Unsupervised Learning for AoD Estimation in MISO Downlink LoS Transmissions

TL;DR

This work addresses AoD estimation for downlink LoS MISO transmissions in SLAC by proposing an unsupervised learning framework that unifies deterministic ML (DML) and stochastic ML (SML) training. The approach feeds both received signals and available pilot-sequence information into a common neural-network architecture that outputs $\hat{\theta}_W$, $\hat{\sigma}^2_W$, and $\hat{\xi}_W$, and optimizes a LS-type loss for DML or a covariance-based loss for SML. Numerical results show superior AoD accuracy with significantly reduced observation requirements and competitive latency compared to traditional methods, validating the framework's potential for low-overhead AoD estimation in IoT/SLAC scenarios. The work suggests practical adaptability via fine-tuning for non-LoS or out-of-distribution conditions and points to future work on priors, domain adaptation, and transfer learning to handle dynamic environments.

Abstract

With the emergence of simultaneous localization and communication (SLAC), it becomes more and more attractive to perform angle of departure (AoD) estimation at the receiving Internet of Thing (IoT) user end for improved positioning accuracy, flexibility and enhanced user privacy. To address challenges like a large number of real-time measurements required for latency-critical applications and enormous data collection for training deep learning models in conventional AoD estimation methods, we propose in this letter an unsupervised learning framework, which unifies training for both deterministic maximum likelihood (DML) and stochastic maximum likelihood (SML) based AoD estimation in multiple-input single-output (MISO) downlink (DL) wireless transmissions. Specifically, under the line-of-sight (LoS) assumption, we incorporate both the received signals and pilot-sequence information, as per its availability at the DL user, into the input of the deep learning model, and adopt a common neural network architecture compatible with input data in both DML and SML cases. Extensive numerical results validate that the proposed unsupervised learning based AoD estimation not only improves estimation accuracy, but also significantly reduces required number of observations, thereby reducing both estimation overhead and latency compared to various benchmarks.

Figures (3)

• Figure 1: Illustration of the proposed unsupervised learning framework.
• Figure 2: The MAE versus the transmit power $P$ with $L=6$, $Q=4$, $\theta=23.4^\circ$ and $r=32.1$ m.
• Figure 3: The MAE versus the number of the time slots $L$ in one block of pilot sequence with $P=15$ dBm and $Q=4$.