Two problems, one network
Coverage planning asks whether there is enough signal to serve a location at all. Capacity planning asks whether there is enough resource to serve everyone in that location at once. A network is dimensioned against whichever binds first, and in practice that changes over the life of the site — most urban sites are built for coverage and rebuilt for capacity.
Choosing a propagation model
The model has to match the clutter. Applying an open-terrain model to a dense urban grid produces an inter-site distance that will not survive first drive test.
| Environment | Model | Typical ISD, LTE 1800 |
|---|---|---|
| Dense urban | COST 231, or ray tracing where building data exists | 200–500 m |
| Urban | Okumura-Hata, extended | 500 m – 1.5 km |
| Suburban | Hata | 1–3 km |
| Rural and open | Free space or Hata | 3–10 km |
Model output is a starting point, not a conclusion. Calibrate against measured data from the actual environment wherever it exists — a model tuned with local drive test is worth far more than a default parameter set, and the correction is often several dB.
Capacity: what sets the ceiling
C = B × log₂(1 + SINR)
B in hertz, SINR as a linear ratio rather than in dB. This is a theoretical bound that no real system reaches, but it fixes the shape of the problem: capacity scales linearly with bandwidth and only logarithmically with SINR.
That asymmetry drives most capacity decisions. Doubling bandwidth roughly doubles capacity; doubling SINR does not. Once a cell is already clean, more power buys very little — the remaining levers are more spectrum, more spatial streams, or more cells.
C_cell = BW × SE × N_streams
SE is spectral efficiency in bits per second per hertz: roughly 5–7 bps/Hz for LTE with 2×2 MIMO, and substantially higher for 5G NR with massive MIMO, where spatial multiplexing rather than raw bandwidth carries much of the gain.
| Configuration | Bandwidth | SISO | 2×2 MIMO | 4×4 MIMO |
|---|---|---|---|---|
| LTE Cat 4 | 20 MHz | 75 Mbps | 150 Mbps | 300 Mbps |
| LTE-A, 256QAM | 20 MHz | — | ~200 Mbps | ~300+ Mbps |
| 5G NR, FR1 | 100 MHz | ~400 Mbps | — | ~1 Gbps+ |
| 5G NR, FR2 mmWave | 400 MHz | — | — | Multi-Gbps |
Peak rates are not planning figures
Every number above is a single user, at the cell centre, with the channel to themselves. Dimension against the cell edge rate and the busy hour load, not against the marketing peak. The gap between the two is routinely an order of magnitude.
Dimensioning to load
N_users = Target throughput per cell / Per-user data rate
Voice is dimensioned with Erlang B or Erlang C against a blocking target. Data is best effort, so it is dimensioned instead to keep resource block utilisation below roughly 70–80 per cent at busy hour.
Resource block utilisation is the number that actually constrains a site. Above about 70 per cent, scheduling delay rises sharply, cell edge users are starved first, and the throughput KPI degrades before any accessibility or retainability KPI moves. A cell can be fully covered, entirely clean, and still failing its users.
The trade-off
Coverage and capacity pull in opposite directions, and nearly every planning decision is a position on that axis:
- Lower frequency covers further but carries less spectrum. Low band is a coverage layer.
- Higher frequency carries more spectrum but covers less. High band is a capacity layer.
- More downtilt contains a cell and reduces interference, at the cost of edge coverage.
- More power extends one cell and interferes with its neighbours — a net loss once the network is dense.
- More sites solve both, and cost the most.
There is no configuration that optimises both at once. The planning job is to state which one is binding in a given area, and to be explicit about what is being traded away in exchange.