Spectrum Sensing With Correlation-Image CNNs for Low-SNR Occupancy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing spectrum sensing technologies struggle to accurately determine frequency band occupancy, especially in low signal-to-noise ratio environments, due to their reliance on rule-based methods like energy detection.
Innovation Solution
A spectrum sensing apparatus utilizing a convolutional neural network (CNN) to analyze a signal correlation array, which includes a receiver, ADC, correlator, and controller to generate image patterns from correlation functions, and a band occupation estimator to identify frequency band occupancy based on trained image patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If rule-based spectrum sensing (energy detection) is used, then the system is simple to implement, but detection accuracy degrades in low signal-to-noise ratio environments
Solution Approach 1:
The patent introduces an intermediary processing stage between signal reception and detection decision. A correlator computes the correlation function of the received signal, and an image pattern generator converts this into a visual representation (image pattern) that serves as input to the CNN. This intermediary transformation enables deep learning-based detection while maintaining a structured, modular system architecture.
Solution Approach 2:
The patent replaces traditional rule-based detection mechanisms (energy detection thresholds, statistical tests) with a data-driven CNN model. The CNN learns optimal detection features and decision boundaries from training data, substituting manual rule design with automated machine learning-based pattern recognition, thereby improving accuracy in low SNR conditions.
2Measurement precision
If deep learning-based spectrum sensing is used, then detection accuracy improves in low signal-to-noise ratio environments, but system complexity increases
Solution Approach 1:
The patent segments the spectrum sensing system into distinct functional modules: signal reception, correlation computation, image pattern generation, CNN-based detection, and decision output. This segmentation allows the complex deep learning functionality to be integrated in a structured manner, where each module performs a specific transformation, making the overall system more manageable and implementable despite the sophistication of the CNN component.
3Measurement precision
If correlation function computation is performed, then signal processing capability is enhanced, but computational complexity increases
Solution Approach 1:
The patent extracts the essential spectral characteristics from the full received signal by computing only the correlation function at specific time lags. Instead of processing the entire signal spectrum, the system focuses on extracting correlation features that are most relevant for detection, thereby reducing computational burden while maintaining detection accuracy. The image pattern generator then further extracts visual features from this correlation data for CNN input.
Data Source
AI summary
The present invention provides a spectrum sensing apparatus including a receiver which receives an analog signal in a frequency band of interest; an ADC(Analog Digital Caonverter) which samples the analog signal to output a digital signal; a correlator which performs the auto-correlation of the digital signal to output a correlation function; and a controller which generates an image pattern corresponding to the correlation function and identifies whether to occupy the frequency band of interest according to the image pattern.