Nonlinear LLR Scaling for Quantized Turbo Decoder Inputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional turbo decoding techniques face challenges in maintaining Bit Error Rate (BER) performance due to non-optimal scaling of log-likelihood ratios (LLRs), particularly with linear scaling methods that do not effectively distribute LLR values for improved processing in receivers like UMTS systems.
Innovation Solution
A non-linear scaling method is introduced to reshape the LLR distribution by applying a specific function to LLRs generated by the inner receiver, ensuring a more uniform frequency and improved processing for the outer decoder, thereby enhancing BER and Block Error Rate (BLER) performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If linear scaling method is used to scale LLR values, then the processing is simple and fast, but the BER performance is degraded due to non-uniform distribution of LLR values
Solution Approach 1:
The patent applies a non-linear scaling function that transforms the LLR values based on the noise variance parameter. This parameter change approach reshapes the LLR distribution to be more uniform, improving BER performance while maintaining computational feasibility through a mathematically defined transformation rather than complex iterative processes
Solution Approach 2:
The patent performs preliminary scaling of LLR values before they are quantized and processed by the outer decoder. By pre-shaping the distribution of LLR values through non-linear scaling, the system prepares optimal input for subsequent processing stages, ensuring better BER performance without adding complexity to the decoding algorithm itself
2Quantity of substance
If LLR values are quantized to smaller number of bits, then the memory requirement is reduced, but the decoder performance degrades without proper scaling
Solution Approach 1:
The patent changes the distribution parameters of LLR values through non-linear scaling before quantization. This transformation ensures that even when quantized to fewer bits, the scaled LLR values maintain a more uniform distribution that preserves decoder performance, effectively decoupling memory requirements from performance degradation
Solution Approach 2:
The patent performs preliminary non-linear scaling of LLR values before the quantization step. This pre-processing action reshapes the LLR distribution to be more uniform, ensuring that when the values are subsequently quantized to smaller bit representations, the decoder still receives optimally distributed inputs that maintain performance while reducing memory requirements
3Ease of operation
If conventional linear scaling is used, then the implementation is straightforward, but the LLR distribution remains non-uniform leading to suboptimal decoding
Solution Approach 1:
The patent modifies the scaling transformation from linear to non-linear based on noise variance parameters. This parameter change enables the system to adaptively reshape LLR distributions to be more uniform, improving decoding accuracy while maintaining implementation feasibility through a closed-form mathematical solution rather than complex iterative algorithms
Data Source
AI summary
Methods and apparatus are provided for non-linear scaling of log likelihood ratio (LLR) values in a decoder. A decoder according to the present invention processes a received signal by generating a plurality of log-likelihood ratios having a first resolution; applying a non-linear function to the plurality of log-likelihood ratios to generate a plurality of log-likelihood ratios having a lower resolution; and applying the plurality of log-likelihood ratios having a lower resolution to a decoder. The non-linear function can distribute the log-likelihood ratios, for example, such that the frequency of each LLR value is more uniform than a linear scaling.