Zaph's latest entry: non-linear distortion

[fjhuerta]

What an excelent post. Between him and Jeff Bagby they shed light on one of the strangest things I've experienced in DIY speaker building.

Specifically, the sound of a low non-linear distortion speaker.

After having grown up with JBL's, and then upgrading to other brands, and then going the DIY route, I was always intrigued as to why cymbals didn't have enough presence, or there wasn't enough "air" when I compared my own designs to commercial speakers - even when frequency response was similar.

The theory behind the post - which, in one part deals with the Fletcher-Munsen curves, if I understood correctly, and in other part explains the part non-linear distortion plays when listening to speakers- is very, very interesting.

But, in my opinion, it also means well designed speakers will never catch people's attention. Since distortion is what we are used to.

It's very tought-provoking. At least, now I understand why my designs aren't so "exciting" as commercial ones.

[Jed]

A low non-linear distortion speaker catches my attention because it sounds more like music. :W

Geddes' research was brought up when I was doing some harmonic distortion testing on my Seas DXT H1212 motor hybrid tweeter and Seas W15CY versus Accuton C79 thread. You should check that out as well.

[joecarrow]

I think that what catches my attention with a low distortion speaker (I guess I'm talking about all types) is if I get confused about whether sounds are coming from the speakers or the real world. It's one of those things that doesn't jump out at you, it just sort of creeps in here and there, and when you do notice it you just have to say, "Wow, I thought there really WAS someone else in the room".

For example, I have a boom box in my kitchen (nice for when I'm washing dishes). If I leave that on and walk away, there's no confusion that there might really be someone speaking in the kitchen. The Modula MT setup, on the other hand.... I'm sure that an even better set of speakers would just take that to the next level.

[JRT]

What an excelent post. Between him and Jeff Bagby they shed light on one of the strangest things I've experienced in DIY speaker building. Specifically, the sound of a low non-linear distortion speaker...

Geddes' research was brought up when I was doing some harmonic distortion testing on my Seas DXT H1212 motor hybrid tweeter and Seas W15CY versus Accuton C79 thread. You should check that out as well.

Got some links guys?

[tabasc07]

Here is Zaph's post http://www.zaphaudio.com/nonlinear.html

[JSG]

A minor nit-pick in an excellent article: Linear distortion does usually change the shape of a waveform. It cannot introduce new frequency components, as had been said. It can, in theory, completely remove frequency components. I suspect that given the rest of the article, this is an oversight.

John

[augerpro]

This should get interesting :E I do have an experiment everyone here can try and hear for themselves how different distortion profiles sound. The inspiration came from this article: http://www.stereophile.com/reference...ard/index.html
It's a brief on how we can correlate the distortion we hear with measurements. Mentioned is Wigan's and Geddes' formulas. Wigan's formula is easy to calculate and is better indication of audiblity than the previous THD formula used for so long, and still is. Geddes took Wigan's idea a step further and produced what is currently the most accurate way to quantify audibility of distortion IMHO. It is ironic that he later pretty much discounted non linear distortion as a factor in speaker design, which is what started this whole recent internet blow up on the subject. To be fair to Geddes though, he did say that to a designer using good drivers non-linear distortion is not important. Which of course changes the argument completely. Geddes uses high quality B&C drivers where I'm sure distortion is well controlled. He's not digging through $30 tweeters to find the gems like we are. At some level decreasing levels of non-linear distortion probably does enter an area of
diminishing returns. And if all you ever use are drivers in that top 10%, then I suppose distortion probably isn't that big of deal, compared to all the other factors.

But the rest of us mortals it probably is a big deal. But don't just take the gurus word for it. Check it out yourself! And here is a tutorial on how to that :B From Stereophile article above download AddDistortion.zip and channelutilities.zip Also download Audacity The article and help file in the AddDistortion folder lay out the process pretty well. Basically AddDistortion will add any distortion profile you can dream up to a wav file that you have made from, say, your favorite Diana Krall song. The only requirements are that it be 1) a wav 2) mono 3) less than 60 seconds long.Then you can compare the original to the distorted version and see for yourself the difference.

1)First rip the song from the disc. I used the latest Windows Media Player. Before you do this go to Tools>Options>Rip Music and make sure the format selected is Wav (uncompressed). Rip the song.
2) Now we have to take a 60 snip out of that song that we will aplly distortion to. Open Audacity and open the wav file. I'm using Shadows part 2 from the Blue Man Group's Complex cd. First go to the Edit tab and set the Snap To to On.
3) On the bottom of the screen you will see the time marker. By clicking on the timeline above the wave plot you will select first the left mark, than 60 seconds later (I used 59 just to prevent any problems later) you make the right mark. This part is intuative but hard to describe so you'll have to figure it out. When you are done you should have a shaded area that is 59 or 60 seconds long. The time is marked at the bottom, I selected the time from 1:30 to 2:29.



4) Now we want to remove everything before and after that shaded area. Click the button shown below, Trim outside selection:



5) Go to Edit>Preferances>File formats and select WAV (Microsoft 16 bit PCM) as shown below:



6) Now go to File>Export as WAV and export the file to your working folder.



7) Now open the ChannelIsolate.exe program. It's command line, so there is no browse button, you have to type in the filepath yourself where it says "Wave file to process?". I made it easy on myself by just saving everything in folder named "wavs" on the C driver so the filepath is short and easy to remember. Click enter and you will see the following screenshot:



See how it wrote two new files "Shadows part 2 left channel.wav" and "Shadows part 2 right channel.wav"? You will apply distortion to each of those files and then recombine them into a stereo file for playback.

8) Open AddDistortion. Type in the name of the file to be distorted. Since I haven't finished doing the Shadows part 2 song I'm going to use the song I feel Love from the Blue Man Group for the rest of this tutorial. You should see this screen after entering the file name:



9) Hit any key to continue. The next screen is where you add the distortion profile. On this I'm going to see if a very high F2 is audible. So I entered -10 dB. Now you input the polarity. There is some info on this in the article and the AddDistortion help file, but I'm not going to give any advice simply because I'm not sure how I can determine what to use :B Anyway, after that if you just hit enter it will fill in the rest of the lines with -250 and then begin adding the distortion. Forget about using your computer for awhile, this step is incredibly resource intensive. I did one file on my laptop, but the rest will be done on my desktop, which is running an overclocked dualcore adn has 2gig of ram. Screenshot of all this:



10) After the processing is done with right channel file do the same to the left channel file. After you are done you will see two new files with "distorted" in the filename, like so:



11) Open ChannelCombine.exe and enter the left and right channel distorted file. Hit enter you will see the following screen:



12) All done!

[Undefinition]

Brandon,
Thanks for the tip. It seems like a good test, and I am curious to try it. However, the question needs to be asked: how many dB down is the harmonic distortion on most drivers? Is it really just 10 dB down? I mean, in order to make this experiment analogous to loudspeakers, it would be helpful to know exactly how much distortion to add in order to make a fair comparison.

Thanks,
Paul

[jkrutke]

A minor nit-pick in an excellent article: Linear distortion does usually change the shape of a waveform. It cannot introduce new frequency components, as had been said. It can, in theory, completely remove frequency components. I suspect that given the rest of the article, this is an oversight.

Nope, it's not an oversight. The way I see it, the only way to change a waveform is to add or subtract another frequency. Otherwise it's just amplitude or phase which constitutes linear distortion. Well, amplitude does change the height of a waveform I guess, which is to say it changes shape in that way. It's all semantics. If Mark K weighs in on this, I'll defer to his input.

To be fair to Geddes though, he did say that to a designer using good drivers non-linear distortion is not important. Which of course changes the argument completely. Geddes uses high quality B&C drivers where I'm sure distortion is well controlled. He's not digging through $30 tweeters to find the gems like we are.

Also to be fair, I know that Geddes is a smart guy. He does like to make "shock and awe" forum posts though, which work against him. Unfortunately, his post over on DiyAudio is now frozen in eternity since edits for clarity aren't allowed. (my other blog rant) 5 years from now, people will be bringing that post up to the top, reminding him of his statement. You've got to be very careful what you say on DiyAudio. It will haunt you.

I used the AddDistortion program myself, it's pretty damn cool. It's been a few years though. Thanks for the writeup.

$30 tweeters... you big spender. I'm frequently digging through the $10 tweeter pile myself. And yeah, non-linear distortion is pretty freaking important when working with low end stuff. What quantifies as a "good driver" is completely subjective. Consider how many mass production designs (polk, infinity, etc) use $10 or less tweeters.

[jkrutke]

Straying from topic a bit, you added distortion to Blue Man Group's remake of Donna Summer's "I feel love" disco tune. Sung by Annette Strean, one of my favorite female singers. It breaks my heart to distort that voice. :B She's the singer from the band Venus Hum. Electronica isn't my favorite genre of music, but she carries the band. Critics have caller her a Bjork that can actually sing.

[fjhuerta]

I love that "I feel love" tune... it rocked in concert, too!

The Blue Man Group has the hardest-hitting subwoofer system I've ever experienced (at least on their Las Vegas theater).

[---k---]

Veering way off topic, but I was actually disappointed in Blue Man Group in Chicago. I have their Audio DVD, that I've enjoyed for a long time on my home system. It just wasn't the same live.

(Yes, I'm serious)

[Notorious_AK]

Even if speaker has low distortion, it still may have limited dynamic range, creating dull sound. That's why I like horns, very low distortion and dynamic "exciting" sound.

[jkrutke]

Even if speaker has low distortion, it still may have limited dynamic range, creating dull sound. That's why I like horns, very low distortion and dynamic "exciting" sound.

A dynamic range limitation IS distortion, and can be measured.

In the few horns I've measured, the issue has always been a ragged response curve rather than non-linear distortion. I believe that the response curve is what will typically lead to what people have called "horn sound." If Geddes were to say non-linear distortion was not an issue in horns, (rather than the broad generalization of all types of drivers) I'd probably let that one fly and agree with him.

[Dennis H]

If Geddes were to say non-linear distortion was not an issue in horns, (rather than the broad generalization of all types of drivers) I'd probably let that one fly and agree with him.

I think he says it's not a big deal in compression drivers, just like woofers. But again, he's comparing excellent drivers (B&C) to exotic drivers (TAD) and concluded the TADs aren't worth the money. I don't think he ever said you couldn't hear the difference with crummy drivers.

However a big deal in his research is the nonlinear distortions caused by the horn itself. Oldtime horns went for max SPL at the expense of distortion and his waveguides go the other way, minimizing distortion at the expense of SPL. He can get away with that because the drivers are so much better now and powerful amps are cheap.

Edit: I don't know if the horn colorations are technically nonlinear or if that's just how they sound. There's a lot of stuff going on in a horn with a long narrow throat that compresses the air heavily and a body that encourages a lot of reflections before the mouth. It may all be linear phasing effects from the reflections or there may be some modulation distortion happening. Whatever, there's something going on that's clearly audible, even if you EQ the FR flat.

[Saurav]

A dynamic range limitation IS distortion, and can be measured.

A question I asked in a different thread a few days ago...

When a transient comes along that's 20dB above the base signal level, how well does the output track the input? Does anyone test that kind of stuff? Can that behavior be inferred from the low level tests?

This is something I've always been curious about, and any insight you might have on this would be very interesting. Basically, I'm asking about linearity w.r.t. displacement/SPL, for want of a better term.

[M48scout]

Distortion audibility over narrow freq. range versus entire range

On the "Add Distortion" program - does anyone know if you can specify the level of distortion over certain frequency ranges (like occus in real drivers). This would be useful for determining the audibility of different driver's distortion profiles. (of course you would need to know the profile of the drivers you are listening on too)

I wonder if a given amount of distortion is more or less audible when applied accross the entire frequency spectrum or when it occurs only in a 'sensitive' range of frequencies only, with the remainder being very clean. Just thinking out loud.

David

[augerpro]

David that's good point. AddDistortion will not do this. I think Klippel has some similar type programs on his site. Maybe one of those do?

PS- Zaph thanks for the tip on Venus Hum. I've been on a female singers kick lately so I'll have to check that out.

[Kevin Haskins]

Dr. Earl likes to stick his neck out for sure. I think it is fair to say that often people make more of a THD number than it deserves on it's own merit. I've seen people judging a driver based upon a distortion spike well outside of its range of intended use. Lets face it.... sometimes people make a mountain of a molehill.

Also.... the makeup of that THD is just as important as the amount. Dr. Earl catches that in his metric so that in of itself is an improvement in terms of having a metric that relates to audibility.

I'd give him a break though. We have all said things on a forum that are misconstrued. It is just natural that people get defensive once confronted over a statement and being human, dig their hole deeper. The more you participate in forums the more likely it is to happen. It doesn't help that the good Dr. is very dogmatic about the entire subject.

In terms of the "sound" of low distortion. I don't find I miss the "splash" of a higher distortion system. I've always been more sensitive than most people it seems. When I'm at a live event and there is something wrong I'll have my fingers in my ears and look around at the mass of people not noticing anything askew. There is a lot of variability not just in the hearing mechanism, but in how sensitive our brains are to the stimulation.

[Undefinition]

I feel like I have a right to pitch in here, in regards to non-linear distortion of systems.
:rant:
I'm not a professional mix engineer, however I did get As in my recording tech classes in college, and I've clocked in an awful lot of hours in studios, behind mics, and behind consoles. And I have to say that it's Standard Practice, when mixing a recording, to listen to it on a variety of systems... everything from the 5-figure system in the control room, to your car, to your grandma's radio on her porch. As you listen to the mix on all these systems, you learn what is important in the mix--what stands out, and what shouldn't. The final mix of a project is almost always a compromise of what sounds best on all those systems. (Many times I have mixed what I thought was a "perfect kick drum" sound in a control room, only to have it be practically inaudible on other systems I listen to the song on)

hence, it's a safe bet to say that the vast majority of recorded music was mixed with harmonic distortion of a system in mind--even if it was subconscious. Therefore, THAT is why it sounds "dull" on drivers with ultra-low distortion, and "splashy" on the majority of other systems. IT VERY WELL MIGHT HAVE BEEN MEANT TO SOUND THAT WAY!!! That is the way the musicians and the mix engineer (and the mastering engineer) wanted the song to sound.

Now yes, there are definitely some recordings out there that were recorded and mixed with high fidelity in mind, but recordings like that are (as far as I know) definitely the minority, when compared to the rest of commercial music available.

So maybe the only real solution is to build two different speakers:
1. A pair using ultra-low distortion drivers, on which you can truly enjoy Cowboy Junkies "Trinity Sessions" and Diana Krall "Live in Paris"
2. The other pair you listen to everything else.

-Paul Carmody

[h-bar]

I hope I haven't misunderstood the intent of jkrutke's comment, but I thought I'd chime in on the issue of waveform shape.

Nope, it's not an oversight. The way I see it, the only way to change a waveform is to add or subtract another frequency. Otherwise it's just amplitude or phase which constitutes linear distortion. Well, amplitude does change the height of a waveform I guess, which is to say it changes shape in that way. It's all semantics. If Mark K weighs in on this, I'll defer to his input.

If "waveform" is taken to include things other than pure sine waves, then linear distortion can change the waveform's shape. Consider starting with a 500 Hz square wave, which can be thought of as the sum of a 500 Hz sine wave and odd harmonics of that frequency (1500 Hz, 2500 Hz, 3500 Hz...). Each sine wave component must be present with a particular amplitude (the kth harmonic should have an amplitude proportional to 1/k) and phase in order for their sum to yield the square wave. If the relative amplitude or phase of one or more components is changed, the resulting waveform will no longer be a square wave. No new frequencies have to be added in order to "unsquare" the square wave.

Below I've added a few graphs, snagged from here, showing what happens if you only use lower harmonics when trying to construct a square wave. The more of the harmonics you include, the better the approximation. In these examples the harmonics are either included with the correct relative amplitude, or eliminated completely. Similar effects could be achieved by using the wrong amount of one or more harmonic---in other words, by using a device with non-flat frequency response.

h-bar

[Mark K]

Hi h bar,

Your right. I think everyone just has to be careful in their wording.

"changing waveform shape" is somewhat nebulous. Both linear and nonlinear distortion do that. The difference is whether new frequencies are added.

It helps to think of breaking the complex wave into it's expansion series. If you change the amplitude or phase terms only, that's linear distortion.

If you add new terms to the expansion, that's nonlinear distortion.

But both change the shape of a complex waveform, say, if you viewed it in realtime in the time domain (i.e. an oscilloscope etc.)

[jkrutke]

Hi Mark, thanks for weighing in. Yeah, it's all in the wording. I've actually seen it argued somewhere that there is no such thing as linear distortion. I wish I could remember where, but the guy made good points. But among us here, I'm pretty sure we all know what we're referring to when we mention linear distortion.

[jquin]

Hi h bar,

Your right. I think everyone just has to be careful in their wording.

"changing waveform shape" is somewhat nebulous. Both linear and nonlinear distortion do that. The difference is whether new frequencies are added.

It helps to think of breaking the complex wave into it's expansion series. If you change the amplitude or phase terms only, that's linear distortion.

If you add new terms to the expansion, that's nonlinear distortion.

But both change the shape of a complex waveform, say, if you viewed it in realtime in the time domain (i.e. an oscilloscope etc.)

At the risk of being annoying, won't real musical waveforms contain virtually every frequency within the Audio Bandwidth, hence it will acutally be impossible to add a new frequency, it will already be there.

I'm new to this linear distortion stuff, it all sounds like "flux capacitors" and "deafening silence" to me.

[JRT]

At the risk of being annoying, won't real musical waveforms contain virtually every frequency within the Audio Bandwidth, hence it will acutally be impossible to add a new frequency, it will already be there.

I'm new to this linear distortion stuff, it all sounds like "flux capacitors" and "deafening silence" to me.

Different musical instruments can play notes with the same fundamental frequency, but sound different because they do so with much different timbre, much different spectra. Consider that just changing the strings on the same guitar from steel to nylon will make it sound very different when playing the same notes.

Pollute that spectra with additional content, with nonlinear distortion, and you change the timbre of the sound your are listening to. Of course this is all about perception, which depends on what and how much is added.

[edited to add the following]

I think I may have misinterpreted the point of your comment. I was referring to the perceived effect. I now think you were merely commenting that the frequency content is already there and that nonlinear distortion only increases some of it, doesn't inlcude something new that wasn't already there. Thats not really the case.

Adding nonlinearity is just that, adding some new signal content above the noise floor in the spectra, interfereing only with the specific frequencies that match the newly added frequencies. Linear distortion is not adding new content, but is more like stretching or bending the envelope of existing content rather than adding something new.