Well, the truth is that at low currents, that is, in the transition region (sorry for using the T-word again) MOSFETs show a similar exponential behavior. So the X-over between the two output devices isn't that abrupt. Admittedly, WRT to BJTs, the gain 'wobble' is higher, but equally important, it is also wider. That means that the spectrum of harmonics from this wobble is smaller (i.e. less HF). Consequently, it's easier to linearize it by means of feedback.
PS: You really don't like MOSFEts, do you?
That is correct, and I would agree that Self's discussion on mosfets is lacking, however I am sure you also know there's much more in the mosfet vs. bipolar debate than the spectral distribution of distortions.
PS: You really don't like power bipolars, do you?
Hi Douglas,
Forgive me if I say it a bit bluntly, but you for ask additions and/or improvement on the 5th edition: Chapter 14 on MOSFET output stages should be rewritten, completely.
The simulations of the (vertical) MOSFETs doesn't take into account the so called weak-inversion (or sub-threshold conduction). This has far reaching consequences, not just regarding the the graphs, but also with respect to your conclusions about MOSFETs. The point is that they perform much better then you and your simulator might think.
In this chapter, you stated:
"However, the most important difference may be that the bipolar gain variations are gentle wobbles, while all FET plots seem to have abrupt changes that are much harder to linearize with NFB that must decline with rising frequency. The basically exponential Ic/Vbe characteristics of two BJTs approach much more closely the ideal of conjugate (i.e. always adding up to 1) mathematical functions, and this is the root cause of the much lower crossover distortion.
A close-up examination of the way in which the two types of device begin conducting as their input voltages increase shows that FETs move abruptly into the square-law part of their characteristic, while the exponential behavior of bipolars actually gives a much slower and smoother start to conduction."
Well, the truth is that at low currents, that is, in the transition region (sorry for using the T-word again
Actually, I'm surprised that you didn't revise this chapter already in the 5th edition, as Marcel van de Gevel pointed this already out in 1996, see below.
Cheers,
E.
PS: You really don't like MOSFEts, do you?
Hi Edmond,
You are spot-on in your observations here, and they agree with what I pointed out in my book in the MOSFET chapter. Indeed, this is why I worked so hard to come up with EKV models for some of the power MOSFETs that I posted on my web site. The square-law model for MOSFETs is prehistorically crude. In my real-world designs, the MOSFETs also show a smooth, well-behaved, non-abrupt transition as they turn on.
In some ways, the MOSFETs are the opposite of BJTs. In my 1983 paper, I coined the term "transconductance droop" to describe the class-AB crossover shortcoming of MOSFETs. Later, I believe Doug coined the term "gm-doubling" to describe the crossover problem one runs into when over-biasing a BJT output stage. The silver lining for the MOSFET is that it loves to be biased at a higher current to achieve higher gm. I usually run them at 150-200mA per device. You virtually cannot get a MOSFET stage into gm doubling. This means that you can bias them as hot as you are willing to do, at the same time increasing the class A region of the class AB output stage. Of course, if you are looking for an amplifier that runs cool at idle with small heat sinks, you are better off with a BJT output stage.
The beauty of the MOSFET with healthy bias is that the bias can move around quite a bit and you will not get into dangerous waters with crossover distortion. You cannot say that about BJTs. I believe that many people grossly underestimate the sonic deficiencies brought on by BJT output stages due to dynamic bias shifts with real program material. This problem is not generally revealed by sinewave THD testing, which is rather static thermally. ThermalTrak output transistors have mitigated this problem in BJT output stages. Once you have mastered ThermalTrak biasing techniques (ignore the OnSemi app note, please), you will use ThermalTraks and never look back.
This is not to say that MOSFETs are universally better than BJTs. Indeed, the ThermalTrak BJTs and high ft have made BJTs much better than they used to be. Although I lean toward using vertical MOSFETs, much of what I have said here applies to both vertical and lateral MOSFETs.
Cheers,
Bob
Response to DS and question for both authors.
I am curious about a topic that involves both DS's query and Bob's enthusiasm for output triples and MOSFETs.
In the chapter on OutputS there is a CFP + EF triple shown (p 156 pic. 6.17b of 5th Ed.) and the comment is made that it looks to have "promise".
It looks so to me too, that it combines the best points of a CFP that DS advocates with the triple that Bob likes.
So a follow up on this promise would be of interest.
The CFP stability is of course a concern so more material here would help too.
It occurs that a MOSFET as the second transistor could provide some extra bandwidth that would allow more flexibility to compensate the CFP internal loop. Would the BJT buffer the problematic MOSFET capacitance variation?
In view of Bob's comments above and DS's observation that "it is... wise to consider whether BJTs or FETs are the best devices for the job" do either of you have any comments?
Best wishes
David
I am curious about a topic that involves both DS's query and Bob's enthusiasm for output triples and MOSFETs.
In the chapter on OutputS there is a CFP + EF triple shown (p 156 pic. 6.17b of 5th Ed.) and the comment is made that it looks to have "promise".
It looks so to me too, that it combines the best points of a CFP that DS advocates with the triple that Bob likes.
So a follow up on this promise would be of interest.
The CFP stability is of course a concern so more material here would help too.
It occurs that a MOSFET as the second transistor could provide some extra bandwidth that would allow more flexibility to compensate the CFP internal loop. Would the BJT buffer the problematic MOSFET capacitance variation?
In view of Bob's comments above and DS's observation that "it is... wise to consider whether BJTs or FETs are the best devices for the job" do either of you have any comments?
Best wishes
David
Last edited:
I propose this be called Cherry compensation.
He may not have been the first to do this but he was certainly the first to analyse it in detail and show its advantages.
I shall resist chiming in on the pros & cons of TPC vs Cherry etc except to suggest the various pundits look at the open loop operating conditions 'generated' by the various schemes. Some of these go against Self & other Gospels but surprisingly result in less open loop distortion too.
______________________
BTW, where is the thread where Edmond Stuart pontificates on these matters?
Last edited:
I propose this be called Cherry compensation.
He may not have been the first to do this but he was certainly the first to analyse it in detail and show its advantages.
I shall resist chiming in on the pros & cons of TPC vs Cherry etc except to suggest the various pundits look at the open loop operating conditions 'generated' by the various schemes. Some of these go against Self & other Gospels but surprisingly (??) result in less open loop distortion too.
For this reason, I spurn the bastardized TMC as proposed by Self, Edmonds, Uncle Tom Cobley & even Great Guru Baxandall as it doesn't have ALL the advantages of pure Cherry.
______________________
BTW, where is the thread where Edmond Stuart pontificates on these matters?
This shows a complete lack of comprehension and/or a stupid attempt to troll. In either case, best to remove the post within the edit time window then search for "TMC" and read before you post.
Best wishes
David
Last edited:
I propose this be called Cherry compensation.
He may not have been the first to do this but he was certainly the first to analyse it in detail and show its advantages.
I shall resist chiming in on the pros & cons of TPC vs Cherry etc except to suggest the various pundits look at the open loop operating conditions 'generated' by the various schemes. Some of these go against Self & other Gospels but surprisingly result in less open loop distortion too.
As to Mr. Self's assertion that pure Cherry compensation always results in oscillation, I would suggest the truth is that a
"Blameless" amplifier is difficult to get stable with pure Cherry. Other, often simpler topologies can take full advantage of pure Cherry with less aggro.
But to get back to the new book ....
I don't think there is any point in asking Mr. Self to put stuff in that he doesn't believe in. The worth of his pontificating is the care he has put into stuff that he believes.
Though I don't agree with everything he proposes, ALL of it is worth reading for this reason. This is true even for Great Gurus like Baxandall.
For the best expose of CFA, MOSFET O/Ps etc. find someone who believes in such mechanisms and beg them to write a book. It would be perfectly valid to beg this worthy fellow to 'prove' that CFAs etc are far better than the evil VFAs etc that Doug puts his faith in.
______________________
BTW, where is the thread where Edmond Stuart pontificates on these matters?
It's ok to want to sell books primarily about one topology and using primarily one type of semicinductor.
Plenty has been written about other topologies and it is true that some excellent refinements have been shown in DIYAUDIO let alone by the IC industry. It doesnt matter that it is current-mode or voltage in disguise... the topology is interesting and has its set of pro-con. I called it complimentary push-pull before it got called current-mode feedback. And, with simple circuits obtained below .001% THD+N many decades ago. To get below .001 can be done with several topologies. Push pull is one of the family of differential amplifiers composed of differentail, push-pull, paraphase. Explore them all and use the one which best matches your performance priorities.
Thx-RNMarsh
Plenty has been written about other topologies and it is true that some excellent refinements have been shown in DIYAUDIO let alone by the IC industry. It doesnt matter that it is current-mode or voltage in disguise... the topology is interesting and has its set of pro-con. I called it complimentary push-pull before it got called current-mode feedback. And, with simple circuits obtained below .001% THD+N many decades ago. To get below .001 can be done with several topologies. Push pull is one of the family of differential amplifiers composed of differentail, push-pull, paraphase. Explore them all and use the one which best matches your performance priorities.
Thx-RNMarsh
Last edited:
And after ten years, you still hold a grudge against mr. Self for ignoring you in this matter? WOW!
So it's the same but different? What's the point the acronym or name we call something? There is fundamentally different behavior for two amplifiers that are distinguished by what has been an industry accepted nomenclature. Remember there is more than audio out there and we a talking by and large about three terminal general purpose devices as well. Do you honestly think a successful product could be made out of a multi-pin VFA giving the user access to input degeneration and two or three compensation nodes so that it could be customized for constant BW at each closed-loop gain?
A fundamentally useful property an illusion? Better tell all the users that their performance is an illusion. What unity gain bandwidth is unchanged? The open-loop transimpedance? I thought breaking the loop included the feedback network as a load on the input. Of course buffered it hardly matters, but here the input transconductace is modified so with a constant shunt Ccomp the unity voltage gain frequency scales with closed-loop gain. Something has to give. For voltage feedback there is in general a constant GBWP, for constant bandwidth over a range of closed-loop gains this cannot hold.
From a very narrow viewpoint (audio) maybe and even then some here will disagree. Please keep it civil.
Not so far, I'm afraid. I have done a bit of searching but nothing has so far emerged.
Possibly; I have to say I don't recall. Professional matters are rather pressing at the moment, but when I can, I will have another trawl through the archives and see what can be found.
How long is too long? The Blameless amplifiers are certainly quite happy with 6 inches or so, (probably much more) and I would expect any workable design to be not very different. I don't recall all the details of construction after this length of time but I am sure I would remember something eccentric like extra-long leads to the output devices.
I assume you are referring to the two 33pF capacitors in the ETI design? That is not shunt compensation- they would have to be very much larger for that. They look very much as if they were added to suppress parasitic oscillation in the output stage; a connection to ground would be better to avoid the possibility of injecting rail noise. This technique is dealt with on p222 of the Fifth edition of Audio Power Amplifier Design.
Thank you for this David.
TMC appears to most clearly and usefully described (by Great Guru Baxandall) at The Baxandall Papers
I distinguish between the bastardized TMC and pure Cherry compensation. Prof. Edward Cherry analysed his compensation scheme in a number of JAES articles from about the early 80's. All are worth reading even if you don't agree with his final solutions.
I apologise for chiming in before finishing my study of ALL this thread and its links ... I'm only at about page 200 on the thread off Cordell's book. I hope to finish it before the next millenium.
Last edited:
There is some more on this useful technique, and more on bootstrapping in general, which (surprise) is not as straightforward as it looks.
That resistor was pure dim-wittedness on my part. The original idea was to protect the circuitry if the CCS failed short-circuit; actually very unlikely. The disadvantage is that the voltage drop across the resistor means that when you are variac-ing up the supply from zero the amplifier will not work until the rails reach some +/-9V. By that time damage may have been done if there is a fault. Without the resistor, you can get a visibly good sinewave with only say +/-3V and the operation is much safer. This is dealt with on p511 of the 5th edition. I hope I have by now removed all those resistors.
There will be a good deal more on the characteristics of the audio signal.
Hi Bob
I'm sorry, but I cannot get enthusiastic about such a vague acronym. Three letters just aren't enough to convey the information.
I could settle for TOIMC, (output-inclusive) so we can also use TIIMC. (input-inclusive)
Back to suggestions for the next edition ..
I'd like to make a plea that the various schematics be checked for missing details that might lead to self destruction on overload. My particular beef is the emitter follower driving the VAS.
(all references to the 4th edition
eg fig 6.16 : TR12's collector current is unlimited on overload so will kill TR4 on HF overload.
A simple 'cure' is a resistor in TR12 collector.
Some of the circuits shown are OK. eg fig 5.24 where Q9 protects both Q10 & Q11.
I feel this point needs to be stressed ad nauseum as the emitter follower is an essential part of "Blameless".
I'd like to make a plea that the various schematics be checked for missing details that might lead to self destruction on overload. My particular beef is the emitter follower driving the VAS.
(all references to the 4th edition
eg fig 6.16 : TR12's collector current is unlimited on overload so will kill TR4 on HF overload.
A simple 'cure' is a resistor in TR12 collector.
Some of the circuits shown are OK. eg fig 5.24 where Q9 protects both Q10 & Q11.
I feel this point needs to be stressed ad nauseum as the emitter follower is an essential part of "Blameless".
Edmond, I wholly agree with you. That chapter does indeed need a complete makeover. However, the structure of the next edition is gelling as we speak, and I am afraid it's not going to happen this time round.
I am painfully aware that both the FET chapter and the chapter on Class-D have nothing remotely like the depth of the rest of the topics. But if they did, consider the impact on the size of the book- it would be something like twice the length. As it is, I have pushed the limits on size as far as I could, having persuaded my publishers to increase it several times. Sometimes compromise is inevitable.
Both these chapters could easily be separate books. Anyone fancy tackling Class-D?
PS: You really don't like MOSFEts, do you?
No. Variable Vgs, transconductance...not helpful. You know where you are with a BJT.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.