PDD Anomaly Detection
Post-dial delay varies naturally by time of day, destination, and carrier load. GSNOC builds hour-of-day and destination-aware baselines so PDD anomalies represent real problems — not expected variation.
The PDD Monitoring Problem
High post-dial delay degrades user experience — callers hear silence before ring-back and often hang up. But PDD varies significantly by time of day, destination, and network load, making simple thresholds generate constant noise.
- Long PDD causing callers to abandon calls before connection, degrading effective ASR and customer experience
- Time-of-day PDD variation generating false alerts during legitimate peak periods
- Destination-specific PDD differences making universal thresholds inappropriate for international routing
- Carrier-specific PDD degradation going undetected when mixed with routes that have naturally higher PDD
How PDD Anomaly Detection Works
Eight analysis methods combine hour-of-day bucketing, destination awareness, and trend analysis to separate signal from noise.
Separate PDD baselines are maintained for each hour of the day per carrier and route, so peak-hour PDD increases don't trigger off-peak alert thresholds.
PDD baselines are maintained per destination prefix so that international routes with inherently higher PDD are not compared against domestic baselines.
Sudden PDD spikes that exceed the hour-adjusted baseline by a statistical threshold flag potential carrier congestion or routing issues.
PDD that remains elevated over multiple evaluation periods — even below spike thresholds — is flagged as a sustained degradation event.
Compares PDD across multiple carriers serving the same destination to identify which carrier is the source of observed delays.
Identifies step changes in PDD that correlate with routing configuration changes — distinguishing intentional changes from accidental degradation.
Identifies PDD patterns consistent with carrier congestion — elevated delay during specific time windows that repeat across days.
Tracks PDD trends over days and weeks to identify carriers whose PDD is slowly worsening before it becomes a visible problem.
Why GSNOC PDD Anomaly Detection Is Different
- Hour-of-day bucketed baselines eliminate the constant false positive problem of simple PDD thresholds
- Destination-aware baselines correctly handle international routes that have inherently higher PDD than domestic
- Carrier comparison identifies which carrier is responsible when multiple carriers serve the same destination
Key Metrics
Average post-dial delay measured in the current evaluation window in milliseconds
Expected PDD for this hour, destination, and carrier based on rolling historical data
How many milliseconds the current PDD exceeds the hour-adjusted baseline
Where current PDD sits in the historical distribution — 95th percentile and above indicates a significant anomaly