Last week I finalized an alternative DSP filter which optimizes the predicted sound power response while maintaining nearly the same on-axis response. I went through 7 iterations, so for clarity in my graphics I call it "Sound Power Optimized DSP (v7)".
Thanks to some hints from Kimmo, I tried non-symmetric 3rd order slopes on the mid and tweeter. Actually, I tried a lot of combinations, but this is what worked the best. A classic BW3 response got me close, but to really optimize it, I had to combine a 1st order slope with a 2nd order slope, and then vary the Q and frequency of the 2nd order filter.
As can be seen in the graphs, the Sound Power, Predicted in-room, and DI curves are all very nicely behaved.
To make a fair subjective evaluation, it was important that the baseline filter and the new Sound Power optimized filter have very similar tonal balance. The second graphic show that I did not succeed perfectly, but I think it is close enough. Just to be clear, the baseline filter is LR4 at 200 Hz and 1.6 kHz, as shown in posts 156 and 157.
Thanks to some hints from Kimmo, I tried non-symmetric 3rd order slopes on the mid and tweeter. Actually, I tried a lot of combinations, but this is what worked the best. A classic BW3 response got me close, but to really optimize it, I had to combine a 1st order slope with a 2nd order slope, and then vary the Q and frequency of the 2nd order filter.
As can be seen in the graphs, the Sound Power, Predicted in-room, and DI curves are all very nicely behaved.
To make a fair subjective evaluation, it was important that the baseline filter and the new Sound Power optimized filter have very similar tonal balance. The second graphic show that I did not succeed perfectly, but I think it is close enough. Just to be clear, the baseline filter is LR4 at 200 Hz and 1.6 kHz, as shown in posts 156 and 157.
Attachments
So how does it sound?
I have been listening to the new filter for about a week. The Hypex Fusion amps make it very simple and fast to switch between the baseline and the new filter. I had to adjust the tweeter level up on the new filter by 0.3 dB to get it to tonally match the baseline, but once I did, it sounded quite good.
I was really expecting a noticeable improvement. I was somewhat perplexed that after an hour of listening, I honestly could not tell them apart. After a couple of days, I learned what to listen for and I could hear a very slight difference, but it was very subtle. I did not even bother with blind evaluations because I knew I would never be able to tell them apart under blind conditions. In my experience, blind testing will quickly reveal the big differences, but will often mask the subtle differences. Those subtle differences tend to reveal themselves over days or weeks of listening. The differences I was detecting between the baseline and the new filter were most definitely of the subtle kind.
After a week, I have developed an opinion on the new filter versus the baseline, and it is not what I was expecting. I prefer the baseline by a small margin. The baseline is able to reveal more low-level detail during loud complex program material. The baseline seems just slightly more 3 dimensional. As I said, most of the time I could not tell them apart, but on those occasions when I could hear a difference, I preferred the baseline.
And now I am perplexed by what this all means.
I have been listening to the new filter for about a week. The Hypex Fusion amps make it very simple and fast to switch between the baseline and the new filter. I had to adjust the tweeter level up on the new filter by 0.3 dB to get it to tonally match the baseline, but once I did, it sounded quite good.
I was really expecting a noticeable improvement. I was somewhat perplexed that after an hour of listening, I honestly could not tell them apart. After a couple of days, I learned what to listen for and I could hear a very slight difference, but it was very subtle. I did not even bother with blind evaluations because I knew I would never be able to tell them apart under blind conditions. In my experience, blind testing will quickly reveal the big differences, but will often mask the subtle differences. Those subtle differences tend to reveal themselves over days or weeks of listening. The differences I was detecting between the baseline and the new filter were most definitely of the subtle kind.
After a week, I have developed an opinion on the new filter versus the baseline, and it is not what I was expecting. I prefer the baseline by a small margin. The baseline is able to reveal more low-level detail during loud complex program material. The baseline seems just slightly more 3 dimensional. As I said, most of the time I could not tell them apart, but on those occasions when I could hear a difference, I preferred the baseline.
And now I am perplexed by what this all means.
I wonder if this is what Floyd Toole called the Circle of Confusion.
As I'm sure you know jim, you're evaluating your speaker by using recordings which were made by using microphones, EQ, reverb, and effects which are evaluated using professional monitor speakers, which are evaluated by using recordings that were made for consumer loudpspeakers that were made by...
Let's go back, few speakers in the past few decades have flat frequency response, on and off axis. Many had all kinds of peaks and dips, limited LF, or even worse exaggerated highs and lows. Some of the better ones that could get the frequency response smooth and flat sometimes had limited LF extension, due to budget constraints.
Unfortunately I can think of more manufacturers whose speakers had frequency were all over the map, than manufacturers who who have been aiming for smooth on and off axis response and good DI (previously called dispersion).
There were all kinds of "classic" speakers that used an 8 or 10" midwoofer crossed over to a tweeter above 2 or 3KHz because tweeters just weren't that good enough to be crossed lower than 2KHz. Which means so many speakers would have power response dip in the 1-3KHz response.
So if music is listened to, recorded through, and monitored through speakers that had a 7" + 1 tweeter combo or 8-10" woofers of a bygone era, well they're probably going to sound not too bad on those. And it's possible your Sound Power Optimised DSP V7 may be the most "correct" crossover, but it may not be the preferred one, because you're probably listening to music recorded and mixed and mastered through speakers which are NOT correct.
If I take the red pill then I probably going to be unsettled and realise that there's no perfection to listening to recorded music because there's so many variables, and technology that's state of the art, and what's understood as the best available science today is archaic 50 years into the future.
If I take the blue pill then I should fine tune your crossover to the one that sounds best for most of the music that you listen to, and accept that crossover tuning is equal parts art and equal parts science...
Last edited:
The main directivity of the design is set by the geometry of the drivers used. The rapid fall in soundpower due to the increasing directivity of the tweeter and the general slope is pretty much the same in both. There is a 3k resonance which is more pronounced in the later version which could explain why you like it slightly less.
Lipshitz and Vanderkooy found that power response holes can be quite severe without them necessarily being audible.
Your experience would seem to bear that out.
AES E-Library >> Experiments in Direct/Reverberant Ratio Modification
No it does not. It seems that, with my baffle layout and driver directivities, the only way to get both a flat on-axis response and a flat power response is with a 90 degree phase difference between the mid and the tweeter. If the tweeter had a waveguide, it would be a different story. If the drivers had been spaced with 1/4 wavelength spacing, different story.
Yes, and I am not sure of the source of that resonance. It is in the mid driver raw response but not the tweeter. However, when I measured the mid driver in a 60cm x 60cm foam board baffle last october, I found no resonance. It could be a cabinet resonance perhaps? I had assumed that such a high resonance, and with such a high Q, would be inaudible, but perhaps with very critical listening it is revealed? In any case, the 1.6 kHz LR4 filter eliminates it.
Now that is very interesting indeed. I had been viewing the power response of the baseline system (post 156) as a problem to be solved, but perhaps it is not. Thanks for that information.
I was enamored with the preference rating for about a month... then I got over it... When I looked at S. Olive's paper, there was a fairly low correlation between individual preference and the final score. If it were rounded to the nearest even number, it might be useful in segregating into three general categories, the bad, the acceptable, the very good. Calculating it down to a decimal place is misleading... calculating it down to 2 decimal places is silly.
I don't know which equation to select, so I am showing the score from both.
Attachments
There is another option which is to create a faceted trapezoidal baffle so there is minimal baffle surrounding the tweeter. I would strongly consider making a foam mockup to see the effect if I had these these drivers. I don't think the dished face plate is doing any favours either.
Looks like these things circled here could be combining, you could try a couple of EQ tweaks to see if you can make a worthwhile change and listen to the difference. I think it is a big enough bump in a sensitive hearing area and likely to be audible.
I big power response hole is not ideal but perhaps looks worse than it sounds. Trying to fix it with a non acoustic solution probably just shifts the problem instead of fixing it.
I would look at trying to flatten the listening window more than the power response and see what that does.
Attachments
I appreciate your thoughts, Fluid. This current project is done, I won’t be making a new cabinet. The next project will benefit from any new knowledge of diffraction and simulation, but this cake is baked. With the baffle dimensions and driver spacing fixed, there is only so much that can be done. And honestly, I am so pleased with the performance of the system with the baseline filter. I have no complaints, at all.
I tried notching out the 3k resonance, but I could not get a simulation I was happy with, and a quick listening test convinced me I was making the situation worse.
I have continued to listen to the baseline DSP filter (200 LR4, 1.6k LR4) and the new one (Sound Power Optimized v7). I continue to be astonished at how similar they sound on most program material. However, in this post I want to focus on how they are different. On any movie or video soundtrack I have listened to, the two filters are indistinguishable. Almost all music sounds the same. It is only on certain tracks that I get a very slightly enhanced sense of realism with the baseline. It would be more accurate to say that with the Sound Power Optimized DSP filter, at certain times the sense of space and realism recedes slightly, almost as if the sound collapses back into the two speakers for just a moment.
This is very similar to the situation where I was choosing between a 1.6k, 1.8k, or 2.0k LR4 crossover several months ago. All three options sounded nearly the same. But the 1.6k crossover never lost the sense of space and presence, but the 2.0 k occasionally did.
My working theory is that I am hearing the character of the drivers. I know some will disagree with this statement, but I believe that drivers can sound different even when operating in their pistonic region, a difference that goes beyond what the frequency response and dispersion would suggest. The new filter is asking the MW16TX to operate up to 2200 Hz, a half octave higher than the baseline asks. If a metal dome tweeter and a silk dome tweeter can sound different (even with similar response curves), then how much more different will a midwoofer and tweeter sound? One has a moving mass of 12.6 g and a resonance of 29 Hz, the other has a moving mass of 0.51 g and a resonance of 650 Hz. Is it plausible that they will sound different when playing from 1600 to 2200 Hz? Yes, I think it is plausible. I could almost say, how could they not sound different?
So the bottom line is that I believe the Sound Power Optimized v7 filter is just asking too much from the 6 inch driver.
Headshake wrote:
I don't have the right data to show an apples-to-apples comparison of waterfall data. But with the LR4 crossover at 1.6k, the mid driver is quite suppressed at 3k.
I am glad we delayed measuring your system, I have learned some important things about measurements in the last 6 months or so...
j.
So now let’s talk about why the baseline filter and the Sound Power Optimized v7 filter sound so similar on the majority of material. Here is my theory...
Several months ago, I made a room response curve. I used 5 different microphone positions and 2 different speaker positions for a total of 10 separate measurements. The 5 mic positions were spaced out quite a bit further laterally than the normal listening position, at different heights and distances from the speakers. When all 10 curves are averaged together, the result is smooth and even. I show this in the first graphic. The next graphic shows the same thing, but with 1/6 octave smoothing.
The third graphic shows the average room response curve compared to the measured left and right on-axis curves. I was pleased to see the close correlation between the room curve and the on-axis curves.
The fourth graphic compares the average room response curve to the VituixCad2 simulation. My measured room response curve is much closer to the simulated listening window curve than it is to the predicted in-room response or the power response. So, this is the explanation. Since both the baseline filter and the new filter have very similar on-axis responses and listening window responses, they tend to sound very similar in my unique room. Optimizing the sound power response had no effect on my listening experience because in my room, I don’t hear the sound power response.
This is not so surprising in hindsight. In his book, F. Toole discusses how the sound power response is a good indication of how a speaker will sound in an average room, but he acknowledges that every room is different, and some are quite different. Does this make sense? or am I completely dished
?
j.
Several months ago, I made a room response curve. I used 5 different microphone positions and 2 different speaker positions for a total of 10 separate measurements. The 5 mic positions were spaced out quite a bit further laterally than the normal listening position, at different heights and distances from the speakers. When all 10 curves are averaged together, the result is smooth and even. I show this in the first graphic. The next graphic shows the same thing, but with 1/6 octave smoothing.
The third graphic shows the average room response curve compared to the measured left and right on-axis curves. I was pleased to see the close correlation between the room curve and the on-axis curves.
The fourth graphic compares the average room response curve to the VituixCad2 simulation. My measured room response curve is much closer to the simulated listening window curve than it is to the predicted in-room response or the power response. So, this is the explanation. Since both the baseline filter and the new filter have very similar on-axis responses and listening window responses, they tend to sound very similar in my unique room. Optimizing the sound power response had no effect on my listening experience because in my room, I don’t hear the sound power response.
This is not so surprising in hindsight. In his book, F. Toole discusses how the sound power response is a good indication of how a speaker will sound in an average room, but he acknowledges that every room is different, and some are quite different. Does this make sense? or am I completely dished
?j.
Attachments
Last edited:
I'm a little late in replying to this one, almost a year to the dayThis is not so surprising in hindsight. In his book, F. Toole discusses how the sound power response is a good indication of how a speaker will sound in an average room, but he acknowledges that every room is different, and some are quite different. Does this make sense? or am I completely dished
I wonder if you would hear more of a difference between the two filters if you were sitting in a different position. While the average of both is quite similar with only a slight change in overall smoothness (in favour of the soundpower one), the individual curves are really quite different with a lot of combing in the baseline.
Could be an interesting experiment?
I actually went through one more development of the Sound Power Optimized filter. Sound Power Optimized v8 had some notching of the 3k resonance and a few other minor tweaks. I went back and forth between baseline and Sound Power Optimized v8 for several months. Most of the time there was so little difference between them. In my normal listening position (fairly close), I preferred the baseline, but only by a very small degree and with certain program material. Interestingly, if I moved to the back of the room where I would get much more reflected sound versus direct, I had a slight preference for the v8.
In addition to Headshake, I had three other dedicated listeners do an extended audition. One person was neutral, i.e. the two speakers sounded slightly different, but this person did not have a preference. The second person preferred the v8, but it was a difficult choice.... Only after I pressed her to make a choice did she choose the v8. The third person preferred the baseline, and did not hesitate to say the baseline was superior. So I did not get a statistically valid "group preference" ranking from this.
I don't draw any conclusions from all of this, except that for this speaker, in this room, I prefer the baseline DSP filter. I refuse to believe that a -3 dB dip in the power response from 1k to 2k is a sound-improving feature I should strive for in future projects.... I think the baseline sounds just slightly better for reasons that have nothing to do with directivity. I think that the directivity performance of the two filters is so similar that its not much of a factor in preference.
Too bad you are 20,000 km away, I would invite you over for a beer and we could listen and evaluate these two filters, and my newer project...
j.
In addition to Headshake, I had three other dedicated listeners do an extended audition. One person was neutral, i.e. the two speakers sounded slightly different, but this person did not have a preference. The second person preferred the v8, but it was a difficult choice.... Only after I pressed her to make a choice did she choose the v8. The third person preferred the baseline, and did not hesitate to say the baseline was superior. So I did not get a statistically valid "group preference" ranking from this.
I don't draw any conclusions from all of this, except that for this speaker, in this room, I prefer the baseline DSP filter. I refuse to believe that a -3 dB dip in the power response from 1k to 2k is a sound-improving feature I should strive for in future projects.... I think the baseline sounds just slightly better for reasons that have nothing to do with directivity. I think that the directivity performance of the two filters is so similar that its not much of a factor in preference.
Too bad you are 20,000 km away, I would invite you over for a beer and we could listen and evaluate these two filters, and my newer project...
j.
This is what I thought would be the case, when you were far enough away or more off axis, the more erratic room response would play a bigger role.
It's always dangerous to try and draw hard and fast conclusions when there can be so many factors affecting the outcome. A simple preference choice between two filters is one thing but trying to work out why is a slipperier slope for sure. I think how the dip is created is central to whether it matters so much, because that will determine what might be done to "fix" it. If the listening window is well controlled and quite similar, the near axis responses aren't changing much. If the slope of the power response is the same, the drivers are the same and nothing else has been done outside the crossover to affect directivity the Lipshitz and Vanderkooy research would suggest a power dip in those circumstances is hard to hear.
What I have come to feel more strongly about over time is that the vertical response is important in a way that might not be immediately obvious. A narrow vertical directivity or one where where the main vertical null is managed to produce a smoother in room and power response has an effect on music that might be considered badly produced or difficult. I like a lot of this sort of music and on many good speakers at a standard 3m listening distance it sounds woeful. If you only like or want to listen to well regarded audiophile music those good speakers do it better, but if you want to run the full gamut on one speaker... When I bought my first good pair of speakers I did the done thing and went to the local audio store to listen and evaluate the speakers. I took an Audioslave CD with me amongst others. The speaker I bought made Audioslave sound good and not awful like all the others did. That speaker had a ring Radiator tweeter. I knew nothing about what that meant at the time but it makes sense now.
It really is.
I appreciate the offer
Given my current status as enthusiastically and happily retired, I doubt I will be visiting either the FINAF in Jyväskylä nor the RAAF Amberle in Queensland ever again... That is not to say I would never visit Finland or Australia as a tourist, but it is not in the current plans.
This makes sense. I think the patterns of vertical directivity are areas that need more explorations. What may work well in one room with a 12' ceiling and a 10' listening distance may not work well at all in another room with an 8' ceiling and a 15' listening distance. My experiences with line arrays has been limited to large magnepans, and when properly set up, I always thought they sounded spectacular. Vertical directivity on these is obviously much narrower than a cone-dome speaker...
The closest I have to that is what I posted in #171 above, which shows the average of 10 different mic/speaker positions. I would have to make a new set of measurements. However, I am somewhat doubtful that a mic at the listening position will actually capture what I perceive. Ungated FR curves (i.e. RTA spectrum analysis of a pink noise signal) include the full range of direct and reflected sound. A lot of research has shown that our ear/brain is able to discriminate the direct sound from the early reflections, and the early reflections from the late reflections.
It's important to remember, that long gating-measurements (of sine sweep) shows the highest amplitude of each frequency. So does noise, I'm not sure about MLS chirp. REW has many choises for smoothing and gating of sweeps, psychoacoustic, ERB or fixed cycles are interesting to study.
And, after all like Jim said our hearing system works differently! And then add room effects (reflections, reverberance/decay, modes), we don't listen in an anechoic chamber or outdoors...
And, after all like Jim said our hearing system works differently! And then add room effects (reflections, reverberance/decay, modes), we don't listen in an anechoic chamber or outdoors...