Variable-Length Encoding for All-Zero Quantized Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing encoding technologies face reduced coding efficiency due to long code lengths allocated to symbols with low frequencies of occurrence, particularly when all quantized values are zero, as they are not excluded from encoding targets.
Innovation Solution
An encoding apparatus and method that normalizes input values, generates quantized values, and uses a code table to allocate variable-length codes, excluding combinations of all zero quantized values by replacing them with different combinations, thereby reducing the probability of zero occurrences and optimizing code lengths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If Huffman code table is generated based on frequency of occurrence of quantized data, then coding efficiency is improved for frequent symbols, but code length becomes excessively long for symbols with low frequency of occurrence
Solution Approach 1:
The patent extracts and removes the combination of all-zero quantized values from the encoding targets. By excluding this specific low-frequency symbol combination from the Huffman code table generation, the system avoids allocating long code lengths to it, thereby improving overall coding efficiency without compromising the representation of meaningful data patterns.
Solution Approach 2:
The patent changes the parameter of the code table by generating a Huffman code table that excludes the all-zero quantized value combination. This modification to the code allocation parameters ensures that the code length for this specific symbol is not excessively long, while maintaining optimal code lengths for other frequently occurring symbol combinations.
2Loss of information
If all quantized values are encoded using variable-length codes, then complete data representation is achieved, but maximum code length increases due to inclusion of low-frequency symbols
Solution Approach 1:
The patent selectively extracts the all-zero quantized value combination from the set of encodable symbols. This extraction allows the system to maintain complete representation of meaningful data patterns while excluding the problematic low-frequency case that would otherwise require an excessively long code, thus reducing the maximum code length in the system.
3Measurement precision
If normalization coefficient is selected to be larger than maximum value, then quantization accuracy is improved, but probability of all-zero quantized values increases
Solution Approach 1:
The patent converts the harmful effect of increased all-zero value occurrences (caused by using larger normalization coefficients for better quantization accuracy) into a benefit by excluding these all-zero combinations from the Huffman code table. This allows the system to maintain high quantization accuracy while avoiding the penalty of long code lengths for the resulting frequent zero patterns.
Data Source
AI summary
An encoding apparatus includes a unit that calculates a plurality of normalized values by dividing input values in an input signal by either a normalization coefficient that is closest to a maximum value of absolute values of the input values or a normalization coefficient that is closest to the maximum value from among normalization coefficients that are larger than the maximum value; a unit that generates a plurality of quantized values by quantizing the plurality of normalized values; a unit that stores a code table in which the smaller the probability of occurrence of the plurality of quantized values, the longer the code length of a variable-length code allocated to the plurality of quantized values; and a unit that outputs, when the plurality of quantized values are all zero, a variable-length code allocated to a combination of a plurality of quantized values in accordance with the code table.