Rotation Matrix Quantization for Low-Bitrate Spatial Audio Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current encoding methods for ambisonic signals, such as multi-mono encoding, fail to effectively utilize channel correlations, leading to spatial deformations and artifacts like phantom sound sources and diffuse noises, especially at low bit rates.
Innovation Solution
Optimized encoding of rotation matrices using quaternion-based quantization, where quaternions are converted to spherical coordinates and quantized over a half-length interval, minimizing the need for complex conversions and dictionaries, and ensuring efficient decorrelation of channels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If multi-mono encoding is used for ambisonic signals, then encoding simplicity is maintained, but spatialization quality deteriorates due to failure to utilize channel correlations
Solution Approach 1:
The patent combines multiple ambisonic channels into a decorrelated representation using rotation matrices, merging the channel information in a way that preserves spatial correlations while enabling efficient encoding. This resolves the contradiction by maintaining encoding simplicity through a unified transformation approach while improving spatialization quality through proper channel decorrelation.
Solution Approach 2:
The patent changes the parameter representation by applying rotation matrices to transform the ambisonic channel parameters into a decorrelated domain. This parameter transformation enables better utilization of channel correlations, improving spatialization quality while maintaining encoding efficiency through the structured parameter change.
2Device complexity
If rotation matrices are quantized using traditional methods, then encoding complexity is reduced, but spatialization accuracy deteriorates due to spatial deformations and artifacts
Solution Approach 1:
The patent applies preliminary decorrelation transformations using rotation matrices before quantization. This preliminary action prepares the data in an optimal form that reduces the negative impact of subsequent quantization, thereby maintaining spatialization accuracy while keeping encoding complexity manageable through pre-processed data.
Solution Approach 2:
The patent introduces an intermediary decorrelated representation as a intermediate step between the original ambisonic channels and the final quantized encoding. This intermediary form acts as a mediator that preserves spatial relationships while facilitating more accurate quantization, thus resolving the contradiction between complexity and precision.
3Loss of energy
If bit rate is reduced for efficient transmission, then bandwidth consumption is lowered, but spatialization quality deteriorates with phantom sound sources and diffuse noises
Solution Approach 1:
The patent changes the parameter domain by applying rotation matrices to transform ambisonic channels into a decorrelated representation. This parameter change enables more efficient bit allocation, allowing lower bandwidth consumption while maintaining spatialization quality by encoding the decorrelated parameters that better represent the spatial information.
Solution Approach 2:
The patent discards redundant correlated information through the decorrelation transformation and recovers the essential spatial characteristics in the transformed domain. This selective discarding of redundant data reduces bandwidth requirements while the recovery of spatial relationships in the decorrelated form maintains spatialization quality even at lower bit rates.
Data Source
AI summary
A method for encoding a multichannel sound signal, a corresponding decoding method, an encoding device and a decoding device. The encoding method includes: forming a transformation matrix in the form of a rotation matrix to be applied to the input signals; quantifying the rotation matrix; and encoding the transformed signals after application of the rotation matrix. The quantifying of the rotation matrix includes the following operations: converting the rotation matrix in the quaternion domain with at least one first quaternion; forcing the first quaternion to have a positive component; converting the at least one first quaternion into spherical coordinates, one of the spherical coordinates being associated with the forced positive component of the first quaternion; and quantifying the resulting rotating spherical coordinates, the spherical coordinate associated with the forced positive component of the first quaternion being quantified in a half-length interval.