Floating-Point Residual Encoding for Stable CAD File Compression

The patent applies preliminary action by computing predicted values for floating-point data before storage. The method calculates expected values based on spatial relationships and stores only the differences (residuals) between actual and predicted values. This preprocessing step enables efficient compression while maintaining data integrity, as the original values can be perfectly reconstructed by adding the stored residuals to the predicted values.

The patent substitutes the conventional compression approach with a prediction-based method that uses spatial relationships and mathematical models to predict floating-point values. Instead of applying generic compression algorithms that may introduce platform-specific variations, the method uses deterministic predictions based on geometric relationships, ensuring consistent results across different platforms while achieving high compression ratios.

The patent applies segmentation by dividing the computational task into two distinct phases: a compression phase that computes predicted values and stores residuals, and a decompression phase that reconstructs original values by adding residuals to predicted values. This segmentation allows the system to achieve high compression rates while managing computational complexity through efficient algorithms in each phase, avoiding the need for complex real-time processing.