Skip to content

real_drive_test_ingestion.md

Latest commit

 

History

History
98 lines (74 loc) · 3.68 KB

real_drive_test_ingestion.md

File metadata and controls

98 lines (74 loc) · 3.68 KB
Raw

Real Drive-Test Ingestion Design

This project currently analyzes synthetic data generated in the internal schema. Real drive-test support should be added through explicit parser adapters rather than by forcing vendor exports directly into the analyzer.

Current validation status: synthetic-data validated; real operator drive-test validation pending.

Target Sources

Future adapters could support CSV or Excel exports from tools such as:

  • TEMS Investigation / TEMS Discovery
  • Nemo Outdoor / Nemo Analyze
  • QualiPoc
  • Rohde & Schwarz scanner exports
  • Generic UE diagnostic exports where the required KPIs are available

No real parser is included yet because vendor exports differ by product version, license options, technology layer, and export template.

Expected Internal Schema

Parser adapters should map vendor columns into these internal fields:

Internal Column Meaning
timestamp Measurement timestamp
latitude, longitude WGS84 GPS coordinate
serving_cell_id Anonymized serving cell or sector identifier
frequency_band LTE/NR band label
pci LTE PCI or NR PCI
technology LTE, 5G NSA, or 5G SA
rsrp_dbm RSRP in dBm
rsrq_db RSRQ in dB
sinr_db SINR in dB
rssi_dbm RSSI if available
dl_throughput_mbps Application or PHY downlink throughput
ul_throughput_mbps Application or PHY uplink throughput
latency_ms Ping/RTT or session latency
handover_event Boolean serving-cell change or decoded HO event
dropped_session Boolean dropped call/session/RLF indicator
speed_kmh GPS or UE-reported speed
route_distance_km Calculated cumulative route distance

Optional but useful real-world fields:

  • neighbor PCI/SSB information
  • EARFCN/NR-ARFCN
  • cell global identity after anonymization
  • event type, for example A3, A5, B1, RLF, or re-establishment
  • serving beam or SSB index for NR
  • device model and logging tool version
  • test script or service type

Column Mapping Strategy

Each vendor adapter should:

  1. Read the export with a strict input encoding and delimiter strategy.
  2. Normalize column names into a vendor-neutral mapping table.
  3. Validate units, for example Mbps vs kbps and milliseconds vs seconds.
  4. Convert timestamps to ISO format.
  5. Preserve only analysis-safe columns.
  6. Add derived fields such as route_distance_km, RF score, and coverage class.
  7. Fail loudly if required KPIs are missing.

Do not guess silently when a vendor export has ambiguous columns. A bad parser is worse than no parser because it can create misleading RF conclusions.

Anonymization Requirements

Real drive-test data may include confidential network or subscriber information. Before committing or sharing any real export:

  • remove IMSI, IMEI, MSISDN, device serial numbers, and user identifiers
  • anonymize operator name, PLMN, TAC, eNB/gNB IDs, cell names, and site names
  • generalize or mask exact customer routes if they reveal sensitive locations
  • remove internal comments, engineer names, ticket IDs, and project identifiers
  • confirm that sharing is allowed by employer, operator, customer, and NDA terms

For a public portfolio, synthetic data is the safer default.

Validation Needed For Real Data

Real drive-test ingestion should be considered incomplete until it is validated against known-good exports. At minimum:

  • compare row counts before and after parsing
  • verify KPI units and ranges
  • manually inspect several known poor and good locations
  • confirm handover events against serving-cell changes
  • compare map output with the original route
  • document any vendor-specific assumptions

Only after this work should the project claim real drive-test support.