Detailed Description
This artifact is part of a multi-agent architecture for next-generation autonomous networks. It detects uplink interference and abnormal RAN behavior in 5G networks using multivariate time-series anomaly detection. The agent continuously ingests RAN telemetry through the deployed monitoring xApps. The detection core is based on an LSTM Autoencoder model trained on normal network behavior to learn temporal and cross-KPI correlations. During inference:
- KPI sequences are reconstructed by the LSTM Autoencoder
- Reconstruction error is computed per time window
- An anomaly is triggered when the error exceeds an adaptive threshold
- Per-KPI residual analysis identifies the most deviating metrics
The agent further:
- Performs severity scoring based on magnitude, persistence, and KPI impact
- Generates explainability reports highlighting top contributing KPIs
- Maps deviation patterns to likely root causes (e.g., uplink interference)
- Recommends mitigation actions
It operates in both real-time mode for streaming monitoring and historical analysis mode for post-mortem investigation. This enables predictive maintenance and proactive fault management in intent-driven autonomous RAN frameworks.
Key Features
- LSTM Autoencoder-based multivariate anomaly detection
- Real-time uplink interference detection
- KPI-level explainability via residual contribution analysis
- Historical replay and root cause analysis
Use Cases
- Uplink interference detection in 5G networks
- Predictive maintenance in Open RAN deployments
- Autonomous RAN observability within intent-driven architectures
- Root cause analysis for degraded uplink performance
