Error-Bound Floating-Point Compression for Variable IoT Data

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing lossy floating point compression techniques fail to provide satisfactory compression for IoT data with significant variances in consecutive values, leading to inefficient resource usage in data storage.

Innovation Solution

A method that rounds floating point values to a specified number of decimal places within an error bound, followed by further integer compression, to achieve improved error-bound compression by reducing the difference in the sum total of rounded values compared to the original values.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If known lossy floating point compression is applied to IoT data with significant variances, then processing speed is improved, but compression ratio deteriorates

Engineering Contradiction:
Improveprocessing speedVSAvoidcompression ratio
Core Design Contradiction:
SpeedVSLoss of substance

Solution Approach 1:

The patent applies preliminary rounding of floating-point values to a specified number of decimal places before compression. This preprocessing step groups values that are within the error bound, creating runs of identical rounded values that can be efficiently compressed by subsequent compression algorithms, thereby achieving both fast processing and high compression ratios.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent changes the parameter representation by rounding floating-point values to a controlled number of decimal places. This parameter transformation maintains values within the error bound while creating patterns amenable to compression, resolving the contradiction between processing speed and compression ratio.

Inventive Principle:
Principle #35Parameter changes

2Ease of manufacture

If known lossy floating point compression is applied to IoT data with significant variances, then processing simplicity is improved, but storage efficiency deteriorates

Engineering Contradiction:
Improveprocessing simplicityVSAvoidstorage efficiency
Core Design Contradiction:
Ease of manufactureVSLoss of substance

Solution Approach 1:

The patent introduces a preliminary rounding step that simplifies the data structure by creating runs of identical values. This maintains processing simplicity while dramatically improving storage efficiency, as compressed representations of rounded values require far less space than uncompressed original values with significant variances.

Inventive Principle:
Principle #10Preliminary action

3Loss of substance

If floating point values are rounded to fewer decimal places, then compression ratio is improved, but measurement precision deteriorates

Engineering Contradiction:
Improvecompression ratioVSAvoidvalue accuracy
Core Design Contradiction:
Loss of substanceVSMeasurement precision

Solution Approach 1:

The patent optimally determines the number of decimal places to round to, balancing compression ratio and measurement precision. By controlling the rounding precision parameter, the system achieves satisfactory compression while maintaining values within the acceptable error bound, thus resolving the contradiction between compression ratio and measurement precision.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3835939B1Error-bound floating point data compression system
Publication Date: 2024.02.14 SAP SE
  • EP3835939B1 patent drawingFigure 1
  • EP3835939B1 patent drawingFigure 2~3
  • EP3835939B1 patent drawingFigure 4

AI summary

A system includes reception of a plurality of original floating point values, rounding of each of the plurality of original floating point values to a respective rounded floating point value, where each rounded floating point value is within an error bound of its respective original floating point value, compression of the plurality of rounded floating point values, and storage of the plurality of compressed rounded floating point values in the data storage system.