Bob Cordell Interview: Error Correction

Jump to Latest
#2,502
Edmond Stuart said:
Hi Jcx,

First, I'm glad that you too will break a lance for Cherry. This man deserves far more recognition than he gets from those AH's on this forum who obviously don't understand him or even accusing him of fraud.
Having said this, let's come to the point. Using Bob's words "trying to come up with reasonable apples-apples comparisons among them, and let the best technique win". In other words, we must make a fair comparison between HEC and NFB topologies.

In Cherry's example (smplCherry) there are two methods applied to reduce distortion:
1. A nested differentiating feedback loop (NDFL)
2. The inclusion of the output stage into the Miller compensating loop.
BTW, let's define an acronym for his 2nd trick: "OMC" (everybody okay?)

As each of these two methods can reduce the THD20 maximally by about 20dB, it's not 'fair' to compare his circuit with HEC. However, as several stages of Cherry's amp are less sophisticated than in Bob's amp, this lead in advantage will not fully exploited.

Anyhow, to make a fair comparison, one should only apply one of both of these techniques and compare the performance of such an amp with HEC.

Cheers, Edmond.
Click to expand...


Hi Edmond,

I also have a lot of respect for Ed Cherry, and have communicated with him in the past.

We do indeed face multiple challenges in making a fair comparison among the techniques. Let me try this on for size. I think most of us agree that the most difficult "bad guy" here is the output stage, both because of output transistor bandwidth limitations and the fact that it is Class AB. It is also most potentially vulnerable to de-stabilizing influences of the outside world (depending on how much coil, if any, is used for HF isolation).

So it seems to me that in our comparisons of complete amplifiers using different distortion-reducing techniques, we do not want to start with amplifiers where the input/VAS might dominate distortion. For that reason, I would suggest that all the comparisons be done with the same very good input/VAS stage (to the extent allowed by the feedback-reducing topologies). This could be a simplified version of mine or something else entirely that is at least as good, for example.

Whether one is allowed to use more than one of Cherry's tricks at the same time may be a matter of judgment. For example, if we use both of Cherry's tricks, can I then combine NDFL with HEC?

BTW, am I correct in believing that the Leach circuit for dispensing with the output coil could be considered a form of TMC?

BTW, Bruce Candy of Halcro also seems to assert that other NFB approaches are just as good as HEC. I don't have the number in front of me, but he did subsequently get another patent after the one wherein he claimed HEC. In that later patent, there is no HEC, but gobs of high-order NFB loops. Indeed, he even complains about the "need" for a pot in HEC designs. I don't know if any of his products use this later approach.

Cheers,
Bob
 
#2,504
Bob Cordell said:
Hi Edmond,
[snip]
So it seems to me that in our comparisons of complete amplifiers using different distortion-reducing techniques, we do not want to start with amplifiers where the input/VAS might dominate distortion. For that reason, I would suggest that all the comparisons be done with the same very good input/VAS stage (to the extent allowed by the feedback-reducing topologies). This could be a simplified version of mine or something else entirely that is at least as good, for example.
Click to expand...

:yes::yes::yes:


Whether one is allowed to use more than one of Cherry's tricks at the same time may be a matter of judgment. For example, if we use both of Cherry's tricks, can I then combine NDFL with HEC?
Click to expand...

Sure, why not? As a matter of fact we (Ovidiu and I) have done just that and indeed, the distortion was not only considerably reduced by HEC, but also by about the same amount due to NDFL.

BTW, am I correct in believing that the Leach circuit for dispensing with the output coil could be considered a form of TMC?
Click to expand...

Hmm......Basically, the idea is based on the same principle, but the practical implementation is completely different. With all respect, I don't favor his approach.

BTW, Bruce Candy of Halcro also seems to assert that other NFB approaches are just as good as HEC. I don't have the number in front of me, but he did subsequently get another patent after the one wherein he claimed HEC. In that later patent, there is no HEC, but gobs of high-order NFB loops. Indeed, he even complains about the "need" for a pot in HEC designs. I don't know if any of his products use this later approach.
Click to expand...

Sorry Bob, it's hard to comment on that issue. I had a look at his patents, I even spiced one of his circuits, but I got the impression that he enjoys misleading us with his 'inventions'

Cheers, Bob
Click to expand...

Cheers, Edmond.
 
#2,507
Bob wrote:
We do indeed face multiple challenges in making a fair comparison among the techniques.
Click to expand...
I think tying a few things down will help. This thread is about output stage EC rather than designing complete amps, so perhaps we should limit the designs to approx. unity gain, OS circuits?

All comparisons are going to have to be in simulation so they won't be entirely realistic. Even so I think educational comparisons can be made provided a few basic things are specified, like design constraints and performance measures:

For example, design constraints:
1) Must use IRFP244 and FQA12P20 models
2) +/- 35V psu for FETs
3) ideal PSUs, voltage and current sources allowed
4) Up to 500nH output series L can be used
5) 150mA to 200mA FET dc bias current, unloaded
6) dc output offset <20mV, unloaded

You've already suggested THD20 at 50W average into 8-ohms resistive load as a measure. What other performance measures would you like to see?
 
#2,508
traderbam said:
Bob wrote:
I think tying a few things down will help. This thread is about output stage EC rather than designing complete amps, so perhaps we should limit the designs to approx. unity gain, OS circuits?
Click to expand...

Hi Brian,

That's impossible. You can't build (approx.) unity gain OS circuits with NDFL, OMC or TMC.


[snip]
For example, design constraints:
1) Must use IRFP244 and FQA12P20 models
[snip]
Click to expand...

At this moment, that's not a good idea, as (as far as I know) nobody has the correct models for these trannies which also mimic the weak inversion. For reliable simulations this is an absolute must.

May I suggest to use the EKV models of the 2SJ201 and the 2SK1530 in stead? These models are based on real measurements (by me) and developed by Andy_C.
See also the spice thread (and the attachment of course).

Cheers, Edmond.

PS: Remove the .zip extension before unzipping.
 

Attachments

#2,510
traderbam said:
Maybe you can't. :cheeky: What's the problem?
Click to expand...

You need at least a VAS etc. That's not just an output stage, rather an almost complete amp.


Ok by me. The main thing is that a standard pair is chosen so that only the differences in EC circuit are a factor. (I cannot open your file attachment)
Click to expand...

I understand that. Perhaps you have good models for this standard pair.

For unzipping the file you need Winrar. Do you have it?

Cheers, Edmond.
 
#2,511
2sk1530 model broken?

replaced .zip with .rar

sucessfully extracted with 7zip

copied the LtSpice subcircuits into my amp sim

I already had n&p mosfet subcircuit symbols with D G S pin order

but I get >1 A Id with 1 V negative Vgs with the N channel model in LtSpice

Edmond's 2SK1530-Y sub is broken

I downloaded Andy's from:
http://www.diyaudio.com/forums/showthread.php?postid=1316183#post1316183

and I can bias my sim now, distortion results hardly changed from the VMOS models I was using
 
#2,512
Win what??? Can't you use the standard Windows zipper?

You need at least a VAS etc. That's not just an output stage, rather an almost complete amp.
Click to expand...
Isn't the idea of HEC to augment a near unity gain emitter follower and reduce its distortion without changing its gain? This can be done using fewer parts and only NFB loops. You don't need a miller gain stage. Then again, why not have a miller gain stage and an LTP if you wish?
 
#2,513
#2,516
traderbam said:
Win what??? Can't you use the standard Windows zipper?
Click to expand...

NO. :no: Whenever possible, I avoid using Windows s/w.
BTW, winrar too is a standard, used by zillions of people. 😀

Isn't the idea of HEC to augment a near unity gain emitter follower and reduce its distortion without changing its gain? This can be done using fewer parts and only NFB loops. You don't need a miller gain stage. Then again, why not have a miller gain stage and an LTP if you wish?
Click to expand...

Indeed, that's the idea of HEC, but we were also discussing Cherry's tricks, i.e. NDFL and OMC (and TMC if you like). So why limiting the discussion to only output stages (and ignore Cherry)?

Cheers, Edmond
 
#2,518
So why limiting the discussion to only output stages
Click to expand...
Mainly because that is the topic of this thread. Otherwise we end up with a much broader scope...lots of issues.

I'm not sure exactly what "Error Correction" is supposed to mean in analogue audio power amp design. It was originally a bit of spin by Hawksford. The phrase "Hawksford EC" seems to be used mostly in this thread to refer to a NFB loop around a unity gain OS where the excess forward gain is generated by a PFB loop. But using a PFB is a detail as far as I'm concerned.

So I'm arbitrarily calling "EC" just local FB around a unity gain OS stage. Just my attempt to pin it down.

I'm not too bothered what this thread becomes. Can you suggest an alternative set of constraints that makes the discussion more specific than, say, "make a low distortion power amplifier by any means you like using these two FETs"?

Mind you, there are so many parts in some of these "EC" circuits that you could well mistake one for a "complete amp". Or two. 🙂
 
#2,519
lumanauw said:
Hi, Edmond, Traderbam,

It will be very interesting (and important, IMHO) to discuss about the possibilit(ies) of making as good as possible of gain=1 classAB (low biased) output stage. Exploring the use of EC or something else.
Click to expand...

Hi lumanauw,

As I said before, why limiting the discussion to one specific class of output stages and excluding all other (more promising) possibilities?

This is the most used output stage in the whole world 😀
Click to expand...

So what? Sorry, this is a non-argument. In an attempt to invent something new, the worst thing you can do is conforming to the majority.

Cheers, Edmond.
 
You must log in or register to reply here.
Top Bottom