Krell KSA 100mkII Clone

[Stuart Easson]

my experience with the ksa50 was that a higher value of Re made the tolerances in everything less important...

To make a WAG: the amp will become more sensitive to thermal tracking and transistor differences in inverse proportion to the value of these resistors...ie using 0r3 will double imbalances between transistors.

I'd be careful going down to 0r1, at this point a 10mv difference in the Vbe of the outputs will show up as 100ma difference in the current split (with 0r68, the difference would be 14ma).

Since most people are talking about matching outputs perhaps it doesn't matter...though if matching is necessary you'd better match drivers, their Re, outputs, their Re...life's too short, and/or I'm too lazy...

If you make the outputs run in AB and they are essentially cold under normal operating conditions you can probably use half as many anyway. Someone should do the math, but (AFAICT) the only reason to use 4 TO3 pairs in the KSA100 is 'cos they are hot all the time...

A single pair of mj15003/4 are more than adequate for a 50w class AB amp...

HTH

Stuart

[AndrewT]

Hi Stuart,
how did you measure those imbalanced currents for mismatched Vbe?
Since my Klone is still in assembly stage I cannot measure.

I reckon as follows, (this is based on reading the Ic vs Vbe graph).
standard Iq=400mA and Vbe650mV and substituting a transistor with 640mV @ the same 400mA Ic.
Re= 1r0, gives 409mA. Vbias=1050mV, Vbe=641mV
Re=0r33, gives 422mA. Vbias=782mV, Vbe=643mV
Re=0r1, gives 450mA. Vbias=690mV, Vbe=645mV

note that Vbias across the transistor + emitter resistor is fixed.
The Vre varies as bias current changes. This is the self correcting effect.

I had reckoned that going from 400 to 409 for the 1r0 case and becoming 400 to 422 for the 0r33 case gave a reasonable match between case temperatures. Your numbers give the impression that the case temperature mismatch is much worse.
Do you or anyone care to explain.

[Stuart Easson]

Hi,

My reasoning is as follows: If the Vbe difference between a pair of output transistors is 5mv, the current through Vre is where the imbalance has to be equalized, so you would see a difference of 5mv across a pair of 0r1 resistors, hence 50ma difference between them.

All the figures you show are perfectly valid, but two (Vre, Vbe) of the three variables are modified with each line, I think if Vbe had been held constant, you'd have differences that match what I'm calculating. You have adjusted the Vbe of the mismatched transistor for the current flowing through it, but AFAICT you have not done this for the control. Hence by the last measurement you only 'see' a 5mv difference in Vbe and a corresponding 50ma difference.

As for the measurements, I used a wheatstone bridge to find a set of 10 of the 0r68 resistors that were identical, then I measure them in series and divide the result by 10 to get an accurate value of Re, then I simply measured the voltage across it in circuit. Maybe +/-1% accurate.

Clearly the 50ma (or 100ma) difference is only important if one is trying to reduce the overall output stage current to minimal levels.

I measured temperature using a 'gun' (no-contact) thermometer, AFAIK not super accurate, but it seemed consistent. Short term I got differences of ~10C from the hottest to the coolest device. IIRC Vbe changes at 2mv/C...20mv is a 200ma difference if Re is 0r1 or 29ma if Re is 0r68...

Stuart

[AndrewT]

Hi Stuart,

if Vbe had been held constant, you'd have differences that match what I'm calculating

I agree.
But when the output stage is set up with sets of identical transistors each feeding identical emitter resistors then when the bias pot is adjusted all the transistors+ their respective emitter resistors are fed identical bias voltage. If the transistors are identical then for each transistor Vbias=Vbe1+Vre1=Vbe2+Vre2 ...=Vben+Vren and Vbe1=Vbe2=.... =Vben and Vre1=Vre2= etc.


Let's assume for a 6pair output stage that each 1r0 emitter resistor is carrying 400mA, Vbe=650mV, Vre=400mV, Vbias=1050mV and Iq=2.4A.
Now replace one transistor with an unmatched version. Let's choose one that during the selection process gave Vbe=640mV when Ic was set to 400mA. i.e. 10mV low compared to it's partners.

The low Vbe will allow an extra 10mV across Re since by construction Vbias must be the same for all the parallel pairs.
The extra 10mV now leaves 410mV across the emitter resistor and first calculation indicates 410mV is passing. But @ 410mV the Vbe will no longer be 640mV. The transistor must have a higher Vbe than it's selection figure of 640mV to allow it to pass 410mA.
If you now add Vre=410mV to the raised value of Vbe of about 641mV (from the Ic vs Vbe graph) then Vbias =1051mV. But that is not available so the current through the rogue device MUST be lower than first guess of 410mA. I estimate it is approximately 409mA.
This confirms that for 10mV spread in Vbe @ Ic=400mA then using Re=1r0 gives a spread of Ib of about 400 to 409mA.

Redoing this calculation for alternative emitter resistors gives the results I posted earlier.
It seems to show that if Vbias is held fixed for all of the parallel pairs then the unbalanced Ib is reduced from the 100mA figure you quoted to about 50mA due to the feedback action of a 0r1 emitter resistor.
Our different results are purely down to whether Vbias or Vbe is fixed.

Which applies to the operational circuit?

It would appear that the rogue (10mV low Vbe) transistor will dissipate about 15% more quiescent power than it's partners.
Assuming that the heatsink is the same temperature under each transistor then Tj will be raised by about 4.7Cdeg [34.7-30 from 0.45A*50V*(1+0.54C/W)-0.39A*50V*1.54C/W]

For those considering matching transistors that hFE has not entered any of the calculation above. I believe that Vbe is the major selection criteria.

[Stuart Easson]

...the Vbe of the output transistors is, IMHO, the critical parameter, of course it can't hurt to match the other parameters as well...

I think we are in violent agreement here, you have chosen a slightly different interpretation for the Vbe mismatch, if you had chosen transistors that mismatch by 5mv in situ under the required load, the difference would, by definition be 50ma with Re=0r1 and 7.3ma with Re=0r68.

My point is still: A higher value of Re mitigates against the cumulative errors (thermal, Vbe, Re, hfe...) becoming a problem, in fact the total 'current sharing' errors are reduced in the inverse ratio of possible Re...0r1 is 6.8 times worse than 0r68, and 10 times worse than 1r0. I like a simple solution that reduces circuit errors and imbalances by a factor of 10.

I'm going to offer another WAG and postulate that Dan would have chosen 0r1 if he had felt comfortable with the tolerances and stability etc, after all the amp would have produced more power for the same components, not a small thing in the retail world. Not having to hand match all the output components is probably a nice saving for a manufacturer too.

Stuart

[AndrewT]

Hi Stuart,
agreed.
Now back to my questions.

Does anyone want to comment on reducing Re from 1r0 to something lower?
How much lower?
How will it affect thermal stability?
Will it affect sound quality?
Will it affect surviveability on shorted output line?

Would thermal stability be degraded if lower value emitter resistors were fitted. eg when ambient temperature varies. Would the location of the Vbe multiplier become more critical if Re were lower?

[crypt0wind]

I have 2 untouched spare KSA 100 blue/gold boards for sale, $35 plus S&H . Contact me if you're interested.

[jacco vermeulen]

Do a search for the comments of Hugh Dean on the ETI466, it contains a few lines on the value of Re.
Lower Re values improves sound quality, but even pro designers such as Mr Curl need to raise the Re values sometimes, see comments on the Halo JC-1.

[jacco vermeulen]

Houston, we have a go on the Krell space project but Weird Jaccowitz is wearing teflon socks again.

Hit it, boys

[QSerraTico_Tico]

HAHAHAHA really funny!
And you think it sounds better with Teflon sleeving? How can we take you serious after this clip?