Well, 5-6MHz is a believable resonance for 100nF caps and you've got a bunch of them in a row. Also, negative slew rate is dependent on current mirror degeneration isn't it? Once your current mirror is saturated, LTP CMRR fails, so any bounce in your LTP CCS will get into the signal path. If your rails bounce at 5MHz, your CCS will bounce at 5MHz and after CMRR fails this may show up in the output.
That's my highly speculative theory for today, so I had better go to bed now. 🙂
EDIT: That wouldn't really explain why output base stoppers fix it, huh?
That's my highly speculative theory for today, so I had better go to bed now. 🙂
EDIT: That wouldn't really explain why output base stoppers fix it, huh?
You may have coupling thru the base collector capacitance and up onto the rails. So high value base stoppers would help to kill this.
Some posts above I wrote that 2 different PCB layouts (1st one was using many small decoupling caps and 2nd one is using many small electrolytics) and also the first prototype hand wired shows the same "wiggles"...
You know I have found out that the too high current mirror degeneration killed my negative slew rate but have had no influence on the "wiggles".
This explanation may help. 8 parallel devices mounted on the heat sink add some extra capacitance and inductivity for interaction.
But has anybody who has built a 5 - 8 parallel devices output stage some oscilloscope screen shots from a nearly full power 20kHz square wave for comparison? Something like > 100Vpk2pk@8R
Given the explanation above, you can see why you need the damping components mentioned earlier to stop the oscillations.
3.6 ohms is fine. Base stoppers that are too large will tend to degrade thermal stability, since the dc drop across them is idle current divided by beta of the output transistor, which is a positive function of temperature. Of course, beta is also a function of collector current and Vce.
Large base stopper resistors also reduce the effective speed of the output transistors and can introduce excess phase lag. Large base stopper resistors will also tend to make the output impedance of the final emitter follower more inductive.
Cheers,
Bob
Yes - 47R. Also tried Cbc 22p - 100p. Not really helped. But I have never inserted e.g. a 10R in the driver collectors. Thats worth a try!
These are very good and interesting points. The bandwidth issue also touches on the conflict between specs and sound. For example, if we were to bandlimit to, say 20kHz, we would be down 3dB at 20k and we would have to put it on the spec sheet. Even if that sounded better, how would the reader of the spec sheet react?
There is a parallel to this. It is soft clipping. Put an adaptive soft clipper in front of the amplifier (like the Klever Kippler) and all your concerns about clipping behavior are gone. The amp never ever clips. No nasties or recovery issues. But a soft clip will necessarily introduce distortion, albeit soft (third, typically). Those low-ppm THD numbers are way gone. Now you are looking at numbers like those of tube amps, like in the 0.1 to 1% range that you must put on the spec sheet. Many cringe when they see a 0.1% number for THD at 1kHz for a solid-state amplifier.
So we have the paradox. Oh well....
BTW, if we put the soft clip in front of the amp, we should still shoot for very good distortion numbers in the amp itself, especially in regard to nasties like primary and secondary crossover distortion.
Cheers,
Bob
Hi Toni,
Your quote from my book does not conflict with my statement about the size of the base stopper resistors. It is all a matter of degree. It seems we always need base stopper resistors when more than one output pair is used (sometimes they are not needed if only a single pair is used). But my point was that if base stoppers of value greater than about 5 ohms are needed for stability, then there must be something wrong somewhere, HF-wise, in the output stage. There are so many possibilities it is hard to be specific.
One thing I also recommend is serious decoupling of the rails between the output transistor collectors and the driver collectors, and then another stage of R-C decoupling back to the pre-drivers and everything else. I am talking about very low-impedance decoupling, where the voltage drop across the decoupling resistor is no more than 1V, and the size of the shunt decoupling capacitor is significant and consists of an electrolytic paralleled with a film. In addition to providing high attenuation at HF, the R-C decoupling is of low enough impedance to act as a Zobel that damps the rails.
Cheers,
Bob
This is quite correct, and I have been using this technique for quite some time, with a resistor in series. You can see an example of this in Figure 11 of the application note for the LSK489 on the Linear Integrated Systems web site (Linear Systems || Home). Figure 11 is a power amplifier design that uses the LSK489 in the IPS. The app note was published in February 2013.
I have found that using this is especially important when using a helpered current mirror load on the IPS in combination with a 2 EF VAS. There are interactions at HF in such toplogies that can reduce stability of the VAS and make it more easily distorbed at HF by the output stage load, especially when Triples are used. Even in simulation, a key test is to evaluate the net output impedance at the VAS output node by injecting a signal current there. This is one of the stability tests that I mention in my book.
Finally, it is VERY important to recognize that the addition of the shunt R-C network to the output of the VAS is NOT part of the global compensation, and does not take the place of the Miller compensation. It is NOT lag compensation of the amplifier. Having said that, it is important to bear in mind that this shunt R-C load on the VAS can in some cases reduce the slew rate that the VAS can achieve if the capacitor is too large or the VAS standing current is too low. This is a case where a push-pull VAS is superior, since it can deliver twice the signal current for a given standing current.
Cheers,
Bob
Dear Bob,
thank you for response. I meant on reading your sentence
"interactions among the paralleled output transistors"
that maybe the higher number of output devices and higher emitter resistor values have the requirement for higher base stoppers as there is more interaction ...
But a R-C decoupling between driver and ops transistors is definitely a good idea as I never have tried this. Currently only IPS/VAS is strong R-C decoupled from OPS.
BR, Toni
Excellent advice , Bob. I wish some other DIY triple designs (designers) would
heed this very valuable tip.
This advice has given my Indonesian friends VERY stable EF3's . We use
4.7R stoppers/.22Re plus 4.7R/100u-.1u decoupling for the pre+driver.
Also, B-C shunt of 33p at the predriver.
Rock solid with many IPS's , just as well behaved as an EF2. You gave this
advice long ago (trouble with triples). Now ... we all have excellent OPS's.
REMOVING the R/C in both simulation and in reality turns the EF3 into a
"touchy", unstable mess 🙁 . Thanks (for past advice). It works VERY well.
PS - with the decoupling 3/5 pair OPS's are "happy" with 2.2 - 4.7R stoppers ,
Sanken epitaxial and ON NJWxxxx's are the same.
OS
Last edited:
+1
Using this decoupling technique has helped get my HEC MOSFET (4 pairs) output stage stable and free of oscillations.
I find that quibbling between 10 ohms and 5 ohms as base stopper resistors is more political than practical. Of course, less is best, but not too small or problems will happen that can be almost random in nature and hard to protect against any other way. However, I agree that 10 ohms is the 'outside' value and 5 might be better.
30 year old knowledge....
Bob has spotlighted this technique.
BUT , My 32 year old toshiba driver/sanken planar OP EF3 (Z5900) has the R/C.
As does my 21 year old H/K 680.
Both amps use 4.7-6.8R / 47-100u at pre/driver and 4.7R stopper/.22 -.33Re.
Decades of use (abuse) determine merit (for this technique). 🙂
Both also shunt B-C of the predriver with 22-33pF and load the
VAS with 100-220K to ground.
OS
Bob has spotlighted this technique.
BUT , My 32 year old toshiba driver/sanken planar OP EF3 (Z5900) has the R/C.
As does my 21 year old H/K 680.
Both amps use 4.7-6.8R / 47-100u at pre/driver and 4.7R stopper/.22 -.33Re.
Decades of use (abuse) determine merit (for this technique). 🙂
Both also shunt B-C of the predriver with 22-33pF and load the
VAS with 100-220K to ground.
OS
No. you would have to contact the mfr. There may be several ps film makers but only one produces the best quality film. But, they wont give up their sources of materials. it was researched, sampled, measured, tested and inspected every which way. I am just trying to save you the time to find out these things. The Chinese film source supplies other mfr but the only one I know is RELiable Capacitor and they also do other things beside use the best film to make a top cap for military, communications, oil exploration equip... high rel places that Must hold spec for decades under harsh conditions. You want the best?
Thx-RNMarsh
Last edited:
I can easily hear the bass affected by changing the base-stopper value. Lower value always sounds more like 'real'.... it takes a surprising amount of current drive to do the sound justice. Best to depend more on the inductor rather than Rs on the base from my experience. What's your experience on this?
[maybe this belongs on your site not on CFA?]
THx-RNMarsh
Last edited: