[diy_freak]

What kind of component tolerance do you see as necessary for loudspeaker filters? This question was triggered by a short discussion at one of the manufacturers circles where I was kindly asked to leave. Seems fair in the current settings of that part of the forum. I won't provide a link to the thread to avoid stirring things up.

There was mention of tweaking a crossover by ear with increments of 0.01uF on a base value of 100uF. That's a 'tolerance' of 1/100% I thought that was stretching it beyond belief, so I prepared a simple simulation to show the effects of a mere 2% (already considered a tight tolerance) in steps of 0.5%, seen on a 10uF first order highpass on a tweeter:



Not much to see here. Zooming in at around 1K, at about 15dB below the tweeter's passband spl, where the minute differences start to show:


So it seems that 0.5% difference in the component value leads to a deviation of about 0.04dB. Mind you, this is a simple first order filter. Things are different for higher order filters, that have a tendency to react more violently to component variances. The worst case scenario would probably be a sharp (high Q) notch filter to counter the effect of say the breakup of a metal cone.

There was another interesting notion about constant directivity designs and component tolerance. My questions was: "Why would a design being CD influence this behaviour?", with the behaviour being extreme reactions to components tolerances. To the best of my knowledge, CD has no influence to the variances due to component tolerances, but maybe I'm missing something...? In fact, when I think if it, a low order CD design which was discussed would probably be very resistant to tolerances because of the gentle slopes and the off-axis freq behaviour that closely follows the on-axis feq curve.

So what tolerances do you aim for? And in which applications?

Cheers,
Hans.

[Ethan Winer]

Hans,

> So what tolerances do you aim for? And in which applications? <

I'm not a speaker designer, but I agree with your empirical approach to this. If a particular component variation has an affect of 0.04 dB it is clearly irrelevant. When I (used to) design parametric equalizers and other circuits I was thrilled when I was able to find 2 percent capacitors.

--Ethan

[James Romeyn]

The answers run everywhere between one manufacturer who stops at 1/100th of one percent & a 2-way boombox at 20%.  High-end speakers probably run around 1-3%.  Wilson chooses 1/10th of one percent.  

I hope readers will forgive me for expanding on something diy mentions in the first post.  This subject was hammered to death at least a couple times already.  Probably, between ac & aa, a good 100 pages on the subject were written, very possibly more.  That is the main reason the speaker forum facilitator desired to avoid a "debate".  Diy ignored the facilitator's direct, plain & reasonable request to cease & decist.  Diy only ceased & started this thread after the police appeared w/ a warning to move on (man I pee in my pants every time I see that borg's head...)  This paints a more complete picture of the events.  I mean this as no attack, & only bring it up because diy addressed it himself first.  He could have chose to omit any mention of the earlier thread.      

For general info: What happened in the earlier "debates" is that people who believed in higher tolerances (let's say 2% for clarity) decided that was an adequate  tolerance & it was disengenuous or worse for a manufacturer to pick a lower tolerance (like a fraction of 1%) & to state that this was audibly superior.  The high tolerance fans would not accept this at face value, especially professional designers selling speakers w/ higher tolerances.  It devolved into harsh personal attacks & was very ugly.

My 2c.  Hopefully this thread will be more enlightening.  

One last thing: The manufacturer who chooses 1/100th of 1% chooses his tolerances for his speakers only.  I am almost positive he never stated that such tolerances would necessarily improve any other speakers (though I might be wrong).  The point is, unless you were in his lede soundroom, on his speakers, removing &/or adding a .001uF capacitor to the jumper leads exiting the rear of his speakers (this jumper parallels the main capacitor bank), you yourself will never ever know if you could audibly discern the difference.  You could debate it till hell freezes over & you won't know.  You could list every type of evidence & type till your hardrive stops, you still won't know if you could discern the difference, & if you would prefer the bank w/ or w/o the extra .001 capacitor.  

The lunacy in the earlier threads started when the proponents of higher tolerances stated that the lower tolerances could not possibly make an audible difference or advantage in the speakers made by the lower tolerance proponent, w/o having ever experimented to determine if their hypothesis was true or not.  That's were it became like any other religious or political argument, something that makes a lot of heat but no light.

[Brian Cheney]

Actually, the lowest value from which I can hear an audible difference is 0.01uF on 100uF of total capacitance.

I buy 200pc sacks of the Sprague "orange drop" polypropylene caps for trim, bypass, and Zobel use.  I start the trim with the average crossover value of the last 100 systems, less .1% of average capacitance value, and then approach it in increments of 0.01uF while listening.  One value yields optimum clarity and definition, and 0.01uF on either side of this value does not.

I have spent 20 years refining the series first order filter.  The last step in filter design is listening, even after measurements show no more improvement from any change.  Those who scoff at this procedure have not tried it, or lack my decades of experience doing it.

[JoshK]

diy_freak has the right to ask the question in this circle, IMO.  Its a good question, again IMO, if you ask me as long as it stays non-personal and informative.

Some may know I have owned Brian's speakers for over 5 years and recently sold them to move on to more DIY projects.  They are very good speakers, but I still don't buy into Brian's argument.  If he is convinced it is so, so be it, more power to him and his customers.  Certainly some modicum of matching is going to yield better results than none.

My problem is I fail to see the logic in how matching from one side to the other of .01uf can matter when the reactance (capacitive in this case) and mismatch of the other parts, cables and drivers themselves dwarf this value.  This belies logic to me.   The only logical explaination I can see if we assume Brian is hearing a real affect is that it is a bypassing affect and not a matching affect.  That is, could it possibly be that trimming the aggregate value of the capacitor in series with small trim caps has a nice affect on the sound, due to "speed" lack of spectral overlap, etc?  Could be, but personally I am not convinced that it is due to matching, 20 years of experience or not (no disrespect Brian).

[_scotty_]

I agree with JoshK's bypassing effect statement.  I also have heard and used this to my advantage on own DIY speakers that have had series and parallel networks. The effectiveness of bypassing is not limited to only series
networks.
Scotty

[Rick Craig]

What kind of component tolerance do you see as necessary for loudspeaker filters? This question was triggered by a short discussion at one of the manufacturers circles where I was kindly asked to leave. Seems fair in the current settings of that part of the forum. I won't provide a link to the thread to avoid stirring things up.

There was mention of tweaking a crossover by ear with increments of 0.01uF on a base value of 100uF. That's a 'tolerance' of 1/100% I thought that was stretching it beyond bel ...

Studies have shown that .2db differences tend to be the audible threshold.

[Brian Cheney]

It is not a bypassing effect, since I usually replace the numerous small values with one large one after trimming.  The influence on the sound of the system is the same.

The trimming procedure actually compensates for all the driver, wiring, and parts differences, which is why no one value is correct even for matched parts and drivers.  I find the differences run in the -.05% to 0.15% range from system to system.

The trimming has to be done with each pair of speakers after 50 hrs breakin.  It's a lot of work, but rewarding.

[JoshK]

It is not a bypassing effect, since I usually replace the numerous small values with one large one after trimming.  The influence on the sound of the system is the same.

With all due respect, that statement doesn't refute that it could be a bypassing affect.

[Brian Cheney]

If I replace ten 0.01uF caps with one .1uF cap, and the sound does not change, that is adequate proof in my mind that I am dealing with a trim effect and not bypass.  You're welcome to your opinion, of course.

[diy_freak]

Actually, the lowest value from which I can hear an audible difference is 0.01uF on 100uF of total capacitance.

I buy 200pc sacks of the Sprague "orange drop" polypropylene caps for trim, bypass, and Zobel use.  I start the trim with the average crossover value of the last 100 systems, less .1% of average capacitance value, and then approach it in increments of 0.01uF while listening.  One value yields optimum clarity and definition, and 0.01uF on either side of this value does not.

Can you describe how you do the tweaking and what the conditions are like room (acoustics), audio equipment, loudspeaker setup, listening position, listening levels, source (music or test tones) and others? Which frequency range are (mostly) influenced by the adjustments? Do you verify the results with measurements?

When you find an optimum, how do you replace the 0.01uF trimcaps with one cap? I would think the tolerance of the single replacing cap (if you can find the exact value somehow) would mean you would have to start all over again?

[diy_freak]

Studies have shown that .2db differences tend to be the audible threshold.

What you happen to have references to any of those studies? Preferably online...  

[Rick Craig]

Studies have shown that .2db differences tend to be the audible threshold.

What you happen to have references to any of those studies? Preferably online...  

I don't know the exact AES papers but I originally read about it in the Audio Critic. You may want to do a search under "ABX" testing because that's why they match levels to .1db so that a level difference won't skew the blind test results.

Having taken part in some blind tests I would say it really changes the way you think about what things are really important in audio. I would challenge anyone who claims that a .01 mfd bypass cap or .01 mfd change makes an audible difference to do a blind test. Your ears may not be as golden as you think  

[JoshK]

If I replace ten 0.01uF caps with one .1uF cap, and the sound does not change, that is adequate proof in my mind that I am dealing with a trim effect and not bypass.  

There enlies the crux of our disagreement.  But anyway...  then what happens when you change the pot a little?

[Ethan Winer]

Jim,

> unless you were in his lede soundroom, on his speakers, removing &/or adding a .001uF capacitor ... you yourself will never ever know if you could audibly discern the difference. <

That's not really true, and it directly relates to Rick's comment:

> Studies have shown that .2db differences tend to be the audible threshold. <

Forget studies for the moment. As a professional recording engineer I know that 0.2 dB is about the smallest change one could discern in either overall level or changes in frequency response. And this is at the most sensitive midrange frequencies. At low and high frequencies (let's say < 200 Hz and > 5 KHz) we can't hear changes that small.

But assume that folks with really good ears, in an ideal listening environment, really can hear 0.2 dB at all frequencies. I didn't check DIY's math but let's accept his 0.04 dB conclusion for now. A change of 0.04 dB is not audible by anyone in any situation.

So now the question shifts to "Why do people sometimes think they hear a change even when none could possibly exist?" I believe this is the real issue, and it has nothing to do with adding tiny bypass caps, or auditioning speaker cables and replacement power cords, or demagnetizing audio CDs, etc.

--Ethan

[Brian Cheney]

As a former professional recording engineer, I agree with you.  Please maintain your line of thinking.  I certainly don't want anyone else doing what I do, and it's comforting to think diy-ers and my competitors in the speaker business know for sure what's audible to others, and what's measurable, and what measurements exactly correlate to the listening experience.

You're also right about the audibility of absolute polarity, speaker breakin, 0.2dB level changes,  listening quality of amplifiers, double blind testing, interconnects and power cords, speaker THD under 5%, supersonic and subsonic frequency response, you name it!

I'm serious,  as I sit here, admiring my two Best of CES awards for High End audio (2002,2003).  It's all delusion.

If you're not an AES member, please join.  Only mainstream thinking allowed there, and I speak as a 15 year elected Secretary of the SF section.

[diy_freak]

You're also right about the audibility of absolute polarity, speaker breakin, 0.2dB level changes, listening quality of amplifiers, double blind testing, interconnects and power cords, speaker THD under 5%, supersonic and subsonic frequency response, you name it!

I'm serious, as I sit here, admiring my two Best of CES awards for High End audio (2002,2003). It's all delusion.

If you're not an AES member, please join. Only mainstream thinking allowed there, and I speak as a 15 year elected Secretary of the SF section.

Many of the things you mention in this post are audible, or at least under certain circumstances. They're also quite different from the topic we are discussing here.

The CES is a fun consumer show to help promote the industry. For those who don't know what the AES is: it's an institute that encouraged the research in acoustics with renowned scientists and engineers like Small, Thiele, Klippel, Toole, Olson, Klipsch, Augspurger, Baekgaard, Leach, Kates, Vanderkooy, Lipshitz, Kantor, Geddes and Linkwitz. These people don't think the earth is flat and don't have the habit of showing a closed mind. Just about all of the people named above have contributed to the science of acoustics and/or loudspeaker engineering in a groundbreaking way.

I'm fairly sure that if you ask any of these people about the audibility of differences in loudness, none of them will mention numbers lower than 0.1dB, not even under lab conditions. Dr. Klippel is the world's foremost authority on measurements and ow to interprete them and would be the first person one would want to ask, which I did a few minutes ago via the Klippel site. Hopefully the email will come through.

I did a websearch on this question and a few interesting links on this very subject popped up:
http://www.paudio.com/Pages/presentations/audibility/sld008.htm (research presented here by Moulton)
http://www.moultonlabs.com/index.php/weblog/more/yeah_i_think_i_can_hear_it (a weblog, also by Moulton)
http://www.audioholics.com/techtips/roomacoustics/HumanHearingAmplitude.php (overview of research through the years)

Since you mentioned serie filters, I'm even more convinced the 0.01uF difference won't audible because serie filters are, by design, particularly robust against components tolerances. The plots I posted which show a difference of 0.04dB are based on a difference of 0.5% (in a first order parallel filter), which is still 50 times greater than the tolerance you presented. That is why I chose the words 'stretching it beyond belief'. I'm sorry if that upset you.

If you are certain your claims are valid, why didn't you answer my challenge where I offered to buy your top-of-the-line loudspeaker in case you could prove repeatably the audibility of a 0.01% value difference on a 100uF cap in a series filter in a controlled and verified test? I would think that would make one heck of a marketing tool if you were able to prove your claim. And you would actually get paid for succeeding in it.

Apart from the tremendously small numbers we're discussing, I'm still puzzled why, as you say, a CD (constant directivity) design would be more susceptible to component tolerances than a 'conventional' design.

[Danny Richie]

Hi Guys.

I am not joining in to be argumentative but to add some meat to the discussion so please take this constructively.

I am actually going to agree with Rick and with Brian about one issue but not completely. I'll explain.

Also I have responded to questions about this before, and this old post complete with measured responses will give some insight.

http://www.audiocircle.com/index.php?topic=11132

I'll take the posts in order and respond to them that way. This one first.

Actually, the lowest value from which I can hear an audible difference is 0.01uF on 100uF of total capacitance.

There is no way one can detect, hear, or even measure a difference in response such as this in the value of a cap. In other words, if the cap value changes by this much you will NOT hear the difference. See my link above. Even a .1uF change is hardly measurable.

However, by-passing is a completely different issue and I have no doubt the Brian can hear the difference in adding a small value by-pass cap to a 100uF cap.

Here is a brief explanation on the effects of by-passing (generalizing a little for clarity).

Caps are actually an energy storage device. They can store and hold a charge. The larger the cap value the more energy it can store. As an audio signal passes through the cap it actually stores then releases it, and there is time involved in this. It is a charge and discharge rate.

Large cap values charge and discharge more slowly than a small cap. So small caps have a much faster dissipation rate. Some caps don't discharge all of its energy right away either. After discharging there can still be some residual charge left.

What all of this does is cause a smearing effect in the sound.

By-passing a cap with a vary small value cap can have a vary positive effect on the sound. The small cap has vary little lag time. Charge or discharge is nearly instantaneous. So when you by-pass a larger cap value with the really small one, the small cap shorts the larger one and discharges it much more quickly, and it will improve the sound qualities.

Higher quality caps and by-pass caps can make a considerable difference as well. Not all by-pass caps are the same.

My problem is I fail to see the logic in how matching from one side to the other of .01uf can matter when the reactance (capacitive in this case) and mismatch of the other parts, cables and drivers themselves dwarf this value.

This is correct. Driver tolerances are much larger in the variances.

Studies have shown that .2db differences tend to be the audible threshold.

I have to agree with Rick here, but only in relation to a change in output due to an increase in input. In other words .2db louder across the board.

For a speaker to have a dip or peak in one area it will take quite a bit larger change for it to be noticed with music playback.

The trimming procedure actually compensates for all the driver, wiring, and parts differences, which is why no one value is correct even for matched parts and drivers.

I am sorry Brian, but trimming a cap value one way or another will NOT compensate for differences in the drivers or other parts.

If I replace ten 0.01uF caps with one .1uF cap, and the sound does not change, that is adequate proof in my mind that I am dealing with a trim effect and not bypass.

While you may not be able to hear a difference in a handful of .01 by-pass caps verses a single .1uF by-pass cap, there is still a really big difference in the sound of using any small by-pass caps verses not using any by-pass caps on a large value cap. Unquestionably, the effect of adding a by-pass cap makes a difference in the sound, but NOT because the value has changed.

Josh makes a really good point here.

There enlies the crux of our disagreement. But anyway... then what happens when you change the pot a little?

A variable pot placed on each driver may be fun to play with, but the possibility of adjusting them on each speaker so that they are exactly the same and to do so without measuring equipment is an impossibility. Even a quarter of an ohm difference in adjustment will make a far greater difference in output level than what these small cap variances can make.

[/quote]Since you mentioned serie filters, I'm even more convinced the 0.01uF difference won't audible because serie filters are, by design, particularly robust against components tolerances. [/quote]

This is vary true. I have used a lot of series networks myself. If you change a cap value, for instance, from one value to the other, it can take away from one driver and at the same time give to another. It can shift the balance of how much of the signal goes to each driver and not effect the overall response.

[Brian Cheney]

While I disagree with Danny on virtually every point he makes, I see no reason to continue this discussion, which is becoming circular.

[Danny Richie]

I don't mind if someone disagrees with me, but Brian can you please substantiate your position a little better for us?

No need to take all your marbles and go home.

I think a nice civil discussion can be vary educational.

If your disagreement is on every point then please present your views in regards to each point. I am sure that we would all like to know what it is that would allow you to think otherwise.