Source Detection With Multi-Label Classification

Abstract

Radio source detection through conventional algorithms has been unreliable when trying to solve for large number of sources in the presence of low SINR and less number of snapshots. We address this by reformulating source detection as a multi-class classification problem solved using deep learning frameworks. Incoming waveforms are sampled using a centro-symmetric linear array with omni-directional elements and the normalized upper triangle of the autocorrelation matrix is extracted as the input feature to a modified convolutional neural network with uni-dimensional filters, trained to detect the sources in the presence of both uncorrelated and correlated signals. Two detection algorithms are introduced and referred to as CNNDetector and RadioNet, and subsequently benchmarked against the conventional source detection algorithms. By including pre-processing in forward backward spatial smoothing, RadioNet can also resolve the number of uncorrelated sources in the presence of correlated paths. Finally, the algorithms are stress tested under challenging operational conditions and extensive evaluations are presented showing the efficacy and contributions of the introduced predictive models. To the best of our knowledge, this is the first time the source detection problem has resolved L-1 sources, for an antenna array of L elements using a deep learning framework.

Joint Identification for Number of Transmit Antennas and Channel Order

“…Identifying the number of transmit antennas can be regarded as an extension of source number detection in the array signal processing. For the identification of the number of transmit antennas, the existing literature [16]- [24] mainly falls into two classes: methods using the information theory criterion and those using the feature extraction. These methods require that the channel order is 1.…”

“…Compared with the information-theoretic methods, the feature-based algorithms [18]- [24] can identify the unknown number of transmit antennas with a lower complexity. Using the pilot signal, an identification method of the number of transmit antennas is proposed for the OFDM system, in which only one receive antenna is required for the receiver.…”

“…Recently, great progress has been made in identifying the number of transmit antennas based on the deep learning, which obtains the source number by formulating the source detection as a multi-class classification problem solved using deep learning frameworks. The normalized upper triangle of the autocorrelation matrix [24], the subspace-rank features [25], the eigenvalues of the normalized covariance matrix [26], the eigenvalue preprocessing [27], and the logarithmic eigenvalue [28] are extracted for the input features. The deeplearning-based methods can obtain a significant performance improvement compared to the methods based on non-deeplearning.…”

“…, p ∈ I P (24) where Γp is a matrix of the first N r −1 rows and the first N r −1 columns of Γ p , and r is the column vector formed by the first N r − 1 rows of the N r th column in the matrix Γ p . Then, we can obtain the equation ( 25) using the EVD of Γp , i.e.…”