It is not so difficult to understand nature of distortion, the high-order is mostly crossover, non-class A.
BUT - in case there was 10x higher low order distortion (like -70dB 2nd), the scale on distortion waveform would be 10x lower and high order distortion would not be apparent, or completely "disappear".
BUT - in case there was 10x higher low order distortion (like -70dB 2nd), the scale on distortion waveform would be 10x lower and high order distortion would not be apparent, or completely "disappear".
john curl said:
John, thanks. Makes sense. From the discussion in this great thraed, I gather that many people (me included) looked too narrowly on the BJT vs MOSFET story. Low drive current, no second breakdown, best thing since sliced bread. If I learned anything in this thread it is that it is a LOT more complex. And that you can build excellent amplifiers either way.
Jan Didden
PMA said:
I agree. My experience is that the lower the distortion is in absolute terms, the more 'ragged' the waveform of the THD+N becomes. One could also say that the ragged shape 'proofs' that the THD is very low.
One could add in 0.05% 2nd and the waveform would look nice and smooth, but the underlaying 'ragged' components would still be there. Food for marketing departments!
Jan Didden
Hi,
since Xpro referred us to the test results, can I ask what folk think about the relative performance into 4ohms cf 8ohms.
The test results showed 19.3dbW 8r and 17.8dbW 4r.
That reduction of 1.5dbV into half value load seems high.
I try to get <-0.8dbV between 8 & 4ohms.
and more often achieve <0.6dbV (using BJT).
I think this has an effect on the apparent ability to play low bass properly. This of course applies to an 8ohm capable amplifier. If I were designing a 4ohm capable amplifier I would be looking to comparing 4r to 2r results.
Would the adoption of BJTs over FETs affect this type of result?
Are FETs up to giving a good result in the low load value situation?
since Xpro referred us to the test results, can I ask what folk think about the relative performance into 4ohms cf 8ohms.
The test results showed 19.3dbW 8r and 17.8dbW 4r.
That reduction of 1.5dbV into half value load seems high.
I try to get <-0.8dbV between 8 & 4ohms.
and more often achieve <0.6dbV (using BJT).
I think this has an effect on the apparent ability to play low bass properly. This of course applies to an 8ohm capable amplifier. If I were designing a 4ohm capable amplifier I would be looking to comparing 4r to 2r results.
Would the adoption of BJTs over FETs affect this type of result?
Are FETs up to giving a good result in the low load value situation?
john curl said:
John, it is a residual of some crossover distortion and level is very low. By increasing the idle current you can easily get rid of these completely but it did result in reduction in the sound quality and so the bias was set at a certain point deliberately on sonic grounds. The linearity of the circuit could be seen clearly from the IMD graph (fig. 8) . There is nothing to be ashamed of in a budget amplifier
.AndrewT said:
Andrew, have a look at the fig. 10 and you'll see that the circuit is quite happy even into 2 Ohm. However in a continuous sinewave test both channels driven into 4 Ohm the limit is the power supply - mostly the size of the transformer and filter capacitors. It is not an expensive amp and corners have to be cut somewhere
.Cheers
Alex
Hi Xpro,
I am not critising your design nor the compromises you have incorporated to allow them to build it down to a price. It was a convenient example for my purpose.
I am asking about the merit of voltage drop, or lack of it, in a power amp when feeding continuous low impedances. Whether this affects low bass quality and whether this aspect of performance is affected by choosing FETs or BJTs for the output stage.
I am not critising your design nor the compromises you have incorporated to allow them to build it down to a price. It was a convenient example for my purpose.
I am asking about the merit of voltage drop, or lack of it, in a power amp when feeding continuous low impedances. Whether this affects low bass quality and whether this aspect of performance is affected by choosing FETs or BJTs for the output stage.
AndrewT said:
The low bass quality in my view is not directly connected with this parameter as long as the amp in not clipping. And I can see no real reason why this drop should be very different for a properly designed amp with either BJTs of MOSFETs. In real life this drop is mostly due to the power supply size, unless the output is heavily current limited or has unusually high output impedance.
PMA said:
Thanks, Pavel!
Alex
john curl said:
John, nothing could be further from the truth. I often ask questions and don't assume up front that I know the answers. It is better to hear what the person has to say, and we might learn something more. In this case, not only did I not know what the rails were or whether the amplifier was balanced, but I certainly did not know that Charles was using only a single transistor pair in the output stage, and that part of the reason for that was wanting to avoid matching. The whole story makes sense. That probably would not have come out had I not simply asked the question.
It continues to be a refreshing pleasure to be able to engage in candid technical discussions with Charles without a personal issue being injected, even if we would not engineer things the same way.
Language and the use of it can get us all into trouble, and no-one is perfect. Just a little further back I was remiss in accusing you of being condescending when you use the term "you folks".
Sometimes we just get it wrong. I actually thought that you answered that question in the way you did because you saw an opportunity to take a shot at me. Guess I'm just paranoid.
Bob
Bob, I simply answered the question. It would have gone no further, IF you had not taken exception to my answering the question.
I also wanted to point out that 60V devices were rather popular 15-20 years ago, and cheaper too. In fact, usually, lower voltage devices performed better as fets than high voltage devices, which always lose transconductance, all else being equal.
I also wanted to point out that 60V devices were rather popular 15-20 years ago, and cheaper too. In fact, usually, lower voltage devices performed better as fets than high voltage devices, which always lose transconductance, all else being equal.
Charles Hansen said:
Hi Charles,
Thanks for the information and clarification. The upward slope of the IRCP054 Id at higher Vds is a mystery, but it looks like that amplifier may have had a thermal runaway problem even without that transistor behavior.
I plugged the numbers into my equation for intrinsic thermal stability and saw evidence of a problem.
Recall from my earlier post, that equation defines the positive thermal feedback coefficient Beta as:
Beta = Theta_JS * Vds * TCvgs * gm
For the IRCP054 operating at 550 mA, I estimate gm to be about 3.1 S and TCvgs to be about 6 mV/C. With Theta_JS = 1.3 C/W (including insulator thermal resistance) and Vds = 28V, we get:
Beta = 0.68.
This number is well into the danger zone (which I arbitrarily define as anything greater than 0.5). At Beta = 0.68, we get thermal gain enhancement from positive feedback by a factor of 3.1. Of course, if Beta hits unity, we are into thermal runaway, and it can happen in the blink of an eye, given the fairly fast thermal time constants of the die and the package.
Now let's recognize that the 550 mA bias value was probably set under quiescent, no-load conditions after thermal equilibrium was reached.
Now somebody plays program through the amplifier at modest levels into a load, and by doing so suddenly increases junction temperatures by 20 C (this doesn't take that much additional device dissipation to happen, maybe only 15 watts or so). You have a nice big heatsink, so its temperature doesn't move, and there is no change in bias temperature compensation (if you use it). At 6 mV/C, the 20 C junction increase corresponds to a Vgs change of 120 mV, which, against a starting gm of 3.1S, increases bias current by about 370 mA.
For reasonably small changes, gm is proportional to current in this region, so gm as a result increases to about 5.2 S at the new operating bias current (even though the music may have stopped briefly). We now calculate Beta under the new conditions:
Beta = 1.14 => GAME OVER - we are in thermal runaway.
I know there is a lot of approximation and handwaiving here, but I suspect that this is what happened.
Best regards,
Bob
Bob Cordell said:
Sorry Bob, I didn't read your earlier post, and unfortunately don't have time to fully digest the math in your current post. But this I can assure you -- your theoretical analysis does not reflect the real world.
First of all, we only had a handful of failures. We finally traced them down to high AC line voltage (causing the rails to increase to the point where the curves of the IRCP054 inflected upwards). When we changed to a different transistor that didn't have this problem, the remaining handful of failures disappeared. QED.
Like I've said in a previous post, there's nothing like putting a part on a curve tracer and seeing how it *really* behaves.
Charles Hansen said:
Hi Charles,
First, thanks very much for your candor about your designs and your patience throughout this discussion. I think we have all learned from it.
Obviously, there still remain some questions and areas of possible disagreement.
When you have a chance, I really would appreciate your feedback on my theoretical analysis. If in some way it does not reflect the real world, I'd really like to know in what way. We'll all learn from that.
If you merely mean that it doesn't reflect the real world with respect to what actually happened in your amp, that is entirely possible. Indeed, your replacing the transistor with one not having the upward swing in Id, and then having the problem go away, is pretty indisputable evidence. I would still speculate (and indeed it is only speculation based on the figure of merit I explained) that my analysis was correct in showing that there was not enough thermal safety margin, and that the presence of the upward swing in Id pushed it over the edge.
Again, thanks for discussing this and I would very much appreciate any criticism from you (or anyone else) in regard to the figure of merit for thermal bias stability that I use. I recognize that it is a very simple figure of merit and, as such, at the least invites caveats.
If you would be willing to send me one of those transistors so that I can study it, please send me email at bob@cordellaudio.com.
Cheers,
Bob
Tico, IF you are in class A mode exclusively, THEN there is no optimum value of voltage drop across the emitter resistor. Back in 1967, I built a small power amp with 1 ohm emitter resistors and 0.5A bias. Guess what? Up to 4 watts, (the class A region) it worked great, BUT then in the Class B region (up to 16 watts) it became class AB and THEN the distortion was not nearly so good. I can still remember, after virtually 40 years when the technician measuring the amp, noticed that the amp did NOT like the transistion from class A to class AB. In a few years, I dropped the value to .05 ohm resistors, instead, with all the problems with thermal tracking, added on.
I then had to develop a more accurate thermal tracking method, that I still use today to keep it thermally stable.
I then had to develop a more accurate thermal tracking method, that I still use today to keep it thermally stable.
