Or maybe he doesn't want to tell you. Trade secret and all that. An incentive do to do some investigating...
Bob Cordell's Power amplifier book
- Thread starter ihan
- Start date
Or maybe he doesn't want to tell you. Trade secret and all that. An incentive do to do some investigating...
Then, he should not write the book and charge so much for that!! I found good articles here free. I am reading Bonsai's article, he hit it right on the nail with the problem I faced today. I had to discover the hard way.
I am going to try lowering the resistor soon as I see good result with 0.22ohm with my matching transistors. Only 2mV spread. I might be able to take it down to 0.15 or lower. I'll report back.
Alan, Jan is correct. There will be no substitute for burning the midnight oil. You should be able to get DS or RC methodologies to work - but you still have to work at it. I took a slightly different tack, but this in no way detracts from their solutions.
Right. In addition to giving his experience here, and writing books about it, he should provide unlimited free consulting to anybody here with a personal blank spot.
You really crack me up.
Jan
Hi Alan,
I'm sorry to say so, but your bias generator is wrong. The main culprit is Q18, which lowers the tempco of this circuit (unless it's also mounted on an O/P device, of course). Try fig. 22.28-b of APAD6.
Cheers, E.
Yes, I think Mr. Self should take his rotten books off the shelves without haste and refund the money to all the lemmings that were suckered into buying such tosh... His nonsensical drivel has set audio design back a few decades and misled many wayward young designers - this must be stopped!😀
Thank you for the reference, I was just about to revise sensitivity analysis so it was most opportune.
But I have studied it and Damir's amp and see no problem.
Only two relevant poles and the Q is low.
Why should this cause concern?
Do you have any analysis or simulation that shows a problem?
Best wishes
David
(6.5mV quiescent bias of the Audiolab 8200 may be a tad low)
Doug Self's thermal comp circuit has two compensation mechanisms at work.
The first one is the conventional bias spreader due to the action of Q12 - this provides 1st order correction. The second order correction is provided by R12.
During the design phase there would be a few iterative adjustments between the pot setting and the value of R12, but once calibrated it would be fixed for that specific heatsink and layout configuration. I suspect you could then churn them out in their hundreds with good repeatability.
Not quite two point thermal comp as I ended up doing, but an elegant solution IMV.
The first one is the conventional bias spreader due to the action of Q12 - this provides 1st order correction. The second order correction is provided by R12.
During the design phase there would be a few iterative adjustments between the pot setting and the value of R12, but once calibrated it would be fixed for that specific heatsink and layout configuration. I suspect you could then churn them out in their hundreds with good repeatability.
Not quite two point thermal comp as I ended up doing, but an elegant solution IMV.
Last edited:
R.Cordell has a big section on arranging different Tempco for the multiplier.
I think part of your problem and it affects all of us, is that the temperature range of the sensor is much lower than the temperature range of the output stages (driver + Output).
If the driver + output has 4 Vbe and each sees a temp range of 20C degrees, and the Multiplier/sensor sees 10Cdegrees of temperature range, then a tempco for the sensor needs to be equivalent to 8Vbe*50%, to apply the required 4Vbe correction.
If the drivers have a temperature range that is significantly less than the outputs, then your sensor tempco can be a bit less.
If the sensor is inside the outputs, then this location is very likely to have the highest temperature range cf all the other "external sensor locations"
Last edited:
Indeed. I'm surprised it's taken this long for someone to point that out.
Alan needs to read chapter 14 of Bob's book more carefully. In particular, section 14.8 talks specifically about manipulating the temperature coefficient of the bias spreader.
Perhaps for his second edition, Bob could consider addressing some of this earlier in the chapter, as the issue of tempco is relevant whether the output stage is using ThermalTrak transistors or not.
I don't get what you are trying to say here? Are you talking about thermal attenuation? If the bias spreader is on the back of an output device and then surrounded by insulation, thermal attenuation should be minimal and the spreader temperature should track output device temperature very closely. Even without the insulation, I don't think you'd get a factor 2 attenuation.
Let's not forget we have 6 delta VBe's stacked up in a triple - that's a lot of compensating to do!
Hi Andrew,
>" the temperature range of the sensor is much lower than the temperature range of the output stages (driver + Output)."
Alan uses a diamond driver, which is already temperature compensated. So he needs only to compensate the OPS, in the right way of course. The trouble with his circuit is Q18, a PNP tranny, which makes it necessary to lower the resistor ratio at the base of Q9 (the sensing tranny), hence the tempco of the bias generator is too low.
Using a NPN tranny for Q18 (and adapting the circuit accordingly!) has the opposite effect and hopefully matches the tempco of the OPS much better.
Cheers, E.
>" the temperature range of the sensor is much lower than the temperature range of the output stages (driver + Output)."
Alan uses a diamond driver, which is already temperature compensated. So he needs only to compensate the OPS, in the right way of course. The trouble with his circuit is Q18, a PNP tranny, which makes it necessary to lower the resistor ratio at the base of Q9 (the sensing tranny), hence the tempco of the bias generator is too low.
Using a NPN tranny for Q18 (and adapting the circuit accordingly!) has the opposite effect and hopefully matches the tempco of the OPS much better.
Cheers, E.
>6 delta Vbe ? No! Just 2.
Thermal attenuation sounds like we are in the same ballpark.
If thermal insulation were perfect then I would agree, but it isn't.
The sensor will not change by the same temperature range as the junction of the output, Even the TT/Sanken can't achieve that.
BTW, I used the *2 factor just to make the arithmetic simple and arrive at the Vbe*50%
Last edited:
Edmond, can you explain why? I am not getting your point on this.
I'm not sure if R12 has that big a role to play in the thermal compensation, it's much more to do with keeping the bias spreader's voltage constant in the face of varying current through it.
In reference to Doug's use of 0R1 Re in multi-pair output stages, contrary to Bob's recommendation, this does suggest that we must be missing something in terms of the analysis presented in Bob's book. This analysis would suggest that 0R1 Re is extremely risky, in terms of the possibility of a single device in a multi-parallel-device output stage hogging all/most of the current and thereby exceeding its SOA.
Some comments in regards to this:
1. Doug has stated that 0R1 presents no problem in terms of thermal stability. This is, however, not surprising. What one might expect, in the absence of device matching, is current hogging in a multi-pair output stage. Perhaps Doug meant for the phrase "thermal stability" to cover this aspect of thermal design also, but thus far, he has not stated explicitly that current hogging is not a problem.
2. Perhaps his designs "get away with it" because audio amps are used to amplify music signals and not sinewaves. Perhaps with a 1/3 rated power sinewave test, the amplifiers would not fare so well.
3. Perhaps Bob's analysis is overly conservative and the thermal resistance numbers are much lower, giving a below unity (hence safe) thermal loop gain in the Re=0R1 case.