Encoder Residual Quantization Optimizes Bit Stream Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing transform encoders face challenges in optimizing quantization due to uncertainties in bit consumption, leading to suboptimal performance and unexploited bits in the bit stream.
Innovation Solution
The proposed solution involves an encoder with a quantization stage using a dead zone, followed by an entropy encoder and a residual quantization stage that refines the quantization noise based on the dead zone. This system forms a coded signal that optimizes bit stream utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a first quantization stage is used to quantize the input signal, then the quantization process can be performed with a defined dead zone, but the bit consumption is only known after entropy coding, making optimization difficult
Solution Approach 1:
The patent applies preliminary action by performing a first quantization stage before entropy coding to establish a dead zone and initial quantized values. This preliminary quantization allows the system to define quantization parameters in advance, and then use a residual quantization stage to refine the results based on the actual bit consumption after entropy coding, thus resolving the timing conflict between quantization optimization and bit consumption measurement.
2Reliability
If the bit consumption is over-estimated to avoid truncation, then the bit stream can be completed without hard truncation, but few bits are unexploited in the final bit stream
Solution Approach 1:
The patent implements feedback by using an entropy coding bit estimation unit to determine the actual bit consumption after entropy coding, then feeding this information back to a residual quantization stage. The residual quantization stage uses this feedback to refine the quantized values and utilize any remaining bits in the bit stream, thus avoiding both truncation and bit waste by dynamically adjusting based on actual bit consumption.
3Measurement precision
If an optimization loop is used to converge quantization optimization, then the quantization can be optimized, but the complexity increases and the number of iterations must be strongly limited
Solution Approach 1:
The patent applies segmentation by dividing the quantization process into two distinct stages: a first quantization stage that performs initial quantization with a dead zone, and a residual quantization stage that refines the results. This segmentation eliminates the need for complex iterative optimization loops by structuring the process as a sequential two-stage approach, reducing computational complexity while maintaining optimization effectiveness.
4Productivity
If residual quantization is added to exploit unused bits, then bit stream efficiency can be improved, but the device complexity increases with an additional quantization stage
Solution Approach 1:
The patent introduces an intermediary residual quantization stage that acts as a mediator between the first quantization stage and the entropy coding stage. This residual quantization stage receives the initial quantized values, refines them based on available bit budget information, and produces improved quantized values. The intermediary nature of this stage allows it to coordinate between the quantization and entropy coding processes, improving bit stream efficiency while maintaining a manageable system structure.
Data Source
AI summary
Embodiments of the present invention provide an encoder including a quantization stage, an entropy encoder, a residual quantization stage and a coded signal former. The quantization stage is configured to quantize an input signal using a dead zone in order to obtain a plurality of quantized values. The entropy encoder is configured to encode the plurality of quantized values using an entropy encoding scheme in order to obtain a plurality of entropy encoded values. The residual quantization stage is configured to quantize a residual signal caused by the quantization stage, wherein the residual quantization stage is configured to determine at least one quantized residual value in dependence on the dead zone of the quantization stage. The coded signal former is configured to form a coded signal from the plurality of entropy encoded values and the at least one quantized residual value.