Vector Decoding with Multi-Level Permutation for Low-Complexity Audio Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing multi-rate lattice vector quantization approach in audio coding, such as in the TCX part of the AMR-WB+ algorithm, involves complex calculations with multiple irregular factorials, leading to high calculation complexity during coding and decoding.
Innovation Solution
A vector decoding method that decomposes an initial vector into a sign code and an absolute vector code, performs multi-level permutation-based decoding on the absolute vector, and combines the sign and initial absolute vectors to obtain the initial vector, reducing calculation complexity by avoiding complex factorial calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multi-rate lattice vector quantization is used for audio coding, then quantization accuracy is improved, but calculation complexity increases due to multiple irregular factorial computations
Solution Approach 1:
The patent segments the vector quantization process into multiple stages: initial vector generation, codebook selection, and index determination. By dividing the complex quantization task into manageable segments, the calculation complexity is reduced while maintaining quantization accuracy. The multi-rate structure processes vectors at different resolutions in separate stages rather than requiring all computations simultaneously.
Solution Approach 2:
The patent performs preliminary actions by pre-defining codebooks (Q0, Q2, Q3, Q4) with specific leaders and combinations before the actual quantization process. These pre-prepared codebooks contain all possible quantization outcomes, allowing the decoder to simply look up results rather than performing complex factorial calculations in real-time. The leaders and combination tables are computed in advance and stored for efficient retrieval.
2Measurement precision
If complex codebook structures with multiple leaders and combinations are used, then vector quantization precision is improved, but storage requirements increase
Solution Approach 1:
The codebook structure is segmented into multiple independent codebooks (Q0, Q2, Q3, Q4), each with its own leaders and combination tables. This segmentation allows the system to store codebooks in a structured manner where only relevant portions need to be retained in memory during decoding. The patent organizes codebook data into tables with specific columns (Leader, Combination, Sign, Size) that can be efficiently stored and accessed.
Solution Approach 2:
The patent uses simplified representation copies of the full codebook structures. Instead of storing complete multi-dimensional codebook data, it stores compressed representations including leader vectors, combination indices, and sign information. The decoder reconstructs the necessary codebook portions on-demand using these compact copies, significantly reducing storage requirements while maintaining quantization precision.
Data Source
AI summary
A vector decoding method includes decomposing a code of an initial vector to obtain a sign code and an absolute vector code, decoding the sign code to obtain a sign vector, performing multi-level permutation-based decoding on the absolute vector code to obtain an initial absolute vector, and combining the sign vector and the initial absolute vector to obtain an initial vector, wherein each respective vector ha a plurality of elements. A vector decoding apparatus includes a decomposing unit configured to decompose a vector code to obtain a sign code and an absolute vector, a sign decoding unit configured to decode the sign code to obtain a sign vector, a multi-level permutation-based decoding unit configured to perform multi-level permutation-based decoding on the absolute vector to obtain an initial absolute vector, and a combining unit configured to combine the sign vector and the initial absolute vector to obtain an initial vector.