Actually CFP is MUCH worse than EF2 for this. I've described a test which shows THD goes up to maybe 1%; all nasty crossover stuff for a minute or two, until the amp cools to 'normal' temperature. Bob Cordell has reported similar results.
That's not to say it is impossible to design CFP O/P without this behaviour .... but I haven't found any in real life.
There's another evil in CFP O/Ps ... at least as Self does them.
In one version of his early books, Self shows EF2 THD drops with level but CFP THD tends to increase at low level below 1W. (This is on top of the usual measured 'increase' due to S/N) I asked him to investigate this further for a later edition but got no answer
Last edited:
@PMA - I realized that making the simulator lopsided would be easier than straightening the amplifier.
I adjusted the parameters to match your 23mV measurements in post #47. I found that I could replicate the distortion spectrum through the fifth harmonic by using 0.1 ohm and 0.34 ohm emitter resistors. This mimics NPN/PNP Hfe mismatch times source impedance in the amplifier.
And the optimum bias is... more is better.
The curves are for different peak amplitudes in volts.
I feel good about the simple model (this model has negative feedback).
Ed
I adjusted the parameters to match your 23mV measurements in post #47. I found that I could replicate the distortion spectrum through the fifth harmonic by using 0.1 ohm and 0.34 ohm emitter resistors. This mimics NPN/PNP Hfe mismatch times source impedance in the amplifier.
Code:
harmonic % dB
0 0.004641 -86.668312
1 99.992177 -0.000679
2 0.001541 -96.245626
3 0.000031 -130.180912
4 0.000289 -110.772811
5 0.000033 -129.681139
6 0.000104 -119.652462
7 0.000033 -129.573269
8 0.000042 -127.470751
9 0.000025 -132.000111
thd 0.001573%And the optimum bias is... more is better.
The curves are for different peak amplitudes in volts.
I feel good about the simple model (this model has negative feedback).
Ed
Last edited:
Gee, and I thought that measuring real life amplifiers with real equipment actually meant something. My god! Decades utterly wasted by not knowing how to drive a simulator!
Give me a break!
Give me a break!
Actually Ed ... you do. Or at least borrow one.
Ebers-Moll is obviously correct, but there are so many other factors and assumptions made in simulations. A simulator is much more useful in an integrated IC, like an op amp. Outside of that you have too many factors to take into account.
I matched the parts in a Symasym. It was a "utility amplifier" I didn't expect huge things from. I also used different devices for outputs and drivers. My minimum THD point was 5 mA of bias current at 1 watt into 8 ohms. The crossover notch disappeared before that value, I built four, they were all consistent. Non-inductive wire wound resistors for emitters. Many amplifiers I see work the best from 10 to 30 mA of bias current. Some need more. Some designs are simply never happy, typically "designed by ear" models. Matching the outputs (multi-output) and drivers reduces crossover notches.
That is real life, measured performance. No assumptions, no "should be". Real life.
Ebers-Moll is obviously correct, but there are so many other factors and assumptions made in simulations. A simulator is much more useful in an integrated IC, like an op amp. Outside of that you have too many factors to take into account.
I matched the parts in a Symasym. It was a "utility amplifier" I didn't expect huge things from. I also used different devices for outputs and drivers. My minimum THD point was 5 mA of bias current at 1 watt into 8 ohms. The crossover notch disappeared before that value, I built four, they were all consistent. Non-inductive wire wound resistors for emitters. Many amplifiers I see work the best from 10 to 30 mA of bias current. Some need more. Some designs are simply never happy, typically "designed by ear" models. Matching the outputs (multi-output) and drivers reduces crossover notches.
That is real life, measured performance. No assumptions, no "should be". Real life.
I am okay with getting the first-order effect right and designing in a manner that minimizes all other effects. Audio is just an old hobby and I already have enough amplifiers. 🙂
ETA: Here is another check. The simulator needed a loop gain of 700 to produce the right absolute values. @PMA - What was the actual loop gain?
Ed
ETA: Here is another check. The simulator needed a loop gain of 700 to produce the right absolute values. @PMA - What was the actual loop gain?
Ed
Last edited:
Yup, already addressed the issue and clearly stated.
If the amplifier frontend is not anything special then 30 to 40ma is all that is needed.
I am taking the " VAS" into account. Majority of all the 2nd harmonic in these test you show is the = VAS
THD will never change much, until that 2nd harmonic goes down.
Also mentioned THD in bias is more observed at higher frequency.
This particular amplifier, isn't good enough to observe any notable outcome.
Since the frontend and compensation wont allow any higher bias to show high frequency improvements
Your observations of lower signal and higher signal THD in this particular amplifier.
Is not taking the pre drivers bias current into account as well.
And contributing as well to most the observations.
The effects of crossover distortion at low and high signals, and wingspread graphs 3D or not
is well known.
Post#1 shows MJE15030/31 they are also = under biased
Oversimplifying is not a " big deal" your measurements show .0017 and .0016 THD
And including noise, has = no change
You are correct, at low power and high power there will be notable change in THD
from 40 to 120 ma.
If the amplifier loads the VAS to much and the bandwidth is limited.
At higher power expect 100ma + bias to be no benefit. The VAS loading creates more THD than any bias will fix.
Exceeding 40ma is nothing but heat and has no benefit.
Again especially at high frequency, where high bias actually has use.
Which has been and will always be 100 ma which will increase to 118 to 140ma over a timeline
depending on the thermal tracking.
Me saying 100ma then posting 105ma yadda yadda yadda.
It all changes over a timeline, and people coming up with mystical magic bias circuits.
Can go waste time with that too.
Last edited:
The only way to use CFP is strong A-class.
Just imagine, last OPS device are switched off, local intrinsic in-stage feedback loop becomes broken, driver interacts with load straightly and OPS input resistance drops dramatically (two orders magnitude) overloading VAS.
No way to correct such a high-frequency effects in a global feedback loop, it just lacks bandwidth.
CFP should be avoided in any AB-class designs.
additional Folded drivers , dedicated current sources and bootstrapped to the output
Should eliminate most that issue.
From those curves i would say that the optimum is at 35mA, there s not much that is gained at higher bias, and more NFB
easily bring you at those lower levels, indeed commercial japanese amps using fast devices like the 2SA1302/2SC3281
were usualy biaised at 30-35mA.
Indeed then the noted observation at " low levels" like one watt
100ma would improve distortion measurements.
Then also as noted with a fronted that allows it.
High frequency also improved.
Any issues with the second gain stage in the frontend solved with EF3
and with CFP a additional folded driver introduced.
The current sources operating at DC dont give a darn.
Folded drivers bootstrapped to the output. Dump any other nonsense
straight to the output.
Sounds like Lochonthi solved most these issues with what he did for the JBL amplifier.
Including the symmetrical 2nd gain stage.
even though both simplified, basically Hitachi figured it out further to improve it with minimal parts.
The overall device quality was the only threshold Bart was against.
Silly for sure. With you on that, have done it too with Class A , and agree never to be built.
It strangely does and doesn't make class A " better" at the same time.
It just further enforces the actual original claim of being superior for low distortion.
Actual low distortion Class A would be a unbeatable sea of grass for harmonics, buried in the actual noise floor.
The noise floor itself would be the limit.
And entirely within reason, not complicated to achieve. Just too good for peoples imagination to not accept it.
As being the end of where you can go.