I know what you were getting at but, yikes, that's pretty broad brush (and therefore not universally applicable) statement.
Distortion reduction is very important if echo cancellation is required in the system. This could be for a PA system going out on simulcast, or a group audio room conferencing system, or even a basic telephone.
The reason is that echo cancelers use linear modeling techniques to model the echo path from speaker to mic, and therefore can not model non-linear effects in this echo path. If the distortion is high enough, and the end-end delay is long enough, the distortion becomes audible as echo.
We measured distortion for all our telecom application speakers in a production environment for this reason.
Dave Dal Farra
You are kidding of course! Either that or you are young. People don't want the truth. They want you to reinforce what they already believe. Give them what they want and you will get no objections. Challenge their beliefs and you are in for a fight.
I turned 60 today - that makes me old and wise.
Just because we're DIY doesn't mean we have to be Luddites. Some of us actually want to apply the science but still respect that others don't. Can we get back to talking about why speakers may not sound flat even if they "measure" flat?
Dave
Dave - you get all tied up in your background. Linearity in an echo canceller is critical, but not because its audible. I learned this back in 1994 when doing Active Noise Control. I am surprised that you would not understand this.
Adaptive cancellation is based on linear theory and it fails to converge when the "plant" is not linear. But this in no way indicates that such a nonlinearity is audible. In fact, in case you guys haven't figured it out yet, the convergence algorithm will spend its "degree's of freedom" on linearizing the "plant" first, before it can even begin to find a model of the echo.
LOL, I think we've discussed this before. I disagree. In the signal processing world, 90 deg = decorrelated. The fact that it's in phase quadrature doesn't change this.
All I can say is that I have a number of patents in this area and spent years tweaking these algorithms and working on the industry specs, along with the auditory psychologists that ran the DBTs to develop these specs. You can throw all the taps you want at it, it can't model the non-linearity. Even with a perfect cancellation of the linear transfer function, the distortion generated in the echo path comes right through. If the distortion level is high enough, and the delay is long enough, it's audible as echo. Or was the whole ITU-T wrong on this one? ;-) Or our considerable testing that led to:
US Patent 6201873 - Page 1 Image - Loudspeaker-dependent audio compression
Dynamic optimization of handsfree microphone gain - Patent 5796818
I find it funny that you will often defend your position with your experience, but I'm "too caught up in my background". For everything else, there's Mastercard. 🙂
We're getting way off topic again, so I'll drop this line of discussion but... All I'm saying to you Earl is that a little perspective would help your arguments go a long way.
Thanks,
Dave
Last edited:
No I don't. Your turn. 😉
Edit. I did a 2 minute google search, and lo and behold. David Griesinger needs to hit his books too? (LOL)
http://www.davidgriesinger.com/note_to_sengpiel.txt
Last edited:
Happy Birthday.
You may be wise but no I'm not kidding! 🙂 Of course most would often prefer that they not be proven wrong but most people do value "truth" very highly, in fact I believe this is almost universal.
In addition, many will react positively when presented with the truth in a respectful fashion. I think you may be confusing the reasons for a negative reaction to "the truth" if it is presented with a lack of respect. Not so many textbooks on the subject however. 😉
Happy Birthday.
You may be wise but no I'm not kidding! 🙂 Of course most would often prefer that they not be proven wrong but most people do value "truth" very highly, in fact I believe this is almost universal.
There is cognitive dissonance. I think this especially applies to audio and especially to this conversation:
Too Smart And Informed To Be Suckered? Think Again
I was most amused by: "People who see through flim-flam are self-critical". That sentence just lays there like so much raw fish.
Dave
Very nice... I think I should write a letter to my conservative MP. 🙂
so, 2 methods to resolve the dissonance:
- study and perhaps change of attitudes, beliefs, and then of course I would take some action. http://en.wikipedia.org/wiki/Cognitive_dissonance#cite_note-Festinger1957-1
or
- reduced by justifying, blaming, and denying
I think the reason most of us are here because we do hold some conflicting information on the mysteries of great audio.... and of course I aim to resolve multiple sources of conflicting information (I haven't made any changes to my system for some time now).
Last edited:
Dave, Earle, You guys are losing us. I want to know about decorrelation as it applies to a home listening room.
My understanding is primitive.. I've been reading Kendall
http://www.garykendall.net/papers/Decorrelation1995.pdf
and Usher
http://www.jar-lab.com/papers/PhD/Usher_PhD_twosided.pdf
It's going to be a while before I get it under my belt, but it's pretty clear that added reverberance to a small listening space is more effective the more decorrelated it is from the direct signal.
So what's the the best way to get to this in the DIY home listening world? Starting quick and dirty is always good. One can always get refined later.
Is there existing software one can "pervert"?
BTW, is it the stuff p.2 and ff in Kendall's paper you are arguing about? edit: or are you talking past each other?
My understanding is primitive.. I've been reading Kendall
http://www.garykendall.net/papers/Decorrelation1995.pdf
and Usher
http://www.jar-lab.com/papers/PhD/Usher_PhD_twosided.pdf
It's going to be a while before I get it under my belt, but it's pretty clear that added reverberance to a small listening space is more effective the more decorrelated it is from the direct signal.
So what's the the best way to get to this in the DIY home listening world? Starting quick and dirty is always good. One can always get refined later.
Is there existing software one can "pervert"?
BTW, is it the stuff p.2 and ff in Kendall's paper you are arguing about? edit: or are you talking past each other?
some would just prefer to hear good sound
oh yeah, some of us would really like to see the science applied:
http://www.diyaudio.com/forums/mult...nce-back-into-loudspeakers-4.html#post1572845
but some other - those with their minds made up on the issue - just want to apply what they have in their minds and present it as a state of the art science
I tell them - go to church and preach because that's obviously Your vocation
and there you can even name the dissenters monkeys and dogs and be applauded for this
but I have always thought that this is not how one has to behave on a public discussion forum...
Thanks for clarifying what the black and red lines on the graph represent. In my defence with neither labelled it was rather ambiguous as to what they were, especially with an unfamiliar design.
Since the power response from ~600Hz - ~7Khz is nearly flat on this design, the DI curve will largely mirror the measured frequency response in this region, and could easily be confused with a plot of the tweeter when unlabelled.
Since my comments were made after misinterpreting the graphs, I take back most of what I said about the design, with the exception of my concerns relating to the extreme variations between 2-6Khz over a very small angular range centred on the on axis response, and the suggestion that designing for a flat response at a small arbitrary angle whilst allowing an extremely non-flat on axis response just a few degrees away is good design practice.
As much as I am on the side of constant directivity designs in general, (and believe controlled directivity designs are superior to extremely wide / omni designs) I'd rather see one with a much more uniform response within the constant directivity lobe, especially at and near on-axis. But again, if it's a sound reinforcement / Pro Audio design rather than an Audiophile design then it may be deemed an acceptable tradeoff.
This is how Kendall arrives at correlation numbers between -1 and +1:
An externally hosted image should be here but it was not working when we last tested it.
How is this not voicing ? Just because you might be implementing a "generally accepted curve" by measurement only it's still voicing in the sense that all the people who "agreed" on that curve in the first place must have done so because they thought it sounded right.
In other words they admitted a deliberate deviation from flat response at the top end was required (at least for that design topology) for the best sounding result. This is the very definition of voicing - a small but deliberate shift away from a nominal flat response to give the most satisfactory subjective result, especially when the amount or type deviation required is different for different speaker designs and topologies.
How would it be any different to if you were just starting out and had never heard of such "generally accepted curves" - built a speaker that was flat at the top, but no matter what you did it just sounded a bit too bright, so you decided to finally give in and taper down the top end slightly until it sounded right, even though the measurement is now "worse".
The only difference is other people have already done the work of realizing that the change is necessary and worked out roughly how much is needed, and for a given similar design topology (in this case a CD horn) a similar correction curve will be "about right", and could perhaps be applied without listening.
Try applying that same curve to a vastly different type of driver with vastly different polar response and you could quickly run into trouble.
Sure, let me clarify.
You're correct that any normal passive design is highly unlikely to achieve this level of flatness, certainly not narrow band, if only for reasons of baffle diffraction let alone driver imperfections, and I wasn't suggesting it would, or that it's necessary.
(With modern drivers and FIR DSP techniques that can apply very narrow band corrections you can actually build a speaker that is flat on axis to less than 1dB, but personally I think this is a wasted effort in over optimization, and it wasn't what I was getting at)
Here's how I'm defining voicing relative to "Flat".
After correcting the response of a speaker to measure as flat as possible based on the limited number of filter coefficients typically available in a fully passive or simple analogue active filter (eg correcting for broad band errors, not the extremely narrow band corrections available with DSP) you can get a response that has uncompensated narrow band ripples that well exceed 1dB, but the overall broadband trend could be pretty flat.
Voicing is then making small deliberate changes to the response which deviate away from this "optimized by measurement" response, and I suggested that in most cases the changes are quite subtle and on the order of less than 1dB, especially broad band shelving adjustments where very small changes on the order of 1/4 to 1/2 dB are noticeable.
I should have been more clear that it's not 1dB from a perfectly flat response, because no speaker has such a response without elaborate DSP, but less than 1dB from the "previously flattest measured response obtainable from the speaker with the filter coefficients available". Does that make more sense ?
Because speakers are not perfectly flat, and some frequencies are perceptually much more important than others, it could well be that voicing is needed because not all frequency response aberrations of the same amplitude are equally audible or important.
For example even if you could design a speaker to say +/- 1dB narrow band, depending on where the errors were and whether they were narrow band or broadband would cause them to have a different perceptual effect. Some errors of this magnitude could be almost completely inaudible, while others (like a 2dB shelf) would be as plain as day, yet from a purely measurement perspective both would meet a target of +/- 1dB.
Perhaps it's a delicate balancing act to balance the various small errors in the response against each other in a way that is most perceptually correct.
At the end of the day it has to sound right, and if the "corrections" are of a significantly smaller magnitude than the basic errors in the response of the speaker, I don't see what all the fuss is about. 🙂
If you define voicing as a deviation from flat then you're right. But if you define voicing as a deviation from reference you're wrong.
Bold claims indeed. Distortion is of no consequence is what you're basically saying ? Paul Klipsch would have disagreed with you:
http://www.elpee.info/Documenten/KlipschVervorming.pdf
One of the three major claims to fame of a large well designed horn system is vanishingly low distortion, particularly IM. The others of course being directivity control and headroom / minimal dynamic range compression.
Good horns sound amazingly good for a reason, and it's not just directivity and headroom.
Sure, this is a very old article now, and the direct radiators he's comparing to are considerably more non-linear than those we have available today thanks to improvements in motor design, but isn't that kind of the point ? That there was a lot of improvement that needed to be made in distortion of direct radiators in the time between that articles publication and now?
If it was so unimportant why has so much effort been expended in the last 20-30 years in linearising drivers, particularly with improved motor linearity ?
Another one who pays attention to distortion, particularly in the midrange is Linkwitz:
Midrange distortion
And in this section he discusses his thoughts on the perceptual effects of distortion:
Frontiers
Some interesting quotes from the second link:
"I have found these and other types of multitone test signals useful for sorting and relative ranking of drivers. Their subjective evaluation when part of a complete speaker correlates closely with the measurements. It should be understood that distortion has audible effects long before it is perceived as distortion. For example, it might enhance sonic detail, analogous to contrast enhancement in a photo."
I couldn't agree more. Another excellent quote:
"I think a recent experience while attending a San Francisco Symphony performance gave me further insight into why we are able to tell a loudspeaker from a live instrument. Loudspeakers reproduced pre-recorded material, interspersed with the musicians' playing, during the performance of several modern pieces of music. It seemed, that coming out of the loudspeaker a sound carried a comet like trail of newly created sounds with it, that formed a continuum of background sounds, which was completely absent from the live instruments and made the speaker immediately recognizable as such. Live instruments had a space between tones, like a black background, and even en mass always remained articulate. I believe the observation had nothing to do with recorded ambience, which was present occasionally, or with hall reverberance. It reminded me of other incidences, where I had easily recognized live instruments, though there was no reflection free path between the source and my ears, and where I then struggled to describe to myself what would have been different, if a loudspeaker was playing instead. It seems to me that the ongoing-ness of sound is one of the major problems with speakers. It shows less with recordings of a single voice with small accompaniment, so often favored by audiophiles, but in complex passages of classical orchestral music and choral works."
Only high levels of distortion are evident as "distortion". Small to moderate levels of distortion which are typically present on small to medium sized direct radiator home/audiophone speakers played at moderate to loud listening levels have more subtle effects such as the "comet trails" SL describes, and a general clutter and confusion to the sound when a large number of discrete instruments are present in the recording.
I think he's spot on when he says that a live performance has clean spectral gaps between instruments whilst those gaps tend to get filled up with "crud" as soon as you introduce even moderate levels of IM distortion.
The more independent instrument specta there are within a recording, the more of a problem distortion becomes, and the more of a cluttered mess it becomes.
Another apparent effect with distortion is that up to a point, the lower distortion levels are the louder you can comfortably listen to music, even if the distortion percentage wasn't increasing with level, which it does.
At low playback levels even relatively high distortion percentages result in distortion products that are below the ambient noise level or the threshold of hearing - so no difference might be noted between a system with 1% or 2% distortion. But at much higher playback levels the distortion products come well up into the audible range and become offensive in a way that makes you want to turn the volume down, even if it's not perceived as distortion as such.
In my experience much lower than typical distortion levels mean the volume can be a lot higher before distortion products become audible and so the urge to turn the volume down isn't there.
Perhaps its "easy" to design a large system that has inaudible levels of distortion at normal playback levels, and to a large extent I agree with you that this is the case, but a lot of speaker systems aren't large. Attempt to get "large" performance from small drivers and boxes, and distortion is a very real problem.
Particularly small 2 / 2.5 way systems where IM between the bass and midrange is always a problem. Nobody in their right mind would claim that bass/midrange IM in a small 2 way system is not a problem well within comfortable listening levels.
Longer throw more linear drivers can only go so far, (and can't deal with FM distortion) and the constant attempts to miniaturize speaker designs is causing designers to butt their heads well and truly against audible distortion, despite good drivers being radically more linear than they were 30-40 years ago. Basically people are just expecting too much out of small drivers.
I'm of the school of thought that says don't try to expect miracles out of small drivers, if you want good performance build a large speaker (with good drivers) that has plenty of head room (at all frequencies) and distortion will pretty much take care of itself.
I guess that is what you're also trying to say, but you're over generalizing it to the point you're saying distortion doesn't matter.
I can't even begin to fathom how you can suggest that distortion in speakers would have to be as high as 15-20% to be audible, and yet .01% in an amplifier would be...if you have any research papers that could back this up I'd love to see them, otherwise I think that's just an opinion of yours.
Last edited:
- Status
- Not open for further replies.