confidence in results

Hi,

I'm new to Pachyderm - I'm trying it out to see if I can use it to help with architectural design. I've studied masters level acoustics at university, and have used Odeon before, though don't have access to it now I'm out of uni.

I did some simple testing, modelling a simple rectangular space, applying some basic materials, and using Sabine's formula to derive RT60 as a baseline, then ran the same model through Pachyderm, using a few different methods and method parameters. I got wildly varying results - to the point where I'm really not sure if it is even close to reasonable. I'm hoping someone in this group might be able to tell me if I'm making an elementary mistake, or how to make sense of the results.

I only considered frequencies between 125Hz and 4kHz.

As a starting point, I used a simple rectangular box, 100x30x10m, with plywood panelling on all surfaces (from 125Hz, 0.28,0.22,0.17,0.09,0.1,0.11) except the ceiling, which I applied felt absorptive panels (0.05,0.1,0.25,0.55,0.8,0.95) to.

I did a simple reverb time calc using Sabine’s formula, and this resulted in RT60 figures of:

125Hz 2.81s
250Hz 3.15s
500Hz 2.84s
1kHz   2.24s
2kHz   1.63s
4kHz   1.39s

Does this seem reasonable?

I then tried running simulations of the same space with the same parameters in Pachyderm for Rhino.

Using the Image/Source method (second order) to derive the impulse response and then looking at the T-30 parameter (and here it is not clear if this is RT30, or RT60 extrapolated from T-30 - but I assume the latter, otherwise the times would be even further from the Sabine calcs)

125Hz 3.1s
250Hz 3.12s
500Hz 3.25s
1kHz   3.84s
2kHz   3.84s
4kHz   3.77s

Which is quite different from the figures obtained with Sabine, especially at the higher frequencies.

If I use the Ray-Tracing method, with 10000 rays (not enough for a space of this volume, but just as a starting point) to derive the impulse, and again looking at T-30

125Hz 4s
250Hz 4.9s
500Hz 5.38s
1kHz    5.89s
2kHz    4.21s
4kHz    8.47s

Which is even further from Sabine.

I really have no Idea what to make of these results. If anyone can help me out here, I'd really appreciate it.

Thanks,

Ben

Replies to This Discussion

Permalink Reply by Arthur van der Harten on January 23, 2018 at 6:24am

Hi Ben,

If you've read the documentation on the website, you are familiar with the benchmarks that have been done with pachyderm. You can use to that to establish confidence to a degree. I stand by pachyderm in terms of it's comparability to other software out there of it's type.

That said, geometrical acoustics (especially Monte Carlo methods) have their limitations. Anyone who preaches absolute accuracy is clearly trying to sell you their software.

First off, if your results 'vary wildly' it means poor sampling... Not enough rays.

Second, the only reason you can do the image source method on it's own is in the case that all you wanted was the early reflections, for viewing in rhino. As you should know, image source neglects the late part and scattering, which makes it unsuitable for impulse response analysis. In short, if you want T30, you must use raytracing, and with plenty of rays.

In terms of comparison to Sabine, as you should also know, the Sabine equation assumes a diffuse field, and evenly distributed absorption. You have a very long low, reflective box with absorption applied to just one surface, all parallel surfaces, and (you don't address this, but I assume) very little diffusion (leading to long sustains between parallel surfaces). It should not surprise you that the results aren't the same as the Sabine. I would question Sabine's suitability for this test, as it will not account for flutter echo, or a long echo from 100 m away.

I suggest you try again with something more realistic, and for which the Sabine equation is reasonable to use... And make sure you are using enough rays.

Kind regards,

Arthur

Permalink Reply by Ben Dixon on January 23, 2018 at 3:46pm

Hi Arthur, thanks for the response. Yes - I know Sabine is probably not the right choice for this kind of space. I did try to find some Pachyderm documentation on the website but couldnt - all I could see were the set of basic tutorials. I've run through a few other folk's video tutorials also.

The test case I chose, I picked because it is a super simplification of an actual space I'm trying to model (a large school sports complex - see below). Ive modelled it as a closed volume, with a few solid objects inside it, and it is a much less box-shaped space, with a ceiling that is not flat, and a significant lattice of acoustic panelling that encloses the roof trusses.

the volume of this space is around 50000 cubic metres, which if I followed the guidelines o0f 50-100 rays per cubic metre, would be 2.5 - 5 million rays. I ran a simulation on the test simplified box space with 100k rays, which took about 2 hours running on a macbook pro booted into windows. Perhaps I need to find a much more serious machine to run this on. would it be a reasonable assumption to think that as more rays are added, the results would converge on a particular solution? if so, if you had to take a guess, how many rays/m3 would be required to get a solid estimate of reverb time +/- 0.1s?

I don't mean to imply that Pachyderm isnt up to scratch - simply that I'm trying to find some way of determining whether a given set of simulation parameters are going to give a result that will be enough to make decisions about surface materials and treatments that will be required. I tried a bunch of different methods and simulation parameters to see if they were even remotely similar, and unsurprisingly, they werent. I'm not an acoustic engineer, I'm an architect who has studied some acoustics in addition to my regular subjects. I know enough to be dangerous, but I'm trying to convert that into enough to be useful. :). I'm totally open to any advice anyone might offer.

One last thing, could you confirm that the T-30 parameter is T-30 (and so needs to be doubled to get RT60)

Thanks for responding,

Ben

Permalink Reply by Ben Dixon on January 24, 2018 at 3:49pm

I ran the simulation overnight in the full space shown in the post above, with correct materials (mostly - some I couldnt get absorption data for so had to guess), with 250,000 Rays (about 5 rays per m3 - still way short of the recommended 50-100 rays/m3) and the results do seem to be converging on what seems to be a reasonable looking set of results. The ETC plots look much more plausible.

one possible bug I've run into is that whenever I try to load saved results, Pachyderm crashes rhino completely. this happens as soon as I click on 'open', before the file selection dialog pops up.

and one question - is it possible to install Pachyderm on Rhino for Mac, or is it Windows only?

Permalink Reply by Ben Dixon on January 24, 2018 at 4:31pm

Just had another crash - when I tried to do anything on the Auralisation tab.

Is it possible to save the impulse response as a WAV file for use in other software?

Thanks again,

Ben

Permalink Reply by Arthur van der Harten on March 7, 2018 at 10:20pm

I am working on the auralization interface... but then, you said auralization tab... it hasn't been a tab in the main interface in many years. Are you using the current version?

Arthur

Permalink Reply by Arthur van der Harten on March 7, 2018 at 10:22pm

Yes, you can export as wave file. There should be a button that says "Export IR as Wave File" in the auralization interface.

Arthur

Permalink Reply by Arthur van der Harten on March 7, 2018 at 10:19pm

At present it is windows only. I don't have plans to port it to Mac. As you an see, I have enough to work out in the windows version for now - and too little time to devote, I'm afraid.

Permalink Reply by Arthur van der Harten on March 7, 2018 at 10:17pm

Hi,

I'm sorry that it has taken me so long to find this. The site did not notify me that you had responded.

T-30 - you would not double it to get RT60. Never Never Never. The 30 just implies that it takes the linear regression of the Schroeder integral over 30 dB worth of decay. Whether it is T-15 or T-30, they all seek to estimate the RT, which is always always the time it takes for sound to decay 60 decibels.

The website has benchmarks, for your reference. You can find them under the 'Pachyderm' drop down menu, under 'Benchmarks'.

Your model may well require millions of rays to be accurate. It sounds like a very large space. I'm sorry if that is an unpleasant answer. Sometimes it does help to have a computer with more cores to help with this. I have gotten up to 90% processor usage on a 12 core machine before.

Arthur

RSS