Benefits of Enhanced Event Analysis in Data Center OTDR Testing Mark Miller Dr. Fang Xu AFL/Noyes Test & Inspection
Overview Challenges Topics Techniques and Improvements
Benefits of enhanced event analysis in data center OTDR testing EVENT ANALYSIS OVERVIEW
Data Center Architecture Characteristics Huge number of fiber runs Structured cabling uses short jumpers and contains many connectors Short networks with closely spaced events OTDR Requirements Short Dead Zones are required to characterize fibers and locate faults Auto event analysis needed to cope with large number of cables
What are Events? Additional structures along a fiber optic network, such as connectors, fusion splices, optical splitter or macro-bends are called events These additional structures leave characteristic foot prints on a trace acquired by an Optical Time-Domain Reflectometer (OTDR)
Events on OTDR Trace and their Measurements Event Measurements Location Type of Event Reflectivity Loss
Why Measure Events? Tier 2 OTDR testing required by standards Data Center: Standard ANSI/BICSI 002-2011 2011 Data Center Design and Implementation Best Practices TIA 568-C.3 Connector Spec ISO/IEC 11801:2010 Baseline documentation of networks Certification reports Future troubleshooting
Auto Event Analysis OTDR Event Analysis Computerized dfinding and measuring of events in optical network from OTDR trace is called Auto Event Analysis Fully automatic Event Analysis allows a consistent interpretation of events regardless of user expertise If you can t trust the accuracy of an OTDR s event analysis: Manual Trace Analysis Use of cursors across trace Time consuming Subject to human error
Manual Event Trace Analysis Move cursors and zoom for each event
Benefits of enhanced event analysis in data center OTDR testing EVENT ANALYSIS CHALLENGES
Event Analysis Challenges Incomplete knowledge of network Time required to analyze trace Closely spaced events Connector followed by connector Connector followed by splice Splice followed by connector Accurate determination of start and end of fiber under test Use of launch and receive cables Necessity of running a separate calibration test to compensate for length variations (+/- 10m) Event types Connectors Splices Splitters Macrobends
Why Event Analysis Misses Events Perfect event analysis does is not achievable It is hard to find all events in practice as we also want limit false events What conditions can trick auto event analysis? Small event magnitude hard to separate from noise Close events limited by used pulse-width Event sequence i.e. non-reflective followed by a reflective event Hardware limitation - event during saturation recovery Algorithm problem usually seen as missing obvious event
The OTDR Trace: Dead Zones Event Dead Zone At 1.5dB down from peak reflection(unsaturated event) where the user can accurately measure the distance between 2 events. EDZ 1.5 db Attenuation Dead Zone Distance from the start of the event to the point where the power following the peak reflectance has returned to within 0.5 db above the level of backscatter
Effect of ADZ on Trace Events Separated and overlapping events Events with similar reflectance Separated Overlapping Events with different reflectance Separated Overlapping
No Industry Dead Zone Specification There is no standard method for specifying the performance of auto event analysis Good event analysis performance is taken for granted by OTDR users Dead zone specifications only apply to the p y pp y trace
Event Analysis Performance Evaluation Matched event rate: Are all actual events being detected? This should be as high as possible, with 100% being perfect. False event rate: Are false events being detected? This is usually due to noise spikes. This should be as low as possible, with 0% being perfect Pass rate: Are the results repeatable when the same network is tested multiple times? This should be as high as possible, with 100% being perfect.
Tools for Event Analysis Evaluation What people see is a tip of an iceberg: Lab and field testing Golden trace library Fiber networks Networks description Traces collected from different units Event Regression tool Performance comparison
Benefits of enhanced event analysis in data center OTDR testing EVENT ANALYSIS TECHNIQUES AND IMPROVEMENTS
Traditional Event Analysis Implementation Based on apparent event magnitude Insertion loss Reflectance User sets threshold for these parameters If greater than threshold, then it is an event Basic methods used inside event analysis Gradient (derivative) Linear regression Correlation Traditional i user model allows the user to chose limits but can miss small loss event in this example 0.08dB 0.11dB Limitations Accuracy depending di on position Noise increases as event getting further from launch Et Extra user it intervention ti needed dd Highly depends on user s knowledge and skill Technical limitationsi i Derivative is sensitive to noise Need smart way to determine where a segment starts and ends to apply linear regression There is no universal model of events to run correlation
Traditional Event Analysis Algorithm Gradient (derivative) Derivative is sensitive to noise Linear regression Determine segment start and end is challenging Correlation There is no universal event model to correlate with before finding events
Traditional Event Analysis Implementation Based on apparent event magnitude Insertion loss Reflectance User sets parameter threshold If greater than threshold, then it is considered an event Limitations Accuracy depending on position Noise increases as event getting further from launch Extra user intervention Highly depends on user s knowledge and skill 0.08dB 0.11dB Traditional user model allows the user to chose limits but can miss small loss event in this example
Dead-end for Magnitude Based EA Example of true event and false event found at different threshold using a major OTDR market player Threshold Matched events % False events % 0.1dB 60.33 11.54 0.01dB 61.63 27.08 A slight increase of few event at smaller magnitude raises chance of exponentially increasing false events 1 : 12 is true event to false event increase ratio! Hard to make further progress due to excessive false event
Wikipedia Introduction of Likelihood Likelihood is a function of how likely an event is, which is weaker than probability (or odds in favor). In statistics, a likelihood function is a function of the parameters of a statistical model, defined as follows: the likelihood of a set of parameter values given some observed outcomes is equal to the probability of those observed outcomes given those parameter values. Wolfram MathWorld Likelihood is the hypothetical probability that an event that has already occurred would yield a specific outcome. The concept differs from that of a probability in that a probability refers to the occurrence of future events, while a likelihood refers to past events with known outcomes.
Likelihood Applied to EA How event Likelihood is measured Expected noise level surrounding event Shape of event Magnitude of event Slope of line segments surrounding event Etc. Smaller magnitude Higher likelihood Larger magnitude lower likelihood
Problem Similar to Weather Forecast What happens if it rains but the forecast predicted a sunny day? OTDR users think exactly the same way about event analysis. Modern weather forecasts use probability of rain instead:
Benefits of New Approach User can trade-off of sensitivity or reliability by setting likelihood threshold after event analysis Found Events (%) 100 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 5 0 Experienced user can choose higher sensitivity if he feels expected event is missing Default value is balanced between sensitivity and reliability High reliability is useful in certain cases. 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 Likelihood (%)
Completely New Event Analysis Event analysis built from ground-up Multiple factors are taken into account True event and false event found at different threshold using new algorithm for same test networks Threshold Matched events % False events % 0.1dB 73.42 12.29 0.01dB 74.83 12.83 True event to false event increase ratio is 1 : 0.38! It is still on it s way for further progress
Event Analysis Quality Comparison Among OTDRs False event % 25.00 20.00 15.00 10.00 5.00 Default loss threshold at 0.1dB Increasing matched events with moderate increase of false events is challenging F @ 0.01dB G O J M F @0.1 db E Using magnitude as event criteria has exponential increase on false events (Slope > 45 ) D @ 0.02 db D @0.1 db M310 @ 0.01dB M310 @0.07 db M310 @0.1 db Using likelihood as event criteria i limits it false events (Slope < 45 ) 000 0.00 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00 100.00 Found event %
Prefect Score on 16-Event Network M310 16 events (prefect score) OTDR A 29 Events
Application: Jumper Cable Testing Using M310 Reflectance and loss measured at both ends of cable using estimation algorithm 2 m jumper cable
Application: Data Center Network Using M310 9-Event Network: Closely spaced events - Short jumper cables Gainer- Splice
Event Analysis Checklist Matched/Missed event occurrence False event occurrence Correct identification of event type Ability to separate closely spaced events Ability to locate beginning and end of fiber under test without a separate launch and receive cable calibration test Provide all relevant measurements for each event
Potential for Future Improvement False event 25.00 Likelihood technique allows Traditional algorithms for ongoing improvement 20.00 15.00 Traditional methods cannot be improved Direction of progress 10.00 Actual path of progress 5.00 Improvement is possible only if false event rate is low False event rate: an indicator of improvement potential 000 0.00 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 90.00 100.00 Found event
Conclusion Introduction of likelihood in event analysis offers balance between sensitivity and reliability New event analysis increases user satisfaction level by offering higher found event rate It reduces user frustration with fewer false events Lower false event rate indicates it s potential for future performance improvement OTDR performance evolution to match Data Center evolution Dynamic Range Short Dead Zones Event Analysis Past Present Future