Model Order Estimation for Sensor Array Observations using a Neural Network

Document Type

Conference Proceeding

Date of Original Version

1-1-2024

Abstract

This research develops a feedforward neural network consisting of one-dimensional convolutional layers and applies it to model order estimation for data received by an array of sensors. The network is trained using simulated Gaussian data from a standard uniform linear array. Results indicate that the trained neural network excels in accurately estimating the model order when the possible values for the order are small. As the number of possible values for the model order increases, the algorithm's performance declines. Nonetheless, the methodology reveals promise in detecting more sources than the number of sensors. Additionally, neural networks tailored for three different sparse arrays - a coprime array, a nested array, and a shift-invariant sparse array - exhibit similar trends to the one designed for the full array.

Publication Title, e.g., Journal

2024 IEEE 5th World AI Iot Congress Aiiot 2024

Share

COinS