Linking signal integrity to probabilistic models of behavioral dynamics

Quantitative analysis of rodent behavior in naturalistic settings is crucial for neuroscience, yet traditional methods often lack precision or scalability. While ultra-wideband (UWB) sensor tracking provides centimeter-level localization, standard metrics like root mean square error (RMSE) fail to capture the probabilistic and sequential nature of behavior. We introduce a probabilistic, information-theoretic framework that leverages high-resolution UWB sensor trajectories to address this gap. By integrating Shannon entropy to quantify uncertainty, Bernoulli modeling to assess accuracy thresholds, and first-order Markov chains to characterize state dynamics, our approach derives interpretable behavioral markers directly linked to signal quality. Empirical evaluation in an open-field arena demonstrated robust tracking under line-of-sight (LoS; RMSE: 20 mm) and non-line-of-sight (NLoS; RMSE: 35 mm) conditions. Crucially, we show that physical-layer impairments propagate to behavioral metrics: NLoS conditions increased entropy from 1.15 to 1.78 bits, reduced the probability of achieving sub-20 mm accuracy from 63.2% to 27.7%, and decreased state persistence, indicating greater behavioral fragmentation. By treating UWB signals as a probabilistic information source, our computationally efficient framework establishes a methodological bridge between engineering performance and neuroscience, enabling scalable, reproducible, and low-bias behavioral quantification suitable for preclinical research in complex environments.