Due to the increase in gain with collector voltage the Early effect causes an increase in current (i.e. output slope) whether the base is current (high impedance) or voltage (low impedance) biased. Real circuits will only "flatten" the output curve if the resistors used create local feedback . A CCS is a prime example of significant local feedback.
Isn't a higher early effect value supposed to be better?
This isn't what I'm finding, now that I tweaked the VAF on a 3055 model on an amp I was testing before, I find the thd increasing, mostly as frequency rises.
No change at the low end, a tiny increase that grows with frequency.
Why is that?
Can we assume a similar value for the mj2955?
Yes, I adjusted the bias. I first kept it as is, and the influence was larger.
Adjusting the bias reduced that influence but didn't eliminate it, so having made the corrections to make things work as they were before tweaking, the end result is more thd with the tweaks in place.
And I have verified this on other circuits. Same effect. Beats me!
In any case, I prefer bringing the models as close as possible to reality, so whatever is happening, I suspect would reflect more the real parts, and not the more ideal ones.
Vaf is not the "early effect value". High Early voltage (Vaf) is better and is considered the ideal. But Early effect in language the deviation from the ideal. Early effect being worse means that Early voltage is lower.
Va here is the Early voltage. There are lots of abbreviations made in the terminology that makes this confusing. Technically Early voltage is negative, but usually we just omit the negative sign, which is also true for the SPICE model.
Vaf is forward early voltage, Var is reverse early voltage, just in case Vce turns negative during simulation (which can happen in overload and switching conditions).
I also see a higher distortion with higher VAF! This needs looking into. If you consider a classic quasi complementary design of the 1970's and use VAF~50V the distortion is lower than using ~150 when simulated. It is typically 0.1 to 0.2% with about 30% higher with the higher VAF. However if you cascode the VAS stage the distortion drops to 0.02% or so but we still see maybe 0.03% (simulated) with VAF=150. This shows that the Early effect is significant in simple common emitter stages as the cascode makes the devices almost ideally flat. So that proves high Early voltage is better. To observe a worse performance with the output transistors higher VAF's could be due to higher gain if the VAF is not contributing much to the overall distortion in the output stage. Needs investigating stage by stage.
That can happen, in my experience it's because the increased H2 of the VAS is canceling H2 that is generated in another stage (probably the quasi-comp output stage in your case). Neither stage is actually more linear, and the harmonics that aren't canceled still add together.
In my experience fractional improvements in THD tend to be related more to incidental harmonic cancellation and not improvements in linearity.
In my experience fractional improvements in THD tend to be related more to incidental harmonic cancellation and not improvements in linearity.
The question that is far more interesting is: "Is your 2N3055 close to the real thing?"
And the 2nd question: "Which one of the real things?"
I mean, there have been 2N3055 with ft < 20 KHz. Anything that fits into TO-3 and
that is probably NPN has been sold with that type stamped on it.
It would make much more sense to spend these efforts on a transistor that deserves it.
regards, Gerhard
If you compare simulations of class A stages with high VAF and low VAF the lower distortion is always with the higher VAF. This suggests it is indeed more linear (at least from the output side; the base emitter junction won't be changed much and is best improved using current mirrors or balanced CCS). In a class (A)B output stage h distortions are generally not cancelled as with a class A push pull. Nevertheless I agree that some degree of h- cancellation is the reason.
It may be worth saying again that the old slow 2N3055's made by RCA had fT's of 800kHz and more critically fhfe as low as 10kHz, which probably explained much of the "listener fatigue" of 1970's ampifiers. It was possible to design good (sort of) circuits with that museum piece but only by increasing the VAS current above the typically low 6mA and using low Rbe's (10 ohms) to increase turn-off speed.
Yes, many devices stamped 3055 were probably not genuine. Today's real 2N3055s will be epi base and although the spec. is largely the same as it was will typically have fT's of 3MHz and fhfe's of 60kHz or so. Therefore, new designs using those will perform far better than 1970 style circuits. But TO-3 cans now seem to be being phased out, though no-one has put the "new" 2N3055 epi into a plastic TO264 (though TI offered a TIP3055) and probably won't as the latest generation devices are indeed a lot better.
It may be worth saying again that the old slow 2N3055's made by RCA had fT's of 800kHz and more critically fhfe as low as 10kHz, which probably explained much of the "listener fatigue" of 1970's ampifiers. It was possible to design good (sort of) circuits with that museum piece but only by increasing the VAS current above the typically low 6mA and using low Rbe's (10 ohms) to increase turn-off speed.
Yes, many devices stamped 3055 were probably not genuine. Today's real 2N3055s will be epi base and although the spec. is largely the same as it was will typically have fT's of 3MHz and fhfe's of 60kHz or so. Therefore, new designs using those will perform far better than 1970 style circuits. But TO-3 cans now seem to be being phased out, though no-one has put the "new" 2N3055 epi into a plastic TO264 (though TI offered a TIP3055) and probably won't as the latest generation devices are indeed a lot better.
Last edited:
Some observations- simulating a quasi output stage only (NPN/PNP drivers +2x2N3055 output) fed from a 1k impedance input shows that if VAF is ~50V the overall distortion is 0.68% (at 1kHz, 24V peak input, 60V single rail, capacitor coupled output) which increases to 0.82% when VAF=150. Fourier shows that H2 is reduced by order of magnitude (That's why higher VAF is better) BUT H3 is doubled. Looks like a distortion component is cancelling H3 for the lower VAF... lower gain at lower Vce offset by exponential base response??
Canceling one harmonic and boosting another is a common story. The same thing happens in an LTP. Compared to a singleton input, an LTP cancels even harmonics, which reduces the majority of it's distortion, but the small amount of odd harmonics is actually doubled. The curves combine well but not perfectly. For the LTP this works well because at low signal levels the transistors hardly produce any harmonics past the 2nd. But for an output stage or VAS, you often have a spray of harmonics and can't make an improvement by just canceling one or two of them.
Last edited:
Maestro keantoken,
How do I properly determine the phase margin from the images. Grounding the input and applying AC source inside the feedback loop then plotting V(out)/V(A). By calculation that would be 180-X=phase margin where X is the phase deg at 0dB gain. Is this correct? I thought X is the phase where gain and phase crosses path...can't remember where did I get this idea,
it was posted in the forum.Regards,
Albert
Attachments
Really? I would expect that the distortion of the odd harmonics does not double because the fundamental is doubled too.
Do you have reference or simulation?
Best wishes
David
In an LTP the difference is amplified. SInce one transistor response can be written (1+dv/Vth+dv**2/2Vth+dv**3/6Vth ...) by expanding the exponential term, where dv is the small change in base-emitter voltage the other side sees the odd terms become negative (dv=>-dv) so we get the difference
(1+dv/Vth+dv**2/2Vth+dv**3/6Vth ...)-(1-dv/Vth+dv**2/2Vth-dv**3/6Vth ...)
which gives 2dv/Vth+2dv**3/6vth so compared with a single transistor, the odd harmonics are not doubled because we also have 2dv's between the bases.
Looks to me that they stay equivalent.
(1+dv/Vth+dv**2/2Vth+dv**3/6Vth ...)-(1-dv/Vth+dv**2/2Vth-dv**3/6Vth ...)
which gives 2dv/Vth+2dv**3/6vth so compared with a single transistor, the odd harmonics are not doubled because we also have 2dv's between the bases.
Looks to me that they stay equivalent.
Maestro Dave Zan,
Can you help me with my query above?...guess Maestro keantoken is busy at the moment.
Middle image, cursor is at 0db gain and that corresponds to 83 deg phase. To calculate that would be 180-83=97 deg and this would become the phase margin [still stable I suppose]. Right image cursor is placed right where gain and phase crosses path and that corresponds to 90 deg phase, and to calculate 180-90=90 deg [still stable]. I remember it now I guess I saw this at one of member gootee's posting, not so sure though if I understood it correctly I do think it was explained that the crucial point is where the gain and phase crosses each others path based on the graph. 180 deg is where the system fails [Bode & Nyquist theorem].
Correct me if Im wrong.
Regards,
Albert
Can you help me with my query above?...guess Maestro keantoken is busy at the moment.
Middle image, cursor is at 0db gain and that corresponds to 83 deg phase. To calculate that would be 180-83=97 deg and this would become the phase margin [still stable I suppose]. Right image cursor is placed right where gain and phase crosses path and that corresponds to 90 deg phase, and to calculate 180-90=90 deg [still stable]. I remember it now I guess I saw this at one of member gootee's posting, not so sure though if I understood it correctly I do think it was explained that the crucial point is where the gain and phase crosses each others path based on the graph. 180 deg is where the system fails [Bode & Nyquist theorem].
Correct me if Im wrong.
Regards,
Albert
The phase margin (PM) is determined at the frequency where the feedback loop gain has fallen to unity, sometimes called ULGF for Unity Loop Gain Frequency. It has also been referred to as the "gain crossover frequency" because that is where the open-loop gain curve crosses over a level line that represents the closed loop gain.
The gain margin (GM) is just as important, and it is determined at the frequency where the phase lag has increased to the point where the phase around the loop has become 0 degrees (or 360 degrees), i.e., perfect positive feedback. This frequency is typically higher than the frequency at which PM is determined.
Cheers,
Bob
I think you're correct. My mistake. However different ways of making the comparison can have different results. I just tried it several different ways and the bias currents and IPS loading both matter just as much as LTP vs single.
Sir Bob,
Thank you for the quick reply, that explains it clearly. Sometimes I get lost testing LTSpice methods. I think there was this one tutorial showing how to determine the same using the .measure directive instead that involves the Tian probe. The result can be seen in numerical format via error log, but I always get "calculation failed".
Regards,
Albert