Diode switching would spoil the entire setup, from a measurement perspective. I have no idea how audible those spikes are, probably not.
You'd probably be better with a class A amplifier with sliding power supplies. The class A amplifier power supplies are provided by a couple of class B amplifiers, driven by copies of the input signal. The power rails of the class A amplifier can therefore dynamically adjust to only a few volts above the output level, dissipation would be acceptable.
Easier said than implemented, plus that I am not convinced it is an economic approach, the thermal budget will be, worst case, only marginally better, the setup would have multiple points of failure, protecting the overall amp is difficult, etc... An interesting concept, nonetheless.
Yeah, that sounds about right. I haven't looked into it in too much depth but my gut feeling is that when driven with a 4ohm load it would just converge towards the B curve as the "XD Const" does and driving a higher impedance would just cause it to be unnecessarily biased into Class A sacrificing efficiency.
No true. As explained in Mr. Cordell's book, on Oliver's optimization. Even if you bias at high current, you can still meet Oliver's condition to avoid gm doubling.
I was going to have 5 output stages in parallel and bias 200mA per stage to get 1A. But the Re is getting down to 0.12ohm to satisfy Oliver's condition to prevent gm doubling. But now I change my mind. I just layout a new 8 stages and use 0.22ohm to get about 120mA per stage to get back the 1A bias current and get 8w of class A power.
Also it is very important to note that in heavy biased Class AB( not self class AB) like mine, with 1A of bias, I have 2A of Class A current. With the class XD, if you use a 1A CCS to pull current, your class A current is ONLY 1A, unlike 2A of my amp.
For a heavy biased Class AB like my amp, even though the crossover supposed to be at the center, but the worst part of the kink in the graph has been moved away towards both +ve and -ve rail where the amp goes from class A to class B. This is like shifting the distortion away from the center EXACTLY like the class XD. The advantage is with the heavy biased class AB, if I use 1A bias current, I shift the kink to 2A away. You use 1A CCS for class XD, it ONLY shift 1A from the center.
Also, to pull 1A, you need to dissipate a lot of power on the CCS and the top NPN power transistor. Your limitation is NO BETTER than doing in simple class AB with heavy bias current. If you lower the CCS current, it would not help as the kink is still too close.
Bottom line, with class AB, I get double the class A current than the bias AND no gm doubling, class XD cannot do that. they both have to dissipate the same amount, but the kink of the class XD is only HALF the distance away from the center compare to the class AB.
Last edited:
A picture speaks a thousand words:
From 'Audio power amplifier design handbook' fifth ed. "XD Const" is basically a Class-B amplifier with a constant current source hooked up to the output. "XD PP" is the addition that the CCS is voltage controlled by the output. I won't post any more than that because I don't want to infringe on Doug's copyright.
This and the graph with comparison with class A are exactly what I am look at. The graph with class A shows the distortion is ALWAYS lower than class XD in all power level. with the heavy biased class AB, I maintain lower distortion to power level 3dB higher( double) than the class XD if I use the same bias current. That is the low distortion will extent to -5dB before it will rise in your graph. And if I conform to Oliver's condition, there will be NO gm doubling and the distortion is not going to be as high as your graph.
This pulling current to avoid crossover is really old idea. I did that at work 15 or 20 years ago before I even read anything about distortion and all. Just common sense that if you don't want the push pull pair to switch, keep one on all the time!! I did it and I did not even tell anyone. It's not even worth mentioning. difference is for low power opamp, their is nothing more to worry, just one resistor to -V. But here, you have to dissipate all the power, being able to work even when output swing close to -ve rail. complication is no better than my 8 stages. You still need to have multiple transistors for CCS and more NPN transistors at the top to handle the power dissipation if you are doing any significant current that can make a difference.
Last edited:
Alan,
8 pairs of output device to end up with a measly 8 watts of class A seems like such a waste of time. I don't buy the whole class A argument myself, but that comes down to hearing acuity I guess. If you are not using at least 90db per watt speakers why bother, you will be crossing over into the class AB range most of the time anyway.
8 pairs of output device to end up with a measly 8 watts of class A seems like such a waste of time. I don't buy the whole class A argument myself, but that comes down to hearing acuity I guess. If you are not using at least 90db per watt speakers why bother, you will be crossing over into the class AB range most of the time anyway.
Ha ha, I layout the board because I am bored!!! I am still waiting for those 0.22ohm resistor to arrive. Without that, I am just sitting on my hands. After seeing Nattawa's board, I want to try layout with transistors on both sides, then why not adding 3 more pairs!!!
I am retired, I have time, better than playing crossword puzzle. Also, with 8 pairs, I can look at lower power transistors with higher fT, higher beta etc. I have no plans to sent it out for fab. I am just surprised how far I've gone in the last few days working on it a few minutes at a time when I walk by the computer. It actually fit into a 2.4" X 11.5" board and 0.3" separation between each of the TO264. If I use TO3-P, the separation is a lot bigger.
But seriously, if you look at doing the 1A class XD, you still run into every bit of problem of a heavy biased Class AB. The heat sink power limitation, the SOA still apply to the XD. You cannot just use a single transistor CCS. You end up having just as many transistors as the heavy biased Class AB.
If you do less than 1A, why even border?
Last edited:
Once again, you got it all wrong. The Barney Oliver optimum bias point is not found by any gm doubling avoidance criteria.
The Oliver bias point represents mathematically the point where the small signal and the large signal output impedances are equal. The Oliver bias condition is the best we can do to linearize and minimize the crossover variation of the output impedance.
Gm doubling appears for biasing beyond the Oliver point. This results in increasing the distortions. When bias is further increased, towards class A, the equivalent gm peak broadens and lowers, the net result being lowering the distortions.
Please brush up your understanding with the Oliver classic paper in the HP journal of February 1971 http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1971-02.pdf
I agree with Kindhornman. 8 pairs it`s ok for wider class A region but for a high power amp >100W. Even for a 100W you will have a lot of heat, and everything else is large. One pair in my amp sounds better than 2 pairs (and they are matched) even at 2 Ohms speakers load so go figure... Listening of the prototype is a must or you could be wasting your time and money for nothing.
I based on Mr. Cordell's book p102-p103 on how to keep the output impedance approx constant at crossover and Re dominates in extreme end as r'e is much smaller.
Also it shows in p278 in self's book using 215mA and Re=0.1. that is closed to Oliver's condition.
Actually self explain the advantage of high current in p278 also. I'll let the two authors explain this to you.
Last edited:
I have found that CFB sound better with low GNFB (6dB) than 40dB, at least in amp I designed. Higher GNFB sounds somewhat "too tight".
Interesting that one pair sounds better than two. As I said, I am doing it out of boredom as I am sitting on my hands waiting for the 0.22ohm resistors. I have not even test my first OPS yet. Just like playing crossword puzzle only.
Copy/pasting from your favourite authors doesn't help you cause, whatever that is.
My beef was with your claim that the Oliver bias point is based on
This is incorrect, and I don't recall any of your favourite authors claiming such...
Yes, just the magnitude of the speaker system's input impedance doesn't show this. For a non-periodic input voltage,
the peak transient input current can be much higher than that for a periodic input, for the same peak input voltage.
The input current as a function of time can be calculated for a specific non-periodic input waveform with
the crossover's transfer function.
Last edited:
I think it might have been Pioneer who used this scheme on one of their amplifiers back in the 70's. I think they made a low-voltage class A amplifier and drove its floating power supply with a class B or AB amplifier. They presented it at an AES convention. I think the amplifier was well-regarded.
Cheers,
Bob
I think you still get gm doubling when the current sourced to the load exceeds 2A.
Cheers,
Bob
I don't know why it would improve matters, but there may be a subtle reason that I am overlooking.
Cheers,
Bob
Note that if you want to run deeply into class AB and incur gm doubling, you can use modest error correction to pretty much wipe out the distortion from the gm doubling and get less overall distortion anyway. The HEC error correction circuit can be relatively easily implemented with about 4 extra small-signal transistors, and probably ends up being no more complex than a class XD-PP arrangement.
Cheers,
Bob
Cheers,
Bob