Definition
Correlation between the RAN optimizer's target metric (e.g., spectral efficiency, RSRP, CQI) and actual measured user experience (application-level quality, session success rate, customer complaint rate) across cell-level observations.
The Measurement Gap
Decision Supported: Is the RAN optimizer improving the right thing, or is it pushing a number that has disconnected from what users actually experience? Every SON operates on a proxy metric (spectral efficiency, reference signal quality, channel quality indicator) because the optimizer needs a fast, computable target. The assumption is that improving the proxy improves experience. EPDI tests that assumption continuously.
Measurement Gap: Legacy RAN performance metrics track the proxy and the experience independently but never measure the relationship between them. A network engineering team can report that spectral efficiency improved by 8% across a market while customer complaints in that market increased by 12%, and no existing metric connects those two facts. The proxy may have been valid when it was chosen (spectral efficiency genuinely correlated with user experience when voice was the dominant service) but as traffic mix shifts toward video, gaming, and IoT, the proxy-experience relationship degrades without anyone measuring the degradation. EPDI quantifies this drift, answering whether the optimizer's proxy is still worth optimizing (Agiwal, Roy & Saxena, 2016; GSMA Intelligence, 2023).
Formula
Formula (Plain Language): For each cell in each measurement period, record two values: the optimizer's target metric (what it is trying to maximize) and the corresponding user experience metric (what actually matters to subscribers). Compute the Spearman rank correlation across all cells. If the correlation is high, the proxy is valid. If it drifts toward zero or goes negative, the optimizer is chasing a metric that no longer represents experience.
Notation:
EPDI = 1 - rho(Proxy_metric, Experience_metric)
Variable Definitions:
- rho: Spearman rank correlation coefficient computed across all active cells in the measurement population. Spearman is used rather than Pearson because the relationship between proxy and experience may be monotonic but non-linear.
- Proxy_metric: The RAN optimizer's target metric per cell per measurement period. Examples: average spectral efficiency, reference signal received quality (RSRQ), channel quality indicator (CQI). Unit: varies by proxy chosen. One proxy per EPDI computation.
- Experience_metric: Measured user experience per cell per measurement period. Examples: video Mean Opinion Score (MOS), session completion rate, application-level throughput at the 10th percentile, customer complaint rate. Unit: varies by experience metric chosen.
- Measurement population: All active cells in the market or cluster being evaluated.
Units: EPDI ranges from 0 to 2. EPDI = 0 means perfect positive correlation (proxy is fully valid). EPDI = 1 means no correlation (proxy is disconnected from experience). EPDI greater than 1 means negative correlation (the optimizer is actively making experience worse by improving the proxy).
Edge Cases: Compute EPDI separately for voice, video, and data sessions. The proxy may be valid for one service type and invalid for another. Segment by user location within cell (center vs. edge), as proxy-experience divergence is typically worst at cell boundaries. Proposed starting point for minimum sample: 30 cells with 1,000 or more sessions each per measurement period.
Proposed formula, pending validation pilots.