The most serious network issues no longer sit neatly inside individual cells. They appear in the transitions between them.
Voice traffic keeps growing, smartphones are becoming common, and users are no longer staying still. Cells show solid coverage and acceptable KPIs, yet dropped calls continue. Almost always during movement. Walking short distances, driving along roads, or shifting slowly indoors triggers failures that static testing cannot reproduce.
Mobility failures are path problems, not cell problems. That single realization changes how the whole analysis is approached.
In several WCDMA clusters, radio link failures happened repeatedly at specific cell edges. The serving cell looked healthy. The target cell looked healthy. The handover transition itself was unstable.
Overlaying handover failure counters with drive-test logs made the pattern clear. Preparation succeeded. Execution failed. The UE returned to the original cell too late to recover the call. Neither cell's individual metrics showed anything wrong.
One common cause was neighbor list asymmetry. Cells advertised neighbors that existed on paper but ranked poorly in practice under traffic load. The UE followed the planned neighbor sequence while actual radio conditions degraded faster than measurement reporting could keep up.
The fix was not in either cell individually. It required checking actual HO success rates per neighbor pair during busy hours, not just the existence of the neighbor relationship in the list.
A frequent secondary cause was inconsistent measurement configuration between adjacent cells. Slightly different reporting thresholds, hysteresis values, or event triggers carried over from rollout templates created gaps along real mobility paths.
| Parameter | Cell A (source) | Cell B (target) |
| 1A threshold | -14 dB Ec/No | -12 dB Ec/No |
| 1B threshold | -20 dB Ec/No | -22 dB Ec/No |
| Hysteresis | 2 dB | 4 dB |
| Effect | UE loses Cell A from active set before Cell B qualifies. Gap window: ~400ms. Enough for RLF under fast fade. | |
Each parameter looked reasonable in isolation. Together they created a gap window along the boundary where no valid active set existed. The UE had nowhere to go at exactly the moment it needed to move.
Fixing these problems required a different analysis approach. Not evaluating handover KPIs cell by cell. Mapping mobility sequences: which cells users came from, which they attempted to move to, and where failures clustered along actual paths.
Once the analysis shifted to this level, problems that stayed hidden in aggregate metrics became visible immediately. The path failure rate on a specific neighbor pair during busy hours was 20%. The cell-level KPI showed 94% success. Both numbers were technically accurate. Only one was useful.
