Adaptive Attention-Based Model for 5G Radio-based Outdoor Localization

TL;DR

This work tackles outdoor radio-based localization under varying urban conditions by proposing an adaptive framework that switches between specialized, low-complexity attention-based localizers using a router. Three scenario-tailored models are paired with a lightweight router (based on a single-layer perceptron) to dynamically activate only the most suitable sub-model in real time, reducing computation while maintaining high accuracy. Validated on real-world massive MIMO CIR data from a vehicle-mobile setup across three LoS/NLoS scenarios, the approach yields clear gains: MEE improves from (0.83 m, 0.81 m, 0.47 m) with a generalized model to (0.78 m, 0.77 m, 0.40 m) with specialization, and active parameters drop by 58–71% with test-time savings around 66%. The router’s ability to operate with minimal input and few parameters, including a 387-parameter variant achieving ~98.9% accuracy on a single delay-bin input, demonstrates practical potential for scalable, low-latency localization in autonomous/urban mobility contexts.

Abstract

Radio-based localization in dynamic environments, such as urban and vehicular settings, requires systems that efficiently adapt to varying signal conditions and environmental changes. Factors like multipath interference and obstructions introduce different levels of complexity that affect the accuracy of the localization. Although generalized models offer broad applicability, they often struggle to capture the nuances of specific environments, leading to suboptimal performance in real-world deployments. In contrast, specialized models can be tailored to particular conditions, enabling more precise localization by effectively handling domain-specific variations, which also results in reduced execution time and smaller model size. However, deploying multiple specialized models requires an efficient mechanism to select the most appropriate one for a given scenario. In this work, we develop an adaptive localization framework that combines shallow attention-based models with a router/switching mechanism based on a single-layer perceptron. This enables seamless transitions between specialized localization models optimized for different conditions, balancing accuracy and computational complexity. We design three low-complex models tailored for distinct scenarios, and a router that dynamically selects the most suitable model based on real-time input characteristics. The proposed framework is validated using real-world vehicle localization data collected from a massive MIMO base station and compared to more general models.

Figures (6)

• Figure 1: Overview of the adaptive model, where S, N, P, and M represent the subset of data, the total number of subsets, the trainable parameters (weights and biases), and the model, respectively
• Figure 2: The attention-based algorithm pipeline with one encoder layer and pooling layer added.
• Figure 3: Bird-eye view of the measurement environment and trajectories labeled S1, S2, and S3.
• Figure 4: Power delay profile of the $4$ dominant beams and relative power of all $128$ beams in a LoS scenario.
• Figure 5: Power delay profile of the $4$ dominant beams and relative power of all $128$ beams in an NLoS scenario.