or Connect
AVS › AVS Forum › Audio › DIY Speakers and Subs › CEDIA 2010 and Distortion
New Posts  All Forums:Forum Nav:

CEDIA 2010 and Distortion - Page 2

post #31 of 68
Quote:
Originally Posted by maxmercy View Post

If I knew how to add the harmonics without reducing the level of the signal, I would, but I just found the plug in, and it does add the harmonics to the levels specified below the fundamental...

If you are interested, Keith Howard has written a freeware utility to apply a given pattern of distortion to a mono .wav file here: http://www.audiosignal.co.uk/Resourc...distortion.zip

It is limited in the length of signal that you can manipulate and it is restricted to a mono file only. I have some equivalent programs of my own that will work with stereo files that I might be persuaded to share. Bear in mind that for a given specified distortion pattern, the stated distortion % is only achieved for signals that are full scale. At lower amplitudes, the distortion level generally reduces (not always true). This is really rather similar to what happens in amp clipping. Low order distortion of music signals which have large dynamics mean that only the peaks are really distorted. This is another reason why hearing distortion in music signals can be harder.
post #32 of 68
Thread Starter 
Quote:
Originally Posted by cjwhitehouse View Post

Bear in mind that for a given specified distortion pattern, the stated distortion % is only achieved for signals that are full scale. At lower amplitudes, the distortion level generally reduces (not always true).

I have found a way to level-match in Audacity using the Amplification function, and you are correct, the distortion % is only for full-scale signals, signals at less than full scale have proportionally less distortion. I'll post up some more files in the coming days....

JSS
post #33 of 68
Using the program cjwhitehouse linked to this is what I came up with.

I used the real THD numbers from the Ilkka SVS PB13-Ultra 10 Hz tune tests from HTS. Now HD is frequency dependant, but this program won't allow that, so it uses the same distortion percentage at all frequencies. This means the distortion should be even more apparent than real life.

As is, I used the THD% at the 30Hz mark for one wave (3%), and the numbers for the 16Hz pass (10%)for the next. The 16Hz pass bumps the CEDIA 2010 limits of the 3rd harmonic.


I ran the original wave through the distortion program at -250dB distortion levels, to account for any extra anomalies the resampling may add. I then took the original wave and split it into two files. One was lowpassed at 80Hz to simulate a sub, and the other was highpassed at 80Hz to represent a main speaker. The HP file was ran through the program at -250dB, and the LP file has the distortion level input to match the PB13. All distortion above H6 was input as -50dB. Afterwords, the samples were re-level matched, and combined back into one wave.

I have also included the original file.

http://rapidshare.com/files/321387642/Test_One.zip.html


"The Whip Theme (Extended Version) Kevin MacLeod (incompetech.com) Licensed under Creative Commons "Attribution 3.0" http://creativecommons.org/licenses/by/3.0/"
post #34 of 68
Here is the original frequency response from the Stereo mp3.


The heavy content below 100Hz is why I picked this audio file to try first.
post #35 of 68
Here are the FR graphs of the different LP files.

Passthrough



3% Distortion



10% Distortion



If you notice in the LP 10% you get a nice increase in kick drum thump.
post #36 of 68
Another thing to bear in mind when adding distortion using these programs is that with a single tone signal you will get harmonic distortion only. If you start with a signal consisting of more than one tone you will also get intermodulation products generated. With a music signal you will get a mass of such intermodulation components. These intermodulation products will be at frequencies that will tend to be discordant. Although auditory masking comes into play to some extent, I suspect that this added "grunge" accompanying a distorted music signal is one of the ways that the ear can sometimes detect quite small amounts of added distortion. The effect though is highly dependent on the specific harmonic content and dynamics of the original music signal chosen.

You might find it instructive to try putting in a signal consisting of two tones of equal amplitude, reasonably closely spaced like 100Hz and 120Hz and then plot the frequency spectrum of the distorted version to get an idea of what happens.
post #37 of 68
That is the reason I only used a lowpassed wave with the added distortion to better simulate a distorting sub.

Since IMD has been brought up,
Here are two sine waves 30 & 73Hz (No bandwidth filters in IMD files)



Here is the same wave with the PB13 16Hz THD numbers added



Here are the waves: http://rapidshare.com/files/321416981/IMD.zip.html
post #38 of 68
Here it is at 70Hz and 73Hz.

No distortion



PB13 10% Distortion added


File is here: http://rapidshare.com/files/321419510/7073.zip.html
post #39 of 68
Last one.
70 & 80Hz



70 & 80Hz with PB13 10% distortion numbers



Link to waves: http://rapidshare.com/files/321425697/7080.zip.html
post #40 of 68
Thread Starter 
Wow,

I have a lot to download and listen to...

Thanks for posting these, soho...

JSS
post #41 of 68
I'm starting to think that another level needs to be added. Say 3-30Hz at one level, 31-80Hz at another with >80Hz left alone.
post #42 of 68
Quote:
Originally Posted by soho54 View Post

I'm starting to think that another level needs to be added. Say 3-30Hz at one level, 31-80Hz at another with >80Hz left alone.

It could be done but it's a bit of a fiddle and the additional filtering to split the signal and stitch it back together afterwards may start to add other unwanted artefacts and noise.
post #43 of 68
Quote:
Originally Posted by cjwhitehouse View Post

It could be done but it's a bit of a fiddle and the additional filtering to split the signal and stitch it back together afterwards may start to add other unwanted artefacts and noise.

This is true, but the effects are minimal with the filter slopes I'm using. It is also why I put the distorted and non-distorted waves through the same filtering process. Anything the filtering adds to the distorted wave is also in the comparison wave.

This should minimize it's effect on the comparison, yes/no?
post #44 of 68
Here is a new version. This is about as real as it will get, without getting crazy.

The wave is converted to mono, and then split into three new waves using Cool Edit filters. One is filtered with a lowpass at 30Hz(LLP), the next is bandpassed at 30-80Hz(LP), and the final is highpassed at 80Hz.

I create a copy of the waves, and run them through the AddDistortion program. The lowest frequncies (LLP) get distorton that mimics the PB13 at 16Hz, the next (LP) gets the PB13s 40Hz distortion profile, and the HPed portion runs through the program with -250dB added, as well as the "clean" copies of the other two waves.

Next I check the levels, and ensure they are the same. They were.

I use Sony Vegas to splice the three newly distorted and clean sets of files back together into two versions of the original wave. Clean and Dirty.

I have included a version of the files at all levels of change as well.

http://rapidshare.com/files/32172026...ion_2.zip.html

EDIT: FWIW, the bass seems more pronounced in the Clean & Dirty waves when compared to the original due to a 2dB rise in the response right above 32Hz and at 184Hz, caused by the xovers.
post #45 of 68
Quote:
Originally Posted by soho54 View Post

This is true, but the effects are minimal with the filter slopes I'm using. It is also why I put the distorted and non-distorted waves through the same filtering process. Anything the filtering adds to the distorted wave is also in the comparison wave.

This should minimize it's effect on the comparison, yes/no?

I was referring to the Cool Edit Pro filters you are using to split the signal into frequency bands rather than the filters used in the upsampling process within Keith's program. What kind of filters are you using? Butterworth, Bessel, Chebychev 1 or 2? What slopes? I'm not saying you are wrong but naysayers will likely treat this as another variable in play.
post #46 of 68
Quote:
Originally Posted by cjwhitehouse View Post

I was referring to the Cool Edit Pro filters you are using to split the signal into frequency bands rather than the filters used in the upsampling process within Keith's program. What kind of filters are you using? Butterworth, Bessel, Chebychev 1 or 2? What slopes? I'm not saying you are wrong but naysayers will likely treat this as another variable in play.

I thought you brought up a great point/question before, so I just answered and asked another. This is a good point as well.

I ruled out Cheby's due to their ripples. Bessel's would have better GD, but there would be more FR interaction/changes.

I cannot remember the slopes I ended up with right off-hand. They look pretty steep, just eyeballing the pics I have posted. I think the LLP to LP xover needs to be relaxed a little. I would love to hear any recommendations here.
post #47 of 68
OK, since you seem to be pretty accomplished with Cool Edit Pro, humour me and try this experiment. Generate yourself a -5dB log swept sine sweep from, say 10Hz -> 200Hz using the Tone Generate function. Then put it through your Cool Edit Pro filter regime, stitch it back together and look at the waveform envelope you end up with. Try it with 6, 4 and 2 order filters if you have time. I'm intrigued to see what you get by this process without even applying any distortion. Sorry if this sounds like a homework assignment but there's nothing like proving these things for yourself.
post #48 of 68
I'll do it, but I'm not sure what you are after here.

A 2nd order grouping this close together should scoop the bandpassed section out, 4th order should give a 2 to 3dBFS spike around the xover frequencies, and 6th should put things within +/-3dBFS through the area. I used a higher order xover to get a narrow notch at the xover points, -3 to -6dBFS maybe over a few Hz.

In my tests both the clean wave and the dirty one went though these filters, so the tests should still be valid.

I am more interested in how to let more of the harmonics caused by the low frequencies that are generated out of it's passband come through. I think the harmonics are ending up around 12dB to low. I think the first test was more true to life. I still used a steep xover at 80Hz, but enough of the 2nd, 3rd, and maybe 4th harmonics made it though to really change the sound. It is the stuff below 30Hz that is the limiting factor I would think in the real world anyway.
post #49 of 68
I am wondering if you might be better to apply a low pass filter only at, say 15Hz with a shallow slope like 1st (or maybe 2nd) order, put that into your distortion program and use the shallow rolloff to provide the distortion "emphasis" to the lower frequencies. This is probably a better model of a typical subwoofer distortion profile than trying to split it into two bands like you are.

As you will no doubt have found, Keith's program tends to increase the signal length (as a result of the filtering/upsampling) so resyncing it back with the original high-passed element can be a bit problematic.

[Edit] PS, are you "normalising" the signal to peak at 0dBfs prior to running the distortion program and then restoring back to the same level afterwards (using average RMS power in Statistics?) so you get the maximum effect from the distortion ??
PPS, when running Keith's program, are you specifying the harmonic cosine polarities as "-"? The default I think is "+" which is probably not what you want.
post #50 of 68
Quote:
Originally Posted by cjwhitehouse View Post

I am wondering if you might be better to apply a low pass filter only at, say 15Hz with a shallow slope like 1st (or maybe 2nd) order, put that into your distortion program and use the shallow rolloff to provide the distortion "emphasis" to the lower frequencies. This is probably a better model of a typical subwoofer distortion profile than trying to split it into two bands like you are..

The problem with crossing that low is that there are not any fundamental tones there. You are just creating discordant noise. I think a few bass notes need to be included to get the effect right. I am also leaning back towards one bass band. The first trial just seems better.

Quote:
Originally Posted by cjwhitehouse View Post

As you will no doubt have found, Keith's program tends to increase the signal length (as a result of the filtering/upsampling) so resyncing it back with the original high-passed element can be a bit problematic. .

That is why I run it through too, at -250. It isn't quiet as bad that way.

Quote:
Originally Posted by cjwhitehouse View Post

PS, are you "normalising" the signal to peak at 0dBfs prior to running the distortion program and then restoring back to the same level afterwards (using average RMS power in Statistics?) so you get the maximum effect from the distortion ??

The signal clips to badly that way. I set the peak at -6dB before the distortion program. Afterwords, I adjust the RMS back to the original levels. On the second test I messed up and went with -3dB, and there is some minor waveform clipping, but not enough to hear.

Quote:
Originally Posted by cjwhitehouse View Post

PPS, when running Keith's program, are you specifying the harmonic cosine polarities as "-"? The default I think is "+" which is probably not what you want.

I found it wants this format "-34" "+".
post #51 of 68
All Butterworths

2nd Order


4th Order


6th Order


16th Order
post #52 of 68
Quote:
Originally Posted by soho54 View Post

The problem with crossing that low is that there are not any fundamental tones there. You are just creating discordant noise. I think a few bass notes need to be included to get the effect right. I am also leaning back towards one bass band. The first trial just seems better.

The point of using the very low crossover and shallow slope is to apply a differential gain to the bass. That way, content at 15Hz, 30Hz and 60Hz will be presented to the distortion process at 0dB, -6dB and -12dB respectively. Because the effect of the static distortion pattern diminishes as the amplitude reduces, distortion will occur mostly at the bottom end as in a real subwoofer. When stitching back with the high-passed version of the signal, the proper level of 30Hz and 60Hz content will be added back provided you can sort the problem of synchronisation otherwise phase differences are going to mess things up with cancellation effects.

Quote:
Originally Posted by soho54 View Post

The signal clips to badly that way. I set the peak at -6dB before the distortion program. Afterwords, I adjust the RMS back to the original levels. On the second test I messed up and went with -3dB, and there is some minor waveform clipping, but not enough to hear.

I suspect this is due to the harmonic polarity issue (below). By using "+" polarity, a normal sinewave cycle will become more "pointed" in shape, tending towards more of a sawtooth appearance. Subwoofers tend to flatten the waveform as excursion limits are reached - this is what happens when the harmonics are in opposite phase to the fundamental. This polarity issue is all too often glossed over on the basis that "we can't hear phase differences" but it does make a significant difference to the wave shape, transfer function and distortion audibility. Big subject.

Quote:
Originally Posted by soho54 View Post

I found it wants this format "-34" "+".

The -34 is the amplitude relative to the 0dB fundamental. You need to type in the "-" manually in the polarity column to force the program to apply the harmonics in inverse polarity. This will have the effect of flattening the waveform and should prevent clipping even if you supply a full 0dBfs signal to it.
post #53 of 68
Thinking more about this, I am surprised you are hitting a clipping issue. I think Keith's program should automatic attenuate the output to avoid any clipping. The test.wav file he provides as an example is recorded at 0dBfs and you can happily apply distortion to that without clipping. So I am puzzled by your problem.

To illustrate what I was saying about harmonic polarity, attached are a couple of charts based on my own measurement of the PB13 in 10Hz tune, delivering 15Hz at about 9%THD. One version has the harmonics in +ve polarity, the other has them -ve. As you can see the resulting transfer characteristic and output sine wave shape are different with the +ve having the "pointed" shape and the -ve "flattened". For info, the GM numbers are the GedLee Metric and would imply that the -ve polarity version is slightly more easily audible than the +ve even though the THD% numbers are identical.
LL
LL
post #54 of 68
Quote:
Originally Posted by soho54 View Post

All Butterworths

2nd Order


4th Order


6th Order


16th Order

Thanks for doing this. When I tried this, the dips were even more pronounced, especially with the 6th order version. I'm not sure exactly what the vertical scale is on your plots but the point I was seeking to illustrate is that the dips in the frequency response around the crossovers are not insubstantial in relation to the distortion effects you are trying to simulate.
post #55 of 68
Quote:
Originally Posted by cjwhitehouse View Post

I'm not sure exactly what the vertical scale is on your plots but the point I was seeking to illustrate is that the dips in the frequency response around the crossovers are not insubstantial in relation to the distortion effects you are trying to simulate.

Sorry, I forgot to include the scale.

I used an old -6dB sweep I had, and recombined them with Sony Vegas. I don't like the CE multi track tool. The light blue/gray line up top is 0dBFS. The darker lines are in 3dB increments.


I used the same 30 & 80Hz xover points.
post #56 of 68
Quote:
Originally Posted by cjwhitehouse View Post

Thinking more about this, I am surprised you are hitting a clipping issue. I think Keith's program should automatic attenuate the output to avoid any clipping. The test.wav file he provides as an example is recorded at 0dBfs and you can happily apply distortion to that without clipping. So I am puzzled by your problem.

I can explain. The program does attenuate the files, but all files are attenuated at different levels depending on the distortion amounts added. (When using the "+" parameter.) When you manually re-normalize all the waves you will get clipping, unless you use a lower signal to begin with, so that you have the headroom for the distortion.

You could start with a higher strength signal and adjust everything down in the end, but I found it easier when doing this to work in the other direction. The minor few dB S/N hit, and ease of application was a worthy tradeoff for these quick test runs to me.


Quote:
Originally Posted by cjwhitehouse View Post

Subwoofers tend to flatten the waveform as excursion limits are reached - this is what happens when the harmonics are in opposite phase to the fundamental.

Good point. I though it was the other way around. That is an easy fix.
post #57 of 68
Quote:
Originally Posted by cjwhitehouse View Post

Because the effect of the static distortion pattern diminishes as the amplitude reduces, distortion will occur mostly at the bottom end as in a real subwoofer.

The distortion doesn't just rise at the bottom end of a subwoofer....you've got the inductance modulation affecting the high frequencies too. The LMS 5400 from Illka's measurements is a perfect example (distortion at 100Hz can be higher than at 30Hz):
http://www.hometheatershack.com/foru...king-200l.html:


Btw, just outta curiosity...what is the ultimate goal of creating this distortion?
post #58 of 68
Quote:
Originally Posted by soho54 View Post

You could start with a higher strength signal and adjust everything down in the end, but I found it easier when doing this to work in the other direction. The minor few dB S/N hit, and ease of application was a worthy tradeoff for these quick test runs to me.

OK, but if you drop the peak input signal to the distortion program to -6dBfs then by my reckoning, your 10% distortion pattern will then yield less than 2% on the peaks and correspondingly less over the rest of the signal.
post #59 of 68
Quote:
Originally Posted by MBentz View Post

The distortion doesn't just rise at the bottom end of a subwoofer....you've got the inductance modulation affecting the high frequencies too. The LMS 5400 from Illka's measurements is a perfect example (distortion at 100Hz can be higher than at 30Hz):
http://www.hometheatershack.com/foru...king-200l.html:


Btw, just outta curiosity...what is the ultimate goal of creating this distortion?

My statement was indeed a generalisation and it is always possible to cite exceptions, but as a rule, most subwoofer distortion behaviour is dominated by the excursion issue down low and this is what soho54 was trying to emulate.

I thought the idea was to try and prove/disprove whether the CEA2010 distortion thresholds are a good measure of distortion audibility and I seem to recall that it was your request to do the test with a distorted music signal rather than sine tones.
post #60 of 68
Quote:
Originally Posted by cjwhitehouse View Post

OK, but if you drop the peak input signal to the distortion program to -6dBfs then by my reckoning, your 10% distortion pattern will then yield less than 2% on the peaks and correspondingly less over the rest of the signal.

I'm not sure I follow you here. I assumed the program based the distortion on the fundamental signals actual levels. It should scale better than that.

If you look back at post 35 you can calculate/guesstimate the the THD% by comparing the 55Hz area to the 2rd harmonic ~110Hz, and so on. Using the base figures from the HD%/Component graph here: http://www.hometheatershack.com/foru...0-hz-tune.html I think it was the 30Hz line as 3% and the ~15Hz area was used as the 10% figures.

The polarity issue could be a factor here. I'll go through all this and setup some test waves to find out exactly how things really mesh together, and get back with the results. These were just done by the seat of my paints. I assumed a -3dB drop by H3 and at a quick glance it looks like it was -6dB. I will agree that it needs to be done the other way around in the end, but it takes longer.

I was just trying to get something out there to try and start some conversation. I planed on working out the details as things went along, but no one else (other than you) seems to care though.

It looks like things might be a little better off than I though regarding the distortion added out of the passband, if the polarity bit pans out the way I think it will.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: DIY Speakers and Subs
AVS › AVS Forum › Audio › DIY Speakers and Subs › CEDIA 2010 and Distortion