Bob Cordell Interview: Error Correction

Jump to Latest
#2,641
Re: annoying leecher

traderbam said:
Bob wrote:
For example, if it is asserted that one cannot achieve 1 ppm with a unipolar input stage and without resort to NDFL, that is an interesting challenge.
Click to expand...
Quite achievable though. Wouldn't you agree that to assert that a unipolar input won't cut the mustard is to assert that a unipolar subtractor cannot be designed and built with <1ppm THD20? Halcro appear to have achieved this.
Click to expand...

I have already answered that question, look here:
:nownow: Nobody has said that.
Click to expand...
and here for Bob's reply:
I'm sorry I mis-interpreted your words in regard to the unipolar input stage.
Click to expand...

Wouldn't it be wise to read the posts more accurately, before making any comment?


NDFL approximately mimmicks a butterworth filter response. So you could use a filter in the forward path or a variety of transfer function modifiers.
You could think outside the HEC box and look at using a different form of NFB around the OS. Your HEC implementation involves a single, "leaky integrator". It doesn't have to.
Lots of interesting possibilities. 🙂
Brian
Click to expand...

If you see lots of interesting possibilities', I would say: go ahead, build such amp and, most important, show us the results, including schematics.
 
#2,642
traderbam said:

Therefore, to compare meaningfully I have proposed more parameters than just THD20 and suggested you and Bob specify them.


Could you please refresh my memory on what parameters you think we should add to the measurements pages on our web site? Honestly, with the latest additions posted here, I can't imagine anything else that we could add.

Therefore, for the sake of a comparison you look so interested to develop, I think you have from our side all the data you need. Again, let us know if we can add anything else.

There are a few parameters (as e.g. the stability margin) that I don't have a clue on how to measure, however I'm ready to listen and learn from anybody's experience on such. For a single pole amp (Miller compensated) or even dual pole (by means of a synthesis step) my analyzer could extract from the closed loop response of an amp the open loop behaviour and that would be one step forward. However, I don't have a clue on how to deal with a NDFL amp...


I have already talked (argued for some reason) with Edmond about stability. I have asked how you two determined acceptable stability in your design and you haven't answered me. That's fine, you don't have to.


Again, I don't like to speak for others, but I think Edmond was looking for some criteria to estimate the stability. I can assure you that or amp is stable (in a resistive load) without any output coil. However, if we are talking about capacitive loads, the 2.5uH output coil inductance was chosen strictly on the criteria to handle a 1uF load without any stability issues (which, to the extend I'm aware of, it's a pretty typical stability criteria). The amp with a 2.5uH coil is able (according to our measurements) to handle a whatever capacitive load between 1nF and 1uF. If you can think of any other stability criteria, just let us know. Unfortunately, I do not have data on Bob's amp behaviour in capacitive loads. You need to ask him about. If I can add anything else on this stability discussion, just let me know.


Edmond has been very bullish in this thread about his use of NDFL and TMC as the only means to get 1ppm THD20, saying that Bob wouldn't be able to achieve this figure without using those methods. That is a bold challenge. Based on Halcro's US patent #5892398 I don't see Halcro using either method.


Well, first of all, according to my measurements on a Halcro DM58, they did not beat the THD20 1ppm barrier. Of course, I have never seen the Halcro schematics, but based on what's in the patents description (admittendly, from my own experience, patents are almost always written in a way to disclose a minimum of information) and mine and Edmond's simulations it is unlikely they will ever beat the 1ppm barrier based on those designs. I recall joking with Edmond in the early design stages that the plan is to "drive Halcro out of business" and "we better also start building a business plan" 🙂 However, in all fairness, an amp design has to also pass the "manufacturability criteria" that is, the amp performance should be reproducible without major adjustments (this is a major drawback of the whole EC concept) and accomodate an as large as possible dispersion of components. I don't have any data about the Halcro's but certainly we did not design our amp to be manufactured. It's a DIY project!


By "nested feedback loops" I mean simply a hierarchy of feedback loops. Cherry NDFL is a particular method, among others, to achieve a steeper rise in feedback gain than, say, a simple integrator below a certain frequency. If we examine what this overall loop gain roll-off actually is in your amp then this will invite alternative methods to achieve a similar effect.


Could you provide another generic topology, together with design equations, of a "hierarchy of feedback loops" with significantly different topology and having comparable performance to the Cherry's NDFL? Although apparently complex in topology, the NDFL has some pretty simple design equations. It's all about peeling the FB loops and always making sure that the 2nd order system is kept as critically dampened. There are a few fine points here (Cherry targeted a linear phase response, to the price of a slightly overshooted response, which we deemed as suboptimal) but the bottom line is that you don't even need a fancy simulator to design a basic NDFL amp. Of course, then you have to optimize and refine the design and take care of the implementation issues and that's where the fight starts.

I'm personally not convinced that NDFL and/or TMC are the only ways to achieve sub-ppm performance (THD20, CCIF IMD, SMPTE IMD, etc...), but I still have to see alternatives to such.


There is a trade-off between stability and distortion reduction in any feedback system. Feed-forward doesn't have this trade-off (but has its own challenges). The THD20 figure, for example, in the way it is normally measured, is not inclusive of sonic performance issues that may arise due to stability issues. That is my opinion. Indeed, persuing only THD20 as a goal can lead to designs that are sonically compromised by the very circuitry required to reduce the THD20. So my general question for readers of this thread is how stable is stable enough and how should it be measured?


Brian, there's nothing to debate here. I fully agree with you and if you would take a closer look on the Measurements section of our web site you will find that we are far away from preaching a THD20 number only. As above, if you can thing of any other measurements (not necessary related to distortions) we could do, just let us know. On the stability part, I think I responded above.


The stability is very much affected by what load you attach to the amp because the load will be reactive. I'm sure you know all about this. The magnitude and phase of a speaker can vary hugely and can become very capacitive at certain frequencies and in certain dynamic conditions. Speaker cables behave like transmission lines and introduce significant reactive effects at certain frequencies (Nelson Pass wrote a paper on this BTW). OS distortion depends on output current and output voltage and both depend on load impedance.


I have responded to this in another post. I have to admit again my ignorance by failing to understand where is the transmission line in an average speaker cable. On the reactive loads, you are right and I think I responded above as well.

Measuring the performance in a reactive load, simulating a real speaker, is not a big technical problem. The problem is which speaker are you chosing to emulate? To the extend I'm aware of (I would be happy to be proven wrong on this) there isn't any standard reactive load for determining an audio amp response.
Click to expand...
 
#2,643
traderbam said:
Ovidiu, check out post #16 in this thread: high Z output amps vs low Z
Click to expand...

Brian,

I'm well aware of Nelson's fine article on speakers cables. However, I fail to understand the relevance in this context. We all know cables have a distributed capacitance and inductance. And that there are good cables and poor cables. But where does the transmission line, standing waves, characteristic impedance, etc... fit into this discussion?

If you would do me a favor in reading again Nelson's page 5 I think you will find most of the answers.

Also check these articles. Please note that I do not want to divert the discussion toward a cable debate.

http://www.audioholics.com/education/cables/debunking-the-myth-of-speaker-cable-resonance

http://www.audioholics.com/reviews/cables/speaker-cable-reviews-faceoff-2

http://www.audioholics.com/reviews/cables/diy-speaker-cable-faceoff
 
#2,644
dimitri said:



Edmond, they DO behave like trasmission lines starting from hundreds of kHz. Go measure.
Click to expand...

This is a nonsense, Dimitrij.

You can think about frequency independent Zc when, according to

Zc = sqrt((R + jwL)/(G + jwC))

you fulfill condition R/G = L/C. Then Zc is frequency independent. This is a different story than reflections, the reflections depend only on wave length compared to length of cable, and relative permitivity, of course. Go measure.
 
#2,645
Set aside the audibility of a (say) 100KHz wave, such a wave has a wavelength of 3Km, or about 1.9 miles. This is about 500-1000 times the length of an average speaker cable.
Click to expand...

The 10 feet cable becomes quarter wave resonator between 13-18 MHz, at the amplifier end this measures as a short circuit. The loudspeaker impedance usually has a peak at hundreds of kHz and dip at MHz range. These will be converted by transmission line to even higher peak and deeper dip. These frequencies are well within unity loop gain frequency.
 
#2,646
Ovidiu wrote:
Also check these articles. Please note that I do not want to divert the discussion toward a cable debate.
Click to expand...
In one of your articles a mfr talks about using ferrite beads to reduce the effects of cable refelctions at HF.
I agree, this is not the place for going over this again so shall we park it here? 🙂

BTW a certain well-known and well respected mfr of power amps specifies limits on characteristic Z of compatible cable and also specifies a minimum length...about 3m I think. I have read owner's reports of serious problems when using out of spec cables (mains light flicker and excess heat). So this is a real life consideration especially when using high levels of feedback and very fast circuits.
 
#2,647
dimitri said:

The 10 feet cable becomes quarter wave resonator between 13-18 MHz, at the amplifier end this measures as a short circuit. The loudspeaker impedance usually has a peak at hundreds of kHz and dip at MHz range. These will be converted by transmission line to even higher peak and deeper dip. These frequencies are well within unity loop gain frequency.
Click to expand...

Your numbers are entirely correct. Now, could you expand on the impact of a 13-18MHz quarter wave resonance on the audio amp performance?

Thanks again for your input!
 
#2,648
syn08 said:
Set aside the audibility of a (say) 100KHz wave, such a wave has a wavelength of 3Km, or about 1.9 miles. This is about 500-1000 times the length of an average speaker cable.
Click to expand...

Are you assuming the speed of light in that calculation?

I wrote an article in 1980 on speaker cables in which I showed
that a 10 ft sample of unterminated speaker cable has a 1/4
wave resonance anywhere between about 1 megaHertz to
10 megaHertz.

http://www.passlabs.com/articles.htm
 
#2,649
Nelson Pass said:


Are you assuming the speed of light in that calculation?

I wrote an article in 1980 on speaker cables in which I showed
that a 10 ft sample of unterminated speaker cable has a 1/4
wave resonance anywhere between about 1 megaHertz to
10 megaHertz.

http://www.passlabs.com/articles.htm
Click to expand...


Yes, I considered the speed of light. The EM wave propagation velocity in a dielectric is inverse proportional to the square of the relative dielectric constant of the material between the two conductors. Assuming a dielectric constant of 4, the equivalent wavelength would be half of what I have mentioned.

I don't know though how this can be reconciliated with an unterminated quarter wave resonance of 1MHz, unless the dielectric is some barium-strontium ceramic with ultra high dielectric constant 🙂 If you can point me to a source of such speaker cable, I would be happy to look myself into. It could have some interesting applications in building audio delay lines!

OTOH, 10MHz resonance looks possible for 10 foot of the crappiest speaker cable available at RadioShack 🙂 Now, what's left is to understand how a quarter wave resonance at 10MHz is affecting the audio performance.
 
#2,650
Nelson Pass said:
Are you assuming the speed of light in that calculation?

I wrote an article in 1980 on speaker cables in which I showed
that a 10 ft sample of unterminated speaker cable has a 1/4
wave resonance anywhere between about 1 megaHertz to
10 megaHertz.

http://www.passlabs.com/articles.htm
Click to expand...

Hi Nelson,

Thanks for your input, and I agree with you that we should reckon with the 1/4 wave resonance (btw, I've down loaded your article and I can ensure you that I'll have close look at it). BUT this thread is not the most appropriate place to start a discussion on that subject. Besides, if a 1/4 wave quirk might cause problems (if any!), we still have our honorable friend Zobel to send this 1/4 wave dwarf back to his own empire.

Cheers, Edmond.

PS: In the light of this discussion, does the difference of speed in vacuum vs cables really matter???
 
#2,652
traderbam said:
Ovidiu, in response to your ealier post.
This particular thread is entitled "Bob Cordell Interview: Error Correction" and the discussion in this lengthy thread has revolved around HEC and Bob's implementation of it in his amplifer. So, in this thread, I have been trying to encourage you and Edmond to make comparisons with Bob's circuit because I think this is pertinent to this thread and may be of wider interest. By highlighting the differences, both advantages and disadvantages, more learning may be achieved. Therefore, to compare meaningfully I have proposed more parameters than just THD20 and suggested you and Bob specify them.
I have already talked (argued for some reason) with Edmond about stability. I have asked how you two determined acceptable stability in your design and you haven't answered me. That's fine, you don't have to.
Edmond has been very bullish in this thread about his use of NDFL and TMC as the only means to get 1ppm THD20, saying that Bob wouldn't be able to achieve this figure without using those methods. That is a bold challenge. Based on Halcro's US patent #5892398 I don't see Halcro using either method.
By "nested feedback loops" I mean simply a hierarchy of feedback loops. Cherry NDFL is a particular method, among others, to achieve a steeper rise in feedback gain than, say, a simple integrator below a certain frequency. If we examine what this overall loop gain roll-off actually is in your amp then this will invite alternative methods to achieve a similar effect.
There is a trade-off between stability and distortion reduction in any feedback system. Feed-forward doesn't have this trade-off (but has its own challenges). The THD20 figure, for example, in the way it is normally measured, is not inclusive of sonic performance issues that may arise due to stability issues. That is my opinion. Indeed, persuing only THD20 as a goal can lead to designs that are sonically compromised by the very circuitry required to reduce the THD20. So my general question for readers of this thread is how stable is stable enough and how should it be measured?
The stability is very much affected by what load you attach to the amp because the load will be reactive. I'm sure you know all about this. The magnitude and phase of a speaker can vary hugely and can become very capacitive at certain frequencies and in certain dynamic conditions. Speaker cables behave like transmission lines and introduce significant reactive effects at certain frequencies (Nelson Pass wrote a paper on this BTW). OS distortion depends on output current and output voltage and both depend on load impedance.
Brian
Click to expand...


Hi Brian,

This is a pretty good summary here, and I generally agree with it. Comparing different approaches is one of the very best ways for people to learn, and I believe it definitely can be done in a non-confrontational way. There will always be some disagreements about what things need to be compared and how those comparisons can be made fair, but there is also a lot of common ground that can be covered.

The key is that none of us gets defensive when asked a question in a reasonable way, or when someone asserts a different way of doing something in a way that shows respect for the other alternatives.

I can get addicted when a good challenge is put forward. A competition usually has only one winner, while meeting a challenge can have many winners.

Cheers,
Bob
 
#2,653
Nelson Pass said:


Are you assuming the speed of light in that calculation?

I wrote an article in 1980 on speaker cables in which I showed
that a 10 ft sample of unterminated speaker cable has a 1/4
wave resonance anywhere between about 1 megaHertz to
10 megaHertz.

http://www.passlabs.com/articles.htm
Click to expand...


Yes indeed. This brings to mind the approach that some have used of placing a series R-C zobel across the speaker terminals, presumably to effectively and properly terminate the speaker cable at high frequencies.

Some loudspeakers are inductive enough to look like nearly an open-circuit at high frequencies. This will often depend on the signal path used to the tweeter and where and how any L pads are used in that path.

I'm guessing that in those cases where the Zobel truly made a sonic difference, the amplifier was engaging in some stability misbehavior in the absence of the Zobel.

Cheers,
Bob
 
#2,654
Bob Cordell said:
Yes indeed.
Click to expand...

Should I understand that you agree it is possible to get a 1MHz quarter wave resonance from a 10ft. piece of speaker cable? If so, I would love to hear your opinion on what is the EM transmission line model that would support such a resonance.

I have in my basement a vintage speaker used early in the 20th century to distribute, from a central station, radio broadcasting across rural areas. I have checked and it has a 0.5W speaker, a transformer and guess what's across the primary? A series R-C zobel, most likely for
effectively and properly terminate the speaker cable [My comment: thousands ft. long] at high frequencies.
Click to expand...
 
#2,655
Bob Cordell said:



This brings to mind the approach that some have used of placing a series R-C zobel across the speaker terminals, presumably to effectively and properly terminate the speaker cable at high frequencies.

Some loudspeakers are inductive enough to look like nearly an open-circuit at high frequencies. This will often depend on the signal path used to the tweeter and where and how any L pads are used in that path.

I'm guessing that in those cases where the Zobel truly made a sonic difference, the amplifier was engaging in some stability misbehavior in the absence of the Zobel.

Cheers,
Bob
Click to expand...

Few years ago I played with resistive termination of speaker cable at speaker terminals. It was a RG-213 cable, so I used 50 ohm. This resistor brought measurable decrease of RFI HF content. IMHO it makes sense to terminate the cable. In case of R-C, beware of resonance in the capacitor, which occurs unfortunately at several MHz.
 
#2,656
totally OT

PMA said:
Few years ago I played with resistive termination of speaker cable at speaker terminals. It was a RG-213 cable, so I used 50 ohm. This resistor brought measurable decrease of RFI HF content. IMHO it makes sense to terminate the cable. In case of R-C, beware of resonance in the capacitor, which occurs unfortunately at several MHz.
Click to expand...

Hi Pavel,

Maybe it makes even more sense if every mfr of loud speaker systems put a suitable termination inside the box. But, as usual, most of them don't bother about impedance issues at all. Happily not the mfr of my boxes. 😀

Cheers, Edmond.
 
#2,657
syn08 said:
Should I understand that you agree it is possible to get a 1MHz quarter wave resonance from a 10ft. piece of speaker cable? If so, I would love to hear your opinion on what is the EM transmission line model that would support such a resonance.
Click to expand...

Yes, it's easy to observe. The speed of the wave down the
transmission line is proportional to its characteristic impedance.
For 10 ft zip cord we see 1/4 wave at maybe 10 Mhz, for lower
impedance types like Mogami coax or Polk's Litz-type wire, 1 Mhz
is more like it.

😎
 
#2,658
Originally posted by Nelson Pass
Yes, it's easy to observe. The speed of the wave down the
transmission line is proportional to its characteristic impedance.
For 10 ft zip cord we see 1/4 wave at maybe 10 Mhz, for lower
impedance types like Mogami coax or Polk's Litz-type wire, 1 Mhz
is more like it.
Click to expand...

Nelson,

I disagree. Though, this is way off topic so I would suggest to stop this debate here. If you or anybody else wants the math details on why such a resonance would lead to absurd transmission line material and/or geometry properties (PMA mentioned something above), just drop me a line.

I am not questioning your experimental results in the article, but perhaps the measurement technique and the results interpretation? What you are mentioning as a 1MHz L/4 resonance may in fact be some sort of artifact? If I would need to measure a L/4 line resonance, rather than measuring the response with a scope (as you did, apparently), which may load the line with a stray reactance, I would keep the line open, measure the input impedance and search (vs. frequency) for minima. Also, to make sure I'm not measuring some sort of artifact, I would check the periodicity of such L/4 resonances.

I have googled "Mogami coax" and got some tips on it's availability. I'll try to get some (hope it's not to expensive), do some measurements and report back in a new thread. If you have any specific information on the Mogami model# you worked with, just let me know.
 
#2,660
You must log in or register to reply here.
Top Bottom