1 OpenAL Soft's renderer has advanced quite a bit since its start with panned
2 stereo output. Among these advancements is support for surround sound output,
3 using psychoacoustic modeling and more accurate plane wave reconstruction. The
4 concepts in use may not be immediately obvious to people just getting into 3D
5 audio, or people who only have more indirect experience through the use of 3D
6 audio APIs, so this document aims to introduce the ideas and purpose of
7 Ambisonics as used by OpenAL Soft.
13 Originally developed in the 1970s by Michael Gerzon and a team others,
14 Ambisonics was created as a means of recording and playing back 3D sound.
15 Taking advantage of the way sound waves propagate, it is possible to record a
16 fully 3D soundfield using as few as 4 channels (or even just 3, if you don't
17 mind dropping down to 2 dimensions like many surround sound systems are). This
18 representation is called B-Format. It was designed to handle audio independent
19 of any specific speaker layout, so with a proper decoder the same recording can
20 be played back on a variety of speaker setups, from quadraphonic and hexagonal
21 to cubic and other periphonic (with height) layouts.
23 Although it was developed decades ago, various factors held ambisonics back
24 from really taking hold in the consumer market. However, given the solid
25 theories backing it, as well as the potential and practical benefits on offer,
26 it continued to be a topic of research over the years, with improvements being
27 made over the original design. One of the improvements made is the use of
28 Spherical Harmonics to increase the number of channels for greater spatial
29 definition. Where the original 4-channel design is termed as "First-Order
30 Ambisonics", or FOA, the increased channel count through the use of Spherical
31 Harmonics is termed as "Higher-Order Ambisonics", or HOA. The details of higher
32 order ambisonics are out of the scope of this document, but know that the added
33 channels are still independent of any speaker layout, and aim to further
34 improve the spatial detail for playback.
36 Today, the processing power available on even low-end computers means real-time
37 Ambisonics processing is possible. Not only can decoders be implemented in
38 software, but so can encoders, synthesizing a soundfield using multiple panned
39 sources, thus taking advantage of what ambisonics offers in a virtual audio
46 Positional sound has come a long way from pan-pot stereo (aka pair-wise).
47 Although useful at the time, the issues became readily apparent when trying to
48 extend it for surround sound. Pan-pot doesn't work as well for depth (front-
49 back) or vertical panning, it has a rather small "sweet spot" (the area the
50 head needs to be in to perceive the sound in its intended direction), and it
51 misses key distance-related details of sound waves.
53 Ambisonics takes a different approach. It uses all available speakers to help
54 localize a sound, and it also takes into account how the brain localizes low
55 frequency sounds compared to high frequency ones -- a so-called psychoacoustic
56 model. It may seem counter-intuitive (if a sound is coming from the front-left,
57 surely just play it on the front-left speaker?), but to properly model a sound
58 coming from where a speaker doesn't exist, more needs to be done to construct a
59 proper sound wave that's perceived to come from the intended direction. Doing
60 this creates a larger sweet spot, allowing the perceived sound direction to
61 remain correct over a larger area around the center of the speakers.
63 In addition, Ambisonics can encode the near-field effect of sounds, effectively
64 capturing the sound distance. The near-field effect is a subtle low-frequency
65 boost as a result of wave-front curvature, and properly compensating for this
66 occurring with the output speakers (as well as emulating it with a synthesized
67 soundfield) can create an improved sense of distance for sounds that move near
74 As a 3D audio API, OpenAL is tasked with playing 3D sound as best it can with
75 the speaker setup the user has. Since the OpenAL API doesn't expose discrete
76 playback speaker feeds, an implementation has a lot of leeway with how to deal
77 with the audio before it's played back for the user to hear. Consequently,
78 OpenAL Soft (or any other OpenAL implementation that wishes to) can render
79 using Ambisonics and decode the ambisonic mix for a high level of accuracy over
80 what simple pan-pot could provide.
82 In addition to surround sound output, Ambisonics also has benefits with stereo
83 output. 2-channel UHJ is a stereo-compatible format that encodes some surround
84 sound information using a wide-band 90-degree phase shift filter. This is
85 generated by taking the ambisonic mix and deriving a front-stereo mix with
86 with the rear sounds filtered in with it. Although the result is not as good as
87 3-channel (2D) B-Format, it has the distinct advantage of only using 2 channels
88 and being compatible with stereo output. This means it will sound just fine
89 when played as-is through a normal stereo device, or it may optionally be fed
90 to a properly configured surround sound receiver which can extract the encoded
91 information and restore some of the original surround sound signal.
94 What Are Its Limitations?
95 =========================
97 As good as Ambisonics is, it's not a magic bullet that can overcome all
98 problems. One of the bigger issues it has is dealing with irregular speaker
99 setups, such as 5.1 surround sound. The problem mainly lies in the imbalanced
100 speaker positioning -- there are three speakers within the front 60-degree area
101 (meaning only 30-degree gaps in between each of the three speakers), while only
102 two speakers cover the back 140-degree area, leaving 80-degree gaps on the
103 sides. It should be noted that this problem is inherent to the speaker layout
104 itself; there isn't much that can be done to get an optimal surround sound
105 response, with ambisonics or not. It will do the best it can, but there are
106 trade-offs between detail and accuracy.
108 Another issue lies with HRTF. While it's certainly possible to play an
109 ambisonic mix using HRTF and retain a sense of 3D sound, doing so with a high
110 degree of spatial detail requires a fair amount of resources, in both memory
111 and processing time. And even with it, mixing sounds with HRTF directly will
112 still be better for positional accuracy.