The way networks are designed, monitored, and managed has barely changed while everything connected to them has. 5G made that gap impossible to ignore, because a 5G design is a capacity and interference problem before it is a coverage problem.
5G network planning is the work of deciding where radios go, which spectrum each carrier uses, and whether the result carries the traffic it was built for. It runs in six stages: model the site, set the service requirement, configure radios and carriers, simulate coverage and quality, resolve interference and cell identity, then validate against measurement. Coverage is the straightforward half. Capacity and interference decide whether the design survives real users.
A 5G cell can show strong reference signal power across an entire floor and still fail the people standing on it. Reference signal received power tells you the serving cell is audible. It says nothing about how much of the carrier's resource grid is left once the users arrive, and nothing about how much of that signal is competing with the cell next door.
This is what catches teams carrying in-building 4G practice into 5G. Bandwidth and numerology set the size of the resource grid. Scheduling and load decide how much of it any one user gets. A plan that stops at coverage has not answered the question the network was built to answer.
Indoors that means floor plans at a true scale, walls with real material properties, doors, windows and the floor to ceiling context. Outdoors it means terrain, clutter, buildings, foliage and the height reference used for both radios and receivers. A plausible looking map is not an RF ready model. Missing high loss walls, wrong roof heights and optimistic foliage assumptions all move the predicted serving edge a long way.
Fix the band and the design KPI before anything is placed. Then configure a representative user equipment profile, including receiver height and device assumptions, because the simulation resolves the design against that profile rather than against an ideal receiver.
For each radio, add a carrier with the required band, bandwidth, subcarrier spacing where it applies, transmit power, duplex behaviour, modulation and MIMO configuration. Add an antenna sector and assign the carrier to it. Confirm antenna model, height, azimuth and downtilt. Placement comes after the configuration, not before it.
Treat the KPIs as a diagnostic set rather than a scorecard. RSRP reads reference signal coverage and serving cell strength. RSRQ reads reference signal quality relative to wideband received power. SINR reads signal quality under modelled interference and noise. Throughput reflects the configured radio, carrier, MIMO and device assumptions. Best server and dominance views show overlap and handover behaviour.
Do not approve a design from RSRP alone. A strong serving signal coexists comfortably with poor RSRQ, excessive overlap, a constrained carrier or an unrealistic device assumption.
Physical Cell ID is an identity, not a frequency. LTE uses 0 to 503 and NR uses 0 to 1007. Neighbouring cells that share a conflicting ID produce collisions and confusions, which surface as handover failures rather than as a coverage hole. A PCI plan is built from the actual coverage overlap between cells, constrained by a minimum reuse distance and any reserved values, and it has to account for mod 3 and mod 30 conflicts as well as direct collisions.
A prediction becomes trustworthy once it has been compared with a walk test or drive test on the same site. Without that loop the model never gets corrected, so it never improves. This is the stage most often skipped and the one that determines whether the next design starts from evidence or from hope.
Capacity planning starts from a population and a per user profile, not from a rule of thumb about square metres. For cellular, demand is checked against carrier resources. FDD evaluates the two directions separately, while TDD shares resources according to the carrier's downlink and uplink split. The output worth recording is not pass or fail. It is which resource saturated first, because that is what makes the next change traceable to evidence.
Backhaul belongs in the same calculation. Served downlink plus uplink traffic has to fit the backhaul limit, and a design that clears the air interface can still be constrained behind it.
Private cellular runs the same workflow with different spectrum rights. CBRS band 48 in the United States, and the shared and lightly licensed allocations elsewhere, moved private cellular out of carrier hands and into the reach of system integrators and enterprise teams. The planning stages do not change. What changes is that you are planning against your own allocation rather than competing for shared airtime, so the interference picture is yours to control and yours to get wrong.
eino is an agentic platform that unifies and monitors all wireless technologies across any site, indoor and outdoor, through a 3D digital twin.
For cellular that means the site model, the radio configuration, the heatmap and the interference plan live in one project. PCI planning reads coverage overlap from the same model that produced the heatmap, scores the current assignment against a proposed one, and applies a conflict free plan across every carrier at once. You can export the neighbour relations as a CSV to seed the OSS before automatic neighbour relations take over on live traffic, and hold any cell at a fixed PCI when the deployment requires it.
Capacity planning runs against demand profiles mapped onto the plan and reports the health score, the SLA failure probability, the lower tail and median user throughput, the leading failure cause and the bottleneck resource. A shortfall overlay colours every grid cell by the reason capacity falls short, so the next change targets the area that actually saturated.
Propagation is modelled with real ray tracing on GPUs in the cloud, so a changed carrier or a moved sector returns a new heatmap in seconds rather than after a simulation cycle. The same project carries Wi-Fi, DAS, fixed wireless and IoT, so a converged site is planned against one model instead of two that never reconcile.
Prediction does not replace spectrum authorisation, regulatory and EIRP checks, structural and mounting review, backhaul and power design, interference coordination with neighbouring operators, or representative field measurement. Plan for all of them, and treat the model as the thing that tells you where to look first.
They describe the same work at different resolutions. Planning usually covers spectrum, capacity and site selection decisions. Design covers the specific radio, antenna and cabling layout that implements them. Both depend on the same site model, which is why splitting them across two tools creates rework.
There is no per square metre answer. The count follows from the service requirement, the band, the material between the radios and the demand. A high band carrier through internal walls needs far more radios than a mid band carrier across open space.
Physical Cell ID identifies a cell to a device. Two cells a device can hear at once must not share an ID, or the device cannot tell them apart. Conflicts appear as handover failures and dropped sessions rather than as weak signal, which makes them expensive to diagnose in the field and cheap to prevent in the plan.
Some can. Historically these were separate products with separate models, which is why converged sites often end up with two designs that never reconcile. Platforms built around a single site model handle multiple radio technologies against shared geometry.
Coverage and quality heatmaps per band and KPI, a radio and sector schedule with carrier configuration, the PCI plan and neighbour list, a bill of materials, cabling and head end assignments, the capacity scenario with its stated assumptions, and the compliance thresholds the design was judged against.
Accuracy is bounded by input fidelity. Ray traced propagation against accurate geometry and correct material properties is a reliable planning tool. The same engine against a poorly traced floor plan is not. Field validation is what closes that gap, which is why it belongs in the workflow rather than at the end of it.
