Pulse Grouping via Quadratic Regression and Mixture Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for grouping intercepted pulses in electromagnetic listening systems face challenges due to ambiguities in DTOA histograms and limitations in computation time and performance, especially when pulses are missed or transmitters are agile, leading to unreliable database formation without prior knowledge of waveforms.
Innovation Solution
A method that selects analysis windows of limited duration to reduce pulses analyzed, using quadratic regression modeling of pulse levels over arrival time, coupled with a mixture model to group pulses based on estimated parameters via an Expectation-Maximization algorithm, without relying on Euclidean distance or prior waveform knowledge.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If DTOA histograms are used for pulse grouping, then periodic pulse patterns can be extracted, but ambiguities arise due to integer multiples of periods and cross-correlations between different periods
Solution Approach 1:
The patent transitions from analyzing pulses in the time domain (DTOA histograms) to the frequency domain (spectral analysis). By computing the spectrum of pulse arrival times and identifying spectral lines, the method resolves ambiguities inherent in time-domain histogram analysis, where multiple periods could produce identical histogram patterns. The frequency domain provides a unique representation that eliminates the integer multiple ambiguities.
2Reliability
If complex variants of DTOA histogram methods are used to improve robustness, then performance improves for certain cases, but computation time increases significantly
Solution Approach 1:
The patent replaces complex iterative histogram-based methods with a direct spectral analysis approach. Instead of repeatedly adjusting histogram parameters and thresholds to achieve robustness, the method computes the Fourier spectrum once and identifies spectral lines directly. This substitution of the analytical approach dramatically reduces computation time while maintaining or improving reliability through the mathematical properties of spectral analysis.
3Measurement precision
If histogram-based methods are used for pulse grouping, then periodic patterns can be identified, but performance degrades when pulses are missed or transmitters are agile with varying pulse frequencies
Solution Approach 1:
The patent employs dynamic spectral analysis that can adapt to varying pulse frequencies. By computing the spectrum and identifying spectral lines at different frequencies, the method naturally handles agile transmitters whose pulse repetition frequencies change over time. The spectral approach is inherently more adaptable than fixed-parameter histogram methods, as it can detect and track multiple frequency components without requiring prior knowledge of the transmitter behavior.
4Adaptability or versatility
If Euclidean distance-based clustering algorithms are used for pulse grouping, then general-purpose clustering can be achieved, but the distance metric is not well-suited to the model of the received signal leading to limited performance
Solution Approach 1:
The patent changes the fundamental parameter used for clustering from spatial/Euclidean distance in the time domain to frequency domain characteristics. Instead of measuring distance between pulse vectors in n-dimensional parameter space, the method compares spectral signatures and identifies pulses from the same transmitter by their common spectral lines. This parameter transformation aligns the clustering criterion with the physical model of radar signal propagation and reception, significantly improving grouping performance.
Data Source
Figure 1a~1d
Figure 2
AI summary
The invention relates to a method which analyses a level nk and a time of arrival tk of the intercepted pulse of index k, in order to group the pulses into C groups, the grouped pulses being considered as transmitted by a same source. The method comprises the steps of: selecting analysis windows of limited duration, a window comprising K" pulses; and, for each window, implementing, from a modelling based on a quadratic regression of the levels of the pulses of a same group according to the time of arrival, coupled with a mixing model of the groups, an iterative algorithm for estimating parameters of the modelling, the parameters consisting of the number of groups C, the coefficients of the quadratic regression for each group, a variance of a measurement noise, and indicator variables of the mixing zk,c, the indicator variables zk,c, having a value of 1 if the k-th pulse belongs to group c and 0 otherwise; and, grouping the pulses according to the estimated values of the indicator variables of the mixting.