This project uses a CNN to perform Radio Signal source detection. A 20 element array is used to estimate the number of signals impinging it.

Estimation of the number of sources is an important step in various array processing technologies. It is a prerequisite in most DoA algorithms. This model was succesfull in recognisng upto 6 sources at 20dB of SNR. Output of the code includes plots of the training and validation error, stats(precision,sensitivity, and specificity), a confusion matrix, and the ROC plot of the test data.

Contact me at [email protected] for the data. Files are too large to be uploaded here. #Updated results published at - https://ieeexplore.ieee.org/document/10137757

The upper triangle of the received auto-correlation matrix is flattened into a one dimensional vector and fed as the input. The output is a one hot encoded vector which denotes the number of sources present. Experiments are performed with the number of signals varying from 0-5.

Code template

Image generation for CNN model