Radar Pulse Deinterlacing Using Hierarchical Optimal Transport
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for deinterlacing radar pulses from multiple transmitters are ineffective due to the complexity of electromagnetic spectra, requiring prior knowledge of transmitters and being sensitive to noise, with AI models trained on truncated data leading to incomplete signal representation and frequent retraining needs.
Innovation Solution
A method involving a series of partitioning algorithms to classify radar pulses based on frequency, duration, and arrival time, using unsupervised density-based partitioning, Kolmogorov-Smirnov tests, and optimal transport distances to group pulses into classes associated with specific transmitters, effectively discriminating between similar signals and mitigating noise impact.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If artificial intelligence models are trained on labeled data to perform deinterlacing, then discrimination capability between radar transmitters is improved, but model performance deteriorates due to truncated or partially observed training data
Solution Approach 1:
The patent segments the deinterlacing problem into multiple hierarchical levels: first grouping pulses by frequency characteristics, then by temporal patterns, and finally by combined features. This multi-stage segmentation allows each processing stage to focus on specific aspects of pulse discrimination, improving overall reliability without requiring complete training data for all features simultaneously.
Solution Approach 2:
The patent introduces additional dimensions for pulse classification beyond traditional frequency analysis. By incorporating temporal dimension (arrival time patterns) and creating hierarchical classification levels, the system achieves better discrimination capability even with incomplete training data, as each dimension provides complementary information.
2Measurement precision
If artificial intelligence models are trained exhaustively on all known radar emitters, then deinterlacing accuracy is improved, but retraining is required each time a new radar emitter is added
Solution Approach 1:
The hierarchical classification structure allows the system to maintain base-level frequency-based groups that remain stable over time. When new radar emitters are introduced, they can be integrated into existing frequency groups without requiring complete retraining of the entire model, significantly reducing retraining time while maintaining accuracy.
Solution Approach 2:
The patent creates a universal classification framework where frequency-based grouping serves as a foundation that can accommodate multiple types of radar emitters. This multi-functional structure allows the same base model to handle diverse radar signals, and new emitters can be added by extending existing groups rather than retraining the entire system.
3Device complexity
If traditional deinterlacing methods use only frequency and pulse duration characteristics, then processing simplicity is maintained, but discrimination fails for transmitters with similar characteristics
Solution Approach 1:
The patent adds the temporal dimension (arrival time patterns) to the traditional frequency and duration characteristics. By analyzing how pulses are distributed over time and their arrival patterns, the system can distinguish between transmitters that have similar frequency and duration characteristics, significantly improving discrimination accuracy while maintaining reasonable processing complexity through hierarchical organization.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
The invention relates to a computer-implemented method (10) for deinterlacing radar pulses comprising: - implementation (12) of a first partitioning algorithm to assign each pulse to a corresponding first class, based on an associated frequency, duration, and arrival time; - for each first class, estimation (14) of a respective average frequency and average pulse duration; - implementation (16) of a second partitioning algorithm to group the first classes into second classes based on the corresponding average frequency and average duration; - for each second class, determination (18) of a distribution of the arrival times of the associated pulses;- implementation (20) of a third partitioning algorithm to group the second classes into third classes, based on optimal transport distances between the distributions of determined arrival times.;