I didn't want to threadjack surf,sun&sound's "New Reference Speakers with Full/Wide-Range Driver" thread, so I'm starting this topic in another thread.
Surf's objective is to find a speaker capable of reproducing the human voice (80hz to ?khz range) without the intrusion of a crossover.
There have been many suggestions in that thread, but often, the inevitable compromise rears its ugly head, that is, a crossover point within that range.
Are there particular frequencies that should be avoided because the human ear is particularly sensitive, and, on the other hand, are there frequencies more suited to accomodating a crossover point because the ear is less sensitive in that region.
Thanks,
Glenn
Surf's objective is to find a speaker capable of reproducing the human voice (80hz to ?khz range) without the intrusion of a crossover.
There have been many suggestions in that thread, but often, the inevitable compromise rears its ugly head, that is, a crossover point within that range.
Are there particular frequencies that should be avoided because the human ear is particularly sensitive, and, on the other hand, are there frequencies more suited to accomodating a crossover point because the ear is less sensitive in that region.
Thanks,
Glenn
One question may be: Hey, I have these two drivers, where should I place the crossover? Another question may be: Hey, I can acquire drivers to ANY specification, where should I place the crossover? The second question is most often not terribly relevant, but it is implicitly the question I address here. The non-existence of the required drivers may present a problem.
A number of issues arise with crossover frequency selection. Phase distortion, amplitude response on and off-axis, power response, transient response, changes in radiation patterns, etc. I will deal here mostly with phase distortion on-axis.
Theory and numbers do not give me satisfactory answers to this question, so I have performed my own experiments to get my own "personal answer" to the question of the audibility of phase distortion (a small subset of the larger question "where can I place a crossover?"). I posted a summary of my conclusions in a post to the "New Reference Speakers with Full/Wide-Range Driver" thread. The conclusions result in design constraints that unfortunately eliminate most conventional design options. For this reason alone I would expect most speaker designers would discount all I have to say about the audibility of phase distortion. Its just TOO INCONVENTIENT to toss out design options that conflict with such stringent phase distortion requirements, so obviously such design constraints CAN'T BE REALISTIC! Hint: I have no patience for those who would like to "correct my views" on this issue, nor will I try to convince others that what I conclude applies to then as well. My conclusions about my perceptions apply to myself only.
Rather than restate my conclusions regarding phase distortion audibility or elaborate on it I suggest that the audibility of phase distortion is best determined by experiment. It is certainly true that not everyone perceives sound the same way. Your own experiments may tell the story for you.
A difficulty with such experiments is a lack of experience with "what it SHOULD sound like", or "what gives me listening pleasure". If all your listening experience is with playback systems with significant phase distortion then you may not have developed the ability to recognize that anything "needs to be fixed". Probably the best reference is the live listening experience (up close, in a good acoustic environment, no sound reinforcement, music you like, and an experience you enjoy). The question is: can you get this enjoyment from a recording? If a recording does not very well match the live experience in terms of "pleasure of the experience", then perhaps something can be improved in the playback system.
1. Experiment setup:
The experiment setup requires high quality sound recordings, a high quality headphone system, a high quality computer-based record/playback system, and the appropriate digital signal processing that mimics analog crossover systems.
Digitize a set of music selections that you enjoy listening to (our would like to enjoy). The source material should be high resolution (well recorded SACD or DVD-Audio on a high quality player, or high quality vinyl using a high quality playback system). If Compact Disc "works for you", you may wish to use it, but I recommend against this since we want to eliminate as many variables as we can, and lower quality source material will introduce more unknowns, and our test would have additional distortions the effects of which are not clearly known. When doing research such as this we don't want to make unnecessary assumptions.
The digitization should be of high quality, most convenient would be A/D conversion using a high quality 96khz sample, 24 bit A/D converter or better. I use the MSB PAD-1 PRO A-to-D CONVERTER (http://www.msbtech.com/). The high sample rate is required so that the subsequent analog filter simulation will be of high quality.
After digitization, create a set of sound files from each recording that have phase distortion that matches the phase distortion of the type and location of the crossover design you wish to test for. This can be accomplished by using digital filters that mimic analog crossover designs (IIR filters). Sum the high-pass and low-pass signals coherently as would take place acoustically (paying attention to polarity reversals as appropriate). Try a number of crossover types, crossover orders, and crossover frequencies. You may also try multiple crossovers in a single test.
2. Observables:
With the unmodified sound files and the sound files that have the phase distortion introduced, perform listening experiments. Careful with this, a mistake with headphone listening can cause permanent hearing damage if, for example, there is a "bug" in the processing software (I speak from experience).
The listening experiments are "tricky" for a number of reasons. When you listen to a recording with phase distortion you will NOT hear a voice say "ha!, phase distortion at 2khz!". What you should pay attention to is how much you are enjoying the music. Forget about the technical aspects, and simply observe your emotional reaction. Once you "get a handle" on this type of reaction, you may spend more time trying to "pin down the effect", but this is not necessary. Take notes. You may try "double blind" listening sessions.
3. Analyze the results
You may or may not notice a change in your level of enjoyment from the phase distorted sound files. If changes are noticed, you may try to "make sense" of the trends. What crossover frequencies reduce your enjoyment? Does it change with the type of crossover? If you perceive a difference, does it even matter for your enjoyment?
I have found my own answers to the "where can I place the crossover" question through such experiments. For me the results imply very strict requirements on crossover location. For me, I cannot perceive any change when a Linkwitz-Riley 4th order crossover is placed at 8khz, but placing one at 300 hz or at 2khz really interferes with my enjoyment of the music. Translated into a plausible hearing mechanism explanation it would seem that "phase distortion curvature" in the 1khz-3khz range is highly undesirable for me. This implies keeping crossovers below 100 hz and above 6-8khz. Higher order crossovers will require crossover frequencies further from the critical region, lower order less. First order crossovers introduce no on-axis phase distortion, but there can be undesirable direction effects.
It would be interesting to hear from others who perform such experiments.
Regarding the comment from Argo "Looking at Fletcher-Munson Equal Loudness Curves I would say between 1.5kHz and 6kHz is not suited for crossover.", an analysis of the phase distortion of an L-R 4th order crossover at 350hz tells me that this is just as bad as a crossover at 2 khz (looking at the second derivative of the phase distortion in the 1-4khz region). The lower crossover frequency, although seemingly quite distant from the critical region, spreads out its phase distortion quite dramatically into the higher frequencies. This conclusion is based upon the above experiments and from an analysis of the phase distortion.
A number of issues arise with crossover frequency selection. Phase distortion, amplitude response on and off-axis, power response, transient response, changes in radiation patterns, etc. I will deal here mostly with phase distortion on-axis.
Theory and numbers do not give me satisfactory answers to this question, so I have performed my own experiments to get my own "personal answer" to the question of the audibility of phase distortion (a small subset of the larger question "where can I place a crossover?"). I posted a summary of my conclusions in a post to the "New Reference Speakers with Full/Wide-Range Driver" thread. The conclusions result in design constraints that unfortunately eliminate most conventional design options. For this reason alone I would expect most speaker designers would discount all I have to say about the audibility of phase distortion. Its just TOO INCONVENTIENT to toss out design options that conflict with such stringent phase distortion requirements, so obviously such design constraints CAN'T BE REALISTIC! Hint: I have no patience for those who would like to "correct my views" on this issue, nor will I try to convince others that what I conclude applies to then as well. My conclusions about my perceptions apply to myself only.
Rather than restate my conclusions regarding phase distortion audibility or elaborate on it I suggest that the audibility of phase distortion is best determined by experiment. It is certainly true that not everyone perceives sound the same way. Your own experiments may tell the story for you.
A difficulty with such experiments is a lack of experience with "what it SHOULD sound like", or "what gives me listening pleasure". If all your listening experience is with playback systems with significant phase distortion then you may not have developed the ability to recognize that anything "needs to be fixed". Probably the best reference is the live listening experience (up close, in a good acoustic environment, no sound reinforcement, music you like, and an experience you enjoy). The question is: can you get this enjoyment from a recording? If a recording does not very well match the live experience in terms of "pleasure of the experience", then perhaps something can be improved in the playback system.
1. Experiment setup:
The experiment setup requires high quality sound recordings, a high quality headphone system, a high quality computer-based record/playback system, and the appropriate digital signal processing that mimics analog crossover systems.
Digitize a set of music selections that you enjoy listening to (our would like to enjoy). The source material should be high resolution (well recorded SACD or DVD-Audio on a high quality player, or high quality vinyl using a high quality playback system). If Compact Disc "works for you", you may wish to use it, but I recommend against this since we want to eliminate as many variables as we can, and lower quality source material will introduce more unknowns, and our test would have additional distortions the effects of which are not clearly known. When doing research such as this we don't want to make unnecessary assumptions.
The digitization should be of high quality, most convenient would be A/D conversion using a high quality 96khz sample, 24 bit A/D converter or better. I use the MSB PAD-1 PRO A-to-D CONVERTER (http://www.msbtech.com/). The high sample rate is required so that the subsequent analog filter simulation will be of high quality.
After digitization, create a set of sound files from each recording that have phase distortion that matches the phase distortion of the type and location of the crossover design you wish to test for. This can be accomplished by using digital filters that mimic analog crossover designs (IIR filters). Sum the high-pass and low-pass signals coherently as would take place acoustically (paying attention to polarity reversals as appropriate). Try a number of crossover types, crossover orders, and crossover frequencies. You may also try multiple crossovers in a single test.
2. Observables:
With the unmodified sound files and the sound files that have the phase distortion introduced, perform listening experiments. Careful with this, a mistake with headphone listening can cause permanent hearing damage if, for example, there is a "bug" in the processing software (I speak from experience).
The listening experiments are "tricky" for a number of reasons. When you listen to a recording with phase distortion you will NOT hear a voice say "ha!, phase distortion at 2khz!". What you should pay attention to is how much you are enjoying the music. Forget about the technical aspects, and simply observe your emotional reaction. Once you "get a handle" on this type of reaction, you may spend more time trying to "pin down the effect", but this is not necessary. Take notes. You may try "double blind" listening sessions.
3. Analyze the results
You may or may not notice a change in your level of enjoyment from the phase distorted sound files. If changes are noticed, you may try to "make sense" of the trends. What crossover frequencies reduce your enjoyment? Does it change with the type of crossover? If you perceive a difference, does it even matter for your enjoyment?
I have found my own answers to the "where can I place the crossover" question through such experiments. For me the results imply very strict requirements on crossover location. For me, I cannot perceive any change when a Linkwitz-Riley 4th order crossover is placed at 8khz, but placing one at 300 hz or at 2khz really interferes with my enjoyment of the music. Translated into a plausible hearing mechanism explanation it would seem that "phase distortion curvature" in the 1khz-3khz range is highly undesirable for me. This implies keeping crossovers below 100 hz and above 6-8khz. Higher order crossovers will require crossover frequencies further from the critical region, lower order less. First order crossovers introduce no on-axis phase distortion, but there can be undesirable direction effects.
It would be interesting to hear from others who perform such experiments.
Regarding the comment from Argo "Looking at Fletcher-Munson Equal Loudness Curves I would say between 1.5kHz and 6kHz is not suited for crossover.", an analysis of the phase distortion of an L-R 4th order crossover at 350hz tells me that this is just as bad as a crossover at 2 khz (looking at the second derivative of the phase distortion in the 1-4khz region). The lower crossover frequency, although seemingly quite distant from the critical region, spreads out its phase distortion quite dramatically into the higher frequencies. This conclusion is based upon the above experiments and from an analysis of the phase distortion.
have you done any experimentation on 2nd order xovers?
Response to question from tiroth :
"have you done any experimentation on 2nd order xovers?"
Yes, my experiments included 2nd order crossovers of the Linkwitz-Riley type.
A second order crossover helps a little, but not much. A second order crossover at 190hz (or 3050hz) is about the same as a 4th order at 125hz (3850hz). These particular numbers are at the thresold of acceptability for me. The inaudibility of the 8khz crossover for a 4th order L-R crossover also holds for the 2nd order crossover, but here the 2nd order does not allow a lower crossover frequency without becoming audible, this is likely due to the "wider spread" of the crossover region for the L-R 2nd order compared to the 4th order. This is consistent with the "phase distortion curvature" analysis and with experiment.
A true 2nd order crossover (acoustic) does not provide much driver protection, but is still a consideration given all the other trades to be made.
Hope this helps
Response to question from tiroth :
"have you done any experimentation on 2nd order xovers?"
Yes, my experiments included 2nd order crossovers of the Linkwitz-Riley type.
A second order crossover helps a little, but not much. A second order crossover at 190hz (or 3050hz) is about the same as a 4th order at 125hz (3850hz). These particular numbers are at the thresold of acceptability for me. The inaudibility of the 8khz crossover for a 4th order L-R crossover also holds for the 2nd order crossover, but here the 2nd order does not allow a lower crossover frequency without becoming audible, this is likely due to the "wider spread" of the crossover region for the L-R 2nd order compared to the 4th order. This is consistent with the "phase distortion curvature" analysis and with experiment.
A true 2nd order crossover (acoustic) does not provide much driver protection, but is still a consideration given all the other trades to be made.
Hope this helps
((chris8 crawls out of his deep burrow))
I feel I must state that I believe people should follow goudey's advice concerning using a software derived sample set in order to test 'audibility' of this effect. While analog line level filters can also be easily made, the software removes additional variables which may be presented. Unless you have the appropriate test equipment set up in order to measure and confirm that the only differences are the ones predicted due to the crossover, please use the software based setup. I recommend a studio monitor grade headphone at least, in order to have a clear representation of the signals.
In addition, I will state that it is a MUST to remove bias from these tests. Use pcabx software(freeware) to compare the samples. Knowledge of which sample is under test simply adds too many additinal variables. The software i referenced will allow you to remove this burden.
Samples of non-filterd vs. LR4th filters can be found here: http://www.pcabx.com/technical/LR-300-3K/index.htm
The PCABX program can be found here:
http://www.pcabx.com/license.htm
I suggest reading this introduction to the comparison method:
http://www.pcabx.com/getting_started.htm
Regardless of your results, keep in mind that this test is only relevant to a filter in a single plane. In the case of combining multiple transducers, you must also consider the resultant changes of off axis drive interactions in the vertial and horiztonal radial fields.
-Chris
I feel I must state that I believe people should follow goudey's advice concerning using a software derived sample set in order to test 'audibility' of this effect. While analog line level filters can also be easily made, the software removes additional variables which may be presented. Unless you have the appropriate test equipment set up in order to measure and confirm that the only differences are the ones predicted due to the crossover, please use the software based setup. I recommend a studio monitor grade headphone at least, in order to have a clear representation of the signals.
In addition, I will state that it is a MUST to remove bias from these tests. Use pcabx software(freeware) to compare the samples. Knowledge of which sample is under test simply adds too many additinal variables. The software i referenced will allow you to remove this burden.
Samples of non-filterd vs. LR4th filters can be found here: http://www.pcabx.com/technical/LR-300-3K/index.htm
The PCABX program can be found here:
http://www.pcabx.com/license.htm
I suggest reading this introduction to the comparison method:
http://www.pcabx.com/getting_started.htm
Regardless of your results, keep in mind that this test is only relevant to a filter in a single plane. In the case of combining multiple transducers, you must also consider the resultant changes of off axis drive interactions in the vertial and horiztonal radial fields.
-Chris
CHRIS8 said:
I would just like to point out that although this will help remove bias, the ABX paradigm has been shown to be statistically invalid except in cases where you can clearly hear a difference. (ie ABX is satistically incapable of confirming no difference) -- i believe that is how it goes, it was a while ago i saw the proof and attempts to find the exact proof in my 10s of thousands of archived emails hasn't turned it up yet.
dave
This is the main compromise with 2 way speakers, most drivers that give any real bass at all will not smoothly go up to even 4KHz, due to cone breakup modes and directionality. This is why the 2.5/3KHz crossover frequency is chosen in 95% of designs, and to be honest, a little crossover distortion is much more bearable than driver breakup.
I wonder because very good speaker system (not 2500 Hz), 2 way like this have crossover in the region here mentioned to avoid:
http://www.pioneerelectronics.com/P...ndustrialTADProductDetails/0,1445,225,00.html
http://www.pioneerelectronics.com/P...ndustrialTADProductDetails/0,1445,225,00.html
Till, the link is taking ages to load, but are they the big horn loaded tweeter ones?
AFAIK, there is no commercial 2 way speaker that does not have a crossover in this range, so it's all down to personal taste and design, most commercial 2 ways sound better than 3 ways, due to the difficulty of optimising a passive 3 way crossover.
Now DIY, that's where it gets interesting
AFAIK, there is no commercial 2 way speaker that does not have a crossover in this range, so it's all down to personal taste and design, most commercial 2 ways sound better than 3 ways, due to the difficulty of optimising a passive 3 way crossover.
Now DIY, that's where it gets interesting
"just like to point out that although this will help remove bias, the ABX paradigm has been shown to be statistically invalid except in cases where you can clearly hear a difference. (ie ABX is satistically incapable of confirming no difference) "
On the premise of proving 'no' difference, yes this would be impossible logically. However, on the premise of attempting to find a difference, the method has not been proven 'invalid'. ONly a repeatable protocol/methodology can be used to ascertain significant differences for practical purposes of efficient design/engineering. For the purpose of identifigin a difference, the only practical method is an audible comparison of the signal. If not comparing the sound, how else would you identify an 'audible' difference? In order to min. biases, knowledge of signal under test must remain undisclosed during testing. This is all the test is, while removing the listener's knowledge of which signal is being heard at any point during the test. In cases where this is not the objective, simply ignore 'practical' methodology and perform a more relaxed test allowig knowledge of item(s) as they are under test ; but this will yeild results of lesser value.
-Chris
On the premise of proving 'no' difference, yes this would be impossible logically. However, on the premise of attempting to find a difference, the method has not been proven 'invalid'. ONly a repeatable protocol/methodology can be used to ascertain significant differences for practical purposes of efficient design/engineering. For the purpose of identifigin a difference, the only practical method is an audible comparison of the signal. If not comparing the sound, how else would you identify an 'audible' difference? In order to min. biases, knowledge of signal under test must remain undisclosed during testing. This is all the test is, while removing the listener's knowledge of which signal is being heard at any point during the test. In cases where this is not the objective, simply ignore 'practical' methodology and perform a more relaxed test allowig knowledge of item(s) as they are under test ; but this will yeild results of lesser value.
-Chris
This was one thought i had. But as you maybe know from my hornthread somewhere on page 4 in loudspeaker forum, it goes more towards horns. Cyclotronguy (thanks) invested a lot of patience and slightly moved me to this. One point pro short horn con front mount like in a clone of the TAD is, i fear my son will damage the expensive driver if he could easily reach it. Anotherone is efficiency. Hope for dynamic. So i´m on building A7 now and try to understand cyclotronguys crossover advice, building equipment for impedanz and SPL measurements, glue DMS into the A7 to watch the resonance of the enclousure....
Hi Glenn
Have you seen the October/November issue of "Australian hi-fi and home theatere technology" ?
There is a test of an active speaker made by the Aussie manufacturer Dan A Digital.
The midrange driver of this speaker runs from 92 Hz to 4096 Hz in order to leave the vocal range "untouched" (another interesting feature may be that the midrange is driven by a 40 watt class-a tube amp).
They use a 3rd order lowpass on the woofer, a 2nd order highpass plus a 3rd order lowpass on the midrange and a 3rd order highpass on the tweeter.
I tink that depending on driver size, driver placement and crossover type (i.e. 2nd order or less, subtractive crossovers...)even crossover frequencies up to 400 Hz should be feasible without being audible.
Regards
Charles
Have you seen the October/November issue of "Australian hi-fi and home theatere technology" ?
There is a test of an active speaker made by the Aussie manufacturer Dan A Digital.
The midrange driver of this speaker runs from 92 Hz to 4096 Hz in order to leave the vocal range "untouched" (another interesting feature may be that the midrange is driven by a 40 watt class-a tube amp).
They use a 3rd order lowpass on the woofer, a 2nd order highpass plus a 3rd order lowpass on the midrange and a 3rd order highpass on the tweeter.
I tink that depending on driver size, driver placement and crossover type (i.e. 2nd order or less, subtractive crossovers...)even crossover frequencies up to 400 Hz should be feasible without being audible.
Regards
Charles
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.