Access network traffic profiles based on data from real German broadband connections (2024/25)

## Dataset Description This dataset contains aggregated traffic profiles and Fourier series parameters derived from 3129 real German broadband connections, as part of a publication at Optical Fiber Communication Conference (OFC) 2026 [1]. ### Source Data - Data rate time series from three established German network operators, collected in 2024/2025 - 3129 access connections - Temporal resolution: 1 s to 10 s (operator-dependent) - Recording duration: 1–2 weeks per connection ### Clustering Methodology - Features per user (computed separately for weekdays/weekends): - mean total throughput, - upload ratio (upstream throughput/ total throughput), - activity level (fraction of time with data rate > 0) - Feature preprocessing: - median imputation of missing features, - infinite values clipped to 5th/95th percentiles, - outlier winsorization via modified z-score (|M_i| > 3.5), - RobustScaler standardization - Clustering algorithm: - k-means, - optimal k = 3 determined by silhouette analysis - Validation: - 50 bootstrap iterations (80% subsampling with replacement), - mean Adjusted Rand Index (ARI) = 0.97 ### Profile Aggregation - Time series resampled to 1-minute resolution - Profiles computed separately per cluster and day type (weekday/weekend) - Sample-weighted aggregation ### Fourier Approximation - Each mean daily profile approximated via discrete Fourier series - Optimal number of harmonics n per profile determined by minimizing the Akaike Information Criterion (AIC) - R²: Downstream R² = 0.65–0.98, upstream R² = 0.26–0.62 - RMSE: Downstream RMSE = 66.47 - 324.75 kbit/s, Upstream RMSE = 11.77–93.05 kbit/s ## Dataset Structure ### 1. Cluster Profiles (`cluster_profiles/`) - 12 CSV files containing minute-resolution traffic profiles (1440 data points per day) - Naming convention: `cluster_[0-2]_[weekday|weekend]_[upstream|downstream].csv` - Each file contains: - `time`: Time of day (HH:MM format) - `mean`: Mean data rate across all users in cluster (kbit/s) - `median`: Median data rate (kbit/s) - `std`: Standard deviation (kbit/s) - `p10`, `p25`, `p75`, `p90`: Percentiles (kbit/s) ### 2. Fourier Coefficients (`fourier_coefficients.csv`) Contains optimized Fourier series parameters for all 12 traffic profiles: - `cluster_id`: Cluster identifier (0, 1, or 2) - `profile_type`: Day type (weekday/weekend) - `direction`: Traffic direction (upstream/downstream) - `a0`: DC component (mean value) - `n_harmonics`: Number of harmonics (AIC-optimized) - `a1...a12`, `b1...b12`: Fourier coefficients - `rmse`: Root mean square error of fit - `r_squared`: Coefficient of determination - `relative_error_percent`: Mean relative error ## Citation [1] M. Börner, K. Obermann, and C. Mas-Machuca, "Traffic Measurements and Models Based on Real User Data from Different German Operators," in Optical Fiber Communication Conference (OFC), 2026 (To be published)