Ray Tracing News

"Light Makes Right"

February 20, 1989

Volume 2, Number 2

Compiled by Eric Haines erich@acm.org . Opinions expressed are mine.

All contents are copyright (c) 1989, all rights reserved by the individual authors

Archive locations: anonymous FTP at ftp://ftp-graphics.stanford.edu/pub/Graphics/RTNews/,
wuarchive.wustl.edu:/graphics/graphics/RTNews, and many others.

You may also want to check out the Ray Tracing News issue guide and the ray tracing FAQ.


Contents:


Introduction, by Eric Haines

Whew, things have piled up! I've culled my comp.graphics findings as best as I can. I've decided to delete everything on the question of what a stored ray-trace image should be called ("image", "bytemap", "pixmap", and "bitmap" were some of the candidates). It's a good question, but the discussion just got too long to recap. Paul Heckbert's original posting advocated not using "bitmap" for 24 bit images, since "bitmap" denotes an M x N x 1 bit deep image in most settings. It would be pleasant to get a consensus on acceptable usage, but it's also interesting to me from a `word history' standpoint. If you have an opinion you'd like to share on this topic, pass it on to me and I'll summarize them (if possible, a 25 word or less summation would be nice). My own is: "I'm a product of my environment. Cornell used bitmap, Hewlett-Packard uses bitmap, so I tend to use `bitmap', `24 bit deep bitmap', or `image'".

I've put all the comp.graphics postings at the end, and the good news is that the queue is now empty. The `Sound Tracing' postings to comp.graphics were many and wordy. I've tried to pare them down to references, interesting questions that arose, and informed (or at least informed-sounding to my naive ears) opinions.

back to contents


New Subscribers

# Turner Whitted
# Numerical Design Limited
# 133 1/2 E. Franklin Street
# P.O. Box 1316
# Chapel Hill, NC 27514
alias turner_whitted gould!rti!ndl!jtw@sun.com

[this mail path is just a good guess - does gould have an arpa connection? The uucp path I use is: turner_whitted hpfcrs!hpfcla!hplabs!sun!gould!rti!ndl!jtw ]

# Michael John Muuss -- ray-tracing for predictive analysis of 3-D CSG models
# Leader, Advanced Computer Systems Team
# Ballistic Research Lab
# APG, MD 21005-5066
# USA
# ARPANET: mike@BRL.MIL
# (301)-278-6678 [telephone is discouraged, use E-mail instead]
alias mike_muuss mike@BRL.MIL

I lead BRL's Advanced Computer Systems Team (ACST) in research projects in (a) CSG solid modeling, ray-tracing, and analysis, (b) advanced processor architectures [mostly MIMD of late], (c) high-speed networking, and (d) operating systems. We are the developers of the BRL-CAD Package, which is a sophisticated Combinatorial Solid Geometry (CSG) solid modeling system, with ray-tracing library, several lighting models, a variety of non-optical "lighting" models (eg, radar) [available on request], a device independent framebuffer library, a collection of image-processing tools, etc. This software totals about 150,000 lines of C code, which we make available in source form under the terms of a "limited distribution agreement" at no charge.

My personal interests wander all over the map, right now I'm fiddling with some animation software, some D/A converters for digital music processing, and some improvements to our network-distributed ray-tracer protocol.

Thanks for the invitation to join!

        Best,
         -mike

back to contents


The BRL CAD PACKAGE

Short Summary

In FY87 two major releases of the BRL CAD Package software were made (Feb-87, July-87), along with two editions of the associated 400 page manual. The package includes a powerful solid modeling capability and a network-distributed image-processing capability. This software is now running at over 300 sites. It has been distributed to 42 academic institutions in twenty states and four countries including Yale, Princeton, Stanford, MIT,USC, and UCLA. The University of California - San Diego is using the package for rendering brains in their Brain Mapping Project at the Quantitative Morphology Laboratory. 75 different businesses have requested and received the software including 23 Fortune 500 companies including: General Motors, AT&T, Crysler Motors Corporation, Boeing, McDonnell Douglas, Lockheed, General Dynamics, LTV Aerospace & Defense Co., and Hewlett Packard. 16 government organizations representing all three services, NSA, NASA, NBS and the Veterns Administration are running the code. Three of the four national laboratories have copies of the BRL CAD package. More than 500 copies of the manual have been distributed.

BRL-CAD started in 1979 as a task to provide an interactive graphics editor for the BRL target description data base.

Today it is > 100,00 lines of C source code:

        Solid geometric editor
        Ray tracing utilities
        Lighting model
        Many image-handling, data-comparison, and other
        supporting utilities

It runs under UNIX and is supported over more than a dozen product lines from Sun Workstations to the Cray 2.

In terms of geometrical representation of data, BRL-CAD supports:

        the original Constructive Solid Geometry (CSG) BRL data
        base which has been used to model > 150 target descriptions,
        domestic and foreign

        extensions to include both a Naval Academy spline
        (Uniform B-Spline Surface) as well as a U. of
        Utah spline (Non-Uniform Rational B-Spline [NURB] Surface)
        developed under NSF and DARPA sponsorship

        a facerted data representation, (called PATCH),
        developed by Falcon/Denver
        Research Institute and used by the Navy and Air Force for
        vulnerability and signature calculations (> 200 target
        descriptions, domestic and foreign

It supports association of material (and other attribute properties) with geometry which is critical to subsequent applications codes.

It supports a set of extensible interfaces by means of which geometry (and attribute data) are passed to applications:

        Ray casting
        Topological representation
        3-D Surface Mesh Generation
        3-D Volume Mesh Generation
        Analytic (Homogeneous Spline) representation

Applications linked to BRL-CAD:

o Weights and Moments-of-Inertia
o An array of Vulnerability/Lethality Codes
o Neutron Transport Code
o Optical Image Generation (including specular/diffuse reflection,
        refraction, and multiple light sources, animation, interference)
o Bistatic laser target designation analysis
o A number of Infrared Signature Codes
o A number of Synthetic Aperture Radar Codes (including codes
        due to ERIM and Northrop)
o Acoustic model predictions
o High-Energy Laser Damage
o High-Power Microwave Damage
o Link to PATRAN [TM] and hence to ADINA, EPIC-2, NASTRAN, etc.
        for structural/stress analysis
o X-Ray calculation

BRL-CAD source code has been distributed to approximately 300 computer sites, several dozen outside the US.

__________

To obtain a copy of the BRL CAD Package distribution, you must send enough magnetic tape for 20 Mbytes of data. Standard nine-track half-inch magtape is the strongly preferred format, and can be written at either 1600 or 6250 bpi, in TAR format with 10k byte records. For sites with no half-inch tape drives, Silicon Graphics and SUN tape cartridges can also be accommodated. With your tape, you must also enclose a letter indicating

(a) who you are,
(b) what the BRL CAD package is to be used for,
(c) the equipment and operating system(s) you plan on using,
(d) that you agree to the conditions listed below.

This software is an unpublished work that is not generally available to the public, except through the terms of this limited distribution. The United States Department of the Army grants a royalty-free, nonexclusive, nontransferable license and right to use, free of charge, with the following terms and conditions:

1. The BRL CAD package source files will not be disclosed to third parties. BRL needs to know who has what, and what it is being used for.

2. BRL will be credited should the software be used in a product or written about in any publication. BRL will be referenced as the original source in any advertisements.

3. The software is provided "as is", without warranty by BRL. In no event shall BRL be liable for any loss or for any indirect, special, punitive, exemplary, incidental, or consequential damages arising from use, possession, or performance of the software.

4. When bugs or problems are found, you will make a reasonable effort to report them to BRL.

5. Before using the software at additional sites, or for permission to use this work as part of a commercial package, you agree to first obtain authorization from BRL.

6. You will own full rights to any databases or images you create with this package.

All requests from US citizens, or from US government agencies should be sent to:

        Mike Muuss
        Ballistic Research Lab
        Attn: SLCBR-SECAD
        APG, MD  21005-5066

If you are not a US citizen (regardless of any affiliation with a US industry), or if you represent a foreign-owned or foreign-controlled industry, you must send your letter and tape through your Ambassador to the United States in Washington DC. Have your Ambassador submit the request to:

        Army Headquarters
        Attn: DAMI-FL
        Washington, DC  20310

Best Wishes,
-Mike Muuss

Leader, Advanced Computer Systems Team
ArpaNet: <Mike @ BRL.ARPA>

________

p.s. from David Rogers:

If you have the _Techniques in Computer Graphics_ book from Springer-Verlag the frontispiece was done with RT the BRL ray tracer. It is also discussed in a paper by Mike Muuss in that book.

________

p.s. from Eric Haines:

Mike Muuss was kind enought to send me the documentation (some two inches thick) for the BRL package. I haven't used the BRL software (sadly, it does not seem to run on my HP machine yet - I hope someone will do a conversion someday...), but the package looks pretty impressive. Also, such things as the Utah RLE package and `Cake' (an advanced form of `make') come as part of the distribution. There are also interesting papers on the system, the design philosophy, parallelism, and many other topics included in the documentation.

back to contents


_Illumination and Color in Computer Generated Imagery_ by Roy Hall

Springer-Verlag, New York, 1989, 282 pages

(article by Eric Haines)

Roy Hall's book is out, and all I'll say about it is that you should have one. The text (what little I've delved into so far) is well written and complemented with many explanatory figures and images. There are also many appendices (about 100 pages worth) filled with concise formulae and "C" code. Below is the top-level Table of Contents below to give you a sense of what the book covers.

The "C" code will probably be available publicly somewhere sometime soon. I'll post the details here when it's ready for distribution.

    1.0 Introduction                             8 pages
    2.0 The Illumination Process                36 pages
    3.0 Perceptual Response                     18 pages
    4.0 Illumination Models                     52 pages
    5.0 Image Display                           40 pages
    Appendix I - Terminology                     2 pages
    Appendix II - Controlling Appearance        10 pages
    Appendix III - Example Code                 86 pages
    Appendix IV - Radiosity Algorithms          14 pages
    Appendix V - Equipment Sources               4 pages
    References                                   8 pages
    Index                                        4 pages

back to contents


Uniform Distribution of Sample Points on a Surface

[Mark Reichert asked last issue how to get a random sampling of a sphere]

How to generate a uniformly distributed set of rays over the unit sphere: Generate a point inside the bi-unit cube. (Three uniform random numbers in [-1,1].) Is that point inside the unit sphere (and not at the origin)? If not, toss it and generate another (not too often.) If so, treat it as a vector and normalize it. Poof, a vector on the unit sphere. This won't guarantee a isotropic covering of the unit sphere, but is helpful to generate random samples.

--Jeff Goldsmith

________

One method is simply to do a longitude/latitude split-up of the sphere (and randomly sampling within each patch), but instead of making the latitude lines at even degree intervals, put the latitude divisions at even intervals along the sphere axis (instead of even altitude [a.k.a. theta] angle intervals). Equal axis divisions give us equal areas on the sphere's surface (amazingly enough - I didn't believe it was this simple when I saw this in the Standard Mathematics Tables book, so rederived it just to be sure).

For instance, let's say you'd like 32 samples on a unit sphere. Say we make 8 longitude lines, so that now we want to make 4 patches per slice, and so wish to make 4 latitudinal bands of equal area. Splitting up the vertical axis of the sphere, we want divisions at -0.5, 0, and 0.5. To change these divisions into altitude angles, we simply take the arcsin of the axis values, e.g. arcsin(0.5) is 30 degrees. Putting latitude lines at the equator and at 30 and -30 degrees then gives us equal area patches on the sphere. If we wanted 5 patches per slice, we would divide the axis of the unit sphere (-1 to 1) into 5 pieces, and so get -0.6,-0.2,0.2,0.6 as inputs for arcsin(). This gives latitude lines on both hemispheres at 36.87 and 11.537 degrees.

The problem with the whole technique is deciding how many longitude vs. latitude lines to make. Too many longitude lines and you get narrow patches, too many latitude and you get squat patches. About 2 * long = lat seems pretty good, but this is just a good guess and not tested.

Another problem is getting an even jitter to each patch. Azimuth is obvious, but you have to jitter in the domain for the altitude. For example, in a patch with an altitude from 30 to 90 degrees, you cannot simply select a random degree value between 30 and 90, but rather must get a random value between 0.5 and 1 (the original axis domain) and take the arcsin of this to find the degree value. (If you didn't do it this way, the samples would tend to be clustered closer to the poles instead of evenly).

Yet another problem with the above is that you get patches whose geometry and topology can vary widely. Patches at the pole are actually triangular, and patches near the equator will be much more squat than those closer to the poles. If you would rather have patches with more of an equal extent than a perfectly equal area, you could use a cube with a grid on each face cast upon the sphere (radiosity uses half of this structure for hemi-cubes). The areas won't be equal, but they'll be pretty close and you can weight the samples accordingly. There are many other nice features to using this cast cube configuration, like being able to use scan-line algorithms, being able to vary grid size per face (or even use quadtrees), being able to access the structure without having to perform trigonometry, etc. I use it to tessellate spheres in the SPD package so that I won't get those annoying clusterings at the poles of the sphere, which can be particularly noticeable when using specular highlighting.

--Eric Haines

back to contents


Depth of Field Problem

From: Marinko Laban via Frits Post dutrun!frits@mcvax.cwi.nl

First an introduction. I'm a Computer Graphics student at the Technical University of Delft, The Netherlands. My assignment was to do some research about distributed ray tracing. I actually implemented a distributed ray tracer, but during experiments a very strange problem came up. I implemented depth-of-field exactly in the way R.L. Cook described in his paper. I decided to do some experiments with the shape of the f-stop of the simulated camera. First I simulated a square-shaped f-stop. Now I now this isn't the real thing in an actual photocamera, but I just tried. I divided the square f-stop in a regular raster of N x N sub-squares, just in the way you would subdivide a pixel in subpixels. All the midpoints of the subsquares were jittered in the usual way. Then I rendered a picture. Now here comes the strange thing. My depth-of-field effect was pretty accurate, but on some locations some jaggies were very distinct. There were about 20 pixels in the picture that showed very clear aliasing of texture and object contours. The funny thing was that the rest of the picture seemed alright. When I rendered the same picture with a circle-shaped f-stop, the jaggies suddenly disappeared! I browsed through my code of the square f-stop, but I couldn't find any bugs. I also couldn't find a reasonable explanation of the appearance of the jaggies. I figure it might have something to do with the square being not point-symmetric, but that's as far as I can get. I would like to know if someone has experience with the same problem, and does somebody has a good explanation for it ...

Many thanks in advance,
Marinko Laban

back to contents


Query on Frequency Dependent Reflectance

From: hpfcla!sunrock!kodak!supra!reichert@Sun.COM (Mark Reichert x25948)

Hello.

I'm adding fresnel reflectance to my shader. I'm in need of data for reflectance as a function of frequency for non-polarized light at normal incidence. I would like to build a stockpile of this data for a wide variety of materials. I currently have some graphs of this data, but would much prefer the actual sample points in place of the curve-fitted stuff I have now. (not to mention the typing that you might save me).

If you have stuff such as this, and can share it with me, I would be most appreciative. Also, if there is some Internet place where I might look, that would be fine too.

Thanks,

Mark

back to contents


"Best of comp.graphics" ftp Site, by Raymond Brand

A collection of the interesting/useful [in my opinion] articles from comp.graphics over the last year and a half is available for anonymous ftp.

It contains answers to most of the "most asked" questions from that period as well as most of the sources posted to comp.graphics.

Now that you know what is there, you can find it in directory pub/graphics at albanycs.albany.edu.

If you have anything to add to the collection or wish to update something in it, or have have some ideas on how to organize it, please contact me at one of the following.

[There's also a subdirectory called "ray-tracers" which has source code for you-know-whats and other software--EAH]

--------
Raymond S. Brand                 rsbx@beowulf.uucp
3A Pinehurst Ave.                rsb584@leah.albany.edu
Albany NY  12203                 FidoNet 1:7729/255 (518-489-8968)
(518)-482-8798                   BBS: (518)-489-8986

back to contents


Notes on Frequency Dependent Refraction

Newsgroups: comp.graphics

In article 3324@uoregon.uoregon.edu markv@uoregon.uoregon.edu (Mark VandeWettering) writes:
< }< Finally, has anyone come up with a raytracer whose refraction model
< }< takes into account the varying indices of refraction of different light
< }< frequencies? In other words, can I find a raytracer that, when looking
< }< through a prism obliquely at a light source, will show me a rainbow?
< }
< } This could be tough. The red, green, and blue components of monitors
< }only simulate the full color spectrum. On a computer, yellow is a mixture
< }of red and green. In real life, yellow is yellow. You'd have to cast a
< }large number of rays and use a large amount of computer time to simulate
< }a full color spectrum. (Ranjit pointed this out in his article and went
< }into much greater detail).
<
< Actually, this problem seems the easiest. We merely have to trace rays
< of differing frequency (perhaps randomly sampled) and use Fresnel's
< equation to determine refraction characteristics. If you are trying to
< model phase effects like diffraction, you will probably have a much more
< difficult time.

This has already been done by a number of people. One paper by T. L. Kunii describes a renderer called "Gemstone Fire" or something. It models refraction as you suggest to get realistic looking gems. Sorry, but I can't recall where (or if) it has been published. I have also read several (as yet) unpublished papers which do the same thing in pretty much the same way.

David Jevans, U of Calgary Computer Science, Calgary AB T2N 1N4 Canada uucp: ...{ubc-cs,utai,alberta}!calgary!jevans

________

>From: coifman@yale.UUCP (Ronald Coifman)

>> This could be tough. ...
>
>This is the easy part...
>You fire say 16 rays per pixel anyway to do
>antialiasing, and assign each one a color (frequency). When the ray
>is refracted through an object, take into account the index of
>refraction and apply Snell's law. A student here did that
>and it worked fine. He simulated rainbows and diffraction effects
>through prisms.
>
> (Spencer Thomas (U. Utah, or is it U. Mich. now?) also implemented
>the same sort of thing at about the same time.

Yep, I got a Masters degree for doing that (I was the student Rob is refer- ring to). The problem in modelling dispersion is to integrate the primary sample, over the visible frequencies of light. Using the Monte Carlo integra- tion techniques of Cook on the visible spectrum yields a nice, fairly simple solution, albeit at the cost of supersampling at ~10-20 rays per pixel, where dispersive sampling is required.

Thomas used a different approach. He adaptively subdivided the spectrum based on the angle of spread of the dispersed ray, given the range of frequen- cies it represents. This can be more efficient, but can also have unlimited growth in the number of samples. Credit Spencer Thomas; he was first.

As at least one person has pointed out, perhaps the most interesting aspect of this problem is that of representing the spectrum on an RGB monitor. That's an open problem; I'd be really interested in hearing about any solutions that people have come up with. (No, the obvious CIE to RGB conversion doesn't work worth a damn.)

My solution(s) can be found in "A Realistic Model of Refraction for Computer Graphics", F. Kenton Musgrave, Modelling and Simulation on Microcomputers 1988 conference proceedings, Soc. for Computer Simulation, Feb. 1988, in my UC Santa Cruz Masters thesis of the same title, and (hopefully) in an upcoming paper "Prisms and Rainbows: a Dispersion Model for Computer Graphics" at the Graphics Interface conference this summer. (I can e-mail troff sources for these papers to interested parties, but you'll not get the neat-o pictures.)

For a look at an image of a physical model of the rainbow, built on the dispersion model, see the upcoming Jan. IEEE CG&A "About the Cover" article.

                                        Ken Musgrave

Ken Musgrave                    arpanet: musgrave@yale.edu
Yale U. Math Dept.
Box 2155 Yale Station           Primary Operating Principle:
New Haven, CT 06520                             Deus ex machina

back to contents


Sound Tracing

>From: ph@miro.Berkeley.EDU (Paul Heckbert)
Subject: Re: Sound tracing
[source: comp.graphics]

In article (239@raunvis.UUCP) kjartan@raunvis.UUCP
(Kjartan Pierre Emilsson Jardedlisfraedi) asks:
> Has anyone had any experience with the application of ray-tracing techniques
> to simulate acoustics, i.e. the formal equivalent of ray-tracing using sound
> instead of light? ...

Yes, John Walsh, Norm Dadoun, and others at the University of British Columbia have used ray tracing-like techniques to simulate acoustics. They called their method of tracing polygonal cones through a scene "beam tracing" (even before Pat Hanrahan and I independently coined the term for graphics applications).

Walsh et al simulated the reflection and diffraction of sound, and were able to digitally process an audio recording to simulate room acoustics to aid in concert hall design. This is my (four year old) bibliography of their papers:

    %A Norm Dadoun
    %A David G. Kirkpatrick
    %A John P. Walsh
    %T Hierarchical Approaches to Hidden Surface Intersection Testing
    %J Proceedings of Graphics Interface '82
    %D May 1982
    %P 49-56
    %Z hierarchical convex hull or minimal bounding box to optimize intersection
    testing between beams and polyhedra, for graphics and acoustical analysis
    %K bounding volume, acoustics, intersection testing

    %A John P. Walsh
    %A Norm Dadoun
    %T The Design and Development of Godot:
    A System for Room Acoustics Modeling and Simulation
    %B 101st meeting of the Acoustical Society of America
    %C Ottawa
    %D May 1981

    %A John P. Walsh
    %A Norm Dadoun
    %T What Are We Waiting for?  The Development of Godot, II
    %B 103rd meeting of the Acoustical Society of America
    %C Chicago
    %D Apr. 1982
    %K beam tracing, acoustics

    %A John P. Walsh
    %T The Simulation of Directional Sound Sources
    in Rooms by Means of a Digital Computer
    %R M. Mus. Thesis
    %I U. of Western Ontario
    %C London, Canada
    %D Fall 1979
    %K acoustics

    %A John P. Walsh
    %T The Design of Godot:
    A System for Room Acoustics Modeling and Simulation, paper E15.3
    %B Proc. 10th International Congress on Acoustics
    %C Sydney
    %D July 1980

    %A John P. Walsh
    %A Marcel T. Rivard
    %T Signal Processing Aspects of Godot:
    A System for Computer-Aided Room Acoustics Modeling and Simulation
    %B 72nd Convention of the Audio Engineering Society
    %C Anaheim, CA
    %D Oct. 1982

Paul Heckbert, CS grad student
508-7 Evans Hall, UC Berkeley           UUCP: ucbvax!miro.berkeley.edu!ph
Berkeley, CA 94720                      ARPA: ph@miro.berkeley.edu

________

>From: jevans@ucalgary.ca (David Jevans)
Subject: Re: Sound tracing
[source: comp.graphics]

Three of my friends did a sound tracer for an undergraduate project last year. The system used directional sound sources and microphones and a ray-tracing-like algorithm to trace the sound. Sound sources were digitized and stored in files. Emitters used these sound files. At the end of the 4 month project they could digitize something, like a person speaking, run it through the system, then pump the results through a speaker. An acoustic environment was built (just like you build a model for graphics). You could get effects like echoes and such. Unfortunately this was never published. I am trying to convince them to work on it next semester...

David Jevans, U of Calgary Computer Science, Calgary AB T2N 1N4 Canada uucp: ...{ubc-cs,utai,alberta}!calgary!jevans

________

>From: eugene@eos.UUCP (Eugene Miya)

May I also add that you research all the work on acoustic lasers done at places like the Applied Physics Lab.

________

>From: riley@batcomputer.tn.cornell.edu (Daniel S. Riley)
Organization: Cornell Theory Center, Cornell University, Ithaca NY

In article (572@epicb.UUCP) david@epicb.UUCP (David P. Cook) writes:
>>In article (7488@watcgl.waterloo.edu) ksbooth@watcgl.waterloo.edu (Kelly Booth) writes:
>>>[...] It is highly unlikely that a couple of hackers thinking about
>>>the problem for a few minutes will generate startling break throughs
>>>(possible, but not likely).

Ok, I think most of us can agree that this was a reprehensible attempt at arbitrary censorship of an interesting discussion. Even if some of the discussion is amateurish and naive.

> The statement made above
[...]
> Is appalling! Sound processing is CENTURIES behind image processing.
> If we were to apply even a few of our common algorithms
> to the audio spectrum, it would revolutionize the
> synthizer world. These people are living in the stone
> age (with the exception of a few such as Kuerdswell [sp]).

On the other hand, I think David is *seriously* underestimating the state of the art in sound processing and generation. Yes, Ray Kurzweil has done lots of interesting work, but so have many other people. Of the examples David gives, most (xor'ing, contrast stretching, fuzzing, antialiasing and quantization) are as elementary in sound processing as they are in image processing. Sure, your typical music store synthesizer/sampler doesn't offer these features (though some come close--especially the E-mu's), but neither does your vcr. And the work Kurzweil music and Kurzweil applied intelligence have done on instrument modelling and speech recognition go WAY beyond any of these elementary techniques.

The one example I really don't know about is ray tracing. Sound tracing is certainly used in some aspects of reverb design, and perhaps other areas of acoustics, but I don't know at what level diffraction is handled--and diffraction is a big effect with sound propagation. You also have to worry about phases, interference, and lots of other fun effects that you can (to first order) ignore in ray tracing. References, anyone? (Perhaps I should resubscribe to comp.music, and try there...)

(off on a tangent: does any one know of work on ray tracers that will do things like coherent light sources, interference, diffraction, etc? In particular, anyone have a ray tracer that will do laser speckling right? I'm pretty naive about the state of the art in image synthesis, so I have no idea if such beasts exist. It looks like a hard problem to me, but I'm just a physicist...)

>No, this is not a WELL RESEARCHED area as Kelly would have us believe. The
>sound people are generally not attacking sound synthesis as we attack
>vision synthesis. This is wonderful thinking, KEEP IT UP!

Much work in sound synthesis has been along lines similar to image synthesis. Some of it is proprietary, and the rest I think just receives less attention, since sound synthesis doesn't have quite the same level of perceived usefulness, or the "sexiness", of image synthesis. But it is there. Regardless, I agree with David that this is an interesting discussion, and I certainly don't mean to discourage any one from thinking or posting about it.

-Dan Riley (dsr@lns61.tn.cornell.edu, cornell!batcomputer!riley) -Wilson Lab, Cornell U.

________

>From: kjartan@raunvis.UUCP (Kjartan Pierre Emilsson Jardedlisfraedi)
Newsgroups: comp.graphics

We would like to begin by thanking everybody for their good replies, which will in no doubt come handy. We intend to try to implement such a sound tracer soon and we had already made some sort of model for it, but we were checking whether there was some info lying around about such tracers. It seems that our idea wasn't far from actual implementations and that is reassuring.

For the sake of Academical Curiosity and overall Renaissance-like Enlightenment in the beginning of a new year we decided to submit our crude model to the critics and attention of this newsgroup, hoping that it won't interfere too much with the actual subject of the group, namely computer graphics.

The Model:

We have some volume with an arbitrary geometry (usually simple such as a concert hall or something like that). Squares would work just fine as primitives. Each primitive has definite reflection properties in addition to some absorption filter which possibly filters out some frequencies and attenuates the signal. In this volume we put a sound emitter which has the following form:

        The sound emitter generates a sound sample in the form
        of a time series with a definite mean power P.  The emitter
        emits the sound with a given power density given as some
        spherical distribution. For simplicity we tessellate this
        distribution and assign to each patch the corresponding mean
        power.

At some other point we place the sound receptor which has the following form:

        We take a sphere and cut it in two equal halves, and then
        separate the two by some distance d.  We then tessellate the
        half-spheres (not including the cut).  We have then a crude
        model of ears.

Now for the actual sound tracing we do the following:

        For each patch of the two half-spheres, we cast a ray
        radially from the center, and calculate an intersection
        point with the enclosing volume.  From that point we
        determine which patch of the emitter this corresponds to,
        giving us the emitted power.  We then pass the corresponding
        time series through the filter appropriate to the given
        primitives, calculate the reflected fraction, attenuate the
        signal by the square of the distance, and eventually
        determine the delay of the signal.

        When all patches have been traced, we sum up all the time
        series and output the whole lot through some stereo device.

A more sophisticated model would include secondary rays and sound 'shadowing' (The shadowing being a little tricky as it is frequency dependent)

pros & cons ?

                                Happy New Year !!

                                        -Kjartan & Dagur

Kjartan Pierre Emilsson
Science Institute - University of Iceland
Dunhaga 3
107 Reykjavik
Iceland                                 Internet: kjartan@raunvis.hi.is

________

>From: brent@itm.UUCP (Brent)
Organization: In Touch Ministries, Atlanta, GA

Ok, here's some starting points: check out the work of M. Schroeder at the Gottingen. (Barbarian keyboard has no umlauts!) Also see the recent design work on the Orange County Civic Auditorium and the concert hall in New Zealand. These should get you going in the right direction. Dr. Schroeder laid the theoretical work and others ran with it. As far as sound ray tracing and computer acoustics being centuries behind, I doubt it. Dr. S. has done things like record music in stereo in concert halls, digitized it, set up playback equipment in an anechoic chamber (bldg 15 at Murry Hill), measured the path from the right speaker to the left ear, and from the left speaker to the right ear, digitized the music and did FFTs to take out the "crossover paths" he measured. Then the music played back sounded just like it did in the concert hall. All this was done over a decade ago.

Also on acoustic ray tracing: sound is much "nastier" to figure than pencil-rays of light. One must also consider the phase of the sound, and the specific acoustic impedance of the reflecting surfaces. Thus each reflection introduces a phase shift as well as direction and magnitude changes. I haven't seen too many optical ray-tracers worrying about interference and phase shift due to reflecting surfaces. Plus you have to enter vast world of psychoacoustics, or how the ear hears sound. In designing auditoria one must consider "binaural dissimilarity" (Orange County) and the much-debated "auditory backward inhibition" (see the Lincoln Center re-designs). Resonance?? how many optical chambers resonate? (outside lasers?) All in all, modern acoustic simulations bear much more resemblance to Quantum Mechanic "particle in the concert hall" type calculations than to simple ray-traced optics.

Postscript: eye-to-source optical ray tracing is a restatement of Rayleigh's "reciprocity principle of sound" of about a century ago. Acoustitions have been using it for at least that long.

        happy listening,

                brent laminack (gatech!itm!brent)

________

Reply-To: trantow@csd4.milw.wisc.edu (Jerry J Trantow)
Subject: Geometric Acoustics (Sound Tracing)
Summary: Not so easy, but here are some papers
Organization: University of Wisconsin-Milwaukee

Some of the articles I have found include

Criteria for Quantitative Rating and Optimum Design on Concert Halls

Hulbert, G.M.  Baxa, D.E. Seireg, A.
University of Wisconsin - Madison
J Acoust Soc Am v 71 n 3 Mar 83 p 619-629
ISSN 0001-4966, Item Number: 061739

Design of room acoustics and a MCR reverberation system for Bjergsted Concert hall in Stavanger

Strom, S.  Krokstad, A.  Sorsdal, S.  Stensby, S.
Appl Acoust v19 n6 1986 p 465-475
Norwegian Inst of Technology, Trondheim, Norw
ISSN 0003-682X, Item Number: 000913

I am also looking for an English translation of:

Ein Strahlverfolgungs-Verafahren Zur Berechnung von Schallfelern in Raemem [ Ray-Tracing Program for the calculation of sound fields of rooms ]

Voralaender, M.
Acoustica v65 n3 Feb 88 p 138-148
ISSN 0001-7884, Item Number: 063350

If anyone is interested in doing a translation I can send the German copy that I have. It doesn't do an ignorant fool like myself any good and I have a hard time convincing my wife or friends who know Deutch to do the translation.

A good literature search can discover plenty of articles, quite a few of which are about architectural design of music halls. With a large concert hall, the calculations are easier because of the dimensions. (the wavelength is small compared to the dimensions of the hall)

The cases I am interested in are complicated by the fact that I want to work with relatively small rooms, large sources, and to top it off low (60hz) frequencies. I vaguely remember seeing a blurb somewhere about a program done by BOSE ( the speaker company) that calculated sound fields generated by speakers in a room. I would appreciate any information on such a beast.

The simple source for geometric acoustics is described in Beranek's Acoustic in the chapter on Radiation of Sound. To better appreciate the complexity from diffraction, try the chapter on The Radiation and Scattering of Sound in Philip Morse's Vibration and Sound ISBN 0-88318-287-4.

I am curious as to the commercial software that is available in this area. Does anyone have any experience they could comment on???

______

>From: markv@uoregon.uoregon.edu (Mark VandeWettering)
Subject: More Sound Tracing
Organization: University of Oregon, Computer Science, Eugene OR

I would like to present some preliminary ideas about sound tracing, and critique (hopefully profitably) the simple model presented by Kjartan Pierre Emilsson Jardedlisfraedi. (Whew! and I thought my name was bad, I will abbreviate it to KPEJ)

CAVEAT READER: I have no expertise in acoustics or sound engineering. Part of the reason I am writing this is to test some basic assumptions that I have made during the course of thinking about sound tracing. I have done little/no research, and these ideas are my own.

KJEP had a model related below:

> We have some volume with an arbitrary geometry (usually simple such
> as a concert hall or something like that). Squares would work just
> fine as primitives. Each primitive has definite reflection
> properties in addition to some absorption filter which possibly
> filters out some frequencies and attenuates the signal.

One interesting form of sound reflector might be the totally diffuse reflector (Lambertian reflection). It seems that if this is the assumption, then the appropriate algorithm to use might be radiosity, as opposed to raytracing. Several problems immediately arise:

        1.      how to handle diffraction and interference?
        2.      how to handle "relativistic effects" (caused by
                the relatively slow speed of sound)

The common solution to 1 in computer graphics is to ignore it. Is this satisfactory in the audio case? Under what circumstances or applications is 1 okay?

Point 2 is not often considered in computer graphics, but in computerized sound generation, it seems critical to accurate formation of echo and reverberation effects. To properly handle time delay in radiosity would seem to require a more difficult treatment, because the influx of "energy" at any given time from a given patch could depend on the outgoing energy at a number of previous times. This seems pretty difficult, any immediate ideas?

> Now for the actual sound tracing we do the following:
>
> For each patch of the two half-spheres, we cast a ray
> radially from the center, and calculate an intersection
> point with the enclosing volume. From that point we
> determine which patch of the emitter this corresponds to,
> giving us the emitted power. We then pass the corresponding
> time series through the filter appropriate to the given
> primitives, calculate the reflected fraction, attenuate the
> signal by the square of the distance, and eventually
> determine the delay of the signal.
>
> When all patches have been traced, we sum up all the time
> series and output the whole lot through some stereo device.

One open question: how much directional information is captured by your ears? Since you can discern forward/backward sounds as well as left/right, it would seem that ordinary stereo headphones are incapable of reproducing sounds as complex as one would like. Can the ears be fooled in clever ways?

The only thing I think this model lacks is secondary "rays" or echo/reverb effects. Depending on how important they are, radiosity algorithms may be more appropriate.

Feel free to comment on any of this, it is an ongoing "thought experiment", and has made a couple of luncheon conversations quite interesting.

Mark VandeWettering

________

>From: ksbooth@watcgl.waterloo.edu (Kelly Booth)
Organization: U. of Waterloo, Ontario

In article (3458@uoregon.uoregon.edu) markv@drizzle.UUCP (Mark VandeWettering) writes:
>
> 1. how to handle diffraction and interference?
> 2. how to handle "relativistic effects" (caused by
> the relatively slow speed of sound)
>
> The common solution to 1 in computer graphics is to ignore it.

Hans P. Moravec,
"3D Graphics and Wave Theory"
Computer Graphics 15:3 (August, 1981) pp. 289-296.
(SIGGRAPH '81 Proceedings)

[Trivia Question: Why does the index for the proceedings list this as starting on page 269?]

Also, something akin to 2 has been tackled in some ray tracers where dispersion is taken into account (this is caused by the refractive index depending on the frequency, which is basically a differential speed of light).

back to contents


Laser Speckle

>From: jevans@cpsc.ucalgary.ca (David Jevans)

In article (11390016@hpldola.HP.COM), paul@hpldola.HP.COM (Paul Bame) writes:
> A raytracer which did laser speckling right might also be able
> to display holograms.

A grad student at the U of Calgary a couple of years ago did something like this. He was using holographic techniques for character recognition, and could generate synthetic holograms. Also, what about Pixar? See IEEE CG&A 3 issues ago.

David Jevans, U of Calgary Computer Science, Calgary AB T2N 1N4 Canada uucp: ...{ubc-cs,utai,alberta}!calgary!jevans

________

>From: dave@onfcanim.UUCP (Dave Martindale)
Organization: National Film Board / Office national du film, Montreal

Laser speckle is a particularly special case of interference, because it happens in your eye, not on the surface that the laser is hitting.

A ray-tracing system that dealt with interference of light from different sources would show the interference fringes that occur when a laser light source is split into two beams and recombined, and the interference of acoustic waves. But to simulate laser speckle, you'd have to trace the light path all the way back into the viewer's eye and calculate interference effects on the retina itself.

If you don't believe me, try this: create a normal two-beam interference fringe pattern. As you move your eye closer, the fringes remain the same physical distance apart, becoming wider apart in angular position as viewed by your eye. The bars will remain in the same place as you move your head from side to side.

Now illuminate a target with a single clean beam of laser light. You will see a fine speckle pattern. As you move your eye closer, the speckle pattern does not seem to get any bigger - the spots remain the same angular size as seen by your eye. As you move your head from side to side, the speckle pattern moves.

As the laser light reflects from a matte surface, path length differences scramble the phase of light traveling by slightly different paths. When a certain amount of this light is focused on a single photoreceptor in your eye (or a camera), the light combines constructively or destructively, giving the speckle pattern. But the size of the "grains" in the pattern is basically the same as the spacing of the photoreceptors in your eye - basically each cone in your eye is receiving a random signal independent of each other cone.

The effect depends on the scattering surface being rougher than 1/4 wavelength of light, and the scale of the roughness being smaller than the resolution limit of the eye as seen from the viewing position. This is true for almost anything except a highly-polished surface, so most objects will produce speckle.

Since the pattern is due to random variation in the diffusing surface, there is little point in calculating randomness there, tracing rays back to the eye, and seeing how they interfere - just add randomness directly to the final image (although this won't correctly model how the speckle "moves" as you move your head).

However, to model speckle accurately, the pixel spacing in the image has to be no larger than the resolution limit of the eye, about half an arc minute. For a CRT or photograph viewed from 15 inches away, that's 450 pixels/inch, far higher than most graphics displays are capable of. So, unless you have that sort of system resolution, you can't show speckle at the correct size.

back to contents


Eric Haines / erich@acm.org