Signal Detection via Spectrogram Correlation and Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting and classifying wanted signals in electromagnetic signals struggle when multiple signals are present in a small frequency band or when signals collide, making it difficult to reliably detect and classify the wanted signal.
Innovation Solution
The method involves receiving an electromagnetic signal, determining its spectrogram, calculating correlation parameters such as frequency-to-frequency correlation coefficients, and using these parameters as input for a machine learning module to detect and classify the wanted signal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional spectrogram analysis methods are used to detect and classify wanted signals, then the detection process is simple and straightforward, but the method fails to reliably detect and classify signals when multiple signals are present in a small frequency band or when signals collide
Solution Approach 1:
The patent segments the signal analysis process into multiple stages: initial spectrogram analysis, identification of candidate signal portions, calculation of correlation parameters for these candidates, and final classification. This segmentation allows the system to focus computational resources on promising candidates rather than processing all frequency components equally, thereby improving reliability in complex scenarios while managing overall complexity.
Solution Approach 2:
The patent introduces correlation parameters as an intermediary between the spectrogram and the final signal classification. These correlation parameters serve as additional features that capture relationships between different frequency components, enabling the system to distinguish colliding signals and multi-tone signals that would be indistinguishable using spectrogram analysis alone.
2Measurement precision
If correlation parameters are calculated and used as input for machine learning module, then colliding signals can be separated and identified, but the processing complexity and computational requirements increase
Solution Approach 1:
The patent performs preliminary identification of candidate signal portions based on spectrogram analysis before calculating correlation parameters. This preliminary action filters out non-promising candidates, so that computationally expensive correlation calculations are performed only on a small subset of potential signals. This approach achieves high measurement precision for signal separation while controlling processing complexity by avoiding unnecessary calculations on all frequency components.
3Measurement precision
If multiple correlation parameters are used to detect and classify different signal classes, then signal classification accuracy improves, but the size and complexity of the machine learning structure increases
Solution Approach 1:
The patent designs the machine learning module to process multiple types of correlation parameters (frequency-to-frequency, time-to-time, frequency-to-time) through a unified architecture. The same neural network structure handles different signal classes (modulated signals, multi-tone signals, colliding signals) by analyzing different correlation parameter sets, rather than requiring separate specialized structures for each signal type. This universality maintains high classification accuracy while avoiding the complexity multiplication that would result from having separate models for each signal class.
Data Source
AI summary
A method for at least one of detecting and classifying a wanted signal in an electromagnetic signal is described. The method includes the following steps: the electromagnetic signal is received; a spectrogram of the electromagnetic signal is determined; at least one correlation parameter, for example a correlation multi-dimensional algebraic object including several correlation parameters, is determined based on the determined spectrogram; the at least one correlation parameter is used as an input for a machine learning module; the wanted signal in the electromagnetic signal is detected and/or classified via the machine learning module based at least partially on the at least one correlation parameter. Further, a signal detection and/or classification system and a computer program are described.