New error correction amp

Inspired by NP in a "Hawksford" thread, I have developed a new error correction circuit. The schematic of working amp is attached. I have been testing it for more than a month (both measuring and listening) and results seem to be promising.
 

Attachments

  • pm_ab.gif
    pm_ab.gif
    84.6 KB · Views: 7,412
I really wanted to understand how an "Error Correction" works. It gives good sound (even we do not listen in Prague).
I havent understand about Hawksford type, then PMA gives another topology.
This topology has no bias adjustment. In the prototype, the swing is limited to opamp's supply?
Which topology has the lowest voltage rail inefficiency? Etalon type has R to ground to set the input impedance.
 
Nelson Pass said:


I suppose you could consider dropping diodes 1 - 4.

Most probably yes, provided change of resistor values, especially R7. These diodes play interesting role in maintaining thermal stability of the circuit (they are located on PCB, not heatsink).

Anyway - good idea ;) (again ;) ), as I have couple of spare PCB's and it is possible to make direct comparison.
 
Ummm...

Sorry if this sounds crude but in this schematic I see just an exotic error amplifier with high open loop gain that uses classic feedback to linearize the output transfer characteristic dividing the error by the feedback amount

So using a standard LTP and VAS would do exactly the same job I think, but in a less fashionable way

Any amplifier with global feedback is just an 'error correction' amplifier
 
Eva said:
Ummm...

Sorry if this sounds crude but in this schematic I see just an exotic error amplifier with high open loop gain that uses classic feedback to linearize the output transfer characteristic dividing the error by the feedback amount

So using a standard LTP and VAS would do exactly the same job I think, but in a less fashionable way

Any amplifier with global feedback is just an 'error correction' amplifier

Maybe not exactly. This is a Hawksford-like error correction of the crossover region and it reduces distortion of the output stage by amount of 40dB approx. (without opamp, only the error correction of the output stage). The opamp is used to drive the prototype of the output stage, just for evaluation reason, of course it reduces distortion even more. The AD844 is a high speed CFB opamp (2kV/us) capable to source/sink some 60-80mA. It has to drive the output stage that might be considered as 1.6kOhm//1nF load. Various VFB opamps were tried with unsatisfactory sonic result, leaving trace of the typical opamp sound. The AD844 does not and it also measures in this circuit very well.

The circuit in principle was invented by Nelson and was inspired by Nelson's hand drawn image (see attachement). It is an ingenious mix of negative error-correction feedback (left half) and smoothing positive feedback (right half). The right half is responsible for smooth turn-on and turn-off of the output devices, as can be seen on my web page (measurement well corresponds to simulated image).
 

Attachments

  • bias.gif
    bias.gif
    8.3 KB · Views: 5,165
Eva said:

So using a standard LTP and VAS would do exactly the same job I think, but in a less fashionable way

Any amplifier with global feedback is just an 'error correction' amplifier

Again - not exactly. The major drawback of your approach lies in very high order harmonics (strongly Iq dependent - hard to maintain optimum Iq for whole temperature and signal range), produced by standard a la Self class B/AB output stage. Feedback factor reducing with increasing frequency is unable to correct them and what is worse, it creates new compounds of harmonic distortion, not contained in a signal before closing feedback. And - low THD is only a very small part of audio design ;) . Necessary, but not sufficient.

Error correction also considerably improves behavior of the output stage at higher frequencies - again - less work for NFB.
 
Re: changes for bipolar output transistors?

capslock said:
Hi Pavel,

how would you change R7-R11 for bipolar output transistors? Would you place a complementary emitter follower before or after this stage?

Regards,

Eric

Eric,

the circuit's behavior depends on components values considerably. I have simulated it for diferent MOSFET output pairs and I have to tell that resistor values matter if you change for example to 540/9540 from the 413/118 used. Please see different resistors for N-MOS and P-MOS, this results from different characteristics of both types. Simulation result worked well for 418/113 with 1 exception - R7 had to be decreased from 10k (simulation result) to 2k (experimental result). Decreasing R7 increases Vbias and Iq.

I assume that the circuit could be used for BJT's. Then we should omitt D1-D4 diodes and change values of all resistors, or decrease current of both CCS's (high CCS current helps to drive MOSFET's capacitance). I do not have enough courage to recommend the actual values of resistors for BJT's. I would suggest to simulate the circuit first and then to build it. It should be easier to use double EF as the output stage, to have more "space" for error voltage correction function.

Regards, Pavel
 
PMA said:


Again - not exactly. The major drawback of your approach lies in very high order harmonics (strongly Iq dependent - hard to maintain optimum Iq for whole temperature and signal range), produced by standard a la Self class B/AB output stage.

Error correction also considerably improves behavior of the output stage at higher frequencies - again - less work for NFB.

You plan to recreate very high order harmonics(to cancel those generated by amplifier) with an OP-AMP? How's the slew rate? Hope there isn't too many transistors in that OP-AMP chip...miller will sqander any very high frequencies.
 
cunningham said:


You plan to recreate very high order harmonics(to cancel those generated by amplifier) with an OP-AMP? How's the slew rate? Hope there isn't too many transistors in that OP-AMP chip...miller will sqander any very high frequencies.

I did not catch your point well. The opamp, as mentioned, is the AD844, current feedback opamp with slew rate of 2000V/us. I do not plan to recreate very high order harmonics, but trying to prevent their creation, that's why the error correction is done. The high order harmonics are produced by operation of standard class B and AB output stages.
 
PMA is right in this point :

That 'four-transistor-exotic-op-amp' acting as a gate driver has potentially very high bandwidth due to its 'single-transistor-in-the-signal-path' nature and thus linearizes crossover distortion to some extent up to higher frequencies that any global feedback system

However, it provides relatively poor linearization of output devices in its linear region at audio frequencies, but this job is accomplished by global feedback

Anyway, the switching spikes I've seen in VAS current waveforms of real circuits [not simulation] when driving AB output stages with slow Ft=3Mhz bipolar output transistors had less than 1uS width so we are talking about harmonics above 500Khz

Never trust simulations of the switching or saturation behavior of bipolar or MOS devices. Models doesn't reflect at all actual device behavior in these circumstances
 
This is a test amplifier I designed and built last week :
An externally hosted image should be here but it was not working when we last tested it.


And these are some actual measurements. Supply rails are +-40V and load is 3.4ohms resistive. Signal is 10Khz sine. There is actually a single set of drivers and oputput devices with 50mA bias. Blue trace is the output voltage, and red trace is the voltage drop across R32 so it clearly shows the input signal plus the error-correcting signal coming from the LTP

The circuit has global feedback up to 2.5Mhz and uses pole-zero compensation to mantain high open loop gain up to 10Khz [102dB@1Khz, 97dB@10Khz] with stability. Note that even such a small 68pF C.dom capacitor strongly dominates the VAS input impedance over the output stage base drive current. Note also that PSPICE simulation is worthless here since it shows absolutely different waveforms

Overall waveforms :
An externally hosted image should be here but it was not working when we last tested it.


Detail of PNP to NPN switch spike :
An externally hosted image should be here but it was not working when we last tested it.


Detail of NPN to PNP transition, no clear spike :
An externally hosted image should be here but it was not working when we last tested it.


Note also the width of the spikes that clearly reveal harmonics starting at 1Mhz but no distortion below 1Mhz

Finally, the questions are :

- Do you think that these >1Mhz non-linearities may be ever audible?
- Do you think that Hawksford's 'output stage linearizing op-amps' are worth trying for such a circuit?
 
Eva said:


Do you think that these >1Mhz non-linearities may be ever audible?
- Do you think that Hawksford's 'output stage linearizing op-amps' are worth trying for such a circuit?

According to a large number of our listening tests there is an audible difference between the same stage with and without error correction. There is also an audible difference between well HF filtered and non-filtered signal from a CD player. There is a considerable improvement when using audio link signal transmission with buffered output and properly terminated signal cable at both ends. These phenomenae should be inaudible. I am an EE with 25 years experience in analog measuring instruments design. I have no exact answer. The only think I know is that our traditional measurements have very very low correlation with audible results. Our biggest fault is that we mostly measure in repetitive, periodical and selective manner.