[diy speaker guy]

There's a lot of information here so for the people that don't want to follow the process to see how I conducted this experiment I'm showing my conclusions at the beginning (and also at the end). The conclusions are all you really need to read if that's all you are interested in.


Conclusions


When I reverse engineered the test subject plans using the advanced centerline method and the centerline method, the advanced centerline method resulted in a simulation that was closer to the measurements of the actual horn.

For this reason I believe the advanced centerline folding method is superior to the centerline folding method for both engineering and reverse engineering horns.

It should be noted that this study was conducted using a horn with no reflectors in any of the bends. Another study needs to be done to figure out how to properly address the situation of reflectors.

 

[diy speaker guy]

Introduction


Soho54 has been MIA for a couple of years now but while he was around he was one of the best resources for horn info in the forums. From his EZ horn design spreadsheet to his Akabak tuturial to his horn folding using sketchup tutorial, this guy did all the heavy lifting and gave us his results on a silver platter. And although I've only actually read a small percentage of his posts on the various forums he participated in, I've never seen anything to make me question any of the information he shared. So when he makes a comment like this I immediately start using his method without question.

I came up with it (the advanced centerline folding method) through a lot of research, and trial and error. About two years ago I got feed up with the "common knowledge" surrounding horns, so I threw it all out and started over from scratch. I read everything I could by Keele, Edgar, Leach, and Danley looking for common threads. I have pages and pages of excel work exploring their different papers, and ideas from posts. I then went back to Rayleigh, found measured data on acoustic pressure around bends, and sourced fluid dynamic, FEA, and BEM simulations.

I then spent a lot of time trying different ideas out, coming from the view point of getting the simulation to match reality. I would work until I found something that worked for a certain folded horn, and then tried it on another. You can find different ways to sim a single horn, but most of them fail when you try it on another. I kept redoing things until I found a way that works on all the horns I have tried it on.

In truth, every method you have there along with the 45deg, and SQRT of the product of the inside and outside path lengths works well enough on smaller bends in horn with only one to three corners. It is when you start to add more corners, or larger flares that they start falling away. The .707 one is more of a mid horn up thing.

Up until ~6months ago I actually used different methods for different corner setups, but in the end after dozens of more tests I have found that using the single method produces the most consistent results no matter the corner number, or geometry.

The problem with most of them is that they are inflexible, don't scale well, and do not account for the extra volume in a non-squared corner that will add path length. The bass horn system I came up with does.

The Danley quote is actually a key piece to the puzzle. If the volume around the corner is exactly the same as the straight line distance, then the acoustic path will be shorter than the straight line distance. That means a corner should have slightly more volume than it's straight sectioned counterpart, and adding even more volume will increase it's acoustic length beyond the straight sections length. It also means that if you know the volume is the same, and you path length sims longer than the real deal, you haven't measured the path through the corner correctly.

The nice thing about this method I found is that you can take a measured horn, create a sim from the plans that is equivalent. Then use the simulation output to refold the horn, and the results will be within a tolerance of mm to the original horn. It works in reverse as well. The only trick is learning where/when reflectors are needed, but it isn't hard to figure that part out. It is just a continuation of the line of thought stemming from the quote, and applying the same solutions.

I don't think it is the only way to do it, or even the best way. It just seems to work for me, and a few others.

That quote came from here - http://www.diyaudio.com/forums/subwoofers/175658-tham15-compact-15-tapped-horn-21.html#post2452977

And this is a picture that soho54 posted here - http://www.diyaudio.com/forums/subwoofers/171747-spreadsheet-folded-horn-layouts-2.html#post2270926


FIGURE 1

The centerline method is on the far left and soho54's advanced centerline method is in the middle.

 

[diy speaker guy]

The test subject



Lilmike offered a blank Sketchup file of the plans for his Cinema F20 front loaded horn as a subject for me to study. When opened it Sketchup it looks like this.

FIGURE 2


And he had built it and measured it too. Since his measurements are trustworthy, this is the perfect example to study, it means I wouldn't have to design a test protocol and build and measure my own example horn. Here's the F20 measurement.

FIGURE 3


And a link to the F20 thread - http://www.avsforum.com/t/1329971/lilmikes-cinema-f-20

 

[diy speaker guy]

Reverse engineering the F20 using the centerline and advanced centerline methods



Since I started with a plan the first thing I had to do is reverse engineer it back to an accurate simulation. Lilmike did post his Hornresp inputs but I needed to reverse engineer my own to create two simulations, one based on the centerline method and one for the advanced centerline method.


To begin the reverse engineering I used Sketchup to chop the blank F20 drawing into 37 segments. When that was done it looked like this. The red lines and numbers were added after to keep track of the many segment markers and their locations.


FIGURE 4


I used Sketchup to find out the cross sectional area at each segment marker and find the lengths of each segment. The path length through the bends will be different depending on whether I reverse engineer according to the centerline or advanced centerline method. Here's the information I obtained for the advanced centerline method from my notes.

Quote:
S1 - 4.7358 cm x 46.99 = 222.535242 sq cm

S2 - 6.0443 cm x 46.99 = 284.021657

S3 - 11.5067 cm x 46.99 = 540.6998

S5 - 16.9035 cm x 46.99 = 794.2954

S7 - 12.3825 cm x 46.99 = 581.8537

S8 - 12.5131 cm x 46.99 = 587.9906

S10 - 18.2194 cm x 46.99 = 856.1296

S12 - 13.2427 cm x 46.99 = 622.2745

S13 - 15.661 cm x 46.99 = 735.9104

S15 - 21.4491 cm x 46.99 = 1007.8932

S17 - 16.4034 cm x 46.99 = 770.7958

S18 - 16.7264 cm x 46.99 = 785.9735

S20 - 24.7882 cm x 46.99 = 1164.7975

S22 - 17.0692 cm x 46.99 = 802.0817

S23 - 19.3082 cm x 46.99 = 907.2923

S25 - 27.9930 cm x 46.99 = 1315.3911

S27 - 20.2683 cm x 46.99 = 952.4074

S28 - 20.4233 cm x 46.99 = 959.6909

S30 - 29.3228 cm x 46.99 = 1377.8784

S32 - 21.0407 cm x 46.99 = 988.7025

S33 - 29.1271 cm x 46.99 = 1368.6824

S35 - 42.0262 cm x 46.99 = 1974.8111

S37 - 30.2954 cm x 46.99 = 1423.5808

S38 - 36.0667 cm x 46.99 = 1694.7742


L12 - 19.486 cm (segment length)

L23 - 79.5009 cm (segment length)

L34 - 3.0956 cm (first part of first bend)

L45 - 2.1129 cm (second part of first bend)

L56 - 2.1129 cm (third part of first bend)

L67 - 2.8767 cm (fourth part of first bend)

L78 - 1.9005 cm (panel thickness)

L89 - 3.3107 cm (first part of second bend)

L910 - 2.2774 cm (second part of second bend)

L1011 - 2.2774 cm (third part of second bend)

L1112 - 3.1283 cm (fourth part of second bend)

L1213 - 35.2009 cm (segment length)

L1314 - 3.4957 cm (first part of third bend)

L1415 - 2.5564 cm (second part of third bend)

L1516 - 3.3159 cm (third part of third bend)

L1617 - 4.3455 cm (fourth part of third bend)

L1718 - 1.9118 cm (panel thickness)

L1819 - 3.7313 cm (first part of fourth bend)

L1920 - 2.5096 cm (second part of fourth bend)

L2021 - 3.4047 cm (third part of fourth bend)

L2122 - 4.9491 cm (fourth part of fourth bend)

L2223 - 27.4478 cm (segment length)

L2324 - 5.0671 cm (first part of fifth bend)

L2425 - 3.4991 cm (second part of fifth bend)

L2526 - 3.4991 cm (third part of fifth bend)

L2627 - 4.8270 cm (fourth part of fifth bend)

L2728 - 1.9003 cm (panel thickness)

L2829 - 5.2602 cm (first part of sixth bend)

L2930 - 3.6653 cm (second part of sixth bend)

L3031 - 3.6653 cm (third part of sixth bend)

L3132 - 5.1058 cm (fourth part of sixth bend)

L3233 - 97.9547 cm (segment length)

L3334 - 7.5739 cm (first part of seventh bend)

L3435 - 5.2533 cm (second part of seventh bend)

L3536 - 5.2533 cm (third part of seventh bend)

L3637 - 7.2818 cm (fourth part of seventh bend)

L3738 - 43.6758 cm (segment length)

Then the difference in length due to the centerline method is a simple matter of remeasuring the distance around the bends.

Quote:
Difference in length between centerline vs advanced centerline through -

first bend - 11.9446 - 10.1981 = 1.7465 cm

second bend - 12.8779 - 10.9938 = 1.8841 cm

third bend - 15.8068 - 13.7135 = 2.0933 cm

fourth bend - 17.1945 - 14.5947 = 2.5998 cm

fifth bend - 19.7883 - 16.8923 = 2.8960 cm

sixth bend - 20.7321 - 17.6966 = 3.0355 cm

seventh bend - 29.7112 - 25.3623 = 4.3489 cm

TOTAL DIFFERENCE - 18.6041

This information would allow me to create two Hornresp simulations, one based on the centerline method of reverse engineering, the other based on the advanced centerline method. But before I could do that I needed to take the 37 segments and combine some of them since Hornresp can only handle 4 segments. To do that I needed to redraw the horn into a straight horn using all this information. That was necessary because I needed to place the middle segment markers (S3 and S4) in the right spots or the simulation would not be accurate.


Soho54 illustrates the importance of placing your segment markers in the right spot when reverse engineering in this series of pictures that he posted here - http://www.diyaudio.com/forums/subwoofers/175658-tham15-compact-15-tapped-horn-21.html#post2451737

When placed in the wrong spot the simulation simply doesn't match the horn you are reverse engineering. In this case he colored the volume of error in blue. Most people don't do this step (redrawing as a straight horn to find the best spot for the segment markers) when reverse engineering, they just place them arbitrarily.




FIGURES 5, 6, 7


Once I had the F20 drawn out into a straight horn (using the advanced centerline data) it was still difficult to figure out where to place the segment markers. I had to greatly increase the vertical scale and spend a good amount of time drawing red lines across the top of the horn illustration to find the best fit. There is no choice as to where S1, S2 and S5 will go, they are set by nature of being the throat, the driver location and the mouth. I'm trying to get the best spot to place S3 and S4. Here's my final drawing and chosen location of my segment markers.


FIGURE 8


Now I was able to finally determine my reverse engineered Hornresp inputs.

Quote:
HORNRESP INPUTS


S1 - 222.54

S2 - 284.02

S3 - 952.41

S4 - 1423.58

S5 - 1694.77


L12 - 19.49

L23 - ac 212.35 / c 223.57

L34 - ac 142.91 / c 150.29

L45 - 43.68

Notice that the advanced centerline (ac) length of L23 and L34 in the quote above are shorter than the centerline (c) length of the same segments by the amount calculated earlier (in the previous quote bubble).


Finally, here's a picture of my Hornresp inputs using the advanced centerline data (on the right) compared to lilmike's original simulation from post 1 of the F20 thread.


FIGURE 9


Notice the incredible similarity. I didn't find out until later that this horn was originally laid out using the advanced centerline method but it makes sense in light of the fact that when I reverse engineered it using the same method my Hornresp inputs were almost exactly the same as the designer's original Hornresp inputs that were used to fold the horn. There are a few possible explanations for the very slight differences between the original simulation and my reverse engineered version as shown in Figure 9 but the differences are so small as to be negligible and thus not worth spending time on figuring out the cause. The cross sectional area at the worst points of discrepancy between the two sims in Figure 9 amounts to about 2 percent difference and through most of the horn the discrepancy is much much less than that, as can be seen in Figure 8.


Not shown in Figure 9 is my Hornresp input screen for my calculated centerline method simulation, it was not nearly as close to the designer's original Hornresp inputs. The centerline method resulted in a simulation 18.6 cm longer than the advanced centerline method.

 

[diy speaker guy]

Comparing the simulations to the measurements



The next step was to run the sims I made (both the centerline and the advanced centerline versions) to see which matches the measurements of the physical horn the best. Below is a picture showing both sims, advanced centerline on the left and centerline on the right.


FIGURE 10


The advanced centerline sim (left) gives a very slightly higher tuning. When compared to the F20 measurement (Figure 3) the advanced centerline sim matches better than the centerline sim.

 

[LTD02]

"Since his measurements are trustworthy, this is the perfect example to study, it means I wouldn't have to design a test protocol and build and measure my own example horn. Here's the F20 measurement."


was that with a plate amp with built in equalization?

 

[diy speaker guy]

Conclusions restated


When I reverse engineered the F20 plans using the advanced centerline method and the centerline method, the advanced centerline method resulted in a simulation that was closer to the measurements of the actual horn.

For this reason I believe the advanced centerline folding method is superior to the centerline folding method for both engineering and reverse engineering horns.

It should be noted that this study was conducted using a horn with no reflectors in any of the bends. Another study needs to be done to figure out how to properly address the situation of reflectors.

 

[diy speaker guy]

A method even more advanced than the advanced centerline method to simulate a horn



The real professionals probably aren't using the centerline or advanced centerline method. Danley doesn't and soho54 didn't either when he was being serious about accuracy. It's best to let him explain this in his own words with a direct quote from his post here - http://www.avsforum.com/t/1258118/akabak-for-dummies/30#post_19114195

This is also a good post to quote here because he talks about the advanced centerline method too.

Quote:
Originally Posted by soho54  /t/1258118/akabak-for-dummies/30#post_19114195


What TD seems to be talking about is a different (more advanced) way to model a horn. What you do is model the straight sections as waveguides/ducts only, and the bends are modeled as base forces of mass/resistance/compliance, and not waveguide/duct elements.



AkAbak can do it, but getting it right is not easy. My experimental equations for the AcouMass and AcouResistance for one bend are longer than most of the full simple scripts I have posted in this thread.

If you look up those two elements (might as well check out AcouCompliance as well) in the AkAbak manual it will give you the idea, and base equations. Then you have to figure out how to apply it to bends with changing area, different degree corners, and frequencies. Just so you know the simple 90deg corner mass equation will not work right here, so don't waste your time there.



It is much simpler to just physically account for the path length, and model that way.



Something along the lines of this will give you good results with a horn with multiple bends.

You use each triangle section though the bend as a waveguide element. This will adjust the mass/resistance properties though the bends on it's own, and get it pretty darn close. If you do it correctly the FR and impedance peaks and dips will line up the the true measured ones. The Q of the peaks and dips will be a little different, and will generally be a little less in magnitude as the frequency rises, but you can get a very accurate model without you having to become a PhD level math whiz.

EDIT:I should add that this only works at sub frequencies. If you need bandwidth much above 150Hz you will want to learn the hard way. The dimensions through the bends must be less than 1/4WL of the frequencies in question.

 

[diy speaker guy]

Why don't the simulations match the measurement perfectly?



I figured I would include this addendum since I did the work anyway. Lilmike and I had a great conversation about this topic while I was working on the folding experiment. There are a great many things that can make a measured product deviate from a simulation's predictions.


These are some of the factors we discussed, both of us already knew about all these but we found we are in agreement for most of these factors regarding the potential for deviation. Aspect ratio of the horn flare (most horn simulators assume a circular cross section), skin friction (aka boundary layer friction) which can be approximated somewhat by calculating the Reynolds number of a section of the horn flare, the operating temperature of the driver (simulations assume zero power compression), the weather and altitude (barometric pressure) of the measurement site, laminar vs turbulent airflow, the compressibility of air, the inability to accurately simulate the effects of inductance (although Akabak probably could assuming the user knew how), and the lossless nature of simulations (an infinitely rigid material is assumed in most simulators although Akabak is capable of incorporating panel losses if the user is sufficiently advanced) are all factors that can introduce a deviation from simulation to measurement. There's also the possiblility of having erroneous measurements. There's a million different ways you can go wrong here, faulty equipment, procedure, environment, etc.


And there are even more factors that can introduce deviation but I'll only mention one more. As far as I'm concerned the single biggest reason most people's simulations don't match measurements is the fact that most people don't accurately build what they simulated. This can include deviations from the plans by adding (or not including) bracing, errors in construction, placing the segment markers in the wrong location in the simulation, and simulating corners. When I talk about simulating corners I'm talking about ACTALLY simulating them, not just accounting for the path length through the bends using the advanced centerline folding method.


I've already shown that lilmike did a very good job folding this horn with regard to the plans matching the original simulation. And although I don't know anything about his measurement equipment or procedure I can only assume he did a very good job performing the measurements as well. So I was a bit disappointed to see that the measurements (even the low power 1 watt measurement) didn't match the simulation at least a little bit better, especially in terms of predicting the low corner frequency. It's not off by much but since I have more time and computer skills than money and woodworking skills I'd like to be able to have a closer correlation if it's possible. Closer correlation = less failed projects and expensive wood up in smoke for me in the future.

 

[diy speaker guy]

Comparing the Hornresp sim, the Akabak sim and the measurements



Let's see if a simple Akabak simulation matches the results better than Hornresp. The Akaback sim has enough segments to fully capture the horn flare including the corners that stick out above the red line in Figure 8 but otherwise it's still a simple simulation. I'm not advanced enough to simulate acoustic mass, acoustic compliance and acoustic resistance and I can't make Akabak simulate losses due to less than infinitely rigid panels. Not yet anyway.


First up let's look at the F20 with the MFW driver.


FIGURE 11


And now the F20 with the Ultimax driver as measured by lilmike here - http://www.avsforum.com/t/1329971/lilmikes-cinema-f-20/1620#post_23618560


FIGURE 12


And ... that's disappointing. I was hoping the Akabak sim would be a closer match to the measurements than the Hornresp sim. It does look like Akabak was slightly more accurate in predicting the low frequency corner but in the 30 - 60 hz region it looks like Hornresp did a better job. Above 60 hz it looks like both the Hornresp and Akabak sims are close enough to each other that they are both fine.


But I was expecting a better result with Akabak, so it's time to dig a bit deeper. But first here's the Akabak script in case anyone wants to see it.

Quote:
|DATA EXPORTED FROM HORNRESP - RESONANCES NOT MASKED


|COMMENT: F20 akabak starter


|~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


|REQUIRED AKABAK SETTINGS:


|File > Preferences > Physical system constants:


|Sound velocity c = 344m/s

|Medium density rho = 1.205kg/m3


|Sum > Acoustic power:


|Frequency range = 10Hz to 20kHz

|Points = 533

|Input voltage = 2.00V rms

|Integration = 2Pi-sr

|Integration steps = 1 degree ... 1 degree

|Integration method = Cross


|~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


Def_Const |Hornresp Input Parameter Values

{

|Length, area and volume values converted to metres, square metres and cubic metres:


S1 = 222.5352e-4; |Horn segment 1 throat area (sq m)

S2 = 284.0217e-4; |Horn segment 1 mouth area and horn segment 2 throat area (sq m)

S3 = 540.6998e-4;

S4 = 794.2954e-4;

S5 = 581.8537e-4;

S6 = 587.9906e-4;

S7 = 856.1296e-4;

S8 = 622.2745e-4;

S9 = 735.9104e-4;

S10 = 1007.8932e-4;

S11 = 770.7958e-4;

S12 = 785.9735e-4;

S13 = 1164.7975e-4;

S14 = 802.0817e-4;

S15 = 907.2923e-4;

S16 = 1315.3911e-4;

S17 = 952.4074e-4;

S18 = 959.6909e-4;

S19 = 1377.8784e-4;

S20 = 988.7025e-4;

S21 = 1368.6824e-4;

S22 = 1974.8111e-4;

S23 = 1423.5808e-4;

S24 = 1694.7742e-4;


L12 = 19.4860e-2; |Horn segment 1 axial length (m)

L23 = 79.5009e-2;

L34 = 5.2085e-2;

L45 = 4.9896e-2;

L56 = 1.9005e-2;

L67 = 5.5881e-2;

L78 = 5.4057e-2;

L89 = 35.2009e-2;

L910 = 6.0521e-2;

L1011 = 7.6614e-2;

L1112 = 1.9118e-2;

L1213 = 6.2409e-2;

L1314 = 8.3538e-2;

L1415 = 27.4478e-2;

L1516 = 8.5662e-2;

L1617 = 8.3261e-2;

L1718 = 1.9003e-2;

L1819 = 8.9255e-2;

L1920 = 8.7711e-2;

L2021 = 97.9547e-2;

L2122 = 12.8272e-2;

L2223 = 12.5351e-2;

L2324 = 43.6758e-2;


Vrc = 76.68e-3; |Rear chamber volume (cubic m)

Lrc = 34.00e-2; |Rear chamber average length (m)


|Parameter Conversions:


Sd = 823.70e-4; |Diaphragm area (sq m)


Arc = Vrc / Lrc;

}

|~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


|Network node numbers for this offset driver horn system:


| 0-Voltage-1

| |

|4-Chamber-5-Driver-

| |

| 8-Segment-9-Segment-10-Radiator


|~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


Def_Driver 'Driver'


Sd=823.70cm2

Bl=14.15Tm

Cms=4.74E-04m/N

Rms=3.38Ns/m

fs=15.00Hz |Mmd = 223.87g not recognised by AkAbak, fs calculated and used instead

Le=2.07mH

Re=3.04ohm

ExpoLe=1


System 'System'


Driver Def='Driver''Driver'

Node=1=0=5=9


Duct 'Rear chamber'

Node=4=5

SD={Arc}

Len={Lrc}

Visc=0


Waveguide 'Horn segment 1'

Node=8=9

STh={S1}

SMo={S2}

Len={L12}

Conical


Waveguide 'Horn segment 2'

Node=9=10

STh={S2}

SMo={S3}

Len={L23}

Conical


Waveguide 'Horn segment 3'

Node=10=11

STh={S3}

SMo={S4}

Len={L34}

Conical


Waveguide 'Horn segment 4'

Node=12=11

STh={S5}

SMo={S4}

Len={L45}

Conical


Waveguide 'Horn segment 5'

Node=12=13

STh={S5}

SMo={S6}

Len={L56}

Conical


Waveguide 'Horn segment 6'

Node=13=14

STh={S6}

SMo={S7}

Len={L67}

Conical


Waveguide 'Horn segment 7'

Node=15=14

STh={S8}

SMo={S7}

Len={L78}

Conical


Waveguide 'Horn segment 8'

Node=15=16

STh={S8}

SMo={S9}

Len={L89}

Conical


Waveguide 'Horn segment 9'

Node=16=17

STh={S9}

SMo={S10}

Len={L910}

Conical


Waveguide 'Horn segment 10'

Node=18=17

STh={S11}

SMo={S10}

Len={L1011}

Conical


Waveguide 'Horn segment 11'

Node=18=19

STh={S11}

SMo={S12}

Len={L1112}

Conical


Waveguide 'Horn segment 12'

Node=19=20

STh={S12}

SMo={S13}

Len={L1213}

Conical


Waveguide 'Horn segment 13'

Node=21=20

STh={S14}

SMo={S13}

Len={L1314}

Conical


Waveguide 'Horn segment 14'

Node=21=22

STh={S14}

SMo={S15}

Len={L1415}

Conical


Waveguide 'Horn segment 15'

Node=22=23

STh={S15}

SMo={S16}

Len={L1516}

Conical


Waveguide 'Horn segment 16'

Node=24=23

STh={S17}

SMo={S16}

Len={L1617}

Conical


Waveguide 'Horn segment 17'

Node=24=25

STh={S17}

SMo={S18}

Len={L1718}

Conical


Waveguide 'Horn segment 18'

Node=25=26

STh={S18}

SMo={S19}

Len={L1819}

Conical


Waveguide 'Horn segment 19'

Node=27=26

STh={S20}

SMo={S19}

Len={L1920}

Conical


Waveguide 'Horn segment 20'

Node=27=28

STh={S20}

SMo={S21}

Len={L2021}

Conical


Waveguide 'Horn segment 21'

Node=28=29

STh={S21}

SMo={S22}

Len={L2122}

Conical


Waveguide 'Horn segment 22'

Node=30=29

STh={S23}

SMo={S22}

Len={L2223}

Conical


Waveguide 'Horn segment 23'

Node=30=31

STh={S23}

SMo={S24}

Len={L2324}

Conical


Radiator 'Horn mouth'

Node=31

SD={S24}

 

[diy speaker guy]

Akabak can't do PAR segments, is it reliable at all?



This is a good question. The answer is yes, maybe and no. It depends on how many segments you break your horn down into and how long each of those segments are. Here's a picture to illustrate.


FIGURE 13


The two pictures in the top row shows the longest segment in the F20 horn as depicted in Figure 4 as a PAR segment and as a CON segment. This is the segment defined by markers 32 and 33 in Figure 4. There's no difference whatsoever between these two except that one was simulated as PAR and the other as CON. As you can see there is 0.504 liters difference, or a bit less than 0.5 percent difference in volume. That one segment alone accounts for about 26 percent of the total unfolded volume (not including the small amount of extra space added by corners when folded) of the F20 horn. The shorter segments will have even less percentage difference when compared as PAR vs CON. So the difference in overall volume between the Hornresp sim and the Akabak sim is very small, I'm guessing about 1 liter or so total.


The two pictures in the bottom row show that as the segments grow longer there is a much larger discrepancy when compared as PAR vs CON. There is a 33 percent difference in volume between these simulations, and again they are exactly the same except that one was simulated as PAR and the other CON.


But the other thing that the bottom two pics illustrate very clearly is that the volume is distributed in a very different manner resulting in a very different shape for the segment. THIS is problematic.


The only thing you can do about this is to keep the segments as short as possible. For example, a 100 cm single segment horn simulated in Hornresp will be different than the same horn simulated in the same way in Akabak due to the CON vs PAR difference. But if you split that Akabak sim up into 100 segments, each 1 cm long and plotted the cross sectional area vs length according to a PAR guideline (the Hornresp horn data export), even though each of those segments will be a CON segment the collection of segments will form a PAR shape and the difference will be negligible.


Would my Akabak simulation of the F20 horn have been closer to the measured results if I had chopped it up into more segments? I don't know and I don't have time to try.


The only other option to avoid this issue is to use a simulator that can do a lot of PAR segments. There's only one program I know of that can do this. It's called TL.app for now and it's still kind of stuck in beta mode and hardly anyone knows about it, but it's discussed (with a download link) here - http://www.diyaudio.com/forums/software-tools/220421-transmission-line-modelling-software.html


The only problem is that the results don't agree exactly with Hornresp and I don't know which program is more accurate. They are close though, here's a picture of a fairly large and complex 4 segment front loaded horn compared in both programs.


FIGURE 14


So what's the all time best program to use? I don't know. That's a study for another day. Or maybe never. All three of these programs (Hornresp, Akabak and TL.app) give very good results. Maybe not perfect but still very good.

 

[diy speaker guy]

Thanks



Many thanks to soho54, wherever you are. Thanks for introducing me to the advanced centerline folding method, thanks for the Akabak tutorial and the general informative nature of all your forum posts.


Thanks to lilmike for providing the F20 blank Sketchup drawing, the F20 measurements, the good conversation, and all the rest you do. A constantly growing library of designs and info shared freely, something to be proud of.


Thanks to LTD02 for asking a question I couldn't answer. I might look like a moron temporarily when that happens but I actually like it because I always end up learning something. And I learned a couple of things from this week long study.


And finally, thanks to anyone that actually read this. I know it's very long but I had a lot of information to show.

 

[diy speaker guy]



Quote:
Originally Posted by LTD02  /t/1489784/horn-folding-a-brief-study-of-the-centerline-vs-advanced-centerline-method#post_23714986


"Since his measurements are trustworthy, this is the perfect example to study, it means I wouldn't have to design a test protocol and build and measure my own example horn. Here's the F20 measurement."


was that with a plate amp with built in equalization?

Well I certainly hope not, otherwise this was a massive waste of time. But I'm guessing no. I'm sure he would have mentioned that at some point. Also, the measurement with the MFW driver and the measurement with the Ultimax driver both match the sims fairly well and they were measured at different times in different places. I don't think he even owns the F20 with the Ultimax driver.


So no. I hope not.

 

[LTD02]

well here is a quicky, no where near as sophisticated as what you did.


i just plopped his drawing into word, drew in some lengths, added them up and plopped that into hornresp.


the actual has a high pass filter, so rolls off faster under 20hz.


the peak at 72hz is higher in hornresp than in actual, as expected from other tests.


the peaks at 22hz and 41hz are pretty close.


the peak at 100hz is off by a few hz, but that is out of band anyways.


model is lowered by several db so that both lines can be seen.


i'd call the simple centerline method...good enough.




red (top) is actual. black (bottom) is hornresp.




and the rough hornresp model that i ended up with

 

[diy speaker guy]



Quote:
Originally Posted by LTD02  /t/1489784/horn-folding-a-brief-study-of-the-centerline-vs-advanced-centerline-method#post_23715071


well here is a quicky, no where near as sophisticated as what you did.


i just plopped his drawing into word, drew in some lengths, added them up and plopped that into hornresp.


the actual has a high pass filter, so rolls off faster under 20hz.


the peak at 72hz is higher in hornresp than in actual, as expected from other tests.


the peaks at 22hz and 41hz are pretty close.


the peak at 100hz is off by a few hz, but that is out of band anyways.


model is lowered by several db so that both lines can be seen.


i'd call the simple centerline method...good enough.


red (top) is actual. black (bottom) is hornresp.


and the rough hornresp model that i ended up with

Well yeah, you can come pretty close with approximations. I proved that in the thread that started this when I simulated a popular horn as a single CON segment. All I had to go on when reverse engineering that one was a pic of the exterior (basically just showing external dimensions and the size of the mouth) and the driver specs. I got extremely close to measured results.


The whole point of this study though, was to find out which method is better and why.


I'm really curious about the high pass filter issue though. If there was one applied when these measurements were taken I didn't know about it and it would make all my simulations match the measurements better.


I'm sure lilmike will be along soon enough to resolve the hpf question one way or another.

 

[LTD02]

 http://www.avsforum.com/t/1329971/lilmikes-cinema-f-20#post_20330637


and http://www.avsforum.com/t/1329971/lilmikes-cinema-f-20#post_20330934


and http://www.avsforum.com/t/1329971/lilmikes-cinema-f-20#post_20331276


bash 300 boost table: http://www.parts-express.com/pedocs/installation-guides/300-750-bash-installation-guide.pdf

 

[LTD02]

post #5 from above one over the other.


centerline vs. non-centerline.


stated, "When compared to the F20 measurement (Figure 3) the advanced centerline sim matches better than the centerline sim."


i'm just not seeing any material difference here. it is difficult to tell that is actually two plots.

 

[diy speaker guy]



Quote:
Originally Posted by LTD02  /t/1489784/horn-folding-a-brief-study-of-the-centerline-vs-advanced-centerline-method#post_23715156

http://www.avsforum.com/t/1329971/lilmikes-cinema-f-20#post_20330637


and http://www.avsforum.com/t/1329971/lilmikes-cinema-f-20#post_20330934


and http://www.avsforum.com/t/1329971/lilmikes-cinema-f-20#post_20331276


bash 300 boost table: http://www.parts-express.com/pedocs/installation-guides/300-750-bash-installation-guide.pdf

Well that does change things. He didn't mention it and I never thought to ask. Thanks for pointing it out.


I can add active filters to my simulations easily if lilmike tells me what frequency and slope to set it at. But if there's boost involved that's going to be a bit more tricky.


The Ultimax measurements definitely did not use the BASH amp but I don't know anything about the iNuke DSP 3000 amp other than the fact that the name suggests DSP is possible.


Anyway, this is unfortunate but my conclusions still stand. The centerline folding method gives a different path length depending on whether you are engineering or reverse engineering and that's a huge problem that has nothing to do with the measurements. I'll have to investigate the other conclusion later - the one where the advanced centerline method sim matched the measurements better. That was based mainly on the predicted LF knee frequency and if there was a hpf I can't really make any claims on which folding method sim matched the measurements best. I'll leave it unedited for now but if I can't prove that specific claim I'll have to delete it or add a note.


Hopefully the iNuke didn't have any processing applied when the Ultimax was measured.

 

[diy speaker guy]



Quote:
i'm just not seeing any material difference here. it is difficult to tell that is actually two plots.

I thought I was pretty clear about this. The difference is subtle, it's only an 18.6 cm difference in the path length. The longer path (centerline method) has a slightly lower tuning. It's subtle but you can see it, you didn't need to overlay to see it - in fact overlaying makes it a lot harder to see the difference because it just looks like a thicker line at the LF knee where the difference matters the most. Here's my exact quote on this matter.

Quote:
The advanced centerline sim (left) gives a very slightly higher tuning.

The bigger problem here, as I've mentioned over and over, is that the centerline folding method will give a different path length depending on whether you are engineering or reverse engineering. This is a problem IMO. This fact alone makes it "incorrect" in my opinion. The greater accuracy provided by the advanced centerline version is just a happy (but relatively small) bonus.


Let me be very clear here. It was not my intention to imply that the centerline method was massively inaccurate and should be banished from the face of the earth. I was just trying to figure out which method is more accurate and why. Some people appreciate the more accurate method, even if it's only a bit more accurate and the extra accuracy costs a lot of time and work.


The more bends you have in a horn, and the larger those bends are (nearer to the mouth), and the larger the mouth is (meaning the bends will also be relatively larger), the more inaccurate the centerline method will be. But even in a very large, very long horn with a lot of bends and a large mouth you are still only looking at a difference of maybe 70 cm path length difference max between the two methods resulting in maybe 2 or 3 hz max difference in the FR graph. I didn't mean to imply otherwise and I apologize if this was unclear. This horn has relatively few bends and since the mouth is fairly small the bends are all fairly small as well, which doesn't amount to a whole lot of path length difference between the two methods.

 

[LTD02]

let's remember what we were talking about.


you said, "I really don't care what causes the discrepancy (although I would be interested to read about it in proper studies), all I care about is making accurate simulations that will match the measurements of the finished product. The centerline folding method won't get you there."


after all the work, i'm not seeing a dimes worth the difference between the two. the overlay says it all.

http://www.avsforum.com/t/1489784/horn-folding-a-brief-study-of-the-centerline-vs-advanced-centerline-method#post_23715202


i suppose i'll bow out of this one. i commend you on the very thorough analysis even if we disagree on how to interpret the results.