EnergAIze: AI-driven network energy efficiency solution

50% reduction in energy consumption and 99.3% accessibility demonstrated in an urban deployment with 270 antennas
Integration demonstrated with VMware in Deutsche Telekom’s i14y Lab, and suitable for deployment in any type of fixed/mobile network

Microscope: AI-driven traffic decomposition

Real-time service demand estimation
Breaks large traffic aggregates into the service streams active at different locations
Lightweight metadata-based, non-intrusive data processing
Traffic volume predictions (Netflix, Spotify, etc.) accurate to within 1% of measured data
Adaptive solution for intensive traffic analysis

Dedicated neural models learn complex spatiotemporal dependencies
Anticipates the capacity needed to accommodate each service-level flow
Predicts multiple KPI reflective of user quality of experience
Provides operators with supplementary information about the confidence of neural networks' output
Configurable forecasting horizons, temporal and geographical granularities
100x lower model complexity than state-of-the art
Fast training, superior forecasting accuracy

ML pipeline for predicting anomalous patterns across time series of multiple KPIs specific to communications networks
Operates with minute-level granularity on multiple input signals in parallel and produces forecasts with horizons of tens of minutes to hours
Unsupervised data labelling that limits time-consuming human expert input for model training
Proactive incident resolution improving customer experience
Recommendation logic that streamlines network management and reduces OPEX

Harnesses Net AI‘s DeepQoE technology to deliver the anticipated evolution of KPIs reflective of user quality of experience (QoE)
Operates over configurable time horizons with high accuracy
Consumes only metadata and does not interfere with user traffic, carrying low complexity and high scalability
Provides network operators with supplementary information about the confidence of the neural networks‘ output

Embeds Net AI‘s proprietary AI models and uses historic and real-time network traffic data to forecast the upcoming traffic volumes at different network elements
Provides actionable analytics to the near-RT RIC via an O-RAN-compliant interface
AI-driven autoscaling logic produces relative capacity values for gNB Centralized Unit User Plane (gNB-CU-UP) entities, which are used to load-balance traffic between gNB-CU-UPs
Enables CSP to easily allocate resources to guarantee SLAs and improve customer quality of experience, while reducing operational costs