<appendix>
<title>Miscellaneous Annotations</title>
<section id="reverb-objects">
<title>Reverberation Objects?</title>
<para>
In a generalization of the I3DL2 Extension for
listener specific reverberation effects, it might
be best to implement Reverb Objects. A Reverb Object
is one example of a parametrized filter. Each
such object encapsulates a set of attributes (the
filter parameters). Sources and Listener alike
have an attribute that allows the application coder
to choose a reverb object for application either
at the origin of the sound, or at the position of
the listener. Initial implementation would only
support one Reverb Object per Context, applied at
the listener position.
</para>
<para>
The I3DL2 Environment is a filter that alters the
way the user experiences the virtual world. As
filters require DSP operations it is limited by hardware
processing capabilities.
</para>
<para>
The I3DL2 Environment models the surroundings of the
listener by simplifying the presumed acoustic properties
of those surroundings into a small set of parameters.
It allows to reproduce the effects of sound reflections
and reverberation caused by walls and obstacles, and
the muffling effects of obstacles inside environments
or partitions between environments.
</para>
<para>
Environment properties:
Early reflections level and delay.
Late reverberation level and delay, low- and high-frequency decay time.
Reverberation diffusion, density and spectrum.
</para>
<para>
Source properties:
Direct path intensity and spectrum.
Reverberation intensity and spectrum.
</para>
</section>
<section id="al-objects-filters">
<title>On Filters</title>
<para>
RFC/bk000502: Filters as the general concept of modifiers?
Environment as a special case filter?
Can we break down EAX environments into ReverbFilters where we
parametrize late reflections, and ReflectFilters, which fake
early reflections? Do we need this separation if we have
calculated or distinct echo effect reflections instead of
stocastic ones? Does it make sense to superimpose a general
reverb kicking in after a delay t, with reflections (random
or not) or should reverb only kick in after reflections are
discarded?
</para>
<para>
RFC/bk000502: old text.
(Environment) Properties:
Geometry - geometry is specified using an immediate mode API which is
similar to OpenGL. Support for scene lists are also provided on a basis
similar to OpenGL's display lists.
Materials - specify the absorptive and reflective qualities of a piece
of geometry. &AL; should provide a facility for accessing preset
materials, and storing and retrieving new materials at runtime.
</para>
<para>
RFC/nn: Atmospheric/ambient properties?
REF/nn: A3D 2.0 IA3dMaterial
</para>
<section id="al-objects-filters-atmospheric">
<title>Atmospheric Filter</title>
<para>
The atmospheric filter the effects of media with constant density,
on propagating sound waves. The effect of the atmospheric filter
is distance dependent. Atmospheric effects can be parameterized
by specifying attenuation per unit distance, the scale for the
unit distance, for one of a minimum of two frequency ranges
(low frequency and high frequency roll-off).
</para>
<para>
RFC/bk000502: do we specify the atmospheric filter per-source?
The effect is clearly dominated by the most dense medium, but
we have little chance simulating crossings between different
media this way. Distance attenuation in media clearly depends
on source and listener being embedded in the same medium,
without any obstruction along the LOS.
</para>
</section>
<section id="al-objects-filters-listenerreverb">
<title>Listener Reverb</title>
<para>
Listener Reverb is a parameterized filter that modifies the sound at
listener position to emulate effects of the surroundings, namely
effects of late reflections. Without simulating sound propagation
this reverb accounts for the averaged outcome of different arrangements
of reflecting/absorbing surfaces around the listener.
</para>
</section>
<section id="al-objects-filters-sourcereverb">
<title>Source Reverb</title>
<para>
There is currently no support for reverb at the source position.
</para>
</section>
<section id="al-objects-filters-reflection">
<title>Reflection Filter</title>
<para>
First order reflection (and, if support, O(n) reflection for small n)
can choose to simulate the effects of different materials by
parametrizing reflection filters.
There is currently no support for reflections.
</para>
</section>
<section id="al-objects-filters-transmission">
<title>Transmission Filter</title>
<para>
Sound propagation along the LOS can pass through obstructions
specified as convex polygons. The effects of lossy transmission
can be approximated by applying a once-off filtering. Like
atmospheric filters, this can be a frequency-dependent roll-off,
unlike atmospheric filters this does not take distance into
account. Transmission filters can be used to emulate losses
on crossing separating surfaces between different media (water/air
borders).
There is currently no support for transmissions.
</para>
</section>
</section>
<section>
<title>Parameterization over Time</title>
<para>
Fading and cross-fading. There are three ways to handle any kind
of gain control as a function of time:
<itemizedlist>
<listitem>
<para>
manipulate gain per frame/sufficiently often
</para>
</listitem>
<listitem>
<para>
parameterize, i.e. specify a target gain,
a duration over which to interpolate, and an interpolation function
</para>
</listitem>
<listitem>
<para>
provide an buffer that indicates amplitude, stretched over
a duration/by a frequency
</para>
</listitem>
The last mechanism also works for early reflections and echos,
and any other temporal filtering. The first and second approach
also work for attributes like Pitch.
</para>
</section>
<section>
<title>On Geometry</title>
<para>
Both the A3D API and implementation as well as EAX related utilities
like EAGLE seem to indicate that any effort to handle scene geoemtry
at API level will inevitably duplicate modules found in common game
engines for purposes of collision detection, path planning, AI
support, visibility and sound propagation culling.
</para>
<para>
In other words, any such effort will inevitably lead to competing
subsystems and multiple use of processing and memory resources to
implement the same functionality. While it makes sense to provide
templates, examples, and even utilities like EAGLE and SDK's to
developers, it makes no sense to integrate any such functionality
with the API.
</para>
<para>
The geometry based processing inevitably leads to a scene graph
API, with all the resulting problems. On closer examination it
seems that the specification and storage of source and listener
positions is a red herring.
</para>
<para>
Second and higher order reflections seem to be irrelevant.
</para>
<para>
Reflection can be faked by stochastic means, but an actual
presence/immersion effect will require smooth transitions
depending on the continuous change of distance between
sources, listener, and dominant reflectors.
</para>
<para>
Dominant reflectors are presumed to be 1st order,
with material properties that incur little or no loss
(or even provide amplification), and significant
surface area.
</para>
<para>
Transmission loss through dense media is equivalent to
the distance attenuation model.
</para>
<para>
Refraction/reflection loss at border surfaces separating
media....
</para>
<para>
No explicit geometry to check whether there is any indirect
(1st order reflection, multiple reflections) path between
source and listener - the application is usually better
equipped to handle this (portal states, PHS). The benefit
of forcing the AL implementation to check for obstruction
(object inersecting LOS) is questionable at best - LOS
checking is also better done by the main application.
In essence, the application might even handle the 1st
order reflections IFF we provide the means to generate
early reflection instead of rolling dice, and if we
make it cheap to enable a path between a source and
the listener complete with a material. Come to think of
it: the implementation guarantees n paths with m filters
one of which is transmission or reflection, one is
distance attenuation, one is source reverb, one is
listener reverb....
</para>
</section>
<section>
<title>No ALU</title>
<note id="bk000019-01"><title>RFC</title><para>
ALU, like GLU, is a problem: linkage dependencies, multiple drivers
sharing one ALU etc. It would be best to not clutter the specification
with ALU/ALUT. Any support code/template repository/SDK can be
maintained as a separate open source project.
</para></note>
<para>
ALU provides operations that do not affect driver or hardware state.
These can be resampling/conversion methods or other sample data
processing, or utilities for (optimized) filter generation. ALU
does not provide I/O operations. At this time,
ALU is not specified and not implemented.
</para>
<para>
RFC/bk000502: GLU is becoming a bit of a problem right now, with
most applications avoiding it as they load GL DLL's explicitely,
but do not trust the ABI specification enough to link against GLU,
and not bothering to load it. A vendor-neutral open source ALU
works for me, but we can not accept link time dependencies to AL.
ALU (like GLU) is meant for minimal convenience, in small
building blocks, it is not meant as an SDK.
</para>
<para>
RFC/bk000502: old text.
ALU is the AL Utility library, and provide functions for performing audio
conversion, preset material properties, and a compatability layer for
legacy stereo format audio. This includes support for panning and per
channel volume control.
</para>
<para>
RFC/nn: Er, what else does the world of 2D sound usually need?
</para>
</section>
<section>
<title>No ALUT</title>
<para>
Application coders frequently request additional support for sound
handling to the extent of sophisticated SDKs. It is expected that
SDK vendors will provide such products on top of AL. ALUT (in analogy
to GLUT) would constitute an "official" SDK if desired. At this time,
ALUT is not specified and not implemented, and not intended to be part
of &AL; proper.
</para>
<para>
ALUT is a utility toolkit for &AL;. It sits on top of ALC.
It provides convenience functions for accessing files, for playing sounds,
and an API for accessing CDROM functionality.
</para>
</section>
<!-- Driver DDK section
http://www.microsoft.com/directx/
http://www.microsoft.com/DDK/DDKdocs/win98ddk/ds-ddk_8eas.htm
-->
</appendix>
