Hi Everyone,

I've been doing a bunch of reading around here on "QRD's" and "Skyline" type diffusers and I'm going to be making a bunch of my own soon but I have a question.

How important are the well dividers in both the QRD and Skyline type diffusors? I'm a little confused as GiK's new QRD doesn't have well dividers and I've seen many skyline type diffusers without dividers.

I think I'm going to make a bunch of these though as sawing up some 2x2's isn't that hard

PME Records QRD Diffusor Construction

Thanks!

Spencer

I once had the same question:








Screens are from a beta version of AFMG Reflex, posted with permission. More info here:
New EASE Tools Go Beta - AFMG - Ahnert Feistel Media Group


One can make diffusers without fins that perform well but then you need to redesign them in order to achieve even diffusion and good scattering.


Sincerely Jens Eklund

Jens pretty much summed it up!thumbsup

For small prime numbers and 1D, dividers are more important. Large number 2D PRD's are more complex in the response. More like a cloud of energy than the more distinct spatial lobes of 1D small number sequences. The starting point in a 2D PRD is perfectly random. The result of the removal of the well dividers is by nature pretty random as well, though not perfectly random. Summing these effects is still random enough.

The 1D prime 7 QRD shown above only have four distinct depths on the seven steps. A whole world of difference from the large number devices. So for large number devices, at least 2D PRD's, it should work fine without dividers.

I believe there are pros and cons to each approach. PRD's have the property of attenuating specular direction reflections at the build frequencies. This attribute should work better with well dividers as the result is closer to the math ideal. The spread should also be more like the perfectly random math suggests. On the other hand, divider less structures may provide some better performance below the design frequency. I get this idea from the fact that some of the deeper wells are placed next to each other, effectively giving larger pockets for the sound than a single deep well provides. Have seen no references to either of these ideas in the litterature. So please take it for what it worth as half qualified musings.


Regards,

Andreas Nordenstam

Thanks Everyone,

I think im going to build some of those 2d diffusors like on the link I posted, but I'm going to wait to buy some of GiK's all wood QRD's when they come out.

Thanks!

Spencer

sticky! sticky! those pics are so useful!

Jens: Nice graphs indeed! What software is that?

The wider effective width in this situation would result in a lowering of the HF cutoff frequency for that group of wells.

New EASE Tools Go Beta - AFMG - Ahnert Feistel Media Group

/Jens

Thanks for that - it looks like a good tool. I like the idea of using cloud computing resources. I guess ultimately its use to the DIY community will depend on how much they charge when it is released.

I’m already starting to worry. I’m addicted already. heh

Interesting point. Though, there are differences inside those compound wells. Here's an example:


There are still height differences from well to well inside the larger pits. Should result in some effect on the high frequencies.


PS: the graphs are indeed great, Jens!

Glad to help!

I’m currently playing around trying to find a good finless design.

Been having a great time beta testing the Ease/Reflex software Jens recommended. Excellent program!


The dividers vs finless issue seems pretty easy to determine when it comes to QRD's. They really need dividers to be good. PRD's, however.. They do perform pretty good straight out of the calculations without dividers. Interestingly, the results are the opposite of what I expected. I presumed the low end behaviour to be better without the dividers, while the opposite happened.

Here are some results to show one example of the differences. These are prime 17, ~17cm deep, ~60cm wide:

QRD 17 with dividers:


QRD 17 without dividers:


PRD 17 with dividers:


PRD 17 without dividers:



This set of predicitons are mostly included because the QRD with dividers looks much better than the one below. For some odd reason, the prime 23 QRD below looks particularly bad. No optimization applied to the PRD above.


With some optimization of the PRD's using different primitive roots and offsets, performance can be extended quite a bit down into the midrange. So I made my first attempt at a diffusor calculator creating an excel spreadsheet to find the various permutations of PRD's.


Here are some prime 23, ~21 cm deep, ~80cm wide:

23 QRD with dividers:


23 QRD without dividers:


23 PRD with dividers:


23 PRD without dividers:




And lastly, an optimized variant of the prime 23 PRD:

Optimized 23 PRD with dividers:


Optimized 23 PRD without dividers:





Pics are made using the beta version of Reflex. Permission to post the pics here was kindly granted by the AFMG team. There may be discrepancies between these graphs made using the beta and the results from the final release version. The non-normalized diffusion coefficient is particularly tricky and those curves may be disregarded.

Interesting.


How do these perform if multiple periods? I’m currently experimenting with a 1200mm period so we could compare easily with a total width of 2400mm (since yours are either ≈ 600mm or 800mm, 3 or 4 periods to reach 2400mm total i.e.). My effective depth will probably be fixed at 190mm (220mm total depth per panel) due to the material and production technique I’m using.

I know you prefer 1/6 octave graphs for diffusion and scattering coefficients, so the graphs below is in this format but in the appendix of AAaD (Cox/D´Antonio), 1/3 octave is used so I would prefer this option so one can compare (at least to some extent) with these values. Also, it’s a bit quicker to render the graphs if 1/3 oct. heh

As we discussed via PM, the lack of modeling to predict temporal dispersion can be troublesome since the reported diffusion coefficients for some shapes looks “too good to be true”, like the single poly (ellipse) for instance (witch offers superior spatial dispersion but no temporal dispersion and therefore is not a good diffuser). I made some examples to illustrate the problem to other readers of this thread. I think the polar plot is useful get some clue on temporal dispersion. If there are no lobs, there is probably no “diffusion” in the time domain either. A FDTD (Finite Difference Time Domain) model would certainly be useful.



Random incidence for scattering and normalized diffusion coefficients, normal (0 deg) incidence for polar plots. All models 190mm effective depth:


One fined N7 QRD; 600mm, 190mm effective depth (220mm total), 1mm fins:



4 x fined N7 QRD; 600mm period, 2400mm total, 190mm effective depth (220mm total), 1mm fins:



One finless (no dividers) N7 QRD; 600mm, 190mm effective depth (220mm total):



4 x finless N7 QRD; 600mm period, 2400mm total, 190mm effective depth (220mm total):



One poly; 600mm, 190mm depth:



4 x 600mm polys; 2400mm total, 190mm depth:



One poly; 2400mm, 190mm depth:



Freeform (stepped, no fins): positive and negative panel @ 1200mm each, 2400mm total width, 190mm effective depth (220mm total):


Screens are from a beta version of AFMG Reflex, posted with permission. More info here:
New EASE Tools Go Beta - AFMG - Ahnert Feistel Media Group



Sincerely Jens Eklund

Hi!

The ones I'm working on aren't working that well when repeated straight forward. If the repetition consist of a mirrored unit, performance is much better! There are however ways to make much nicer responses from larger arrays if each of the pieces are unique and together form a larger unique piece. This is particularly effective when combining two very similar but quite not the same shapes. Building unique models is not a problem in my case since they won't be machined.

However.. I'm not sure about the array idea. Most diffusers seems to work better alone. Most diffusers works better at some angle to the sound. So even if placed next to each other, it may be benefitial to angle them a bit to avoid perpendicular incidence. And it's often times a good idea to have some air gap around them to let low end diffract around the obstacle to reach absorbers behind.

How important is repeated performance for you? How large arrays do you envision to use?


Was playing around with some ideas that may fit your purpose just for the fun of it. Though I think you're doing just fine on your own..! So I've been concentrating on ideas more suitable for my own builds. So no direct comparisons possible.

There's definitely a learning curve attached to this process. There's so much to consider! The diffuse incidence plots are great for a general view, but they do not reflect performance from the typical diffuser sound receiver angles in typical rooms. There are seldom situations where there truly is a diffuse sound field striking the diffuser. (seems pretty obvious when put that way!) Most of the time, most of the energy will either come from the speakers at a rather steep angle (sidewall placements, ceiling) or nearly perpendicular (rear and front walls). So I've been concentrating more on getting good response in certain directions. The oblique incidence seems easy enough compared to getting good response head on.


The question is how the 1/3 octave data in The Book is calculcated. It would surprise me if each value represents a single sample point. As the 1/3'rd octave graphs do. If you want to have 1/3'rd octave data that is somewhat representative, you'll have to enable the "8 point frequency averaging".

My point is that too few samplepoints makes the graph non-representative of the actual performance. The difference between the true curve and the few lonely samplepoints are too large with 1/3'rd octave by itself. If using frequency averaging, data carries larger value. Then again, if you use 8 points per octave frequency averaging you may as well use a 24 points per octave graph and get the same information amount in a nicer looking graph.


Sorry, but we'll have to find something else to use as a gauge for time dispersion. A flat plate have a pretty large amount of lobing! A pyramid also have a large amount of lobing. There's clearly 0 temporal dispersion going on with those shapes.

Lets hope for an upgrade that provides some means to see temporal dispersion as well.

As for now, I simply look for a nice spread in physical depths. That's bound to translate into a series of lumps of energy comming at different time points.

Got a better idea?


Andreas

Back to this question.

Been playing extensively with simulations on various diffuser configurations. Small prime number 1D diffusers seems to be in dire need for well dividers. Large prime number 1D PRD's works pretty good without dividers. However, it takes a large prime number to get any decent response straight out of the calculator. Have spent a week now optimizing the process and the end result is pretty far off from the standard PRD calculations. These gives significantly improved response as divider-less units.

A skyline, 2D PRD, is an entirely different animal. It's so large and complex that I think it's pretty safe to say that it'll work fine without dividers.

Hi Andreas


Just sent you a PM but anyway:

You are right about the temporal bit. What I meant was that if there are no lobs at all, there’s probably no temporal dispersion but if there is, we just don’t know. I guess I need to measure my upcoming model as I did with my current one (but I was hoping to evade the process this time) .

Below is a pic of a painted (mix of filler and paint) EPS diffuser:




And here’s a 1/24 octave resolution graph of the latest progress of my new model:

Screen is from a beta version of AFMG Reflex, posted with permission. More info here:
New EASE Tools Go Beta - AFMG - Ahnert Feistel Media Group


Good luck with your models!
Sincerely Jens Eklund