Mikeks, Hello. I recently started looking at this site and found this old request for one of my papers. Hope you have found it. You will find it and others at XanalogTech.com. In particular my later papers superced the one you reference. Feel free to contact me for more information through that site. Hope to hear from you.
Regards,
Agustin Ochoa
Regards,
Agustin Ochoa
Mikeks may have left, but your post is perfectly timed for me because I have just started to look seriously at exactly the same problem - how to extend the simplistic feedback model to include load and feed-forward.
I have just read the articles at your site, thanks very much for that.
I will need a little time to think them over but in the meantime I have a few questions.
The main work on the topic that I have studied is by Edward Cherry, do you have any comments on how your approach is related to his?
Is there an implementation of your loop gain / Return Ratio technique in LTspice?
The usual method in LTspice is that of Tian et al, so called Tian probe.
If you are familiar with Tian then have you any comments on your method in comparison?
I have implemented an LTspice version of Middlebrook's GFT method that has some minor improvements over the Tian probe.
Similar question, how does your method relate to Middlebrook's?
Best wishes
David Zan
Last edited:
Hey Dan,
My results are identical to Tian's though I have not fully proven that from his perspective. My developement is independent. Have not looked at Cherry's work. Please send me a reference. As for Middlebrook, sorry, I have not unraveled it though have tried. That is on my list but I think it is too awkward. I of course favor my development. Would love to correspond more on this w you. My problem is that it seems to be intuitive as opposed to developed from first principles. May be wrong and want to seriously look at it. Initially, I thought that it ignored the reverse transmission loop that is inherent in my work and in Tian's results. I am presently working on an article on how to analyze all the variouse types of feedback, series/series, series/shunt. etc. so will delay in looking more into Middlebrook but would love to have a discussion w you on GFT. In my 2013 MWSCAS paper (presentation) I show complete corrections to the 'open loop' method by showing how to terminate replica loading. I also show Bode's return ratio to converge to my result when the reverse loop is included. Thanks for the note and look forward to further discussion. Regards, Auggie
My results are identical to Tian's though I have not fully proven that from his perspective. My developement is independent. Have not looked at Cherry's work. Please send me a reference. As for Middlebrook, sorry, I have not unraveled it though have tried. That is on my list but I think it is too awkward. I of course favor my development. Would love to correspond more on this w you. My problem is that it seems to be intuitive as opposed to developed from first principles. May be wrong and want to seriously look at it. Initially, I thought that it ignored the reverse transmission loop that is inherent in my work and in Tian's results. I am presently working on an article on how to analyze all the variouse types of feedback, series/series, series/shunt. etc. so will delay in looking more into Middlebrook but would love to have a discussion w you on GFT. In my 2013 MWSCAS paper (presentation) I show complete corrections to the 'open loop' method by showing how to terminate replica loading. I also show Bode's return ratio to converge to my result when the reverse loop is included. Thanks for the note and look forward to further discussion. Regards, Auggie
Dave (sorry mis wrote your name in previous reply)
Did not fully answer you question on implementation in SPICE-- I develop test benches to find all relevant partial transfer relations in SPICE (mostly in Cadence), then use their calculator feature to create factors in loop gain, transfer function, numerator, etc. to fully explore contributors to the net response. This gives me insight into design that is ala Middlebrook, low Entropy form.
Regards, Auggie
Did not fully answer you question on implementation in SPICE-- I develop test benches to find all relevant partial transfer relations in SPICE (mostly in Cadence), then use their calculator feature to create factors in loop gain, transfer function, numerator, etc. to fully explore contributors to the net response. This gives me insight into design that is ala Middlebrook, low Entropy form.
Regards, Auggie
The most recent is IEEE Xplore Document - Gain at an Arbitrary Cut in a Linear Bilateral Network, and Its Relation to Loop Gain in Feedback Amplifiers.
You can find quite a bit more with an IEEE search, Author = Edward Cherry, much of it related to the same problems.
Need to be careful of exactly which bit of Middlebrook's work.
His early work (1975) did not include reverse transmission.
Tian does, but not quite completely, perhaps.
Middlebrook's later work (GFT) claims to include reverse transmission completely.
OK, I am definitely keen on that.
Best wishes
David
Last edited:
LTspice is far and away the dominant simulator in this forum and there is a nice implementation of Tian, with all the calculations worked out and results plotted, a similar implementation of your technique would allow direct comparison, be of considerable interest.
One point that I noticed was that your first simulation paper (the OP's request) did not show the separate contributions of the individual components.
That would have been quite educational... but now I read your website and you say there were problems with the method in that paper.
What were the problems?
Best wishes
David
My approach now can show as much detail as one wants. More work but the method allows for fractionalization of the design. For loop gain, I show that one needs to track source and sink of each current in an impedance setup and treat these differently--see mwscas 2012 and 2013. Doing that, you generate the Tian relation that LG=-(i12+i12)/(i11+i22). In this form both directions of transmission are included as well as loading and source impedance effects. In my earlier 'forward*backward' work these currents got co-mingled and the result was approximate. My Z-method is fully developed in my book 'Feedback in Analog Circuits', Springer Pub. So the problem was basically that the form of the result is not sufficient to declare loop gain unless the loop gain factors are kept tractable.
Regards,
Auggie
Regards,
Auggie
Dave, A nice result from application of my Z-method is that I can fully formulate the classic terms 'Aol', 'LG', and 'Beta'. The confusion in these comes from the naming of the terms themselves as in the original sources these terms arise from idealized systems. The amplifier open loop gain 'Aol' for example, arises from the use of an ideal opamp w zero output impedance and infinite input z, plus very high gain. Take these away and the numerator term in the resulting transfer function becomes a function of output impedance, load, and even input impedance (loading effects), and reverse transmission becomes evident. Once you become acquainted with the dpi/sfg and Z-method you can account for the non-ideal aspects of the real design. The factors are in pole-zero form showing effects of circuit elements, Middlebrook's Low Entropy form.
Auggie
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.