# Concept 4E — Building entry and penetration estimation Concept 4E is a bounded analysis for the question: “What is the estimated signal immediately inside a representative building facade?” It is available at 2.6 GHz and 28 GHz. It does not turn the building dataset into an indoor radio map, a floor/room model, or a whole-building coverage claim. The 140 GHz `research_sub_thz` profile remains outside this concept. ## Boundary and no-double-counting rule Concept 4E evaluates a two-stage link: 1. Use the Concept 4D `urban_short_range` UMa outdoor LOS/NLOS median baseline from the serving cell to a representative facade point. 2. Subtract one standard O2I external-wall entry term at a receiver placed at zero indoor depth immediately inside that facade. The facade target is not treated as an outdoor wall event. Outdoor-to-outdoor NLOS already uses the Concept 4D empirical NLOS formula, so Concept 4E does not add the legacy per-wall loss to that outdoor baseline. The response reports `outdoor_wall_loss_db: 0` for this reason. The entry term is applied once, separately, as `low_loss` and `high_loss` scenarios. There is no interior-room, floor, interior-wall, reflection, fast-fading, building-height, or whole-building inference. The receiver height is the configured per-cell receiver height. The result is one deterministic estimate per selected building, served by the strongest eligible cell under the low-loss scenario; the response retains both scenario powers. ## Standard-derived loss model The reference is [3GPP TR 38.901 V19.4.0, §7.4.3.1](https://www.etsi.org/deliver/etsi_tr/138900_138999/138901/19.04.00_60/tr_138901v190400p.pdf), “O2I building penetration loss.” The standard writes the O2I path loss as: ```text PL = PL_b + PL_tw + PL_in + X_sigma ``` where `PL_b` is the basic outdoor path loss, `PL_tw` is external-wall loss, `PL_in` is depth-dependent indoor loss, and `X_sigma` is penetration shadowing. Concept 4E uses the deterministic median (`X_sigma = 0`) and sets `d2D-in = 0 m`, so `PL_in = 0 dB`. Its implemented link budget is therefore: ```text P_outdoor_facade = Concept 4D UMa LOS/NLOS received power P_low_inside = P_outdoor_facade - PL_tw_low P_high_inside = P_outdoor_facade - PL_tw_high ``` For 28 GHz, the standard material terms are evaluated with `f` in GHz: ```text L_glass = 2 + 0.2 f L_IRRglass = 25.4 + 0.11 f L_concrete = 5 + 4 f ``` The two reference compositions are: ```text PL_tw_low = 5 - 10 log10(0.3 * 10^(-L_glass/10) + 0.7 * 10^(-L_concrete/10)) PL_tw_high = 5 - 10 log10(0.7 * 10^(-L_IRRglass/10) + 0.3 * 10^(-L_concrete/10)) ``` At 28 GHz these evaluate to `17.8287874527 dB` and `35.0290195972 dB`. At 2.6 GHz the implementation uses the standard’s UMa/UMi single-frequency backward-compatibility value from Table 7.4.3-3: `PL_tw = 20 dB` and `PL_in = 0.5 * d2D-in`. Because Concept 4E fixes indoor depth at zero, its low/high API scenarios are equal at 2.6 GHz (`20 dB` each). The standard’s reference sigmas are exposed as metadata, but no random draw is sampled. The standard’s material composition is a simulation parameter dependent on metal-coated glass, regional market, and deployment scenario. A.T.O.M. does not invent that mapping from a generic OSM building type. Known tags are returned as optional evidence only; they never select one profile automatically. If material evidence is absent or not defensibly mapped, both profiles remain in the response and `selected_entry_profile` is `unknown`. ## Geometry and deterministic selection The backend evaluates one HTTP request containing the selected cell RF profiles and the building filter. It then: - sorts cells and footprints by stable ID; - selects footprints inside the effective per-cell radius union, unless an explicit `building_ids` filter is supplied; - obtains a representative interior point from the polygon centroid when it is inside, then a vertex average, then a deterministic 15×15 interior grid; - finds the first true outside-to-inside crossing from the cell to that point; - rejects invalid polygons, tangent-only contacts, boundary-only points, and unclassifiable outdoor paths; - classifies blockers before the target facade as outdoor LOS/NLOS and excludes the target building from the blocker test; - evaluates the Concept 4D outdoor baseline and applies the two entry losses. The response identifies the point sources and includes the facade coordinate, outdoor LOS state, outdoor received power, both entry losses, both just-inside powers, serviceability flags, material evidence, and limitations for each building. The `-100 dBm` building-service threshold and each cell’s receiver sensitivity are reported separately; they are not interchangeable. Geometry regression fixtures cover rectangular, concave, multipart/decomposed, centroid-outside, invalid, and tangent cases. GeoJSON holes are not inferred as interior rooms: the existing runtime footprint representation uses outer rings, and Concept 4E makes no claim beyond the selected representative facade point. ## API `POST /api/building-entry-analysis` accepts the network-shaped RF defaults and one to six selected cells. The request is batched; it does not issue one HTTP request per building. ```json { "towers": [ { "id": "cell-28-a", "tower_lon": 32.8541, "tower_lat": 39.9208, "azimuth": 45, "rf_profile": { "schema_version": 1, "network_tech": "5g", "propagation_model": "urban_short_range", "frequency_ghz": 28, "band": "n257", "bandwidth_mhz": 100, "channel_id": "NR-634666", "duplex_mode": "tdd", "tx_power_dbm": 30, "antenna_gain_dbi": 17, "system_loss_db": 2, "radius_m": 400, "beam_width": 120, "antenna_height_m": 25, "receiver_height_m": 1.5, "receiver_sensitivity_dbm": -115 } } ], "rays": 72, "radius_m": 400, "frequency_ghz": 28, "tx_power_dbm": 30, "beam_width": 120, "calibration_offset_db": 0, "residential_only": false, "building_ids": [] } ``` The endpoint returns `model`, `applicability`, `summary`, per-building `results`, per-cell summaries, effective profiles, `rf_contract`, and `diagnostics`. `diagnostics.http_requests_required` is `1`; the frontend stores the response under an RF-derived source key containing effective cell profiles, coordinates, azimuths, dataset revision, and filters. Concept 4E results are invalidated when frequency, power, beam, antenna/RF profile, azimuth, selected cells, tower inventory, or dataset revision changes. Optimization priorities alone do not invalidate them because they do not change the RF payload. A successful azimuth/network optimization does invalidate the prior entry result because it changes effective RF state. The analysis is diagnostic/reporting output only; it is not an optimizer objective, constraint, score term, or candidate-site coupling. ## Data audit The active Ankara Open Planning Dataset is version `2026.07`. Its manifest contains 451 towers and 161,626 building GeoJSON features; the runtime index contains 161,784 outer-ring footprints after multipart decomposition. Dataset validation reported zero invalid, repaired, or dropped building geometries. The material audit scans `building:material`, `facade:material`, `material`, and `roof:material`. The canonical Ankara pack contains no usable material metadata in those fields, so Concept 4E reports 0 known and 100% unknown for that pack. Building type and residential/demand attributes remain useful for filtering and weighting, but they do not become material composition. ## Canonical six-cell Ankara audit The gated regression fixture uses the existing six-cell 28 GHz Ankara network scenario, with deterministic repeated execution and the active dataset. The following values are from the measured run on 17 September 2026. The analyzed building domain is `summary.relevant_buildings`, which includes all selected relevant footprints. `summary.relevant_residential_buildings` is a subset used for residential reporting; it is not the denominator for the all-building service rows. A service numerator includes only buildings with an applicable, valid facade estimate that meets the relevant threshold; unsupported buildings remain in the denominator but cannot enter the numerator. | Metric | Result | |---|---:| | Analyzed buildings (all relevant footprints) | 1,488 | | Residential buildings (subset) | 899 | | Valid facade estimates | 675 / 1,488 | | Outdoor service — all buildings | 675 / 1,488 | | Low-loss entry — all buildings | 675 / 1,488 | | High-loss entry — all buildings | 253 / 1,488 | | Low-loss served demand weight | 2,175.000 | | High-loss served demand weight | 440.000 | | Material-known / material-unknown | 0 / 1,488 | | Unsupported/no eligible serving cell | 813 | | Invalid geometry | 0 | | Candidate cell-link evaluations | 698 | | Analysis runtime (observed) | 34.8 ms | | HTTP requests required | 1 | ### Unsupported reason breakdown The unsupported count is derived from the implementation applicability status, not from a free-text interpretation. The canonical breakdown is: | Applicability reason | Count | Share of unsupported | Representative building | |---|---:|---:|---| | `no_eligible_serving_cell` | 813 | 100.00% | `feature-104140` | There were no `invalid_geometry` records in this run. The dominant status is an intentional aggregate result for buildings where no selected cell produced a valid eligible facade baseline under the configured radius/beam contract; it is not evidence that invalid geometry or a propagation-model fallback was hidden. Representative ledger rows from the same run: | Case | Building | Serving cell | Outdoor facade Rx | Low-loss Rx | High-loss Rx | |---|---|---|---:|---:|---:| | Strong facade signal | `feature-7237` | `LTE-35104` | -33.620 dBm | -51.448 dBm | -68.649 dBm | | Near threshold | `feature-9613` | `LTE-313110` | -79.950 dBm | -97.778 dBm | -114.979 dBm | | Low-loss pass / high-loss fail | `feature-101294` | `LTE-35877` | -87.441 dBm | -105.269 dBm | -122.470 dBm | No known-material row or both-scenarios-fail row occurred among the selected canonical targets; those are intentionally reported as absent rather than fabricated examples. Repeating the fixture produced identical summaries, results, and cell summaries. The complete low-loss-pass/high-loss-fail ledger for `feature-101294` is: | Field | Value | |---|---| | Serving cell | `LTE-35877` | | Representative facade point | lon `32.852901154309`, lat `39.924769672379` | | Outdoor facade Rx | `-87.441 dBm` | | Low-loss entry | `17.829 dB` → `-105.269 dBm`, serviceable | | High-loss entry | `35.029 dB` → `-122.470 dBm`, not serviceable | | Receiver threshold | `-115.000 dBm` | | Material evidence | unavailable; normalized material `unknown`; not used to select a profile | | Model / applicability | `urban_short_range`; `applicable` — representative facade entry estimate at zero indoor depth | | Geometry metadata | indoor depth `0.0 m`; outdoor wall loss `0.0 dB`; outdoor LOS state `nlos` | ## Scope of future refinement The current contract intentionally stops at a representative facade entry estimate. A later concept could add explicit floor/room geometry, material survey evidence, interior depth distributions, or calibrated regional composition parameters. Those additions would require a new model contract; they must not be inferred by silently extending Concept 4E.