G.Kleinschmidt said:
I agree. There are numerous ways to make the EC work with BJTs and give the EC circuit adequate headroom. Most of these ways are no more painful than what must be done when using MOSFETs. Once you bite the bullet and use boosted supplies for the VAS, this is pretty straightforward.
Cheers,
Bob
ingrast said:
The two techniques are different, but in an apples-to-apples simulation their performance is virtually identical. This was startling even to me.
The key point of being apples-to-apples is that the same compensation roll-off in the error path that is needed to stabilize the HEC is also present in the error path when the error is fed forward.
Commonalities
- both have adjustment that provides an error minima
- both depend on accurate amplitude and phase matching
- both produce the same distortion spectra
- both drove THD-1 from 0.08% down to 0.000064%
Differences
- the feedforward summer is difficult to realize in practice
- the HEC requires frequency compensation
The schematic of my simulation circuit is shown below. C1, R4 and R6 represent the compensation circuit. C2 represents finite bandwidth of the summer. Obviously, in a real circuit, your mileage will vary.
I’m sorry I missed that earlier post. I’ll try to reply to it later in a separate post.
Cheers,
Bob
Attachments
Edmond Stuart said:
I should have said that the error correction performance was virtually the same. You are right - the HEC scheme requires compensation and the feedforward scheme does not. Moreover, stability into arbitrary capacitive loads will of course be different. These are reasonable caveats. My amplifier only required a 0.5 uH and 0.5 ohm output isolation network to achieve unconditional stability. This is certainly not unreasonable.
On the other hand, I dare you to realize in a practical way the output summer needed to do the analogous error correction function with error feedforward.
Cheers,
Bob
Bob Cordell said:
Why? What I am missing here? Unless what you are implying is the error processing must be performed actively (true for proper "cancellation") what in turn implies at least one pole.
Neat
Now, this feedforward thing is intriguing. If one could process the error (orders of magnitude larger bandwidth than working bandwidth) and inject passively regardeless of load magnitude and phase angle variations .... Of course for this to be devoid of stability issues, the error pick up point must be as in your test circuit BEFORE the output summing node, otherwise it is feedback bussiness as usual.
Rodolfo
janneman said:
In theory you don't lose the null. I'll put the results of the math at the end.
There's something missing from your block diagram. At the output node, the block diagram should show a summer adding in the error like so:
Vout = N * Vn + err
Then, the goal is to express Vout as a function of Vin, N, P, Q and err, and find conditions such that the term involving err drops out. I did this with a block diagram similar to yours, except that the plus and minus signs were reversed in the summer on the right side, and the summer on the left had a minus term associated with Q. I got this result:
Vout = [1 / (1 + NQ - PQ)] * [N * Vin + (1 - PQ) * err]
So when PQ=1, err goes away. and Vout/Vin = P (=1/Q)
Last year when I was playing around with this in sim, I had the very same problem with losing the distortion null when optimizing EC dynamic range. After playing around with the block diagram equations as above and implementing the diff amp gain fix, the null (using idealized controlled sources for the summers) came back and the distortion was below the ~-180 dB residual of LTSpice.
Edit: In your equation above, it's only necessary that PQ=1. For this condition, Vout/Vin = N / NQ = 1/Q = P. It's not necessary that Q=1.
janneman said:
Sounds right!
Or another way of looking at it is that the final gain is the same as the voltage divider ratio. And the voltage divider makes a great place for the distortion trim pot.I think you could make a good case that I'm overemphasizing the importance of maximizing the dynamic range of the EC. But the idea I had for my design was, in simulation, to minimize the distortion of the output stage itself under very high current conditions (say, 100A peak). So it was just a way to try to make the EC as robust as possible.
andy_c said:
Sounds right!
I think you could make a good case that I'm overemphasizing the importance of maximizing the dynamic range of the EC. But the idea I had for my design was, in simulation, to minimize the distortion of the output stage itself under very high current conditions (say, 100A peak). So it was just a way to try to make the EC as robust as possible.
No I think your emphasis is right to the point. In my implementation I use a current conveyor as the ec element, and that CC runs open loop. I don't use additional nfb around the amp. The final result depends very much on the linearity of that open loop CC. Optimizing things like you did with the 0.95 attenuator decreases the signal levels in the ec circuit and improves overall results.
On a side note, I find it counter-intuitive that the final gain with the 1.05 gain correction comes out to 0.95 . Or maybe my intuition isn't what it used to be
Jan Didden
janneman said:On a side note, I find it counter-intuitive that the final gain with the 1.05 gain correction comes out to 0.95 . Or maybe my intuition isn't what it used to be
I think of it this way. If the voltage divider ratio is the exact same as the linear gain of the output stage, then for a distortionless output stage the error signal will be exactly zero. If the error signal is exactly zero, the EC isn't doing anything at all, so the gain with EC will be the same as the gain without EC.
Of course, if the voltage divider ratio does not equal the linear gain of the output stage, then my intuition fails too. This is where my time-tested approach of using math as a crutch comes in
.
janneman said:
Hi Jan,
I too was thinking about his design. It was most interesting as far as I can remember. Regrettably, I failed to find it back in my large pile of Electronics World magazines. According to D. Self's website it was published in Oct. 1994, p. 818. Does anyone have a pdf copy? TIA.
Cheers, Edmond.