Reinventing the N-channel wheel?

[ilimzn]

Originally posted by Ultima Thule
Ilimzn,
thanks for your detailed answer, as always an excellent teacher! Wasn't that your occupation btw?

No not a teacher but I have been told many times I should have been one

So if I got you right you would due to the explanation given use a fixed cascoding voltage 5 and 20 volts for the cascoded BJT, right?
Cheers Michael

Yes, the voltage is constant. A bootstrap cap with zener in parallel will work just fine, and depending on actual parts used and criteria as I explained above, you end up with a zener voltage of somewhere between 5 and 20V.

[Eva]

First, I have solved the issue of the cross-conduction spikes when coming out of clipping, I think. I used a stupid compensation trick that I will explain later.

Second, I decided to use the old VAS transistor BF471 as a cascode and put a BC546B with a constant Vce of 4.5V in its place. I increased the negative rail bootstrapping from 5V to 9V in order to provide enough voltage for Q18. PSpice must have gone crazy or something, because now it says that my circuit has 125dB open loop gain and still keeps 98dB at 10Khz... The real prototype now shows slightly improved phase margin even with lighter compensation, and the problem of oscillation that I previously had when the lower rail was bootstrapped is fully solved now, even without the RC filter that I was previously applying to the bootstrap voltage (33ohm and 330nF). Apparently, the non-cascoded VAS was not liking to be bootstrapped during the last 5V of the output swing, while the cascoded doesn't mind.

And third, I have tried all the suitable output transistors that I had at hand:

IRF640 - 200V 18A 0.18ohm 1200pF
IRF520 - 100V 9A 0.27ohm 350pF
IRF540 - 100V 28A 0.077ohm 1700pF
IRFZ48V - 60V 72A 0.012ohm 2000pF
HUF75309 - 55V 17A 0.07ohm 350pF
HUF75344 - 55V 75A 0.008ohm 3200pF

SKP10N60 - NPT-IGBT 600V 20A 1.8Vsat 580pF

To my great surprise, all worked fine without any adjustment of bias current or frequency compensation networks. The higher capacitance models (IRFZ48V and HUF75344) showed slightly smaller phase margin while the rest showed more or less the same behaviour as with IRF640.

I have also a pair of IRFP450 and some IRFP460, but I can't insert these directly in the breadboard, so I'll have to find out a way to test them without using wires (stability!!!). I'm thinking about soldering small pieces of resistor legs to the thick TO-247 pins in order to fit them in the TO-220 spaced breadboard holes.

As it can be seen, the circuit may be complex, but it is capable of readily producing sound from any gate driven device, even SMPS switching IGBTs!!!

By the way, I have observed up to 0.4V differences in gate threshold for 100mA between devices of the same batch and model... What a chore if I had to match them

[dimitri]

Eva, several questions, if you don't mind
values of V2-V6?
why C1 and C21 are referenced to rails?
why C28, R64 and C5,R28 are referencrd to rails?
what is the purpose of extra lag C20?

all worked fine without any adjustment of bias current

yes, this is because your splitter Q25, Q17, Q18 is sensing voltage drop across R16,R17 and not Vgs.

[Eva]

Originally posted by dimitri
Eva, several questions, if you don't mind
values of V2-V6?
why C1 and C21 are referenced to rails?
why C28, R64 and C5,R28 are referencrd to rails?
what is the purpose of extra lag C20?

yes, this is because your splitter Q25, Q17, Q18 is sensing voltage drop across R16,R17 and not Vgs.

Supply rails (V2, V3) are +-24V now. Gate cell supplies (V4,V5) are made with 9.1V zeners, 330hm resistors and 100u capacitors. V8 lower rail bootstrap source was made with a 5.1V zener, 1K resistor and 100u capacitor, but later increased to 9.1V when I added cascoding to the VAS.

C1, C21, C28, C5, etc... are referenced to the rails because the circuit does not have a power ground, it only has a "floating" signal ground (not decoupled to the rails) that I don't want to contaminate with any non-signal current. The speaker wire comes twisted with the +V and -V wires to the power supply, and the speaker ground is taken there instead. The signal ground is connected to the same point as the speaker ground in the power supply, thus it's a perfectly clean reference. Since all the speaker current and possible spikes flow through these three twisted wires, inductive coupling to small signal circuits is negligible.

[Eva]

Oh, I forgot to mention that C20 stabilises the complex input LTP, that is inherently unstable due to its negative input impedance. I have already explained it in previous posts.

[dimitri]

thank you,
what about short circuit protection?
upper half if protected at 8-10A by two diodes para b-e of Q17 and R33, but this doesn't work for lower half.
I like the idea to use bc546/556 everywhere, but Q13, Q18 and four input devices should be high-voltage types
I also understand your desire to use your own X connection, but I believe it is not optimal for this application

[Eva]

D15-18 and R33-34 don't have any output current limitation effect, the purpose of the resistors is to control the transconductance at RF (the output stage oscillates without them) and the purpose of the diodes is to protect Q17 and Q18 by limiting their Ic to approx. 30mA when the output clips. However, given the placement of R16 and R17, it's really easy to add a classic SOA V/I limiter. The final version will have it.

Concerning my cross-quad derivative, it's a good alternative because the original circuit will latch up or show phase-reversal effects as the amplifier clips, thus being not practical for that application. My circuit does the same in a more practical way, since it shows flawless clipping behaviour. Note that an excelent degree of cancellation of Vbe effects over gm is obtained even when the currents across the transistors are not perfectly matched, because the d(Vbe)/d(Ic) derivative is what really maters, and around 1mA it's almost the same (no matter if a transistor conducts 0.95mA and the other 1.05mA).

Also, in the original cross-quad circuit the inner transistor is operating at Vce=1.4V while the outer operates at the rail voltage. This ruins the cancellation effect because the pattern of change of Vbe with Ic is absolutely Vce dependent, and one Vce is up to two orders of magnitude higher than the other. In my circuit, both transistors are operating exactly at the same Vce (and the same Ic), thus better correction is expected.

Concerning the BC546B/556B in the input stage, they are good up to +-55V rails or so (Vce=65V max.) and above that I would add cascodes placed half way between ground and the rails The only transistors subject to the full output swing are BF871 and BF872, that are rated at Vce=300V and biased to only 3mA, so they will work fine witn rails over +-100V, even without heatsink because these come in a TO-202 case and already have a small built-in one.

However, I don't like to experiment in breadboards witn +-100V rails because the effect of two component legs touching accidentally may be disastrous, that's why I use a +-18 or +-24 current-limited supply until I make a PCB version.

[Eva]

I hate it...






That's pseudo-class-A stuff. It keeps the ouput devices always biased to an adjustable current level (R79 and R80) and there is another "AC cross-conduction" adjustment (R78). That reduces gate-charging spikes (the ones in Ic(Q17) and Ic(Q18)) to an absolute minimum. These waveforms also show the extreme non-linearity that MOSFET devices show in class-B at high frequencies, even in the harmless PSpice models with linear capacitances.

In PSpice it looks beautiful, as always...

[traderbam]

That's the most complicated amp circuit I have ever seen.

[Ultima Thule]

Originally posted by traderbam

That's the most complicated amp circuit I have ever seen.

No, it's just the strong neonic collors in here schematic that makes it complex!

Eva,

OMG, what are you doing..
Ic Q17, 18 doesn't look very funny..

I have thought myself a lot regarding what I call "dynamic-class-A" (you call it pseudo-class-A, don't know what is the correct name for it), I believe the transistion from the sustain to active region must be made softer to prevent such current jumps.

Cheers Michael