Potential of Robust Regression Methods in Clock Skew Measurement

Clock skew, defined as the difference in clock rates between digital devices, serves as a unique and stable fingerprint for device identification and authentication, particularly in distributed network environments. Traditional clock skew estimation techniques, such as linear regression, are effective under stable conditions but often fail in the presence of data disturbances, such as latency, jitter, and asymmetric delays, which introduce outliers. This study explores the application of robust regression methods to enhance the accuracy and stability of clock skew estimation under such conditions. Three robust techniques are comparatively analyzed: Least Median of Squares (LMedS), Random Sample Consensus (RANSAC), and S-Estimators. LMedS offers high resistance to outliers by minimizing the median of squared residuals, though it is computationally demanding for large datasets. RANSAC achieves a practical balance between robustness and efficiency through iterative model fitting and inlier maximization, while S-Estimators provide strong statistical resistance to both outliers and high-leverage points, albeit with increased implementation complexity. The comparative evaluation considers key parameters such as estimation accuracy, computational cost, and robustness to anomalies. Results indicate that RANSAC is generally preferred for clock skew measurement in distributed systems due to its efficient performance and explicit outlier detection capabilities. However, LMedS and S-Estimators remain valuable in scenarios with more complex anomaly structures or higher noise levels. This study contributes to the selection of appropriate robust regression methods for reliable clock skew estimation in dynamic and error-prone network environments.