Convolutional Neural Network Approach for Emitter Association using Channel Identification in a MIMO System based on Propagation Features

TL;DR

This paper introduces a deep convolutional neural network approach for emitter association in MIMO systems, leveraging channel features like CIR and TF to improve identification accuracy over traditional methods.

Contribution

It proposes a novel use of channel features with deep CNNs for emitter identification, achieving high accuracy and outperforming previous techniques.

Findings

Achieved 97.22% accuracy with DCNN in emitter classification.

Achieved 88.89% accuracy with DCNN-MCM in emitter classification.

Demonstrated superiority over traditional emitter association methods.

Abstract

In this paper, an application of a 1D deep convolutional neural network (DCNN) and 4x4 1D DCNN Multi-channel Model (DCNN-MCM) was developed to predict the probability of a channel being associated with a given transmitter for each emitter in a 4x4 multi-input multi-output (MIMO) system. Counterintuitively, compared to the traditional approach to emitter association (EA), this research argues for the identification of received RF signals based on channel features (CFs) such as the channel impulse response (CIR) and transfer function (TF). Based on the CFs there are unique properties per transmit and receive pair that can be used to identify the different transmitters. More specifically, CFs are often defined by a wide sense stationary (WSS) stochastic process which gives them unique properties such as, being statistically independent and reciprocal for emitter association. Given these CFs, the received RF signals can be used to measure the clutter and objects surrounding the emitters. Safely assuming the surrounding clutter and objects are unique to each emitter, the estimation and tracking of this clutter provides a way of identifying what emitter is transmitting the observed signal. This method successfully classifies and identifies the emitters at 97.22% and 88.89% accuracy for the DCNN and DCNN-MCM respectively. This method was further compared with the accuracy of previous methods to test the validity of the approach and future improvements.