Trellis Likelihood Metrics for Multi-Window Error Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for computing likelihood metrics in trellis-based detection and decoding are inadequate for accurately predicting reliability in communication and storage systems affected by severe defects that span multiple code words or error update windows, leading to overestimation of reliability and potential catastrophic errors.
Innovation Solution
The method involves generating likelihood metrics by considering both converging and non-converging paths in trellis-based detection, using error patterns and error event generators to update metrics based on path metrics and alternate paths that do not necessarily converge to the same winning state, thereby accounting for defects and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If known computations of likelihood metrics rely solely on metrics associated with trellis paths that converge to the same final state, then the computation is simple and fast, but the accuracy of reliability prediction deteriorates when errors span multiple code words or error update windows
Solution Approach 1:
The patent segments the likelihood metric computation into two distinct parts: (1) metrics from converging paths within the current error update window, and (2) metrics from non-converging paths that span multiple code words or windows. This segmentation allows the system to handle different error scenarios separately, improving accuracy without overwhelming computational complexity.
Solution Approach 2:
The patent extends the computation from a single-dimension approach (only converging paths) to a multi-dimensional approach by incorporating non-converging paths that extend beyond the current error update window. This adds a temporal dimension to the metric computation, considering paths that converge at different future states, thereby capturing errors that span multiple code words.
2Reliability
If likelihood metrics are computed using only converging paths, then the processing speed is maintained, but the reliability estimation becomes inaccurate for severe defects causing errors across multiple code words
Solution Approach 1:
The patent performs preliminary computation of path metrics during the trellis decoding process itself, storing these metrics for later use. When computing likelihood metrics, the system retrieves pre-computed metrics from both converging and non-converging paths, avoiding redundant calculations and reducing real-time computation time while maintaining accurate reliability estimation.
3Measurement precision
If metrics from non-converging paths are included in likelihood metric computation, then accuracy for defective bits is improved, but the overall system complexity increases
Solution Approach 1:
The patent applies local quality by treating different bit positions differently in the likelihood metric computation. For bits that are part of non-converging paths spanning multiple code words, the system incorporates additional metric information from these extended paths. For bits not affected by such paths, the computation remains based on traditional converging paths only, thus improving accuracy locally where needed without uniformly increasing complexity everywhere.
Data Source
AI summary
Systems and methods for generating likelihood metrics for trellis-based detection and/or decoding are described. In some embodiments, likelihood metrics for a first subset of bit locations in an error pattern (e.g., bit locations that fall within the error event update window) are updated based on a first metric, such as the path metric difference, associated with an alternate path that converges to the same trellis state as the decoded sequence. In some embodiments, likelihood metrics for a second subset of bit locations in the error patterns (e.g., bit locations that do not fall within the error event update window) are updated based on a second metric, such as a predetermined value of zero, a small metric, or the path metric difference for a path that does not converge into the same winning state as the decoded sequence for the particular error update window of interest.