Recurrence Dynamic Modeling of Metropolitan Cellular Network Traffic

Cellular network traffic analysis is evolving as a pivotal means for detecting anomalous behavior and assisting accurate prediction, which are indispensable for the advancement of automated network management systems. Nevertheless, the existing traffic analysis methods, such as popular wavelet transform and Pearson correlation, exhibit obvious inapplicability owing to the high computational complexity and the data assumptions of distribution, interference or stationarity. In response to these limitations, this paper develops a novel metropolitan cellular network traffic analysis framework based on recurrence plot (RP) and cross-recurrence plot (CRP). Structurally, it is a unified and consistent framework for service traffic analysis with functionalities of phase space reconstruction, dynamic visualization and recurrence quantization. To be precise, the RP and its recurrence quantification analysis reveal the time-dependent evolution law of cellular data in high-dimensional space. The similarity between type-cross traffic or domain-cross traffic can be determined by the CRP and its cross-recurrence quantification analysis. Extensive evaluations are conducted on the single-service RP analysis, type-cross and cell-cross CRP analysis of telecom data of Milan city. Experimental results confirm that our proposal can effectively identify hidden patterns and structures within traffic time series, such as periodicity, chaos, and non-stationarity. Meanwhile, these visual characteristics are measured quantitatively by the recurrence rate, determinism, average diagonal length, and entropy, providing insights into the traffic dynamic and correlation between service traffic, respectively. Furthermore, we illustrate that the proposed framework can effectively support the anomaly detection and accurate prediction of cellular network traffic.