making GM measurements results practically useless unless the output stage node can be guaranteed to remain lower impedance then the feedback node.
Last edited:
Most of the good info on current injection measurements like this will be found under stabilizing switching power supplies. I think there are several good app notes. I have somewhere a set of probes/injector for just this. I forgot who made them.
The current injection becomes the weak part when you don't have a way to reference exactly how much current is injected. Adding a small shunt in series and a differential amp would be the fix but it gets involved.
At 30 MHz the test setup connections become so much of the circuit and its stability you may not get much practical value from the Bode plot. Even on many of the published spec sheets is clear the info above 10 MHz is suspect. As you get into RF land classical solutions need all possible strays incorporated and then you get to the RF solvers. In this case just because you have a tool to see it doesn't mean you should look. I would look for signs of instability of the circuit in transient conditions in its real application, since that is where the problems will surface.
The current injection becomes the weak part when you don't have a way to reference exactly how much current is injected. Adding a small shunt in series and a differential amp would be the fix but it gets involved.
At 30 MHz the test setup connections become so much of the circuit and its stability you may not get much practical value from the Bode plot. Even on many of the published spec sheets is clear the info above 10 MHz is suspect. As you get into RF land classical solutions need all possible strays incorporated and then you get to the RF solvers. In this case just because you have a tool to see it doesn't mean you should look. I would look for signs of instability of the circuit in transient conditions in its real application, since that is where the problems will surface.
The HP3577 has a "Reference input" so you can refer your measurement channel to this input to compensate for the absolute level of stimulus signal coupling 
this references level of current injection loop.
I've found a useful crosscheck (and easier measurement) is to look for peaking in the frequency response - this is a reflection of the Phase margin.
I've found very poor correlation between simulated and measured phase margin - at the end of the day only the measured phase margin matters - I'm surprised that there is so little on the subject with the hordes of "Simulated" designs here on the Forum, I see it as blind faith!
Luckily the most important region of Audio Amplifier PM measurements is around 1MHz to 20MHz... 30MHz would be a dream accept for Gain margin results which bare an even less of a relationship to there simulated results them the PM sims
this references level of current injection loop.I've found a useful crosscheck (and easier measurement) is to look for peaking in the frequency response - this is a reflection of the Phase margin.
I've found very poor correlation between simulated and measured phase margin - at the end of the day only the measured phase margin matters - I'm surprised that there is so little on the subject with the hordes of "Simulated" designs here on the Forum, I see it as blind faith!
Luckily the most important region of Audio Amplifier PM measurements is around 1MHz to 20MHz... 30MHz would be a dream accept for Gain margin results which bare an even less of a relationship to there simulated results them the PM sims
No need for the second coil in the probe, as once you have measured the injected signal via the reference node you have all the information you need (Unless I missing something), I don't see what benefit an indirect measurement of the stimulus signal would bring?
Nothing can be closer to the truth then measuring the induced voltage at the reference.
https://dl.dropboxusercontent.com/u...nt Transformer Injection based Bode plots.pdf
If you study the Maxim App note you can see the "R Port" (Reference port) senses stimulus signal and then serves as a Reference for the A Port (Measurement port) by using the mathematical function (A/R) on the network analyzer.
I'm fairly confident in the measurements (atleast WRT the Phase margin results), although they differ by a disconcertingly large degree to the Sim'ed results.
Nothing can be closer to the truth then measuring the induced voltage at the reference.
https://dl.dropboxusercontent.com/u...nt Transformer Injection based Bode plots.pdf
If you study the Maxim App note you can see the "R Port" (Reference port) senses stimulus signal and then serves as a Reference for the A Port (Measurement port) by using the mathematical function (A/R) on the network analyzer.
I'm fairly confident in the measurements (atleast WRT the Phase margin results), although they differ by a disconcertingly large degree to the Sim'ed results.
anyone serious does know that lots is left out in most Spice Audio Amps circuit modeling
you can find equations for estimating distributed circuit params, L, C even model coupling of your physical layout
use your VNA to verify each wire, trace, package parasitic
then you get a "infinite regress" problem of also verifying, calibrating your fixturing, backing out probe's added parasitic effects
but as mentioned it really isn't "cost effective" use of hobby time when you intend to build, expect to hack/tweak the physical implementation
you can find equations for estimating distributed circuit params, L, C even model coupling of your physical layout
use your VNA to verify each wire, trace, package parasitic
then you get a "infinite regress" problem of also verifying, calibrating your fixturing, backing out probe's added parasitic effects
but as mentioned it really isn't "cost effective" use of hobby time when you intend to build, expect to hack/tweak the physical implementation
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.