Stochastic Computing Circuit Error Reduction in Satellite Relays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Stochastic computing in repeaters for satellite communication systems experiences high operation errors due to conversion and operation errors when using pseudo-random number sequences, particularly in multi-stage weighted addition units.
Innovation Solution
An operation circuit with data conversion, coefficient storage, coefficient conversion, and weighted addition units, along with a control unit to select input sequences, reduces operation errors by optimizing the input sequences to weighted addition units.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If consumer FPGAs are used in cosmic space to improve performance and reduce cost, then performance and cost are improved, but radiation resistance deteriorates
Solution Approach 1:
The patent converts the harmful effect of radiation-induced bit inversions into a beneficial feature by using stochastic computing. In stochastic computing, numerical values are represented by pseudo-random number sequences where the probability of 1s corresponds to the value. Bit inversions caused by radiation only slightly alter the probability distribution, resulting in minimal arithmetic errors. This allows consumer FPGAs with lower radiation resistance to be used in cosmic space while maintaining computational reliability.
Solution Approach 2:
The patent changes the parameter representation from deterministic binary numbers to probabilistic pseudo-random number sequences. By representing numerical values as probabilities through stochastic processes rather than fixed binary codes, the system becomes tolerant to bit flips caused by radiation, enabling the use of cost-effective consumer FPGAs in space environments.
2Reliability
If stochastic computing is used to enhance soft error resistance, then soft error resistance is improved, but operation error level increases
Solution Approach 1:
The patent applies preliminary error compensation by pre-calculating and storing correction values in a correction value storage unit. Before performing stochastic computations, the system prepares correction data that accounts for expected operation errors. This preliminary action allows the system to maintain both soft error resistance and measurement precision by compensating for errors before they significantly impact the results.
Solution Approach 2:
The patent implements a feedback mechanism where operation errors are monitored and corrected through iterative refinement. The correction value storage unit stores pre-computed correction values that are applied based on the detected error levels, creating a feedback loop that maintains measurement precision while preserving the soft error resistance benefits of stochastic computing.
3Productivity
If multiple weighted addition units are arranged in stages to perform complex operations, then computational capability is improved, but operation errors accumulate
Solution Approach 1:
The patent introduces an intermediary correction mechanism between stages of weighted addition units. Correction value storage units are positioned between computational stages to provide error compensation. This intermediary correction prevents error accumulation while allowing multiple stages of computation to proceed, maintaining both computational capability and measurement precision throughout the multi-stage processing.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
A digital filter (100) includes: a data conversion unit (101) that converts data into a first data sequence that is a pseudo-random number sequence; a coefficient storage unit (104) that stores a weighting coefficient that is a value weighted for each of the first data sequences; a coefficient conversion unit (105) that converts the weighting coefficient into a weighting coefficient sequence that is a pseudo-random number sequence; a plurality of first weighted addition units that generate a second data sequence obtained by weighting and adding up the first data sequences using the first data sequences and a plurality of the weighting coefficient sequences; at least one second weighted addition unit that generates a third data sequence obtained by weighting and adding up the second data sequences using the second data sequences and the weighting coefficient sequence; and a control unit (106) that selects the first data sequences to be inputted to the first weighted addition units such that operation error levels in the first weighted addition units and the second weighted addition unit become smaller than a predetermined value.