Accurately predicting directional radio maps is essential for wireless applications, yet prior approaches primarily focus on omnidirectional signals and typically treat transmitter localization and signal map reconstruction as separate tasks. In omnidirectional settings, predicting the maximum signal location often coincides with the transmitter position, which limits the need for explicit joint modeling. However, in directional propagation where angular effects, reflections, and building occlusions play critical roles, this assumption no longer holds. To address this gap, we propose SymNet, a unified framework that jointly predicts directional radio maps and transmitter locations from sparse signal measurements. SymNet incorporates a prediction head for transmitter localization alongside radio map reconstruction, enabling simultaneous learning of both tasks. This joint formulation leverages their complementary information and leads to consistent improvements over treating them separately. Experiments on challenging directional scenarios demonstrate that SymNet outperforms state-of-the-art baselines, achieving superior accuracy in both radio map reconstruction and transmitter localization.

Accurate radio source localization on resource-constrained hardware presents a primary challenge in wireless networks. We introduce EfficientLocNet, a novel architecture achieving superior accuracy with exceptional computational efficiency, driven by two distinct design choices. Its efficiency stems from lightweight Depthwise Separable Convolutions, while its accuracy is enhanced by two components working in tandem: Atrous Spatial Pyramid Attention to capture long-range spatial features, and a Self-Attention module to refine the latent representation. Evaluated against state-of-the-art (SOTA) methods, EfficientLocNet outperforms the top-performing model, DSLoc, on all key metrics: it reduces the mean localization error by at least 5%, possesses a 40x smaller model size, and requires over 100x fewer computations. This compelling combination of performance and efficiency validates EfficientLocNet as a powerful solution for deployment in edge computing environments.

Accurately predicting directional radio maps is crucial for various wireless applications; however, existing methods that primarily rely on Convolutional Neural Networks (CNNs) or U-Nets designed for omnidirectional signals struggle to capture the long-range dependencies and angular sensitivities inherent in directional signal propagation (especially when influenced by environmental factors like building occlusions and reflections). To overcome these limitations, we introduce a novel framework, denoted as "ViT-RefineNet," that is specifically designed for predicting directional radio maps from sparse signal measurements. Our approach integrates a Vision Transformer (ViT)-based model to capture global dependencies with multiple U-Net-like modules for localized refinement at different processing stages. By effectively combining global context understanding with local detail enhancement, ViT-RefineNet offers significant advantages over traditional CNN-based and U-Net-based methods in predicting directional radio maps from sparse directional signal measurements. Experimental results demonstrate the superior performance of our proposed ViT-RefineNet compared to state-of-the-art models.1

This paper addresses the source localization problem by introducing RoLocMe, a multi-agent reinforcement learning system that integrates SkipNet - a skip-connection-based RSS estimation model - with parallel Q-learning. SkipNet predicts RSS propagation of the entire search region, enabling agents to explore efficiently. The agents leverage dueling DQN, value decomposition, and λ-returns to learn cooperative policies. RoLocMe converges faster and achieves at least 20% higher success rates than existing methods in dense and sparse reward settings. A drop-one ablation study confirms each component’s importance and RoLocMe’s effectiveness for larger teams.