Ambisonic Toolkit Kernels

A set of Finite Impulse Response (FIR) filter kernels for use by convolution-based encoders and decoders.

Features

• Sample Rates: 44100, 48000, 88200, 96000, 176400, 192000
• Full band: DC to Nyquist
• Phase Referenced: W, linear phase
• Constant delay: Predictable delay dependent on kernel size
• Channel Order: Gerzon / Furse-Malham, aka, Classic first-order (FOA)
• Normalization: MaxN, aka, Classic first-order (FOA)

The ATK's FIR filter kernels are designed to support a variety of useful encoding and decoding strategies. Encoding means, converting from some other signal format, e.g., monophonic or stereophonic, to Ambisonic B-format. Decoding means converting from Ambisonic B-format.

All kernels are constructed so that the zero order harmonic, W, is linear phase. This gives all encoder and decoder designs a constant delay in samples of:

delay = (kernel_size - 1) / 2

The soundfield's pressure component is delayed by delay, and the phase response of the higher order harmonics then referenced (offset) to preserve the required relationships. The result is, all kernels of the same kernel_size will produce a resulting signal delayed by the same number of samples.

Encoders

• SuperStereo - Classic Ambisonic Super Stereo encoding. Converts standard two channel stereo to pantophonic (Horizontal-Only) B-format by wrapping the input across a wide frontal stage.

  Note: We strongly advise using this encoder as the default stereo encoder.

• UHJ - Classic Ambisonic UHJ format encoding. Converts stereo compatible two channel Ambisonic UHJ format to pantophonic (Horizontal-Only) B-format. Shelf filtering optimized for dual-band psycho-acoustic decoding is included for convenience.

• Spectral Spreading - The frequency spreading technique encodes a monophonic (single channel) input into B-format by smoothly rotating the signal across the soundfield sphere, by frequency. The spectrum is decomposed and then wrapped as planewaves.

• Diffuse Field - Encodes a monophonic (single channel) input into B-format through frequency dependent phase diffusion. Equivalent to applying the spatial characteristics of a diffuse soundfield to the input sound. In the natural world, diffuse soundfields are found as the very late, immersive part of reverberation.

Decoders

• UHJ - Classic Ambisonic UHJ format decoding. Converts B-format to stereo compatible two channel Ambisonic UHJ format.

  Note: We strongly advise using this decoder as the default stereo decoder.

• Binaural - Binaural decoding. Converts B-format to two channel binaural output. Three full-band diffuse field equalized models are implemented:

  • Synthetic, Duda spherical head model
  • Measured, UC Davis CIPIC HRTF database
  • Measured, IRCAM Listen HRTF database

  Discussion: The Synthetic and Measured decoders are implemented as frequency dependent multi-band decoders, and can be regarded as spectrally optimized for hearing in different ranges.

  The Synthetic decoder is 4-in-1:

  1. DC - 700Hz: strict soundfield, aka basic. Retain phase.
  2. 700 - 2000Hz: maximum energy, aka max rE. Retain phase.
  3. 2000 - 7000Hz: strict soundfield, aka basic. Discard phase.
  4. 7000Hz - Nyquist: virtual stereo microphones in cardioid pair. Look angles = +-100deg

  Synthetic spherical head models a sphere only, and is equivalent to placing a pair of spaced, sphere baffled microphones in the soundfield. With no pinnae and no body, this spherical head model is symmetric across elevation. As elevation cues are not present, there is no Z component. That is, this decoder is pantophonic (Horizontal-Only).

  The Measured decoders are 5-in-1:

  1. DC - LF: omni-directional, W only.
  2. LF - 700Hz: strict soundfield, aka basic. Retain phase.
  3. 700 - 2000Hz: maximum energy, aka max rE. Retain phase.
  4. 2000 - 18000Hz: strict soundfield, aka basic. Discard phase.
  5. 18000Hz - Nyquist: response patterns measured from 4., at 18000Hz.

  LF = 35Hz & 60Hz for CIPIC & Listen.

  Measured decoders are designed to compensate for the proximity (near-field effect) of the two HRTF measurement systems. Asymmetry, across all axes, found in the measured HRTFs is preserved.

  Note: You should have received a license notice with these kernels. Please see the included Third Party Notices for further details on the CIPIC and Listen databases.

---

Using Kernels with Reaper

The kernels are already included as part of the ATK for Reaper install, and there is no need to download and install the kernels seperately.

Using Kernels with SuperCollider

Install ATK for SuperCollider. Launch SuperCollider3, and run the following code:

ATK Kernel Installation
(
// Create ATK support directory
// Place unzipped kernels in the directory opened  

Atk.createUserSupportDir;
Atk.openUserSupportDir;
)

If ATK for SuperCollider has been correctly installed, the above code will open the ATK's user support directory. Place the downloaded kernels here.

If you use ATK for Reaper as well as ATK for SuperCollider on Mac OSX, the kernels are shared between the two programs, and they will reside in the same location. We do not expect this to cause any problems.

---

Feedback and Bug Reports

Known issues are logged at GitHub.

If you experience problems or have questions pertaining to the ATK Kernels, please create an issue in the ATK-Kernels issue tracker.

If you use the kernels for an artistic project, please let us know. We plan on adding a gallery of example artistic and creative projects that make use of the Ambisonic Toolkit.

If you wish to use the kernels as part of a technical or software project, we'd like to know about that too. We're planning to link to other projects making use of Ambisonic Toolkit assets.

List of Changes

Version 1.2.1

• Changes:

  • Updates to README

• Bug Fixes:

• NFC radius for CIPIC binaural decoders - Issue #9

Version 1.2.0

• New Features:

• Measured binaural decoders in all supported SRs - Issue #1

• Bug Fixes:

• FOA spherical decoder kernels are silent at 176.4 and 192kHz - Issue #6

Version 1.1.0

• New Features:
• Adds kernel support for 176.4 Hz - Issue #3

Version 1.0.1

• There are no changes to actual sound kernels content in this release.
• Changes:
• Clarifying licensing
• Updates to README

Version 1.0.0

• Initial release after transition to GitHub

Credits

The filter kernels distributed with the Ambisonic Toolkit are licensed under a Creative Commons Attribution-Share Alike 3.0 Unported (CC BY-SA 3.0) License and are copyright the Ambisonic Toolkit Community and Joseph Anderson, 2011, 2016.

• J Anderson : [e-mail]