Another error correction output stage.

[jan.didden]

It becomes clearer to me. Imagine the attached schematic is made with abstract ampifying blocks like janneman's nice diagrams. By varying the values of R1, 2 and 3 and the gain of X2, a continuous range of combinations of EC (HEC included) and normal NFB are possible. The schematic from the previous post is virtually normal NFB, which isn't really what I set out to do.

Right, that's it! Actually, Mr Hawksford started out in a similar way in his paper, and then zoomed in on the special HEC case. One reason was that because only a gain of 1 is required, he could do it with a few transistors: use the emitter to inject a currect proportional to error, and use collector current to develop correction voltage. Since Ie = Ic (almost) using equal resistors gives the HEC gain. And those transistors are already there - in the bias circuit.

But anothe advantage I see is that because of the unity gain around the output stage, there is absolutely no stability issue to worry about. Even if you would have to use some extra components, for me it still is a very attractive topology. I hate oscillating amplifiers, and those compensation things like Vas Cdom and what have you cost you performance.

Jan Didden

 

[jan.didden]

I've been playing around with the circuit a bit, and it seems that I've been overly worried about keeping the amount of EC reasonable to maintain stability.

Removing all feedback around the error amp can yield perfect stability as long as the output stage itself shows well damped frequency response. I was sceptical of the sims at first, but the attached circuit does actually work, and better than the original. It even works with normal speed op-amps, so I think when I get around to building a proper one I will use OPA445 so I can have higher voltage rails (up to +/-45V).

It's also interesting to see what happens as the value of the input resistor (R7) is varied. When it's infinite there is no EC, only normal NFB and distortion is at one maximum. As it's reduced, so distortion reduces until a minimum at approximately the value shown. Then as it's decreased further, so distortion increases again to a maximum when the resistor is zero, shorting out the EC entirely.

Can't run a sim?! I suggest you get yourself over to the Spice page and find youself a free simulator.

However, not having a simulator is no excuse: Use a breadboard! I think it's best to use both, since both can reveal things that the other can't.

Excellent post! The fact that varying that resistor gives you a minimum shows the system works as it should. I have exactly such a curve from my experiments with it a few years ago. The minimum was at gain = 1 as expected.

One thing I noted is that there is a difference in distortion depending on whether you are above or below unity: The phase of the distortion products wrt to the main signal switches 180 degrees when you go through the unity gain point (where they should be zero in theory).

I was thinking at that time to use it to cancel distortion in a bridge type amp, you know, one just below unity and the other just above, thinking that would in practice be easier to realise than exact unity. But I couldn't get it to work satisfactorily.

Jan Didden

 

[Terry_Demol]

But anothe advantage I see is that because of the unity gain around the output stage, there is absolutely no stability issue to worry about. Even if you would have to use some extra components, for me it still is a very attractive topology. I hate oscillating amplifiers, and those compensation things like Vas Cdom and what have you cost you performance.

Jan Didden

Jan,

WRT stability, agreed, however I believe you probably still need a
LR OP decoupling network for stability into cap loads. Anyone
else here know if Hawksfords EC can drive cap loads direct
(obviously without global FB connected)?

IMO, it is a sonic advantage not having to use OP decoupling
network.

Cheers,

Terry

 

[Steven]

I've driven Quad ESL63 loudspeakers with my amplifiers (with Hawksford error correction and no global feedback) without series L in the output and without Zobel (Boucherot) filter. No stability problems.

Steven