Automated ML-Driven Analysis of RF Interference
Spectrum Effect’s revolutionary solution, Spectrum-NET®, provides highly innovative RF interference mitigation capabilities enabling mobile operators to rapidly resolve interference issues, improve network KPIs, improve subscriber QoE, surgically deploy field assets, save significant OPEX and maximize use of valuable spectrum. Protected by numerous patents, Spectrum-NET rapidly identifies, characterizes, classifies, aggregates, locates, predicts and assesses the impact of sources of RF interference.
BDA, CATV egress, DECT, WiMAX, broadcast TV, security cameras, RFID scanner, RF jammer, etc.
Cross Border Interference
Misalignment or inconsistent configuration of the RF band; can be during ducting.
During periods of tropospheric ducting and during loss of synchronization.
RAN Generated Interference
Interference due to UE traffic; requires radio network optimization.
Internal (cabling, connector, combiner, antenna) and External (nearfield to the antenna) sources.
Faulty radio units, Faulty Rx branches and crossed feeders.
Spectrum-NET performs automated ML-driven analysis of RF interference for all bands and operates seamlessly across multi-vendor 5G NR, LTE and UMTS networks on a continual basis. Spectrum-NET is deployed by operators across the globe to drive their RF interference mitigation processes and effectively address RF network interference. Built on a container-based microservices architecture, Spectrum-NET is a web-based, cloud-ready, horizontally scalable solution with extensive mapping, visualization, and report generation capabilities. Spectrum-NET readily integrates with the mobile operator’s other tools (e.g., automation framework, collaboration system, topology database, trouble ticket system and SON) to maximize automation.
RF Interference Detection & Characterization
Spectrum-NET analyzes RAN CM, PM and topology data to automatically identify and characterize RF interference events without any need for external probes. Spectrum-NET performs outlier detection with unsupervised ML techniques to identify behavior that can indicate the presence of interference. The outliers are fed into the feature extraction modules (FEMs) for interference characterization. Spectrum-NET analyzes interference measurements on a per PRB basis and captures the spectral and temporal characteristics of each RF interference event.
Interference Aggregation & Classification
Spectrum-NET performs interference aggregation to identify all cells within a network that are affected by a common interference source. Spectrum-NET incorporates Deep Convolutional Neural Network models that are trained off-line with labeled measurement samples (interference signatures from known interference cases) from mobile operator networks across the globe. The Spectrum-NET CNN models classify RF interference sources (e.g., PIM, CATV egress, WiMAX, DECT, terrestrial broadcast TV, BDA’s, RAN, TDD self-interference, etc.) with high accuracy. Whenever a new interference signature is captured, the labeled data is added to ML training data set and used for future classifications.
For external RF interference sources, Spectrum-NET’s patented Bayesian inference based locating algorithm determines the location of each external interference source with high accuracy. Spectrum-NET provides heat maps showing the likelihood of interference source being located at a given pixel on the map with 5m resolution per pixel at maximum zoom. The interference location precision depends on cell density and number of cells impacted. Spectrum-NET’s interference locating algorithm uses horizontal and vertical antenna radiation pattern information and shuttle radar topography mission data.
Interference Impact Assessment & Interference Severity
Spectrum-NET measures the impact of each interference event on network KPIs (e.g., session drop rate, access failure rate, handover failure rate, downlink/uplink latency and uplink spectral efficiency). Spectrum-NET calculates an interference severity for each interference event based on the characteristics of the event. The interference severity and impact assessment are key objective measures for prioritizing operator resources to mitigate the events.
PIM Interference Mitigation
Spectrum-NET CNN’s have been trained with labeled PIM interference data from many mobile networks for automated classification of PIM interference. Spectrum-NET CNN models detect the presence of PIM interference on cells with high accuracy. Spectrum determines the root cause of PIM interference (i.e., an internal or external source) through analysis of branch imbalance measurements, correlations with other channels at the site, calculating the potential for PIM (IM2 – IM5) via PIM Spectrograph analysis and analysis of the antenna topology mapping. KPI impact from PIM interference and number of cells with PIM at each site are used to prioritize which sites to address.
Future Proof Roadmap
The Spectrum-NET roadmap contains additional highly innovative capabilities including, a regulator portal for collaboration between regulators and operators on interference issues, integration with SON tools for closed-loop actions, a cross-border ML-driven linear regression model, and centralized optimization of the new 3GPP Release 16 RIM-RS feature. Spectrum-NET’s Deep Convolutional Neural Network (CNN) models are trained with RF interference signatures (labeled measurement data) from mobile networks across the globe. As new interference types are identified, additional measurement samples associated with the new interference type are labeled, and the models are re-trained with the new data.