Improving the Accuracy of Direction of Arrival Estimation with Multiple Signal Inputs Using Deep Learning.

Area of Science:

• Signal Processing
• Array Signal Processing
• Machine Learning in Signal Processing

Background:

• Direction of Arrival (DOA) estimation is crucial in various applications.
• Traditional DOA algorithms face challenges with noise and limited snapshots.
• Improving DOA accuracy and reliability remains an active research area.

Purpose of the Study:

• To propose an innovative cyclic noise reduction method.
• To develop an improved CAPON algorithm (Minimum Variance Distortionless Response - MVDR) using Convolutional Neural Networks (CNNs).
• To enhance the accuracy and reliability of DOA estimation.

Main Methods:

• Cyclic noise reduction by processing eigenvalues of the received signal's covariance matrix.
• An improved CAPON algorithm employing a CNN structure with a processed covariance matrix as input.
• Using CAPON spectral values as training labels for the CNN to estimate the spatial spectrum.

Main Results:

• Cyclic noise reduction increases Signal-to-Noise Ratio (SNR) by up to 5 dB, improving DOA accuracy.
• The improved CAPON algorithm demonstrates superior performance over traditional methods across various SNRs and snapshot numbers.
• Root Mean Square Error (RMSE) reduced from 0.86° to 0.8°, with angle estimation error decreasing by up to 0.3°.
• Cyclic noise reduction further decreases angle estimation error from 0.04° to 0.02°.

Conclusions:

• The proposed cyclic noise reduction and improved CAPON algorithm significantly enhance DOA estimation accuracy and reliability.
• The CNN-based CAPON algorithm effectively achieves blind source estimation with improved performance.
• These methods offer a robust solution for DOA estimation in challenging signal environments.