Ambisonics

Scene-based audio: orders, channel ordering, normalization, decoding and when to use it.

This page is a reference for Ambisonics, the scene-based approach to spatial audio that describes sound per direction rather than per speaker. It covers first-order B-format, higher orders and their channel counts, channel ordering and normalisation conventions, encoding, decoding and rotation, how Ambisonics is carried in Eclipsa Audio, and when to use it instead of channels or objects.

It is written for engineers, sound designers, VR and 360 video producers and developers who work with Ambisonics files or with the third-order scenes Spatial9 produces for Eclipsa Audio.

Definition

Most audio formats store one signal per speaker: left, right, rear left and so on. Ambisonics stores a picture of the entire sound field around a single listening point, from every direction, including above and below. Because it is not tied to a speaker layout, one Ambisonics file can be decoded for headphones, for any speaker arrangement, or for VR and 360 video, and it can be rotated as the listener turns.

Ambisonics was developed in the 1970s, principally through the work of Michael Gerzon and Peter Craven. It became mainstream with virtual reality, 360 video and headphone spatial audio, and it is one of the two audio element types in IAMF, the specification behind Eclipsa Audio.

Spatial models compared

A scene-based description has a fixed channel cost. A 300-track session and a 5-track session both become 16 channels at third order.

First-order Ambisonics and B-format

First-order Ambisonics uses four signals, together called B-format.

Combining these four signals in different proportions lets a decoder estimate how loud the sound field is in any direction, which is why four signals are enough to describe a full sphere, at least broadly.

A-format and B-format

An Ambisonics microphone has four capsules arranged in a tetrahedron. Their raw signals are called A-format. Software converts A-format into the W, X, Y and Z signals of B-format, which is the form you edit, store and play back.

Orders and channel counts

Higher orders add signals that sharpen directional detail. The channel count for order N is (N + 1)². Each order adds 2N + 1 components, so third order uses 1 + 3 + 5 + 7 = 16 components.

FOA means first-order Ambisonics; HOA means higher-order Ambisonics, second order and above.

Channel ordering and normalisation

Ambisonics files follow conventions for the order in which signals are stored and how they are scaled. Two conventions are in common use.

Both conventions describe the same sound field, but the signals are arranged and scaled differently. Reading a file with the wrong convention places sounds in the wrong position. IAMF scene-based elements and Spatial9 conversions use ACN ordering with SN3D normalisation, the same conventions as AmbiX.

Encode, decode and rotate

Ambisonics separates making the sound field from playing it.

Encoding

Encoding places sounds into the sphere.

• An Ambisonics microphone encodes a real space directly.
• A mono recording can be panned to any direction in software.
• An existing immersive mix can be converted. Spatial9 encodes Dolby Atmos beds and objects into a third-order scene, following each object's movement over time.

Decoding

Decoding turns the sphere into signals for real speakers or into a binaural signal for headphones. Speaker decoding is a matrix operation that produces feeds for 2.0, 5.1, 7.1.4 or other layouts. The decoder adapts to the speakers it has, so the same file works from stereo to large 3D layouts. Binaural decoding converts the whole sphere into two ear signals with a fixed set of filters. See Binaural rendering, headphones and speakers.

Rotation

The whole sphere can be rotated with a single matrix operation. This is why 360 video can turn the sound with the viewer's gaze and why head-tracked headphones work well with Ambisonics. In the libiamf decoder, head tracking counter-rotates an intermediate Ambisonics bed so sources stay fixed in the room as the listener turns.

Ambisonics in Eclipsa Audio

IAMF supports scene-based audio elements that carry an Ambisonics sound field, which each device renders for its own speakers or headphones.

In mono mode the channel_mapping field can mark channels as silent so they cost nothing. In projection mode a demixing matrix is sent so the decoder can restore the Ambisonics signals. Because Ambisonics channels in a scene are highly correlated, efficient codecs such as Opus keep a full third-order scene within practical streaming bitrates. See Eclipsa Audio and IAMF reference.

Third order in Spatial9 conversions

When Spatial9 converts a Dolby Atmos ADM BWF master to Eclipsa Audio, it uses motion-aware translation to build a third-order scene of 16 components. The conversion must:

1. Read every bed and object, including the time blocks that record movement.
2. Map room-centred Atmos positions to listener-centred directions so front stays front, overhead stays overhead and side stays side.
3. Translate object size into a matching spread in the sound field.
4. Follow movement over time, so a sweeping sound keeps sweeping.
5. Carry low-frequency content deliberately, because Ambisonics has no dedicated LFE channel.
6. Encode to third order to keep enough directional detail for music.

The Intent Fidelity Score then compares placement in the original master with the converted scene. See ADM to Eclipsa Audio.

Ambisonics export options in Spatial9

The Spatial9 Immersive Classroom exports Ambisonics at orders 1 to 4, alongside binaural, stereo, surround layouts and Eclipsa Audio video.

Choose the highest order your destination platform accepts. Real file exports in the classroom are part of the Student Pro plan. See Education and Academy.

When to use Ambisonics

Limitations

• Low orders are blurry. First order gives broad, soft positions; sharp placement needs third order or higher, which means larger files.
• Speaker decodes have a sweet spot. A decode sounds best at the centre of a regular layout; away from the centre, or with uneven speaker spacing, images drift.
• A third-order scene is precise enough for music but is not a perfect point. A very small object can sound slightly broader than in an object renderer.
• Platforms differ. YouTube 360 and VR videos use first-order Ambisonics, so a higher-order mix must be reduced for that route.
• There is no per-instrument control after encoding. Once sounds are in the sphere you cannot extract one instrument to rebalance it. Keep the original stems or objects.
• Converting an Ambisonics scene back into objects would require separating the sources again.

Troubleshooting

Next steps

• Eclipsa Audio and IAMF reference describes how scene-based elements are stored in the bitstream.
• Binaural rendering, headphones and speakers explains how a sound field is rendered for each playback system.
• ADM to Eclipsa Audio shows how an Atmos master becomes a third-order scene.
• Immersive audio formats compared places Ambisonics alongside channel and object formats.