Spectrum-NET Overview

October 2, 2018

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 1
Spectrum Effect™
Pioneering Spectrum Analytics
v Founded in 2015 with technology
purchase from Eden Rock
Communications (acquired by Nokia).
v Innovative technologies with roots in
spectrum sharing programs with the
United States DoD and ONR.
v R&D in Kirkland, WA USA and
Monterrey, Mexico.
v 12 issued patents and 20+ pending
patent applications.
$267B USD*
v Revolutionary spectrum analytics VALUE OF SPECTRUM
solution built on machine learning IN THE US
framework: Spectrum-NET. *Source: Q2 2018 earnings reports from Verizon, AT&T, T-Mobile and Sprint

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 2
Spectrum-NET™
Machine Intelligence Meets Spectrum
v Automated, seamless RF
interference management across
multi-vendor UMTS, LTE and (future)
5G NR networks.
v Innovative ML-based algorithms
applied to operator RAN data.
v Detection, classification, localization
& impact assessment of external and
unintended internal RF interference.
v Web-based, cloud-ready, highly
scalable solution with extensive $7B USD*
mapping, visualization, and reporting GLOBAL COST OF
capabilities. INTERFERENCE ANNUALLY
*Estimate based on trial results, impact projections and extrapolation across networks globally.

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 3
Spectrum-NET Trials
Public Webinar on November 28
The successful trial utilizing Spectrum-NET uncovered numerous
3 instances of external and unintended internal RF interference,
2 3
including previously undetectable PIM interference.
- Kristo Kork, Head of Radio Access Networks and
1 Infrastructure, Elisa

1
Spectrum-NET has proven valuable for us, clearly identifying
interference events and equipping us to improve our network. We
are very impressed with Spectrum-NET...
- Trevor Oster, EVP Engineering & Operations,
Northeast Wireless Networks
10
80k DAY AVERAGE Interference Examples
TOTAL CELLS TRIAL DURATION
Bi-Directional Tropospheric
PIM RF ID Scanner
Amplifier Ducting
2% of cells with severe Interference events
Security Alarm Power Line Industrial Noise Generator
4% of cells with moderate interference events
6% of cells with sporadic interference events Faulty Hardware Misconfiguration Cross Border Grey Market Device

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 4
Spectrum-NET
Advanced Interference Analysis
Data Ingestion Operator Premises Spectrum –NET Private Cloud

v RAN OSS CM, PM and event data and planning tool

UI Micro
topology data are processed and fed into Spectrum- Web UI
Service
NET ML-based algorithms for interference analysis.
Interference Analysis Stages Network Alarms
Services
Management Micro-
Services
v Anomaly Detection Modules for Outlier Detection System Services

v Feature Extraction Modules for Characterization ML-based

Algorithms
v Classification
OSSs &
v Aggregation Planning
Tools Database
v Localization
v Impact Assessment CM, PM, event,
topology date
Containerized Microservices
v Process the ingested data, network configuration, Data
Data
algorithms, analysis results, web UI and reporting. Mediation pre-processing Ingestion
Server

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 5
Spectrum-NET
Machine Learning (ML) to Continually Learn and Refine Results
Outlier Detection
v Unsupervised ML identifies anomalous behavior in PM/KPI/Event
data that can indicate the presence of interference.

ML-based Learning of Interference classification rules

v Spectrum-NET pre-populated with rules set to classify
interference types using features extracted from data.
v Supervised ML techniques use labeled interference examples to
adapt feature classification rules and to create new rules.

Deep learning models for Interference classification

v Deep learning Neural Networks used for further interference
classification.
v Neural Networks trained offline with simulated data and/or data
signatures observed when interference occurs in the network.

Roadmap: Predictive ML techniques to detect the

onset of an interference event in the network
v Trigger preemptive network reconfiguration to offset its effects on
network performance. Flexible and extensible architecture supports product
evolution for future Interference Detection Use Cases

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 6
Spectrum-NET
Localization and Heat Maps
Cell interference measurements
v Determines the RF interference levels seen at each
cell impacted by the interferer.
Localization Heat Maps
v Cell interference measurements used to calculate.
v Shows likelihood of interference source being located
at a given pixel on the map: 50m (1.5 arcsec)
resolution per pixel with max zoom.
v Interference location accuracy depends on cell
density and number of cells impacted.
v Additional innovations to narrow search area.
Playback Function
v Highlights movement of interference sources over
time.

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 7
Spectrum-NET
PIM Analysis
Site Analyzer
v Graphical presentation of all frequencies at
site for each sector.
Spectrograph
v Downlink and uplink frequencies at site.
v 2nd to 5th order potential PIM products.
v Overlap between potential PIM products and
uplink frequencies.

When PIM interference is found, Spectrograph

and Site Analyzer highlight cells and
frequencies causing and impacted by actual
(not potential) PIM interference.

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 8
Spectrum-NET
Interference Event Reporting
Powerful Reporting Framework
v Provides big picture with easy access to
details.
Intuitive User Interface
v On-the-fly data aggregation to instantly
breakdown interference events.
v Advanced search and filtering capabilities of
the detected interference events.
v Geographical representation to easily identify
problematic areas.
Insightful Report Generation
v On demand PDF reports with visualizations
and KPI information.
v Automatic reports in Release 19.X.
© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 9
Conclusion
Realize the Full Potential of Your Spectrum
Impacted UMTS/LTE KPIs Roadmap
v Data Call Drop Rate v Downlink Interference
v RRC Establishment Failure Rate v Cross Operator Analysis
v Handover Failure Rate v Closed-Loop SON Integration
v Uplink/Downlink Throughout v Open ML framework
v Uplink Latency v Evolution to 5G NR
v Uplink Spectral Efficiency
Operator Savings with Spectrum-NET
v Much faster handling and resolution of interference
events and resolution of many more events.
v Improves user quality of experience and network
coverage & capacity.
v Typical 3 year ROI > 1000%.

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 10
AI Meets Spectrum:
RF Interference Demystified
November 28, 2018 9:00 am GMT | 4:00 am EST | 5:00 pm CST

Register now for this must-attend webinar for key

insights from our revolutionary machine learning
spectrum analytics solution, Spectrum-NET.

© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 11
 Thank you!

Frank J. DeJoy, CEO

frank.dejoy@spectrumeffect.com

10510 Northup Way Suite 130

Kirkland, WA 98033

www.spectrumeffect.com
© Copyright 2018 Spectrum Effect. All Rights Reserved. www.spectrumeffect.com Confidential & Proprietary 12