MDD Multi Delays Diffraction (Multi TL, omnidirectional, single drive, ...)

I think some of the dips in measurements can be caused by improper sealing at the driver face. I need to find some good thin gasket material to use.

The frontal acoustic load (MDDFL) does not always have to be sealed. If you want to try, you can leave an open space of a few mm between the driver and the beginning of the waveguides, you would add a ring that emits by 360 degree diffraction.

The open or closed frontal acoustic load at the base changes slightly the sound, after a few open prototypes I am now preparing a closed one. The choice is subjective.
 
I also use version 5.19 but I can see all the graphs, the problem may be related to the JRE. In my pc the 1.8.0_181 32-bit on Windoows 10 version is installed.

You're correct. I needed to reinstall REW to get the right JRE. It must have gotten removed in one of my system updates. I can see all the graphs now and I'll compare the model to the actual measurements. Thanks @claudiogan.
 
I am using the 64bit Linux version downloaded some time ago.

Edit: I think some of the dips in measurements can be caused by improper sealing at the driver face. I need to find some good thin gasket material to use.


Which compression driver did you use. ?

Yes, you should use a gasket to get a proper seal. I use thin (2mm) cork rubber gasket sheets and cut them to size.

.
 
measured data 8.5cm tube

I was looking at the data @pelanj provided and these are a few comments on that data before comparing the simulation results. I made a few adjustments to the measurement data. The IR window was increased to [1,4ms] and the IR were aligned used REW [Controls\->[Gen Min Phase]->[Estimate IR]-[Remove offset] (pic provided). I only used 1Khz-16Khz because most compression drivers (including this one) have a problem below 1Khz.

The second pic has 7 radial subtube measurements and one axial. Radial measurement #6 was suspect and removed. I should add that the time gate should be OK to about 6ms where floor bounce occurs in this setup. However, I see some differences as I adjust the window (3-5ms) suggesting something else was physically near the measurement setup.

It also looks like the compression driver has a high pass filter, however it won't effect these relative comparisons. You can see radial omni behaviour from 1Khz-2.4Khz as the waveforms are in phase and similar amplitudes. You can also see increasing directional behaviour from 8Khz-16Khz as the radial amplitudes decrease with frequency when compared to the axial measurement. In between 2.4Khz-8Khz there are various phase and amplitudes suggesting interactions between the subtubes and driver.
 

Attachments

  • REW - IR offset adj.jpg
    REW - IR offset adj.jpg
    45.4 KB · Views: 174
  • SubTubes Radial vs Axial FRD.jpg
    SubTubes Radial vs Axial FRD.jpg
    115.5 KB · Views: 176
subtube1 compared to sim

This is a comparison of the simulation to the measured results for subtube#1. I adjusted the sim gain to mimic the measured compression driver gain to make comparisons easier (red=actual, green=sim). Simulated in 4Pi space with the mic at distance of 1m at subtube height.

The table has calculated resonance frequencies for a typical tube that is closed one end and open the other end. There should be peaks at these frequencies and for the most part they are present in both actual and sim.

The compared results <2Khz are not so good and something is wrong here. I can only speculate that there may a leak somewhere. There is an occasional missing peak (ie 2.4Khz indicating there is either a leak or damping.

The remainder (2.4Khz-16Khz) generally has the peaks in the correct location. The sim has larger peak-peak excursions than the measured result probably because of differences between the simulation's ideal materials and real world materials.

The observations field is oriented through subtube#1 (right) and#5 (left). You can compare the field patterns to see omni (1K3Hz) , lobing (2K5Hz-3K7Hz), and increasing directivity (10Khz-16Khz. This is also seen in the measured results where the radial subtube outputs are in phase (omni), lobing (phase difference) and increasing directivity (increase freq). The sim is far from perfect but it does give some behaviour insight.

.
 

Attachments

  • ObsField_Subtube1_5@10KHz.jpg
    ObsField_Subtube1_5@10KHz.jpg
    28.6 KB · Views: 63
  • ObsField_Subtube1_5@5400Hz.jpg
    ObsField_Subtube1_5@5400Hz.jpg
    27.3 KB · Views: 61
  • ObsField_Subtube1_5@4100Hz.jpg
    ObsField_Subtube1_5@4100Hz.jpg
    28 KB · Views: 71
  • ObsField_Subtube1_5@3K7Hz.jpg
    ObsField_Subtube1_5@3K7Hz.jpg
    29 KB · Views: 66
  • ObsField_Subtube1_5@2556Hz.jpg
    ObsField_Subtube1_5@2556Hz.jpg
    27.4 KB · Views: 72
  • ObsField_Subtube1_5@1387Hz.jpg
    ObsField_Subtube1_5@1387Hz.jpg
    28.6 KB · Views: 65
  • ObsField-Subtube1+5.jpg
    ObsField-Subtube1+5.jpg
    46.2 KB · Views: 174
  • SubTube1 vs Simulation FRD.jpg
    SubTube1 vs Simulation FRD.jpg
    104.5 KB · Views: 175
  • Tube Resonances.jpg
    Tube Resonances.jpg
    80.2 KB · Views: 484
  • ObsField_Subtube1_5@16KHz.jpg
    ObsField_Subtube1_5@16KHz.jpg
    33.1 KB · Views: 68
I was looking at the data @pelanj provided and these are a few comments on that data before comparing the simulation results. I made a few adjustments to the measurement data. The IR window was increased to [1,4ms] and the IR were aligned used REW [Controls\->[Gen Min Phase]->[Estimate IR]-[Remove offset] (pic provided). I only used 1Khz-16Khz because most compression drivers (including this one) have a problem below 1Khz.

The second pic has 7 radial subtube measurements and one axial. Radial measurement #6 was suspect and removed. I should add that the time gate should be OK to about 6ms where floor bounce occurs in this setup. However, I see some differences as I adjust the window (3-5ms) suggesting something else was physically near the measurement setup.

It also looks like the compression driver has a high pass filter, however it won't effect these relative comparisons. You can see radial omni behaviour from 1Khz-2.4Khz as the waveforms are in phase and similar amplitudes. You can also see increasing directional behaviour from 8Khz-16Khz as the radial amplitudes decrease with frequency when compared to the axial measurement. In between 2.4Khz-8Khz there are various phase and amplitudes suggesting interactions between the subtubes and driver.

My MDD projects with the 3FE25 driver for are oriented towards understanding the psychoacoustic effects of the MDD technology. The pelanj prototype uses acoustic diffraction above all to make the emission a tweeter omnidirectional, with distances of only 85 mm, I don't think you can perceive the MDD effect. I can not do tests or even changes, the following considerations can be useful for those who want to experience diffraction in 2 or 3-way systems.

The interactions between the sub-tubes and between sub-tubes and drivers are influenced by the distance between drivers and waveguides. Raising the guides by 10 mm with respect to the driver activates the L / 2 resonances of the waveguides open on both sides, in my prototypes they help to regulate the response at low frequencies, the resonances are very evident in the progress of the electrical impedance (red line). If the goal is to have an omnidirectional emission, an open ring at the base of the waveguides mounted a few mm above the driver may be useful, the acoustic diffraction would generate a toroidal wavefront. A similar result can be obtained with a hole at the base of each waveguide. I don't have a mathematical model to size a prototype, I can only remember that at 20 KHz the wavelength is only 17 mm and a quarter corresponds to 4 mm. A 10 mm object can influence the directivity of the emission.
 

Attachments

  • 34c9 blue damped L.jpg
    34c9 blue damped L.jpg
    85.6 KB · Views: 67
This is a comparison of the simulation to the measured results for subtube#1. I adjusted the sim gain to mimic the measured compression driver gain to make comparisons easier (red=actual, green=sim). Simulated in 4Pi space with the mic at distance of 1m at subtube height.

The table has calculated resonance frequencies for a typical tube that is closed one end and open the other end. There should be peaks at these frequencies and for the most part they are present in both actual and sim.

The compared results <2Khz are not so good and something is wrong here. I can only speculate that there may a leak somewhere. There is an occasional missing peak (ie 2.4Khz indicating there is either a leak or damping.

The remainder (2.4Khz-16Khz) generally has the peaks in the correct location. The sim has larger peak-peak excursions than the measured result probably because of differences between the simulation's ideal materials and real world materials.

The observations field is oriented through subtube#1 (right) and#5 (left). You can compare the field patterns to see omni (1K3Hz) , lobing (2K5Hz-3K7Hz), and increasing directivity (10Khz-16Khz. This is also seen in the measured results where the radial subtube outputs are in phase (omni), lobing (phase difference) and increasing directivity (increase freq). The sim is far from perfect but it does give some behaviour insight.

.

The DonVK sim is perhaps the first in which the search for the effects of diffraction rather than those of direct emission (front or rear) is the main objective. It is not surprising that there are corrections to be made, both in the model used for the simulation and in the interpretation of the measurement data, they are the results of software applications that are not optimized for the study of acoustic diffraction. The pelanj prototype can also be optimized with respect to directivity. I regret not being able to help in these searches for the moment, I can only list aspects of the system that intrigue me more than others, on which I still have ideas to verify.

The sound field graphs show the directivity but also a general decrease in the sound pressure emitted with increasing frequency. There are no dissipative parts in the system except for any turbulences created by the geometry of the waveguides. In the sim made by DonVK there is also a relationship between L / 4 resonances and filtered frequency response. With this hypothesis it is assumed that only the frequencies in resonance with the L / 4 rule can transfer energy from the driver to the listening environment.

Increasing the space between the driver and the beginning of the waveguides to 10 - 15 mm also activates the L / 2 resonances which provide additional frequencies to transfer sound energy from the driver to the listening environment. In my previous graph with the IR window = 0 msec it makes the frequency response more regular, perhaps in the sim there would be a higher emission at higher frequencies.
 

Attachments

  • mdd pelanj IRWindow left window ms0.jpg
    mdd pelanj IRWindow left window ms0.jpg
    87.3 KB · Views: 69
An integration to the previous post. As seen in the table prepared by DonVK

844461d1589597152-mdd-multi-delays-diffraction-multi-tl-omnidirectional-single-drive-tube- resonances-.jpg


part of the wavelengths between 1 and 20 KHz are comparable with the size of the waveguides. These lengths recreate the same problems as TL resonances at low frequencies.

It would be interesting to see what happens with longer lengths, 300 - 500 mm. Jzagaja sent us the video of the first layer of a 500 mm guide, I hope it will update us soon on the end of the work and the results obtained.
 
The DonVK sim is perhaps the first in which the search for the effects of diffraction rather than those of direct emission (front or rear) is the main objective. It is not surprising that there are corrections to be made, both in the model used for the simulation and in the interpretation of the measurement data, they are the results of software applications that are not optimized for the study of acoustic diffraction.

The ABEC sim is a BEM acoustic simulation and it certainly includes direct emission/reflections/diffractions. Maybe not perfect, but the correlation between the measurements and the sim model is good. Have a look at R&D-Team - Software Development (the tool was renamed Akabak recently).



The sound field graphs show the directivity but also a general decrease in the sound pressure emitted with increasing frequency. There are no dissipative parts in the system except for any turbulences created by the geometry of the waveguides. In the sim made by DonVK there is also a relationship between L / 4 resonances and filtered frequency response. With this hypothesis it is assumed that only the frequencies in resonance with the L / 4 rule can transfer energy from the driver to the listening environment.

No such assumption was made in the simulation. Again, look at @pelanj's measured results. Do you see a nice smooth constant amplitude curve vs frequency ? The IR offset=0 and the window needs to be short (4ms) to exclude room effects and the scale must be meaningful 5db/div or less.

Increasing the space between the driver and the beginning of the waveguides to 10 - 15 mm also activates the L / 2 resonances which provide additional frequencies to transfer sound energy from the driver to the listening environment. In my previous graph with the IR window = 0 msec it makes the frequency response more regular, perhaps in the sim there would be a higher emission at higher frequencies.

Again, look at the measured results provided by @pelanj.
 
Last edited:
An integration to the previous post. As seen in the table prepared by DonVK

It would be interesting to see what happens with longer lengths, 300 - 500 mm. Jzagaja sent us the video of the first layer of a 500 mm guide, I hope it will update us soon on the end of the work and the results obtained.

The sim model has already been completed.

Hopefully @jzagaja will be willing to publish his results (8 radial + 1 axial) at 1m distance and 1m from the floor. I believe he is using a 2" compression driver (closed back) so only sound emits from the tubes.
 
MDD3FE25d project

After a couple of months I managed to complete version d of the MDD3FE25 project. As expected, the increase in length of the rear acoustic load to 2 meters has improved the response to low frequencies. The time available allowed me to try various configurations for the front acoustic load, the best result is that with a front loading 23FL7d (1600 mm with resonances over 3 octaves), the maximum length is close to the maximum length of the rear loading 21BL7d ( 2000 mm with resonances on one octave). The perception of high-pitched sounds has also improved, from the measurements I do not detect the difference that is perceived when listening. The difference in distortion does not justify it, on the contrary it is just higher. Guides up to 2 meters long create more vibration and are more difficult to glue deafly. Pending confirmation or denial, I attribute the credit to the further optimized MDD effect.

In this period I realized that I underestimated the problems related to the omnidirectional emission of frequencies with a wavelength comparable with the size of the waveguides. I cannot simulate the behavior of my prototypes with a numerical model, empirical listening tests show that short frontal acoustic loads give worse results than longer acoustic loads.

The front load 23FL7d in the latest MDD3FE25d prototype is airtight and the 7 frontal emission points have the same diffraction geometry and emit over the whole spectrum. The 7 rear emission points emit frequencies up to 1 KHz. The distribution of resonances on one and three octaves makes the cabinet acoustically neutral.

For some time I will listen to music with these two meter high desk speakers which however have the advantage of occupying only 20 x 20 cm on the ground. They could also be wall mounted with a bracket and replace a driver array.

Those who want to try to make a replica do not forget that the MDD3FE25d project uses a 3 ” driver, it has many advantages but cannot generate high SPL.

Other links:
https://www.claudiogandolfi.it/
https://www.claudiogandolfi.it/mdd3fe25.html#d
https://www.claudiogandolfi.it/mddfl.html#23fl7d
https://www.claudiogandolfi.it/mddbl.html#21bl7d
 

Attachments

  • 3fe25-900.jpg
    3fe25-900.jpg
    33.4 KB · Views: 291
  • 3fe25-b907.jpg
    3fe25-b907.jpg
    150.6 KB · Views: 292
  • 3fe25d left FR-red_avg FL-green BL-blue.jpg
    3fe25d left FR-red_avg FL-green BL-blue.jpg
    84.8 KB · Views: 190
  • 3fe25-vv.jpg
    3fe25-vv.jpg
    133.6 KB · Views: 296
Last edited:
smoothing output ripple

I've been thinking about a concept that was started in the K-tube thread 3D printed K-Tube for 1" compression drivers but after modelling @claudiogan's MDD concept it inspired an idea and here are some results that I'd like to share.

First, I need to correct a formula mistake in the table of calculated resonant freq vs tube lengths that I posted earlier. I discovered the formula mistake in my textbook (Horn Loudspeakers, Kolbrek+Dunker, pg 470, eqn 12.5.10). It is why some of the peaks were "apparently" missing when comparing the sim to the actual. I also simulated a single tube to make sure the table and sim results match (that's how I found the formula error). The avg tube length table is show below (tube dia=20mm, tube mouting plate dia=75mm).

Only the odd numbered freq multiples should have peaks. So in order to spread out the peaks using 8 tubes (from 1freq->3freq) , I used "8th root of 3" (2^0.125) as a multiplier from tube "m" to tube "m+1". I also chamfered the tube mouth to get a less abrupt termination. The "avg" length of the tube was used (mid chamfer length). This is not as good as the K-tube termination, but was just easier in CAD :) . I also oriented all outputs in the same direction to make it less polydirectional.

The peak spacing looks good and all 8 tube outputs are shown below. A point worth noting is the K-tube also throws sound at 45deg and progressively tilts towards axial as the frequency increases. I measured (in the simulator) at 1m and 22.5deg off axis from the shortest tube. Some of the SPL tilt is due to this effect as well as directivity increasing with freq.

.
 

Attachments

  • Seperate Ctubes EqualDrive SPL @1m.jpg
    Seperate Ctubes EqualDrive SPL @1m.jpg
    138 KB · Views: 85
  • Ctube8-SPL@1m.jpg
    Ctube8-SPL@1m.jpg
    39.8 KB · Views: 75
  • Ctube8-RadImp.jpg
    Ctube8-RadImp.jpg
    30.6 KB · Views: 90
  • Ctubex8.jpg
    Ctubex8.jpg
    27.8 KB · Views: 95
  • TubeLengths.jpg
    TubeLengths.jpg
    68.5 KB · Views: 88
Last edited:
damping to reduce ripple

All the tubes were seperately driven with a constant velocity source in the graph below.

The "red" curve is a simple addition of all 8 separate tube's simulated outputs. The tubes still have relatively high Q and you can see that in the peaks.

The "green" curve is from driving all 8 tubes simultaneously (8 sources). The difference between the red and green is the effect of interaction between the physical tube openings. Keep in mind this is still 8 sources, not 1 shared source.

The "black" curve is when damping is applied to the tubes (all driven). I used D=5% Ft=1Khz which low. I'm not sure how much polyfill / fibreglass / cotton balls that is but it would be easy to put loose fill in a tube and measure the nearfield effect. Damping is also commonly used in TLs to smooth out these resonances.

In order to reproduce this from a single driver, the acoustic impedance of each tube needs to be similar because we will have 8 parallel loads on one driver. The impedance of a tube is a function of length and area ~f(L,A) . That part is next.

.
 

Attachments

  • All Ctubes - added - combined - damped.jpg
    All Ctubes - added - combined - damped.jpg
    101 KB · Views: 87
Member
Joined 2008
Paid Member
DonVK, as always, very interesting post. I damped mine with a small cube of open cell foam and I perceived too much high end loss, so I halved the cubes and that is how it stayed. I am currently printing the second holder, the first one was good on first try, the second one looks like cursed.