mikeks said:
I think there are two disadvantages to this circuit in regard to implementing a Hawksford-like EC scheme.
First, notice the collector load resistors to ground on the output of the VAS. This is necessary to establish a known gain for the injected error signal to work with. The need to drive this load taxes the VAS and may cause it to have higher distortion.
Secondly, one cannot apply Miller compensation or any other type of feedback compenasation to this stage, as it will mess up the injection of the error.
Finally, the high impedance of the VAS collector node will allow for substantial addition of phase shift into the Hawksford loop, rendering it either less effective or less stable.
Bob
Mental wrestling is fun, isn't it?
Ah but it doesn't. If you do the maths on H.ec you should see a small difference between Vout and Vin results in an infinite Vn. This is mathmatically provable because my fig. 4 is equivalent to fig 1.
It is provable in simulation too. Andy_c's Spice circuit, which is a mimmick of Hawksford and BC, shows 84dB of loop gain at 1kHz. It would be inifinite if it were not for the integration provided by Cgs of the output FETs and the limited gain of the pre-driver. Any real circuit must be gain limited in some way to ensure stability.
Bonsai said:
I do not know of any commercial implementations of the Hawksford EC scheme with bipolar output transistors, but the two Japanese patents mentioned earlier in this thread show what appear to be practical implementations of the scheme that may have made their way into commercial products.
There are so many different schemes that are forms of sliding bias that it is hard to comment on them as a whole. Some are just a means to keep the bias hot while saving dissipation when not needed, others are arranged to keep either of the output transistors from turning off completely in the Class AB arrangement. I've never used any of these sliding bias schemes myself. They definitely don't seem to directly attack the same issue as the Hawksford EC scheme, namely transconductance variation in the output stage.
In general, even with bipolars, I would say that sliding bias schemes are inferior to Hawksford EC in terms of overall performance.
Bob
janneman said:
Although somewhat greater than unity, there can be no high difference gain, since the later is effectively subjected to the gain of a single BJT operating in common-emitter mode.
Jan, Andy,
I threw a spice model together that attempts to accurately mimmick Hawksford's system as applied to a FET with 8 ohm load.
The voltage source V3 is an addition - it has zero dc value and is just there as a means to measure the loop gain.
What I find in simulation is that the circuit is completely unstable. Even with an "ideal" FET made from a V controlled CCS with fixed Cgs it is unstable.
Brian
I threw a spice model together that attempts to accurately mimmick Hawksford's system as applied to a FET with 8 ohm load.
The voltage source V3 is an addition - it has zero dc value and is just there as a means to measure the loop gain.
What I find in simulation is that the circuit is completely unstable. Even with an "ideal" FET made from a V controlled CCS with fixed Cgs it is unstable.
Brian
Attachments
Bob Cordell said:
Actually, there are no resistive loads to ground at the TIS's output.
Bob Cordell said:
On the contrary, the apparent advantage of this arrangement is that ''error injection'' occurs in the current domain; collector impedances may be blissfully ignored, and the global compensation adopted (series or minor loop) becomes irrelevant; no pots. or sundry tuning required.
lumanauw said:
I think there are some mis-conceptions about crossover distortion. There are two types of crossover distortion. The first is static crossover distortion. This is largely frequency-independent (within the output stage) and is mainly due to output stage total transconductance variation as one goes through the crossover region. Because of the exponential Vbe characteristic in combination with emitter ballast resistors, it is generally not possible to go through the crossover region without experiencing a variation in the net transconductance, which is the sum of the transconductance contributed by the N and P transistors. This has nothing, per se, to do with the fact that one of the transistors ultimately turns off. Douglas Self has described this issue in detail in his book.
Moreover, most sliding bias schemes that prevent one transistor from turning off do not alter this reality. In fact, although they may prevent one transistor from turning off by maintaining some bias current through it, they generalyy do not prevent that transistor from ceasing its contribution to total transconductance. These schemes thus appear to do little to mitigate static crossover distortion.
The second type of crossover distortion is dynamic crossover distortion. It occurs at high frequencies when the rate of change of output current is high when the signal is going through the crossover region. It is often primarily due to the inability of the driver circuits to turn off one of the transistors fast enough. This distortion tends to be reduced in proportion to higher speed of the output devices. Some have argued that dynamic crossover distortion is reduced by keeping one of the output transistors from turning off completely, but it has not been my experience that this makes a big difference in an otherwise well-designed output stage.
Bob
traderbam said:
In fact you found a maximum loop gain in the region of -14dB, which, being much less than unity, hardly qualifies as ''large''.
lumanauw said:
This patent just looks like a current source feeding the load, with a zener diode across the load located in close proximity to the load. The actual circuit he describes appears to be a description of an electronic precision zener; that's all as far as I can see. The net current flowing to and from the load/zener combination is then constant at the value of the current source, which achieves the desired elimination of fluctuating currents in the ground line.
Bob
traderbam said:
Well, I'll certainly look forward to the outcome, but remember, the proof is in the pudding. Theoretical arguments and even SPICE will not necessarily prove the point. At minimum, you need to build it and demonstrate THD-20 well below 0.001% at full power to make a convincing point.
I'll be fascinated to learn exactly why Candy has moved away from Hawksford if indeed he has. Does anybody here know what is really inside his product?? I don't. All I know is what he has patented. I also know that three or four different models of his amplifiers have been reviewed in Stereophile, and not one of them has yet come close to its claimed distortion spec. Look at the Stereophile reviews over the past couple of years and see for yourself.
Vanderkoy and Lipshitz long ago did an AES talk and preprint that showed the inherent high-feedback nature of the Hawksford circuit. I don't have the preprint number handy, but it shouldn't be hard to look up.
What I mean by "tight" is that the number of places where excess phase can build up in the Hawksford loop is pretty much kept to a minimum. Having a "tight" loop in this sense is, in my opinion, one of the keys to the good performance of that circuit architecture.
Bob
Hi, Bob,
About crossover distortion. In DougSelf book, when P and N both conducting (classA region in classAB power amp) is called "GM Doubling". Is this the same as crossover distortion?
Bob, a question about Hawksford EC. Is that Hawksford EC inherently have "anti crossover distortion" mechanism? I read that Hawksford EC automaticly prevents crossover distortion. Is this true?
I also understand the patent like that. In the end I think it is a description of full classA operation (constant draw from supply) rather than immitating battery. What do you think the main advantage/difference when using battery as +/-rail?
About crossover distortion. In DougSelf book, when P and N both conducting (classA region in classAB power amp) is called "GM Doubling". Is this the same as crossover distortion?
Bob, a question about Hawksford EC. Is that Hawksford EC inherently have "anti crossover distortion" mechanism? I read that Hawksford EC automaticly prevents crossover distortion. Is this true?
Hi, Janneman,
Very good idea 😀 Is it working as planned? How do you get the exact B and 1/B to ensure it working properly, especially for reactance gain at higher frequencies?
Your idea can be implemented for output stage. For class B (or AB). Is it possible to make a topology that "nulls" the transfer function between N and P channel, so in all level output (+rail to 0 to -rail) the output stage has always constant GM, eliminating the properties of N and P output halves? (eliminating transistor's transfer function, GM doubling, Xoverdistortion)
Very good idea 😀 Is it working as planned? How do you get the exact B and 1/B to ensure it working properly, especially for reactance gain at higher frequencies?
Your idea can be implemented for output stage. For class B (or AB). Is it possible to make a topology that "nulls" the transfer function between N and P channel, so in all level output (+rail to 0 to -rail) the output stage has always constant GM, eliminating the properties of N and P output halves? (eliminating transistor's transfer function, GM doubling, Xoverdistortion)
mikeks said:
Sorry, I missed that; those two 2K resistors are connected to the output node, not ground as I had first thought.
I still think the second issue is a problem. Look at his compensation - it is plain old shunt rc compensation at the base of the VAS, and just 44 pF shunt C at the VAS collector - so there is no feedback locally around the VAS (as a result of HF compensation) to reduce its high-frequency distortion, as there is with Miller compensation or with feedback compensation from the VAS output node to the input stage.
Bob
Bob Cordell said:
Yes, he uses shunt instead of feedback compensation; i prefer the later.
Incidentally, those two resistors shunting the output stage seem ill advised. Thoughts anyone?
lumanauw said:
I think that the main advantage of a battery is that it is less likely to be able to screw it up in some way, such as by allowing EMI ingress or by having EMI created by rectifier diodes, or by having some kind of an earth-ground noise path. I think that when done with care by an experienced designer, an a.c. supply can be every bit as good.
So-called GM doubling occurs when a Class A-B stage is biased at higher than the optimum Class A-B current. Let's say you have 0.5 ohm emitter ballast resistors and bias the power transistors at a very healthy 200 mA. The re (1/gm) of each output transistor at crossover is 0.125 ohm, quite small compared to the ballast resistance of 0.5 ohm. The net output resistance of the output stage at crossover will be about (0.5 + 0.125)/2 = 0.31 ohms. At high currents on one side, the re of the ON transistor will be negligible, so the output resistance of the output stage will be essentially the 0.5 ohms of one ballast resistor, or on the order of twice that at crossover. It will go to a factor of two in the limit as the stage is over-biased. The difference in output resistance in this example will result in a difference in output stage voltage gain into 8 ohms of about 2.4%, which will result in static crossover distortion.
In a more optimally-biased bipolar output stage, you reduce the bias current to where the re of the output transistor at idle crossover is on the order of the ballast resistor value. In this case, that would be 0.5 ohms. Then, at the middle and at both current extremes you get a net total output stage resistance of 0.5 ohms. This is oversimplified, but you can see the point. For an re of 0.5 ohms, you would bias the bipolar at about 50 mA. In reality, sometimes the optimum current is even lower. Self goes into a lot of detail about this.
In practice, I believe it is better sonically to over-bias (Class AAB) and take some gm doubling, since in reality it is difficult to manage that small optimum bipolar bias current in the face of large current changes and thermal variations that cannot be properly tracked (because the thermal time constant of the output transistor die is much shorter than that of the heatsink, just for starters).
Yes, to first order it is fair to say that Hawksford EC directly attacks crossover distortion.
Bob
Bob Cordell said:
Somehow I got no notifications about developements on this thread till today, neither was I browsing it, so I missed completely the EC discussion.
This is a topic I have been working on now for almost 2 years, with interesting results both in practical terms as well as form a theoretical standpoint.
Jan Didden is well aware about this, having been extremely kind and helpfull, something that deserves to be made public, so much so it is a topic he had been working on to the point of submitting a preprint to the AES and applying for a patent, yet been convinced afterwards by Ed Cherry the topology was in fact no better than conventional NFB. As a gentleman he is, Jan offered to sign a confidentiallity agreement I deemed unnecessary based on the same condition.
In fact it works, and I have test amplifiers measured (better than 0.001% THD, beyond my current measurement capabilites) and working now for hundreds of accumulated hours without problems whatsoever.
The underlying theoretical approach to understand how EC in a generalized form relates to GNF, simulation results and prototype results, was condensed in a paper draft Jan reviewed well over a year ago as well as the faculty at the Institute of Electrical Engineering here in Montevideo.
The paper draft has been reviewed also by Mr. Andres Mayo, VP Latin America, AES, who found merits enough for submission to review for eventual publication in the JAES.
I do not know for sure at this time whether partial or full disclosure is allowed being the paper under review, so I cannot provide more details for the time being, something I hope will not hold for long. If the AES review panel finds merits for publication, then full details will be available there.
As is usually the case, I applied for a patent based not really on the topology itself - pretty much prior art at this time - but on the prerequisites for its succesfull application, which are the core of the draft.
Rodolfo
Jan,
Your idea is very creative.
Your description talks about adding the 1/A positive feedback loop to the existing system. When I solve the equations as a closed loop system I get: Vo = Vi/B
BUT.
If you first calculate the OL system gain, by disconnecting the negative feedback loop, you'll see the gain is infinite. This is because of the positive feedback loop. When the loop is closed the error will be zero by definition: IF IT IS STABLE.
So you sort of cheated.
Who cares as long as it works?
I think it will be very hard to make it work for a practical gain element A that has phase shift and potentially even time delay. The positive feedback element 1/A would have to have corresponding phase lead and time reversal.
And if you could make a perfect 1/A element, you might as well use it as a pre-distortion block ahead of A. Then you have a zero distortion OL system without any feedback, which would be better.
I guess a practical system will necessarily operate with the positive feedback gain being 1/C where C>A. The question is about whether it is better to implement a particular system with a positive feedback loop to generate the OL gain or whether it is better to put a gain block in series with A.
Your idea is very creative.
Your description talks about adding the 1/A positive feedback loop to the existing system. When I solve the equations as a closed loop system I get: Vo = Vi/B
BUT.
If you first calculate the OL system gain, by disconnecting the negative feedback loop, you'll see the gain is infinite. This is because of the positive feedback loop. When the loop is closed the error will be zero by definition: IF IT IS STABLE.
So you sort of cheated.
Who cares as long as it works?
I think it will be very hard to make it work for a practical gain element A that has phase shift and potentially even time delay. The positive feedback element 1/A would have to have corresponding phase lead and time reversal.
And if you could make a perfect 1/A element, you might as well use it as a pre-distortion block ahead of A. Then you have a zero distortion OL system without any feedback, which would be better.
I guess a practical system will necessarily operate with the positive feedback gain being 1/C where C>A. The question is about whether it is better to implement a particular system with a positive feedback loop to generate the OL gain or whether it is better to put a gain block in series with A.