Help! Oscillation and "rail override" problem with my power amp

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi gerenis69
Anyone any idea how the output can swing OUTSIDE the VCC rails by 7 Volts. Simulation shows the clipped output voltage should be INSIDE the VCC rails by about 5 Volts...

C10 is a BOOTSTRAP CAP and you may want to add a diode clamp across Q9 & Q10. This will stop the drive going above the rails. I also add a cap CB of Q11 as a general plan.
Duke.
 
I remembered Dr Leach and DestroyerX recommended in some instances to put the zobel AFTER the output RL.
I did it. I removed the RC and soldered it at the speaker binding posts for convenience. That completelly stoped oscillations and I was able to enjoy the amp since.

In fact if you look at my schematic, I use separate ground connections (from the power supply star-point) to the LTP-VAS (called GND), to the Zobel (called ZGND) and to the output stage decoupling (called PGND). There are also HF RC filters between the OPT and driver transistor supplies (0R68-220n). The capacitors of these HF filters are also connected to the ZGND - maybe it was unwise to connect them there? These HF filters are effective only above 1MHz, their knee frequency is about 1MHz.

OK, let's play around with the Zobel, I'll try perhaps tomorrow.

I have 10R as output and 15R+100nf to ground.
I suspect the Zobel was introducing HF oscillations into ground that excited at peaks the input when it was on board.

I also discovered that the original 2.7 Ohms in the output LR was suboptimal, and replaced it with 10 Ohms.

Maybe an unthoughtful simplification, but I used 6.8R-100nF in the Zobel instead of the usual 10R-100nF just because there's two high-current OPT pairs, not one, and I thought they "should" happily drive a lower impedance than the standard 10R.
 
sound of the amp

I would keep the number of devices and voltage amplification to a minimum.
Active current sources, diodes, ICs, regulators, current mirrors, beta enhancers, Baker clamps and other dirty supplementary parts do not linearize signal transfer. They do just the opposite.

You may be right, but how is this possible? Measurement instruments are much more sensitive than the human ear, and many complicated amplifiers turn out to have very low distortions, at least THD.

How about loading down the VAS with a resistor between C-B to kill the high open loop gain?

As I listen to this amp, it really doesn't sound pleasant, although it may sound "interesting". Dull recordings sound more lively. The cymbals on pop-recordings sound more metallic in spite that my speakers have soft domes. I can here many details, but the sound often becomes "edgy" in the highs. Like acoustic guitar strings sound too metallic and unnatural to me. Sounds like the vibrations or upper harmonics are exaggerated on high pitched instruments like violins. Surely there must be some kind of distortion in the frequency range the ear is most sensitive to. But I don't have a distortion meter, and I'll be able to go to a friend who has one only in two or three week's time. I think we can only do THD measuerements that may not explain everything...
 
@gerenis69
All you need is a Digital Multimeter and your ears. Maybe you could cautiously consider that dynamic events cannot be simulated (measured).

Regarding the circuit in post #17
I would
replace Q2, Q4, Q3-Q6, Q8-Q10 with a resistor
omit Q7, Q12, Q13, Q16, D1, D2, D3, D5
replace R29 with two resistors from emitter to output for better thermal stability
increase the value of R19
reduce the value of C2

I am not urging you to do this, just providing an opinion.
 
Amplification is the distortion. Active devices exhibit considerable distortion. Instability and phase disorder worsen with increase in the number of amplification stages. Resistors are much less distortive than transistors. And so on...This is not much of an explanation, I know.
 
A "fact" is nothing more than an opinion that broadly fits the visible evidence.

An "opinion" in its derogatory sense as being used in this thread is a fact that the evidence is not met or not convincing.

There are no universal truths, although there are widely held facts.
 
You may be right, but how is this possible? Measurement instruments are much more sensitive than the human ear, and many complicated amplifiers turn out to have very low distortions, at least THD.

How about loading down the VAS with a resistor between C-B to kill the high open loop gain?

As I listen to this amp, it really doesn't sound pleasant, although it may sound "interesting". Dull recordings sound more lively. The cymbals on pop-recordings sound more metallic in spite that my speakers have soft domes. I can here many details, but the sound often becomes "edgy" in the highs. Like acoustic guitar strings sound too metallic and unnatural to me. Sounds like the vibrations or upper harmonics are exaggerated on high pitched instruments like violins. Surely there must be some kind of distortion in the frequency range the ear is most sensitive to. But I don't have a distortion meter, and I'll be able to go to a friend who has one only in two or three week's time. I think we can only do THD measurements that may not explain everything...

That doesn't sound right. I think you shall pospone your critical listening until you have a stable, "definitive" version of the amp. Then, every little mod must be followed by extensive listening test, preferably in many systems.

And no, IMHO the hearing system is much better than electronic gear in evaluating what the amps are intended for: music.

Best wishes.
M.
 
Looking at this the emitters of the LTP and current mirror transistors have swamping resistors which are equal in value - creating a situation somewhat like a crank at dead centre.

When the conduction in Q1 increases there is some reluctance for Q2 to pass the expected current due to the checking effect of Q2 emitter resistor on its' base current.

Any Q1 current that is not accepted by Q2 has an alternative path to follow to the Vas which will be driven harder than necessary - the latter a likely impact described by Mark Johnson in post 2.

As suggested by R Dijk in post 29 to resolve this situation the current mirror emitter resistors need to be reduced in value.
 
So you still have a small oscillation?
I would suggest that, if you have not already done so, omit (short) the "base stopper" resistors in the OP transistors (2.2 ohms) and drivers (22 ohms). The problem I see is that they reduce the gain but also cause an additional phase delay which may not help. So the use of these stoppers is one of those conundrums that may help some circuits (if there is enough phase margin when they reduce the gain and possibly compensate lead inductances) - they were necessary for MOSFETS but I'm not convinced for BJTs.
Then, reduce the base-base resistor between the output stages (68 ohms =>33 ohms) and add a 1 ... 10 uF capacitor as mentioned before.
Use a 22 ohm resistor in the vas initially. This will dominate the cut-off frequency of this stage if you also use a colletor-to-ground capacitor. With ~680pF that is 10MHz which should be stable. Slew rate is limited by how quickly this can charge or discharge, and that requires increased current in the VAS (as I have mentioned often).
While the spec's of the KSC3503 look good, I suspect the low Cjc is due to a lightly doped collector measured at 30V. Two consequences are (1) the lightly doped collector may impact the fT quite seriously at low collector voltages (<10 or perhaps 5V) and (2) the capacitance will increase. Both effects will make the phase shifts worse near negative clipping.
Also, you show the design using TMC compensation. Does the oscillation still occur if you use standard Miller?
For 10Mhz unity gain bandwidth and 150 ohm LTP resistors, the Miller capacitor needed would be 100pF: 47 pF equates to 20MHz which would be more sensitive to layout (trace inductances, power supply to PCB trace (unintended magnetic coupling) effects. So try 100pF in addition to the other mods. I also note that you have a stabilised supply for the input stages. -As I can't see the circuit while writing this I'm not sure if it is lower than the - rail. If not, it may help to make it 10V below the neg rail to offset the higher capacitance/quasi-saturation resistance of the high collector epi mentioned, but you may then need to clamp the swing to the neg rail.
Then if the straight Miller works we are on the way to fixing the problem.
BTW I suspect that your concern that Blameless does not seem to exhibit this problem is that the Blameless used 2MHz transistors originally (MJ802 etc). In principle it should work with 30 MHz OP transistors (like MJL 3281A etc) but as these have higher bandwidth there could well be issues perhaps others may comment on.
 
Last edited:
A "fact" is nothing more than an opinion that broadly fits the visible evidence.

A fact is the evidence that fits the subjective opinion. In the mathematical philosophy of Bertrand Russell, the preexistent belief corresponding to the fact and constituting the referential legitimacy of representational truth. However, in the context of mathematical representation everything gets very problematic.

An "opinion" in its derogatory sense as being used in this thread is a fact that the evidence is not met or not convincing.

It is hard to distinguish between the inconceptualisable and unjustifiable notions of opinion, knowledge and belief. I see no reason to derogate any of them.

There are no universal truths, although there are widely held facts.

I would modify it: There are no universal truths, although there are widely held beliefs.


"Nothing exists except atoms and empty space; everything else is opinion." - Democritus

"The truth cannot be decided by a majority vote." - Democritus

“Wrong does not cease to be wrong because the majority share in it.” - Leo Tolstoy

“The fact that an opinion has been widely held is no evidence whatever that it is not utterly absurd; indeed in view of the silliness of the majority of mankind, a widely spread belief is more likely to be foolish than sensible.” - Bertrand Russell
 
It is conceivable that the 5k6 collector resistor for Q7 is having some effect due to the variation in the voltage applied to the collector. This influences the collector to base capacitance - by a rule of thumb the actual value is inversely proportional to the square root of the voltage applied to the collector of Q7.

In terms of voltage swings in opposite directions Q7 collector to base capacitance will not vary in a linear fashion and the usual Miller capacitor to swamp that effect is out of the question.

A possible way to deal with this would be to a add a parallel capacitor 1nF in parallel with the 5k6 resistor to bleed the charges to 0 volts.
 
Hello gerenis69,

To prevent the base of driver Q15 being driven to the collector, you could remove Q13 and flip D6. If the drivers are a distance from the VAS I wouldn't remove the base stoppers from the drivers. Did you install R13 for safety or prototyping? You could do like mjona said or you could short out R13, there must be a "good" current signal at the collector of Q7, but this defeats the reason of using the EF here.

Just my 2 cents.
Mad
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.