TL;DR
- Refarming is a coverage problem before it is a capacity problem: the new carrier's uplink budget is usually worse than the layer it replaces.
- Prove readiness by measuring the replacement layer at the worst-decile locations of the old layer, not by aggregate coverage percentage.
- Guard bands consume capacity that the business case usually ignores, so budget them explicitly before quoting the capacity gain.
What changes when a band moves from 2G/3G to LTE or 5G?
Three things change at once, and only one of them is obvious. The obvious one is capacity: a refarmed 10 MHz block delivers far more throughput as LTE or NR than as GSM or UMTS. The less obvious ones are the link budget and the interference environment. A refarmed block usually moves to a higher frequency within the operator's holdings, or keeps the frequency but changes the multiple access scheme, and either way the uplink budget per user changes.
GSM has an extremely favourable link budget because the channel is narrow and the receiver sensitivity is very low in absolute terms. A voice service on GSM can be delivered at levels where an LTE cell-edge user cannot sustain the minimum modulation and coding scheme. Replacing a GSM layer with LTE therefore removes coverage at the edge of the old footprint even though the new network is technically superior in the centre.
| Refarming case | Uplink budget change | Dominant risk |
|---|---|---|
| 900 MHz GSM → LTE 900 | Roughly neutral; GSM voice reached further on the same band | Rural coverage holes where GSM carried the only service |
| 1800 MHz GSM → LTE 1800 | Several dB worse at the edge | Indoor coverage loss in dense urban areas |
| 2100 MHz UMTS → LTE 2100 | Neutral to slightly better | Capacity gain is small if the block is narrow |
| 900/1800 → NR at 700 MHz | Better, but coverage is limited by the new site grid | New band needs its own antenna and radio investment |
| 2600 TDD → NR | Worse at the edge; TDD uplink duty cycle | Uplink-limited cell edge in dense urban |
How do you prove the replacement layer is ready?
Compare like with like, in the same places, for the same service. The test is not "does LTE cover the area" but "in the locations where the old layer delivered an acceptable service, does the new layer deliver an equivalent or better one". That means measuring accessibility, retainability and a minimum throughput at the specific locations that mattered — not aggregate coverage percentage.
The practical method is a paired comparison: select the worst-decile locations of the old layer, then measure the new layer there. If the new layer fails in a meaningful fraction of those locations, the sunset date is premature regardless of what the aggregate map shows.
What does a guard band actually cost?
Refarming rarely gives you the whole block. Adjacent operators, legacy services and regulatory emission limits impose guard bands, and a 1 MHz guard on each side of a 10 MHz block removes 20% of the capacity before any traffic is carried. On narrow blocks the effect is severe: a 5 MHz refarmed block with 0.5 MHz guards loses a fifth of its spectral resource.
Guard bands also appear inside the operator's own spectrum when two technologies run adjacent to each other during the transition. Coexistence filters, lower order modulation at the block edge and reduced power on edge resource blocks all consume capacity that the planning spreadsheet usually ignores. Budget them explicitly, or the capacity gain from refarming will be smaller than the business case assumed.
How do you sequence a multi-band sunset?
Sequence by coverage criticality, not by spectrum value. The band that carries the most rural coverage should be refarmed last, after the replacement layer has been proven in those exact areas. The band that carries the most urban capacity can be refarmed first, because the urban grid is dense enough that the replacement layer's smaller footprint still overlaps.
Then sequence the physical work. Antenna and radio swaps take longer than parameter changes, and each swap is a site visit. Grouping swaps by site rather than by band reduces the number of visits, which is usually the largest controllable cost in the transition. Finally, keep the old layer's parameter data and measurement baselines: they are the only record of what the network looked like before, and they are what you will compare against when the first complaints arrive.