Real Signal Quantization Using Significant-Bit Distance Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing signal processing technologies face challenges in efficiently quantizing real signals to minimize computational resources and storage needs while maintaining precision, particularly in neural network applications that require numerous operations and large storage capacity.
Innovation Solution
A method for coding real signals into quantized signals by converting them into binary or balanced ternary representations, selecting significant bits, calculating distances, and coding these distances on a limited number of bits, which simplifies multiplication operations and reduces resource requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the real signal is quantized using conventional methods, then the storage capacity and computational resources are reduced, but the precision and performance of signal processing operations are degraded
Solution Approach 1:
The patent changes the parameter representation from conventional fixed-point or floating-point formats to a distance-based coding scheme where real numbers are represented by the distances between their significant bits. This parameter transformation enables more efficient storage and computation while preserving precision, as the distance encoding captures the essential information with fewer bits.
Solution Approach 2:
The patent segments the real signal into significant non-zero bits and represents each signal value by the distances between these segments. By dividing the number representation into discrete distance intervals rather than continuous bit positions, the system achieves compact representation that reduces storage requirements while maintaining the ability to perform precise arithmetic operations.
2Device complexity
If the real signal is quantized to minimize bits, then the hardware structure is simplified, but the computational complexity of multiplication operations increases
Solution Approach 1:
The patent substitutes traditional mechanical multiplication circuits with a distance-based computational approach. Instead of using complex multipliers that multiply binary representations directly, the system computes multiplication by adding the distance codes of the operands and applying a simple decoding function, replacing complex mechanical multiplication with simpler arithmetic operations.
Solution Approach 2:
The patent introduces distance codes as an intermediary representation between the real signal and the hardware implementation. This intermediary encoding scheme serves as a mediator that transforms the multiplication operation into a sequence of simpler steps: distance addition, threshold comparison, and result reconstruction, thereby simplifying the hardware structure while maintaining computational efficiency.
3Quantity of substance
If conventional quantization methods are used, then the signal can be represented with limited bits, but the number of operations required for neural network processing increases
Solution Approach 1:
The patent transforms the parameter representation to distance-based coding, which enables more efficient neural network operations. By representing weights and activations as distance codes rather than conventional binary formats, the system reduces the number of bits required for storage while accelerating processing operations through simplified arithmetic that exploits the distance representation properties.
Solution Approach 2:
The patent performs preliminary conversion of real signals into distance-based codes before neural network processing. This pre-processing step transforms the data into a format that is optimized for subsequent operations, allowing the neural network to process the information more efficiently with fewer computational steps and reduced bit-width requirements throughout the processing pipeline.
Data Source
AI summary
A method implemented by a processor for coding a real signal, for example an image signal, into a quantized signal, comprises the following steps applied to each real sample of the real signal: converting the real sample into a digital representation, selecting, in the fractional part of the number, a predetermined number N of most significant non-zero bits, for each non-zero significant bit i selected, i varying from 1 to N, determining its distance Pi with respect to the neighboring selected non-zero significant bit of higher rank or, for the first non-zero significant bit selected, with respect to the decimal point, deducting from the distance Pi the minimum value of distance between two non-zero bits, coding the modified distance Pi on a predetermined number Mi of bits.