Datasets:

CelfAI
/

COOPER

license: apache-2.0
language:
  - en
tags:
  - mobileNetwork
  - 5G
task_ids:
  - univariate-time-series-forecasting
  - multivariate-time-series-forecasting
configs:
  - config_name: measurements_by_cell
    data_files:
      - split: train
        path: dataset/train_data.csv
      - split: test
        path: dataset/test_data.csv
  - config_name: topology
    data_files:
      - split: main
        path: metadata/topology.csv
  - config_name: performance_indicators_meanings
    data_files:
      - split: main
        path: metadata/performance_indicators_meanings.csv

📡 COOPER

Cellular Operational Observations for Performance and Evaluation Research

An Open Benchmark of Synthetic Mobile Network Performance Indicators for Reproducible Research

🧭 Overview

COOPER is an open-source synthetic dataset of mobile network performance measurement (PM) time series, designed to support reproducible AI/ML research in wireless networks. The dataset is named in honor of Martin Cooper, a pioneer of cellular communications.

COOPER emulates the statistical distributions, temporal dynamics, and structural patterns of real 5G network PM data while containing no sensitive or operator-identifiable information.

The dataset is released together with a reproducible benchmarking framework used to evaluate synthetic data generation methods.

🎯 Motivation

Access to real telecom PM/KPI data is often restricted due to:

Confidentiality agreements
Privacy regulations
Commercial sensitivity

This lack of open data limits reproducibility in AI-driven research for wireless networks. COOPER addresses this gap by providing a realistic yet privacy-preserving synthetic alternative suitable for:

Network monitoring research
KPI forecasting
Anomaly detection
AI-native network automation
5G/6G performance evaluation

🏗 Dataset Creation Methodology

To generate COOPER, three complementary synthetic data generation paradigms were evaluated:

Adversarial approaches
Probabilistic models
Model-based time-series methods

These were benchmarked using a unified quantitative and qualitative evaluation framework considering:

Distributional similarity
Temporal fidelity
Shape alignment
Discriminative performance
Downstream forecasting capability

The generator demonstrating the most balanced and consistent performance across these criteria was selected to produce COOPER.

📊 Source Data Characteristics (Before Anonymization)

The real dataset used to model the synthetic data was:

Fully anonymized to remove operator-sensitive information
Cleaned and standardized for consistency

Property	Value
Radio Access Technology	5G
Number of PM Indicators	45
Total Number of Cells	84
Base Stations	12
Geographic Area	~1.35 km²
Collection Period	31 days
Sampling Interval	1 hour
Data Representation	Multi-cell time series

A cell is defined as a radiating unit within a specific RAT and frequency band. Each base station may host multiple cells.

📡 Network Deployment Characteristics

The modeled network includes two frequency bands and two 5G architectures:

Band	Architecture	Number of Cells
N28 (700 MHz)	Option 2 (Standalone)	6
N28 (700 MHz)	Option 3 (Non-Standalone)	48
N78 (3500 MHz)	Option 2 (Standalone)	6
N78 (3500 MHz)	Option 3 (Non-Standalone)	24

Most cells operate in Option 3 (NSA) mode, reflecting a typical EN-DC deployment where LTE provides the control-plane anchor.

📈 PM Indicator Categories

Indicators follow 3GPP TS 28.552 performance measurement definitions and are grouped into:

1️⃣ Radio Resource Control (RRC) Connection

Procedures for establishing UE radio connections and tracking active users.

RRC.ConnEstabSucc
RRC.ConnEstabAtt
RRC.ConnMax

2️⃣ Mobility Management

Handover and redirection performance across frequencies.

MM.HoExeIntraFreqSucc
MM.HoExeInterFreqSuccOut

3️⃣ Channel Quality Indicator (CQI)

Distribution of downlink channel quality reports (CQI 0–15).

CARR.WBCQIDist.Bin0
CARR.WBCQIDist.Bin15

4️⃣ Throughput and Data Volume

Traffic volume and transmission duration.

ThpVolDl
ThpTimeDl

5️⃣ Availability

Cell downtime due to failures or energy-saving mechanisms.

CellUnavail.System
CellUnavail.EnergySaving

6️⃣ UE Context

User session establishment attempts and successes.

UECNTX.Est.Att
UECNTX.Est.Succ

🧪 Benchmarking Framework

COOPER is distributed with a reproducible evaluation pipeline that allows researchers to compare synthetic data generators using:

Statistical similarity metrics
Temporal alignment measures
Shape-based similarity
Classification distinguishability
Forecasting task performance

This framework enables standardized evaluation of synthetic telecom datasets.

🔬 Intended Use Cases

COOPER is suitable for:

Time-series forecasting research
Network anomaly detection
Root cause analysis modeling
RAN performance optimization studies
Reproducible academic research in 5G/6G systems

⚠️ Data Notice for Dataset Users

Due to the real network nature of the source data, some inconsistent values were intentionally maintained in this dataset.
We recommend preprocessing the data before use (e.g., handling outliers, missing values, or domain-specific inconsistencies) according to your application and methodology.

🤝 Contribution & Reproducibility

This project promotes open and reproducible telecom AI research.
Researchers are encouraged to:

Benchmark new generation models using the provided framework
Share improvements and derived datasets
Compare methods under the same evaluation protocol

📜 License

This dataset is released for research and educational purposes.
(Include specific license here, e.g., CC BY 4.0 / MIT / Apache 2.0)

📖 Citation

If you use COOPER in your research, please cite:

COOPER: An Open Benchmark of Synthetic Mobile Network Performance Indicators for Reproducible Research

(Full citation to be added)