Background: Freezing of gait (FOG) in Parkinson's disease is associated with abnormal beta activity in the subthalamic nucleus (STN), but the temporal structure of burst dynamics remains poorly understood. Objectives: To determine whether temporal features of STN beta bursts distinguish pre-freeze from stable gait and predict freezing onset. Methods: STN recordings during gait from four individuals were analyzed. Temporal features of burst timing, including entropy and variability, were computed across behavioral states. Predictive performance was assessed using leave-one-patient-out classifiers. Results: Entropy of inter-burst intervals was reduced prior to freezing (p < 0.01), with strong predictive performance (AUC = 0.825; threshold AUC = 0.858). During freezing, variability measures decreased and temporal structure increased, while entropy did not differ from pre-freeze. Phase-amplitude coupling showed frequency-specific but heterogeneous effects across comparisons. Conclusions: Reduced temporal variability of STN beta burst timing precedes and predicts freezing, suggesting a transition to constrained neural dynamics.