Sparse Signal Encoding with Dominant-Component Vector Quantization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vector quantization methods for encoding and decoding signals, such as acoustic and video signals, often result in 'spectrum holes' due to insufficient code bit length, leading to musical noise and block noise, especially when the number of quantization bits is less than required, causing discontinuous frequency variations that are perceptible as noise.
Innovation Solution
The proposed solution involves selecting different encoding and decoding modes based on the sparsity of input signal samples, where sparse samples are encoded without noise reduction and non-sparse samples are encoded with active quantization of dominant components to prevent spectrum holes, using a normalization value and quantization index to adjust the encoding process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If vector quantization is applied with limited code bit length, then encoding efficiency is improved, but spectrum holes occur causing musical noise and block noise
Solution Approach 1:
The input signal is divided into multiple blocks, and within each block, only dominant components (samples with large absolute values) are quantized using vector quantization. Non-dominant components are handled differently or set to zero. This segmentation approach prevents spectrum holes by ensuring that significant frequency components are preserved while allowing efficient compression of less important components.
Solution Approach 2:
Different quantization strategies are applied to different parts of the signal based on their importance. Dominant components (identified by threshold comparison) receive active quantization to preserve quality, while non-dominant components are treated differently. This local quality approach ensures that resources are allocated to where they are most needed, preventing musical noise while maintaining encoding efficiency.
2Measurement precision
If all samples with large values are quantized using vector quantization, then encoding accuracy is improved, but code bit length increases
Solution Approach 1:
The invention extracts only the dominant components (samples with large absolute values) from the input signal for vector quantization. By identifying and separating these significant components from the rest of the signal, the method applies quantization only where necessary to maintain accuracy, rather than quantizing all samples. This extraction approach reduces the total number of bits required while preserving encoding accuracy for the most important signal components.
Solution Approach 2:
Instead of applying vector quantization to all samples (excessive action), the invention applies it only to dominant components (partial action). This selective approach uses fewer bits than full vector quantization would require, yet maintains sufficient encoding accuracy by focusing computational resources on the most significant signal components that contribute most to the perceived quality.
3Productivity
If the number of quantization bits is reduced, then transmission efficiency is improved, but discontinuous frequency variations increase causing perceptible noise
Solution Approach 1:
Before applying vector quantization with limited bits, the invention performs preliminary identification of dominant components by comparing sample absolute values against a threshold. This preliminary action ensures that the most critical frequency components are identified in advance, so they can be preserved through quantization while non-critical components are allowed to have discontinuities. This preparation step maintains frequency continuity for perceptually important components even with reduced bit allocation.
Solution Approach 2:
The invention changes the quantization parameter allocation dynamically based on signal characteristics. By adjusting which components receive quantization bits (based on their dominance) rather than using uniform bit allocation, the method maintains frequency component continuity for important components while reducing total bit usage. This parameter change approach preserves transmission efficiency while avoiding perceptible noise.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
When a number of samples which are less than a first reference value is a second reference value or less, a second encoding mode is selected. In the second encoding mode, when a difference value that is obtained by subtracting a value corresponding to the quantized normalization value from a value corresponding to the magnitude of each sample is positive and the sample is positive, the difference value is set as a quantization candidate corresponding to the sample; when the difference value is positive and the sample is negative, the sign of the difference value is reversed and the result is set as the quantization candidate corresponding to the sample; and a plurality of quantization candidates are jointly vector-quantized to obtain a vector quantization index. When the second encoding mode is not selected, a first encoding mode other than the second encoding mode is selected.