Diffraction - more perimeter
Previously, I tried an exaggerated fractal like perimeter to see if the perimeter shape could effect low diffraction patterns.
This time I tried smaller, more regular shapes to see if I could improve things. Unfortunately, only the bigger perimeter changes seems to make a difference. These smaller triangles seemed to make virtually no difference for the woofer.
I did try the triangles on the tweeter disc and mid cone perimeter and they do make a difference for reducing HF diffraction. Probably the shorter wavelength.
Pics#1 show the triangle sizes. I was trying to mimic a 4 cycle sine wave wrapped around the perimeter. Pic#3 shows the comparison effect (none) to 2 different triangle sizes. Pic#4 shows the effect of small circle lobes on the woofer and triangles on the mid-cone that effects HF mostly.
Not finished yet.
Previously, I tried an exaggerated fractal like perimeter to see if the perimeter shape could effect low diffraction patterns.
This time I tried smaller, more regular shapes to see if I could improve things. Unfortunately, only the bigger perimeter changes seems to make a difference. These smaller triangles seemed to make virtually no difference for the woofer.
I did try the triangles on the tweeter disc and mid cone perimeter and they do make a difference for reducing HF diffraction. Probably the shorter wavelength.
Pics#1 show the triangle sizes. I was trying to mimic a 4 cycle sine wave wrapped around the perimeter. Pic#3 shows the comparison effect (none) to 2 different triangle sizes. Pic#4 shows the effect of small circle lobes on the woofer and triangles on the mid-cone that effects HF mostly.
Not finished yet.
Attachments
As a circle is the worst shape for diffraction due to the total symmetry, maybe some of the triangles being bigger than others or making the baffle more oblong might help, making the small ones less regular in position might help too. Or try and use a QRD type pattern to set the relative sizes of them around the edge.
A bit like offsetting a driver on a baffle reduces the diffraction from it.
A bit like offsetting a driver on a baffle reduces the diffraction from it.
QRD is interesting as it has irregularity to it. A repeating circular QRD would be worth trying. Although it is ironic trying to insert irregularity into something inherently trying to be regular and symmetric (omnidirectional)
Last edited:
Diffraction - trying the HF side
This set of tests is on the mid disc and cones used for HF. The wavelengths are shorter and smaller features may work here.
There is an effect but again its not large. I have gone to 1/12 smoothing in some graphs as its easier to see the differences and the diffraction at HF is smaller so the vertical scale is now 2db/div. I am getting the impression that I'm pushing on a cloud. There are probably multiple diffraction sources, and I may have to change multiple areas otherwise the problem just shifts alittle.
Pic#1,#2,#3, #4 show the mid disk perimeter mod configurations. Pic#5 shows the effects using 1/12 smoothing (minimal) for 3/6/9 regular spaced triangles. Pic#6 show 6 triangles with regular and irregular spacing. Pic#7 shows all combinations at 1/24 smoothing.
This set of tests is on the mid disc and cones used for HF. The wavelengths are shorter and smaller features may work here.
There is an effect but again its not large. I have gone to 1/12 smoothing in some graphs as its easier to see the differences and the diffraction at HF is smaller so the vertical scale is now 2db/div. I am getting the impression that I'm pushing on a cloud. There are probably multiple diffraction sources, and I may have to change multiple areas otherwise the problem just shifts alittle.
Pic#1,#2,#3, #4 show the mid disk perimeter mod configurations. Pic#5 shows the effects using 1/12 smoothing (minimal) for 3/6/9 regular spaced triangles. Pic#6 show 6 triangles with regular and irregular spacing. Pic#7 shows all combinations at 1/24 smoothing.
Attachments
Diffraction - cabinet effects
The last item to try is the cabinet. That is the upper 2 discs supporting the waveguide, and the lower woofer BR cabinet. Both these "ends" have transitions between a disc and a box, or a gap between disks. Just like a water pond ripple striking stick or rock the diffraction should happen at any impedance or geometry change.
The transition boundaries were changed by using 6mil plastic to reshape the boundary. This is interesting because it effects the final packaging. My sandbox has many transitions to make it modular, and its been modified many times to solve earlier more pressing issues. I wonder if I hear the difference without a proper double bind test?
Pics#1, #2 show the "wrapping" to change some of the boundaries. Pic#3 shows just the wrapping effect and Pic#4 shows wrapping and 6 irregular spaced triangles.
The last item to try is the cabinet. That is the upper 2 discs supporting the waveguide, and the lower woofer BR cabinet. Both these "ends" have transitions between a disc and a box, or a gap between disks. Just like a water pond ripple striking stick or rock the diffraction should happen at any impedance or geometry change.
The transition boundaries were changed by using 6mil plastic to reshape the boundary. This is interesting because it effects the final packaging. My sandbox has many transitions to make it modular, and its been modified many times to solve earlier more pressing issues. I wonder if I hear the difference without a proper double bind test?
Pics#1, #2 show the "wrapping" to change some of the boundaries. Pic#3 shows just the wrapping effect and Pic#4 shows wrapping and 6 irregular spaced triangles.
Attachments
Baseline myself - expectations
I decided to compare a commercial speaker (2.5 way Paradigm 7v5) in the exact same manner (location, power, method) to see how far I need to go with this. Just trying not to go OCD here.
The Paradigm is clearly more efficient as it only transmits in a relatively narrow cone (<90 deg) for most freq. The Omni transmits 360deg so I expect at least -6dB difference. The question is "are these FR ripples causing me real problems". Usually the FR is 1/3 smoothed because it's most reflective of how our ears respond, but unfortunately it also obscures all analytic detail, so I used 1/24 for most.
Since these are 2 very different speakers, the common parts are probably the room. Whats left are artifacts from the speakers. From my perspective they are different but comparable in graph performance. They have similar LF room responses, about 5db pk-pk hash from artifacts, the Paradigm distortion is better at -35db vs Omni -30db.
Pic#1 is 1/24 FR comparison, Pic#2 is Omni distortion, Pic#3 is Paradigm distortion, and Pic#4 is 1/3 FR.
I decided to compare a commercial speaker (2.5 way Paradigm 7v5) in the exact same manner (location, power, method) to see how far I need to go with this. Just trying not to go OCD here.
The Paradigm is clearly more efficient as it only transmits in a relatively narrow cone (<90 deg) for most freq. The Omni transmits 360deg so I expect at least -6dB difference. The question is "are these FR ripples causing me real problems". Usually the FR is 1/3 smoothed because it's most reflective of how our ears respond, but unfortunately it also obscures all analytic detail, so I used 1/24 for most.
Since these are 2 very different speakers, the common parts are probably the room. Whats left are artifacts from the speakers. From my perspective they are different but comparable in graph performance. They have similar LF room responses, about 5db pk-pk hash from artifacts, the Paradigm distortion is better at -35db vs Omni -30db.
Pic#1 is 1/24 FR comparison, Pic#2 is Omni distortion, Pic#3 is Paradigm distortion, and Pic#4 is 1/3 FR.
Attachments
You can also try the Frequency Dependent Windowing feature for comparison. Click on IR windows, add frequency dependant window choose cycles or octaves and the amount and apply.
The frequency responses are different enough for that alone to account for any preference. If you were to EQ the paradigm to resemble the omni that would be a good test of the differences between them.
The frequency responses are different enough for that alone to account for any preference. If you were to EQ the paradigm to resemble the omni that would be a good test of the differences between them.
Last edited:
I have tried the windowing feature but I don't have a good basis for adjusting it. I agree that EQ can make them similar but the hash cannot be reduced buy EQ alone, its too narrow for the EQ bands. It seems to be the residue.
The previous graphs were to determine if the amount of ripple or hash at 1/24th was abnormal for this Omni. I still think its diffraction based and I'm still trying to reduce it, but it seems like it's similar to a conventional speaker. I was surprised by that.
The previous graphs were to determine if the amount of ripple or hash at 1/24th was abnormal for this Omni. I still think its diffraction based and I'm still trying to reduce it, but it seems like it's similar to a conventional speaker. I was surprised by that.
When I suggested EQ I meant in broad terms, to have a similar low end rolloff and to reduce the peaks that are more prominent in the Paradigm. Not to try and make then both a straight line as there is no chance of that working well. If you moved the mic a little the hash pattern would be different but the peaks and dips would be broadly similar.
If you EQ'd the two based on a FDW version they would likely sound more similar in tonality and the differences from polar pattern, diffraction etc would then be easier to compare.
The idea of the FDW is to apply a progressively longer window as frequency goes down. It is another way of trying to average out the room response (in time). The REW version is quite smoothed when the cycle count is low. The downside to upping the cycle count is that it lets in more of the room. Bit of a catch 22.
For in room measurements spatial averaging by taking multiple measurements at different positions within a defined area is a good way of getting the most persistent room features along with the speaker response but averaging out the rest.
It seems to work well for others (I plan to do it myself when I get the chance), the averaging really does reduce the hash that comes from the room down by quite a large margin, the more measurements you average the more signal and less noise you get.
I imagine a lot of the hash coming from room interaction and the peak and dip pattern from the diffraction underneath.
If you can try the multiple measurements as shown in the rephase thread, that will give you a better idea of what the speaker response is.
If you EQ'd the two based on a FDW version they would likely sound more similar in tonality and the differences from polar pattern, diffraction etc would then be easier to compare.
The idea of the FDW is to apply a progressively longer window as frequency goes down. It is another way of trying to average out the room response (in time). The REW version is quite smoothed when the cycle count is low. The downside to upping the cycle count is that it lets in more of the room. Bit of a catch 22.
For in room measurements spatial averaging by taking multiple measurements at different positions within a defined area is a good way of getting the most persistent room features along with the speaker response but averaging out the rest.
It seems to work well for others (I plan to do it myself when I get the chance), the averaging really does reduce the hash that comes from the room down by quite a large margin, the more measurements you average the more signal and less noise you get.
I imagine a lot of the hash coming from room interaction and the peak and dip pattern from the diffraction underneath.
If you can try the multiple measurements as shown in the rephase thread, that will give you a better idea of what the speaker response is.
Omni V4
I've tried many experiments trying to figure out why I have this FR ripple (hash) and was trying to reduce it. The last changes to V3 were attempts to identify the source as diffraction, mechanical resonances, or reflections, etc. The once thing I could not easily change was the overall outer shape which impacts diffraction.
Omni V4 has rounded, tapered shapes above and below the exit discs to allow wavefront expansion along a smooth single surface.
Pic#1 is the initial V4 mechanical, Pic#2 is the initial measurements with EQ and no damping materials added.
I've tried many experiments trying to figure out why I have this FR ripple (hash) and was trying to reduce it. The last changes to V3 were attempts to identify the source as diffraction, mechanical resonances, or reflections, etc. The once thing I could not easily change was the overall outer shape which impacts diffraction.
Omni V4 has rounded, tapered shapes above and below the exit discs to allow wavefront expansion along a smooth single surface.
Pic#1 is the initial V4 mechanical, Pic#2 is the initial measurements with EQ and no damping materials added.
Attachments
No problem
In that spirit do you think you could bring the black plastic form flush with the baffle or put a roundover on the particle board? Seems like there is a sharp transition there. I think your pool noodle test before wasn't effective because of the type of material and there was a bit of a gap right at the edge.
Put an LED inside and the new clothes make it look like a big lamp
No problem
Yes, and you can tell from the 830Hz null caused by a 0.4m baffle. Its still there only alot smaller now. A proper edge wouldn't hurt. It looks like the 400null-500pk-600null is back from no damping in the woofer chamber. Overall, there are 1/2 the peaks and nulls and alot less hash than V3 and I'm happy so far. I'll publish more results as I add back damping and improve edge radius. I like to make one change at a time to keep some sense of causality.
Maybe, its hard to tell. I find when there are multiple issues the energy just takes the next easiest route. I did change the perimeter on V3 a few times with no measurable effect. No harm in trying again.
Put an LED inside and the new clothes make it look like a big lamp
I was going for "sonic egg" or maybe "warp core" with the right vertical pulse lighting. Add the pool noodle to that light and it could be a harbor buoy when I'm tired of it.
A very sound plan.
I think that is the key, if there is a gap before the radius the sound has already gone down it and some diffraction will happen. One of the reasons that a sphere works so well for diffraction, there is no other path for the sound to take.
I was going for "sonic egg" or maybe "warp core" with the right vertical pulse lighting. Add the pool noodle to that light and it could be a harbor buoy when I'm tired of it.
I can see "warp core" if you chrome the reflector and add some pulse lighting
Now that you have reduced the hash down here is an example (#493 in the convolution thread) of the type of minimum phase EQ DRC can generate if you use a long window (10 cycles)
If you use a highly averaged measurement then you can increase the window of correction more to get a finer grained filter without having to use an awful lot of PEQ filters. Not advisable with a single point measurement though.
Changes from V3 to V4
I should list the changes that occurred from V3 to V4.
The middle reflector cone diameters were reduced (from 40cm to 32cm dia) to make the inner and outer sound path lengths equal, as well as roughly equal for the LF and HF sections.
The LF cone shape was changed to produce an exponential increase in spherical wavefront area. The HF coaxial waveguides already provide exponential increase. This brought the new cone shape closer to the woofer.
A new "skin" has placed over the exposed discs as they were though to be contributing to the ripple + hash in the freq response.
The Omni woofer chamber was changed from 32L ABC back to a 16L BR. I no longer believe I can get the low freq (<80Hz) response I need from an 8inch woofer. I get a reasonable measurement (@1m) but it does not project effectively into the room. I actually want a good response to 30Hz and I also need a boost of nearly 10db to account for less sensitive hearing in the VLF range. So I now use a 10inch active subwoofer XO at 80Hz, and listen to the Omnis in a 2.1 setup. As an added bonus it means a smaller cabinet.
The speakers now require a stand to bring the mid-cone level to ear level while sitting. Reworking the woofer chamber made the box smaller and it was no longer able to sit directly on the floor.
I should list the changes that occurred from V3 to V4.
The middle reflector cone diameters were reduced (from 40cm to 32cm dia) to make the inner and outer sound path lengths equal, as well as roughly equal for the LF and HF sections.
The LF cone shape was changed to produce an exponential increase in spherical wavefront area. The HF coaxial waveguides already provide exponential increase. This brought the new cone shape closer to the woofer.
A new "skin" has placed over the exposed discs as they were though to be contributing to the ripple + hash in the freq response.
The Omni woofer chamber was changed from 32L ABC back to a 16L BR. I no longer believe I can get the low freq (<80Hz) response I need from an 8inch woofer. I get a reasonable measurement (@1m) but it does not project effectively into the room. I actually want a good response to 30Hz and I also need a boost of nearly 10db to account for less sensitive hearing in the VLF range. So I now use a 10inch active subwoofer XO at 80Hz, and listen to the Omnis in a 2.1 setup. As an added bonus it means a smaller cabinet.
The speakers now require a stand to bring the mid-cone level to ear level while sitting. Reworking the woofer chamber made the box smaller and it was no longer able to sit directly on the floor.
ABEC3 BEM sims for Omni V4
These are updated BEM simulation result for V4. I would like to understand how the sound is projected and get some insight into how it interacts with the cabinet. This model uses 1/4 symmetry in BEM and includes the LR4 filters in LEM (AKABAK-ish).
The first figure shows the BR port activity at 70Hz in a Y-plane view. The port is only useful to extend the LF and it does does not contribute above 150Hz. This model uses 2 BR ports (symmetry) that have the same area as a single 5cm x 10cm port.
My sitting listening position height is at Z=85cm. The figures below show the Y-plane@+/-1m x 1.5m and Z-plane@85cm in a isometric view. This is a good way to see the 3D effect of the sound field. You can clearly see the circular polar response in the Z-plane as concentric rings (omni) projected to the sitting listening plane. Also interesting is the Y-plane where you can see vertical lobing occur at the top of the cabinet at 1000Hz. This is the same as you'd expect in the axial response of a woofer at higher frequencies.
You can see the sound pressure roll over all the cabinet components. So a smooth circular exterior is helpful. I still have a few "edges" that do not have nice large radius where some diffraction does occur. There will more posts to see the modelling prediction effects on modifying those edges.
These are updated BEM simulation result for V4. I would like to understand how the sound is projected and get some insight into how it interacts with the cabinet. This model uses 1/4 symmetry in BEM and includes the LR4 filters in LEM (AKABAK-ish).
The first figure shows the BR port activity at 70Hz in a Y-plane view. The port is only useful to extend the LF and it does does not contribute above 150Hz. This model uses 2 BR ports (symmetry) that have the same area as a single 5cm x 10cm port.
My sitting listening position height is at Z=85cm. The figures below show the Y-plane@+/-1m x 1.5m and Z-plane@85cm in a isometric view. This is a good way to see the 3D effect of the sound field. You can clearly see the circular polar response in the Z-plane as concentric rings (omni) projected to the sitting listening plane. Also interesting is the Y-plane where you can see vertical lobing occur at the top of the cabinet at 1000Hz. This is the same as you'd expect in the axial response of a woofer at higher frequencies.
You can see the sound pressure roll over all the cabinet components. So a smooth circular exterior is helpful. I still have a few "edges" that do not have nice large radius where some diffraction does occur. There will more posts to see the modelling prediction effects on modifying those edges.
Attachments
Baffle edge radius vs baffle size - rechecking an experiment
I previously tried different baffle edge shapes before putting a new skin on the Omni. In particular the woofer edge. I got very little return for my efforts and then I thought that the FR ripple was from "other" edges and it was swamping the one item I was testing. So a new skin was added and its time to retry that experiment. How important is the baffle edge radius vs the baffle size.
Initial attempts are at : OmniDirectional - work in progress
The baffle diameter diffraction peaks and nulls calculated at : OmniDirectional - work in progress . Observed 40cm baffle FR peaks [291, 547, 1130Hz] and dips [429, 843, 1350] and changing the edge radius should change the magnitude but not the frequency (baffle diameter dependant).
My edge radius is exaggerated for this test, and the baffle diameter is still 40cm. With the FR hash reduced due to the new skin, you can now see the predicted effect at the predicted frequencies. Its no longer masked.
I previously tried different baffle edge shapes before putting a new skin on the Omni. In particular the woofer edge. I got very little return for my efforts and then I thought that the FR ripple was from "other" edges and it was swamping the one item I was testing. So a new skin was added and its time to retry that experiment. How important is the baffle edge radius vs the baffle size.
Initial attempts are at : OmniDirectional - work in progress
The baffle diameter diffraction peaks and nulls calculated at : OmniDirectional - work in progress . Observed 40cm baffle FR peaks [291, 547, 1130Hz] and dips [429, 843, 1350] and changing the edge radius should change the magnitude but not the frequency (baffle diameter dependant).
My edge radius is exaggerated for this test, and the baffle diameter is still 40cm. With the FR hash reduced due to the new skin, you can now see the predicted effect at the predicted frequencies. Its no longer masked.
Attachments
Corrections using REW, RePhase, EqualizerAPO
I think the basic design is close enough to justify some fine tuning for improvements. There are 3 tools that have been suggested by more than a few members [Fluid, Wesayso, Pano, etc] to correct the speaker response using DSP techniques. They are discussed at length in multiple threads like rePhase, a loudspeaker phase linearization, EQ and FIR filtering tool and SwissBear has an excellent tutorial at https://www.dropbox.com/s/10xdhh83jokzbxv/REW_rePhase_tuto.pdf?dl=0
Multiple measurements were made (using REW) to average out the reflections and (hopefully) separate each speaker from the room. The corrections (using rePhase) were run on a PC (using EqualizerAPO) as well as separately on an RPI3B (using BruteFir).
Bottom line, it's fixed the minor issues (w.r.t detail, clarity) I still had with the speakers. This is the best they've sounded.
Pic#1 shows the avg "pre" FR (green), the avg "post" FR (red) and the rephase correction (blue, was shifted to top). The correction shows its the inverse of the "pre" measurement and the red "post" curve is amazingly flat. Pic#2 show the measured min phase (grey) response of a single measurement. Pic#3 show the impulse response, which is not as clean as I'd like and I'm not sure why. These are all measured at my listening position 2.4m from the speakers with no room treatment and no subwoofer.
I think the basic design is close enough to justify some fine tuning for improvements. There are 3 tools that have been suggested by more than a few members [Fluid, Wesayso, Pano, etc] to correct the speaker response using DSP techniques. They are discussed at length in multiple threads like rePhase, a loudspeaker phase linearization, EQ and FIR filtering tool and SwissBear has an excellent tutorial at https://www.dropbox.com/s/10xdhh83jokzbxv/REW_rePhase_tuto.pdf?dl=0
Multiple measurements were made (using REW) to average out the reflections and (hopefully) separate each speaker from the room. The corrections (using rePhase) were run on a PC (using EqualizerAPO) as well as separately on an RPI3B (using BruteFir).
Bottom line, it's fixed the minor issues (w.r.t detail, clarity) I still had with the speakers. This is the best they've sounded.
Pic#1 shows the avg "pre" FR (green), the avg "post" FR (red) and the rephase correction (blue, was shifted to top). The correction shows its the inverse of the "pre" measurement and the red "post" curve is amazingly flat. Pic#2 show the measured min phase (grey) response of a single measurement. Pic#3 show the impulse response, which is not as clean as I'd like and I'm not sure why. These are all measured at my listening position 2.4m from the speakers with no room treatment and no subwoofer.
