I would not shade a floor to ceiling array. Off axis(vertical) is a virtual infinite length array. There is no off axis. Comb filtering, that's been discussed and controlling the proper frequencies to the proper drivers in a particular configuration is not an issue with me. Widen vertical directivity? Bounce off the ceiling?
I agree, when necessary.
And, there have been others (experts), but, none of them shading a floor to ceiling array
The above is wrong, and here's why:
I assume you will agree that a point source in 4pi space will have your so-called "flat" phase response, right? The wavefront expands as it travels from this point source outwards to the listening/measurement location. You can think of this as an expanding sphere. Now freeze the sphere somewhere between the source and the listener. Next, make a narrow slice of the sphere that is only a few degrees wide and remove the rest of the wavefront. You have something that is akin to a thin wedge of lemon, with the widest part centered on the ray between the source and listener. Now we will fill most of the slice with a vertical line array of drivers, keeping in mind that the surface of the "slice" is curved. This is the concept behind the CBT. It's source strength and alignment are designed to mimic a propagating spherical wavefront. Keele also uses the floor so that he only needs the upper "half" of the slice, which he calls a "spherical cap".
Even though the CBT's drivers are distributed in space, their output mimics a wavefront that is spherical, and thus the phase relationship can be pretty "flat". The same can be accomplished with a truncated line array using delay and shading instead of using a physically curved line array a la Keele, in fact he describes this approach in one of his CBT papers.
The key flaw in your argument was saying that the drivers all radiate the same signal. For the CBT they definitely do not, not with respect to phase and not with respect to amplitude either.
But even in the case that all drivers DO radiate the same signal, when there is an "infinite" line of them (speaking theoretically now) the various amplitudes and relative phases of each driver sum to a coherent wavefront with flat phase as well. The "infinite sum" works in your favor, much like many sinusoids can be added to represent an arbitrary signal, even to closely approximate a square wave.
On the other hand, for "small" line arrays with all drivers reproducing the same signal (phase and amplitude) all bets are off and the limited number of interfering sources and distances causes all sorts of problems. For this reason I could never really see the point of something like a 5-driver array, no matter how it's shaded.
Keyser,
Of course this is a good question. My perspective at this time however is to try to answer the inverse of that question. Why not shade a floor to ceiling array?
My goal, with a line array approach may be quite different than many or most. I want to increase vertical directivity. Over the frequency band that I am using the array I want no minimize / eliminate floor and ceiling reflections.
More generally, my main concern is the delayed wave-fronts arriving from a number of sources. This has an effect on the transient response. I am trying to determine empirically how much this impacts the quality of music reproduction
Hmmm, interesting. That makes sense.
I've measured various ribbons I own, and even the relatively small ones have phase response that's ugly. For instance, even when equalized flat, my Monsoon ribbons have poor phase response.
By comparisons sake, even my sloppy Synergy horns have flat phase from about 500hz to 13.5khz.
As I understood it, the phase issues of the Monsson, which is a flat ribbon that's about 20cm tall, are due to the pathlength differences from the center of the ribbon to the edges. IE, the center of the ribbon "leads" the edge of the ribbon.
Obviously, all of this will depend on whether the Monsoon is generating a truly flat wavefront.
TLDR: I've had a heck of a time getting good phase response from even relatively small (20cm) ribbons.
Sorry about my post regarding Patrick's statements, after every one else. I typed it the evening before and I did not realize what was going on in the morning, when I actually posted it.
However, floor to ceiling vs. CBT...
So, we have gone over, with the experts, that a strait line array, not operating floor to ceiling, cannot function well enough without some sort of shading or physical curve correction. Delay would be another option. This does not preclude a general rule (of thumb, if you will) that you cannot expect 2 or more drivers to radiate in phase, and constructively sum, if they are mounted more than 1 wavelength, of their highest radiation frequency, apart. 1/2 wavelength apart to some, would be the limit. Close spacing of the drivers creates that wave front line arrays are expected to produce. Tapering or shading fixes the polar at the extremes so that a smooth lobeless and very directive shape is obtained from the array.
If a floor to ceiling array can be configured, the outer extremes of the array are different and do not require shading. Curvature, as in the CBT configuration, would be considered equivalent to delay and is necessary for configurations that cannot be floor to ceiling. Amplitude shading has been proven by Keele to be necessary in either configuration if not considered floor to ceiling.
However, floor to ceiling vs. CBT...
So, we have gone over, with the experts, that a strait line array, not operating floor to ceiling, cannot function well enough without some sort of shading or physical curve correction. Delay would be another option. This does not preclude a general rule (of thumb, if you will) that you cannot expect 2 or more drivers to radiate in phase, and constructively sum, if they are mounted more than 1 wavelength, of their highest radiation frequency, apart. 1/2 wavelength apart to some, would be the limit. Close spacing of the drivers creates that wave front line arrays are expected to produce. Tapering or shading fixes the polar at the extremes so that a smooth lobeless and very directive shape is obtained from the array.
If a floor to ceiling array can be configured, the outer extremes of the array are different and do not require shading. Curvature, as in the CBT configuration, would be considered equivalent to delay and is necessary for configurations that cannot be floor to ceiling. Amplitude shading has been proven by Keele to be necessary in either configuration if not considered floor to ceiling.
Last edited:
Great, now we're all agreeing with each other, where's the fun in that?
Don't be too sure about that.
I am not sure there will be any delayed wave fronts if your using a tall ribbon or speaker drivers that are small relative to the wave length they are reproducing. My thinking is that it is only the segment of the speaker that is in ear height (or microphone hight) that will reach the ear or microphone. All sound is moving only in the horizontal plane. Experts, have I got it right?
Array weighting
Here is a little bit of a look at weighting schemes.
These are quick simulations of 16 element arrays, similar to the one I developed at McIntosh.
The plot is of the FFT of the weighting sequence. As such it gives the far field polar response of the array, at least for an angle sequence ranging from the front lobe to the first alias lobe.
If you are not familiar with array theory, arrays with a constant spacing typically have lobes that develop at the 90 degree angles and bend inward at higher frequencies (this is why we like to have good point density, i.e. lots of small drivers). Whatever the frequency of your polar, you will see this pattern mapped between the 0 degree point and wherever the first lobe is, then again from the first lobe to second, etc..
The takeaway is that looking at the weighting sequences we can compare the relative width of the front beam and the level of adjacent side lobes. A clean pattern with low side lobes is best.
Near field performance will be related to this but the lobes will be much wider (good).
Shown are 3 schemes for the shading of the 16 tweeters.
Red is 16 elements of equal strength. It has the highest gain (24dB over the strength of 1 tweeter) but the narrowest front beam and highest side lobes. This is what you get with an unshaded array.
Green dashed is the the McIntosh array. It was a simplified approach with 4 drive levels across the 16 tweeters. The drives were .3 .3 .7 .7 .9 .9 1 1 1 1 .9 .9 .7 .7 .3 3. That is the center tweeters were full strength. The first surrounding group was -1dB. The second surrounding group was -3dB and the outer group was -10dB. You can see it only loses a little bit of level on axis (maybe 3 dB) and is a little wider and also has a first side lobe of over 20 dB down. It comes back a little in the middle because I have the tweeters arranged with paired coefficients (the .3 .3 .7 .7, etc.). If those were split into individual weights then the middle rise would improve, but it is of no consequency in the near field at typical listening heights.
The black curve is a 16 coefficient raised cosine weighting. It has lost nearly 6dB from the unweighted array. It has clearly wider response and exemplary side lobe performance.
You pays your money and makes your choice.
I hope that makes sense and gives you a flavor for how this all works, and again, the far field performance that this represents looks much worse than performance nearer in.
David
Here is a little bit of a look at weighting schemes.
These are quick simulations of 16 element arrays, similar to the one I developed at McIntosh.
The plot is of the FFT of the weighting sequence. As such it gives the far field polar response of the array, at least for an angle sequence ranging from the front lobe to the first alias lobe.
If you are not familiar with array theory, arrays with a constant spacing typically have lobes that develop at the 90 degree angles and bend inward at higher frequencies (this is why we like to have good point density, i.e. lots of small drivers). Whatever the frequency of your polar, you will see this pattern mapped between the 0 degree point and wherever the first lobe is, then again from the first lobe to second, etc..
The takeaway is that looking at the weighting sequences we can compare the relative width of the front beam and the level of adjacent side lobes. A clean pattern with low side lobes is best.
Near field performance will be related to this but the lobes will be much wider (good).
Shown are 3 schemes for the shading of the 16 tweeters.
Red is 16 elements of equal strength. It has the highest gain (24dB over the strength of 1 tweeter) but the narrowest front beam and highest side lobes. This is what you get with an unshaded array.
Green dashed is the the McIntosh array. It was a simplified approach with 4 drive levels across the 16 tweeters. The drives were .3 .3 .7 .7 .9 .9 1 1 1 1 .9 .9 .7 .7 .3 3. That is the center tweeters were full strength. The first surrounding group was -1dB. The second surrounding group was -3dB and the outer group was -10dB. You can see it only loses a little bit of level on axis (maybe 3 dB) and is a little wider and also has a first side lobe of over 20 dB down. It comes back a little in the middle because I have the tweeters arranged with paired coefficients (the .3 .3 .7 .7, etc.). If those were split into individual weights then the middle rise would improve, but it is of no consequency in the near field at typical listening heights.
The black curve is a 16 coefficient raised cosine weighting. It has lost nearly 6dB from the unweighted array. It has clearly wider response and exemplary side lobe performance.
You pays your money and makes your choice.
I hope that makes sense and gives you a flavor for how this all works, and again, the far field performance that this represents looks much worse than performance nearer in.
David
Attachments
This is essentially a polar plot laid out as an XY rectangular plot. The left edge is the main lobe of the array (or at least half of it) and then to the right is the polar response as you move left or right of the on axis beam.
Now the width of this scales with frequency, but the basic shape is constant for all frequencies. For low frequencies the whole polar may just be defined by the first lobe of the plot. For higher frequencies the curve compresses (in polar space) until the full plot is contained in the 0 to 90 degrees of a polar curve (the emergence of the first major side lobe). At still higher frequencies you will see repetitions of the pattern. In theory the back half of the polar is identical, except for the fundamental polar curve of the drivers themselves suppress the back radiation.
The vertical axis is response level for a given angle and is relative to a the strength of a single element (a single element is 0 db).
This is probably hard to absorb if you haven't played with array modeling but looking at the diagrams of my McIntosh paper should clarify it some.
Regards,
David
Now the width of this scales with frequency, but the basic shape is constant for all frequencies. For low frequencies the whole polar may just be defined by the first lobe of the plot. For higher frequencies the curve compresses (in polar space) until the full plot is contained in the 0 to 90 degrees of a polar curve (the emergence of the first major side lobe). At still higher frequencies you will see repetitions of the pattern. In theory the back half of the polar is identical, except for the fundamental polar curve of the drivers themselves suppress the back radiation.
The vertical axis is response level for a given angle and is relative to a the strength of a single element (a single element is 0 db).
This is probably hard to absorb if you haven't played with array modeling but looking at the diagrams of my McIntosh paper should clarify it some.
Regards,
David
16 coefficients as ther are presumed to be 16 elements.
Not pairs of numbers but simply the calculated value of 2*pi/16 steps. That times the cosine curve giving a profile looking like a single cosine cycle but raised with zero at the ends and one in the middle. I guess that is 8 levels due to symmetry.
Is that Hann shading? I would have to look it up.
David
Not pairs of numbers but simply the calculated value of 2*pi/16 steps. That times the cosine curve giving a profile looking like a single cosine cycle but raised with zero at the ends and one in the middle. I guess that is 8 levels due to symmetry.
Is that Hann shading? I would have to look it up.
David
Don't be too sure about that.
I am not sure there will be any delayed wave fronts if your using a tall ribbon or speaker drivers that are small relative to the wave length they are reproducing. My thinking is that it is only the segment of the speaker that is in ear height (or microphone hight) that will reach the ear or microphone. All sound is moving only in the horizontal plane. Experts, have I got it right?
Jeno,
Yes there will be delayed wave fronts. Line arrays are inherently dispersive. This amounts to a non-flat phase response vs. frequency (not sure but the phase frequency relationship may also be non-linear). As one of the original paper authors (Lipshitz / Vanderkooy) put it, the impulse response "leaves a wake".
At every point along a line array you hear the vector sum of all the drivers in the array or in the case of a continuous line the vector sum of vanishingly small sections of the line. When all of these guys sum back together they are no longer a linear (differing only in mag and phase) replica of the original signal
The degree to which this smearing of the impulse response is perceptible has been difficult to find references for. Griffin mentions of assuming a precedence effect (Haas) - shading the outer drivers then has the late comers reduced in amplitude and therefore masked by the louder early arrivals. If you take this to the extreme though, you're left with a point source again. Some treatment of this seems necessary as it has an impact on how long your line can get before this effect swamps the benefit of the long line array.
If I'm reading this right:
1) the unshaded array is sending a significant fraction of the output into the floor and the ceiling, about 80%
2) The McIntosh array isn't sending much into the floor and the ceiling, but there are strong lobes off-axis. IE, there are vertical angles where the output is significant.
3) The newest array is almost like a flashlight. Nearly nothing is hitting the floor and ceiling. But the price is efficiency and maximum output. (I'll bet the crossover and processing is more complex also.)
I'm sorry, I come late on this topic but I have a basic question abour Keele's CBT.
I did simulations with my own software of various Keele's arrays presented in the 2011 AES paper "Performance Ranking of Loudspeaker Line Arrays", using same number of speakers, distances, angles,...
On some of those arrays, I don't get at all same results (see two examples). Especially for the one he ranked as best, the CBT circular shaded. Ie, in that special case, I have to eq considerably to get comparable responses.
Maybe my software does somethings wrong but when I compare with other published simulations (Ureda, Button, QSC), I get same results.
So, I'm interested if somebody has done same simulations (Dave ?) and got same curves as Keele's.
The whole set of Keele's array : here
If someone is interested, I can also give impulse responses of those simulations.
I did simulations with my own software of various Keele's arrays presented in the 2011 AES paper "Performance Ranking of Loudspeaker Line Arrays", using same number of speakers, distances, angles,...
On some of those arrays, I don't get at all same results (see two examples). Especially for the one he ranked as best, the CBT circular shaded. Ie, in that special case, I have to eq considerably to get comparable responses.
Maybe my software does somethings wrong but when I compare with other published simulations (Ureda, Button, QSC), I get same results.
So, I'm interested if somebody has done same simulations (Dave ?) and got same curves as Keele's.
The whole set of Keele's array : here
If someone is interested, I can also give impulse responses of those simulations.
Hi jlo,
Sorry but I have not made cirect simulations of Keele's designs and have tended to take his data at face value. Your simulations look somewhat like his but clearly exhibit differences too.
I do see that Don is plotting normalized plots, i.e. he subtracts/divides every curve by the response of the zero degree curve. Is it as simple as that?
What program are you using? Do you have any code you would want to share?
If your simulations match others well then I assume your approach is good and likely yours and Don's are based on different assumptions. You might contact Don at his web site as he is generally aproachable and would probably be interested in comparing designs.
David
Sorry but I have not made cirect simulations of Keele's designs and have tended to take his data at face value. Your simulations look somewhat like his but clearly exhibit differences too.
I do see that Don is plotting normalized plots, i.e. he subtracts/divides every curve by the response of the zero degree curve. Is it as simple as that?
What program are you using? Do you have any code you would want to share?
If your simulations match others well then I assume your approach is good and likely yours and Don's are based on different assumptions. You might contact Don at his web site as he is generally aproachable and would probably be interested in comparing designs.
David
Thanks!
Are you sure this apply to a virtually infinite line source as well? I guess so. Then there will be no such thing as a "cylindrical wave front".
I'm afraid it is'nt, because even when I normalize, I get different results from DB Keele.
You're right, I should contact him, thanks for this good idea. I remember his study of Bessel's array : I got exactly same curves for some arrays but not for all
Maybe I haven't understood all his assumptions or maybe my simulations have to be improved ?It is my own program and I cannot share it now due to some NDA's.
But if somebody wants me to simulate one special case, I can easily do it for a max of 16 point sources or 16 arrays of each 50 "Huygens" point sources.
Interested in BMR Line array / panel simulation
Hi Jlo,
When I get accurate measurements of the production BMR drivers I would be very happy to see what your software predicts....It would be fascinating to compare simulations to reality and if the two are close I will buy if you are selling!!
I have not had a convincing demo of any software that can accurately predict the BMR as point source, never mind as a line array or panel....
I had a similar problem trying to simulate the Manger driver a decade ago and that was just as point source...But software has moved on in leaps and bounds since then.
I have attached a pic of the LA 16's.
Can your software cope with the following scenario?...:
We are testing large panel arrays for live sound applications with up to 60 BMR's per panel (12 BMR's tall by 5 BMR's wide) with a separate sub below each panel (Xover 80Hz). Depending on size of audience we can stack the panels to create giant line arrays.
Each panel has 1 mic feed ie one panel on vocals, one on drums, one on keyboards, one on lead guitar etc.
Would you be able to simulate this...?
Thanks and all the best
Derek.
Hi Jlo,
When I get accurate measurements of the production BMR drivers I would be very happy to see what your software predicts....It would be fascinating to compare simulations to reality and if the two are close I will buy if you are selling!!
I have not had a convincing demo of any software that can accurately predict the BMR as point source, never mind as a line array or panel....
I had a similar problem trying to simulate the Manger driver a decade ago and that was just as point source...But software has moved on in leaps and bounds since then.
I have attached a pic of the LA 16's.
Can your software cope with the following scenario?...:
We are testing large panel arrays for live sound applications with up to 60 BMR's per panel (12 BMR's tall by 5 BMR's wide) with a separate sub below each panel (Xover 80Hz). Depending on size of audience we can stack the panels to create giant line arrays.
Each panel has 1 mic feed ie one panel on vocals, one on drums, one on keyboards, one on lead guitar etc.
Would you be able to simulate this...?
Thanks and all the best
Derek.
Attachments
I haven't listened to one but I know of someone who has built one. It will help to shade the line regardless of the length.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.