MAP Decoder Recursion Without Normalization Memory Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional maximum a posteriori (MAP) decoders are hardware intensive and require significant memory resources due to the need for normalization and storage of branch metric values, which complicates the decoding process.
Innovation Solution
The proposed method employs a MAP decoder that computes relative log-likelihoods instead of absolute log-likelihoods, eliminating the need for normalization and reducing memory requirements by using forward and backward recursion steps to calculate alpha and beta values from soft-decision values, thereby simplifying the decoding process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional MAP decoder uses normalization devices to maintain branch metric values within quantization limits, then decoding accuracy is improved, but hardware complexity and memory requirements increase
Solution Approach 1:
The patent extracts and removes the normalization devices from the conventional MAP decoder architecture. By computing only relative log-likelihoods instead of absolute values, the normalization step is eliminated entirely, reducing hardware complexity while maintaining decoding accuracy through the preserved relative relationships between path metrics
Solution Approach 2:
The patent changes the parameter representation from absolute log-likelihood values to relative log-likelihood values. This parameter transformation eliminates the need for normalization operations while preserving the essential comparative information needed for accurate decoding decisions
2Measurement precision
If conventional MAP decoder stores all branch metric values M0, M1, . . . , MN−1 and output probability values in memory, then decoding accuracy is maintained, but memory requirements increase
Solution Approach 1:
The patent removes the requirement to store absolute branch metric values and output probability values in memory. By working exclusively with relative log-likelihoods computed through forward and backward recursion, the memory storage requirement is eliminated while maintaining the ability to perform accurate extrinsic computations
Solution Approach 2:
The patent performs preliminary computation of relative log-likelihoods during the forward and backward recursion steps, so that when extrinsic computations are needed, the necessary relative metric values are already available or can be computed on-demand without requiring large memory storage
3Reliability
If conventional MAP decoder implements normalization devices and large memory storage, then decoding performance is improved, but hardware intensity increases
Solution Approach 1:
The patent extracts and removes the hardware-intensive normalization devices and large memory structures from the conventional MAP decoder. By reformulating the decoding process to use relative log-likelihoods computed through recursive relationships, the patent achieves comparable decoding performance with significantly reduced hardware intensity
Solution Approach 2:
The patent replaces the mechanical normalization operation with a mathematical reformulation that computes relative log-likelihoods directly through forward and backward recursion. This substitution eliminates the need for complex normalization hardware while preserving decoding performance
Data Source
AI summary
A method is provided for performing a MAP probability decoding of a sequence R(n) including N bits of encoded data. The method includes the steps of: (a) generating a sequence rn of sot-values by processing the sequence R(n); (b) performing a forward recursion by computing alpha values αS,SG utilizing the soft-decision values; (c) performing a backward recursion by computing beta values βS,SG utilizing the soft-decision values; and (d) performing an extrinsic computation by computing probability values p′k. The alpha values αS,SG are relative log-likelihoods of an encoding process arriving at various states. The beta values βS,SG are relative log-likelihoods of the encoding process arriving at various states. The probability values p′k represent a set of probabilities indicating that each data bit of an input sequence dK had a value equal to zero or one. The sequence R(n) represents an encoded form of the input sequence dK.