Spectrum: FR1, FR2, Bands & Duplexing in 5G NR

Introduction

Spectrum is the raw material of a radio network — the specific slices of the electromagnetic spectrum an operator is licensed to transmit in. Everything else in NR is built on top of the answer to "which frequencies do I have?" 3GPP organises that raw material top-down: two broad frequency ranges (FR1 and FR2), each subdivided into licensed operating bands (the n-numbers), and within each band the discrete raster positions where a carrier and its sync signal are actually allowed to sit.

This matters at every stage of the UE lifecycle. At power-on the UE must know which bands to scan and which raster positions can hold an SSB, or blind cell search would take forever. When a carrier is configured, the band fixes the allowed subcarrier spacings and channel bandwidths, which in turn fix how many resource blocks the UE gets. And when the network plans capacity and coverage, the choice of band — low vs mid vs mmWave — and duplex mode determines reach, throughput, and how hard it must lean on beamforming.

The defining spec for all of this is TS 38.101 (UE RF: TS 38.101-1 for FR1, TS 38.101-2 for FR2) with base-station RF in TS 38.104. Read spectrum as a hierarchy — range → band → channel → raster — and every parameter on this page has a clear home in that tree.

Why NR splits spectrum this way

Concretely, three problems force this structure. First, physics changes with frequency: low bands propagate far but carry little bandwidth, high bands carry huge bandwidth but barely reach and are easily blocked — so NR needs different numerologies and duplexing rules for each, which is exactly what the FR1/FR2 split encodes. Second, spectrum is licensed in fixed blocks by regulators, so 3GPP must name those blocks (the operating bands) and pin each one's uplink/downlink ranges and duplex mode. Third, a UE cannot tune to arbitrary frequencies and cannot afford to search the whole range for a cell — so NR defines a fine frequency raster to place carriers and a much sparser synchronization raster to place sync beacons, making blind cell search tractable.

FR1 and FR2 — the two worlds of NR

All of NR's spectrum (TS 38.101) is divided into two frequency ranges. They behave very differently, so it is worth carrying a clear mental picture of each before anything else.

The dividing philosophy matters. Everything in FR1 is designed around covering area with a modest slice of spectrum, so subcarrier spacings stay small and channels top out at 100 MHz. Everything in FR2 assumes you have bandwidth to burn but a hostile channel — high phase noise, severe blockage, large Doppler — so it uses wide subcarriers, wide channels, and swept narrow beams. Notice the empty stretch between them: nothing operates in the 7.125–24.25 GHz gap in NR today, which is why the two ranges feel like separate universes.

Operating bands: the n-numbers

Within each frequency range, spectrum is licensed in chunks called operating bands. NR bands are named n<number> — for example n1, n3, n78. Each band definition (TS 38.101-1 for FR1, TS 38.101-2 for FR2) fixes an uplink range, a downlink range, and a duplex mode. Many NR band numbers deliberately echo their LTE cousins (LTE band 3 → NR n3) because they occupy the same spectrum and simplify re-farming.

A few reading habits pay off. Bands numbered up to the low hundreds are FR1; the n257/n258/n259/n260/n261 family is FR2 mmWave. A paired band lists two frequency ranges (uplink and downlink); a TDD band lists one range shared by both directions. And a band's number tells you nothing about how wide it is — n78 is 500 MHz of TDD spectrum, while an FDD band like n1 is only 60 MHz per direction.

Duplexing: FDD, TDD, SDL and SUL

Duplexing is how uplink and downlink share the spectrum. NR supports the two classic schemes plus two "supplementary" one-directional modes that exist to patch specific coverage or capacity gaps.

The two supplementary modes are easy to misread, so anchor them to a picture. A normal TDD mid-band cell like n78 has plenty of downlink reach — the base station is powerful — but the handset's uplink is weak and runs out of range first. SUL hands that struggling UE a second uplink carrier down at, say, 700 MHz or 1.5 GHz, where its transmit power carries much farther. The UE keeps downlinking on n78 but uplinks on the SUL, so the coverage-limited direction gets a low-band boost. The two uplinks are mutually exclusive at any instant — the UE transmits on the normal UL or the SUL, selected against a broadcast RSRP threshold (rsrp-ThresholdSSB-SUL). SDL is the mirror-image capacity play: a leftover unpaired block that can only ever carry downlink, bolted on to soak up the DL-heavy traffic of modern networks.

Subcarrier spacing, channel bandwidth and PRB count

NR is flexible about numerology: the subcarrier spacing (SCS) and the channel bandwidth both depend on which frequency range — and often which band — you are in (TS 38.104). The SCS is always 2^μ × 15 kHz for numerology μ = 0…4, giving 15, 30, 60, 120 and 240 kHz. Higher SCS means each subcarrier is wider, symbols are shorter, and you can support wider channels and shorter latency, which is exactly what the higher ranges demand.

Which SCS a specific band allows is fixed per band in TS 38.104 — a low FDD band like n3 supports only 15 and 30 kHz, whereas n78 is a 30/60 kHz band, and FR2 bands center on 120 kHz. The point of numerology is that the same radio grid stretches to fit the channel: pick the SCS the band allows, and the number of resource blocks you get follows. Note the SSB SCS is signalled independently of the data SCS via subCarrierSpacingCommon in the MIB.

The number of resource blocks (equivalently PRBs) you actually receive is a function of bandwidth divided by SCS, because each RB is always 12 subcarriers. Wider SCS packs fewer RBs into the same MHz, but each RB is physically wider.

The exact maxima are tabulated as the transmission bandwidth configuration N_RB in TS 38.101-1 (FR1) and TS 38.101-2 (FR2). Here is the core of that table — the numbers worth memorising:

Read the dashes as "not allowed": a 100 MHz channel is unreachable at 15 kHz because it would need far more RBs than the transmission-bandwidth limit permits, and the 200/400 MHz columns only exist at 120 kHz in FR2. The headline value is 100 MHz at 30 kHz SCS = 273 PRB — the canonical FR1 mid-band carrier. Move to 60 kHz and the same 100 MHz gives only 135 PRB: identical spectrum, wider subcarriers, fewer but larger blocks. Note the maximum 273 PRB is also the largest carrier NR defines at any SCS, so it caps the FFT size the UE must implement.

This is why 30 kHz is the FR1 mid-band workhorse: it packs enough PRBs into a 100 MHz channel for high throughput while keeping the symbol short enough to tolerate the delay spread and Doppler seen at 3.5 GHz. In FR2 the same logic scales up — 120 kHz unlocks the 400 MHz channels, and the 480/960 kHz spacings introduced for FR2-2 are what make GHz-class carriers practical against mmWave phase noise.

Rasters: ARFCN and the SS raster (GSCN)

Spectrum is continuous, but a UE cannot tune to arbitrary real numbers — it needs a discrete set of allowed frequencies. NR defines two rasters for this. The global frequency raster pins where a carrier's center can sit, addressed by an ARFCN (NREF); the sparser synchronization raster pins where an SSB can sit, addressed by a GSCN.

Global frequency raster and ARFCN

Every RF reference frequency FREF maps to an integer channel number NREF (the ARFCN) via a step size ΔFGlobal that grows with frequency (TS 38.104 §5.4.2):

So the raster is fine (5 kHz) in the low bands where channels are narrow, and coarse (60 kHz) up in mmWave where they are enormous. A carrier is signalled to the UE by its ARFCN, and the SIB / RRC field that points at it (for example absoluteFrequencyPointA or absoluteFrequencySSB) carries an NREF value the UE converts back to Hz with the formula above. Note that the applicable ARFCN raster within any given band is further narrowed by a band-specific ΔFRaster in TS 38.104, so not every global-raster point is a legal carrier centre in every band.

The SS raster and GSCN

Searching every ARFCN for a beacon would be brutally slow, so SSBs are only allowed on a much sparser synchronization raster. Each allowed position is a Global Synchronization Channel Number (GSCN), and the UE only checks these "parking spots" during blind cell search (TS 38.104 §5.4.3).

The step jumps from 1.2 MHz-ish in the low range to 1.44 MHz in the mid range to 17.28 MHz in mmWave — again, coarser where channels are wider. Because the SS raster is so much sparser than the frequency raster, the cell's SSB almost never sits exactly at the carrier center; the offset between them is signalled to the UE (as offsetToPointA plus the sub-carrier-level k_SSB from the MIB), which is how the UE bridges from "I found a beacon on this GSCN" to "here is the full resource grid". To shrink the search further, each band lists only a handful of sync raster entries (a GSCN range) in TS 38.104, so the UE scans tens of positions, not thousands.

Guard bands and minimum guardband

A channel's nominal bandwidth is never fully filled with usable subcarriers. Some spectrum at each edge is deliberately left empty as a guard band so the carrier's out-of-band emissions decay before they reach the neighbour's spectrum. That is exactly why the N_RB counts above are a little less than "bandwidth ÷ (12 × SCS)" would suggest — the difference is guard band.

The relationship, in words, is: occupied bandwidth = N_RB × 12 × SCS, and the leftover splits into two edge guards. For the flagship case, 273 PRB at 30 kHz occupy 273 × 12 × 30 kHz = 98.28 MHz inside a 100 MHz channel, leaving about 1.72 MHz total, or roughly 0.845 MHz per side. TS 38.101-1 tabulates this as the minimum guardband per channel bandwidth and SCS.

Two things fall out of this. First, wider SCS needs a larger absolute guardband (its subcarriers spill wider), which is another reason the high-SCS configurations lose PRBs. Second, the guardband is a minimum — an operator may leave more, but never less, or the carrier would fail its adjacent-channel leakage requirements. When two carriers are aggregated inside one band, a similar inter-carrier guard keeps them from interfering. The related metric to watch is the occupied bandwidth ratio (occupied ÷ channel BW), around 98.28% for the 273-PRB / 100 MHz case — high spectral efficiency versus LTE's ~90%.

Unlicensed NR: NR-U and the 5/6 GHz bands

Not all NR runs on licensed spectrum. NR-U (NR-based access to unlicensed spectrum, introduced in Release 16) lets NR operate in the shared 5 GHz and 6 GHz bands, either anchored to a licensed carrier (License-Assisted Access) or fully standalone.

Because this spectrum is shared with Wi-Fi and other users, NR-U must play fair. It uses Listen-Before-Talk (LBT) — the radio senses the channel (an energy-detect clear-channel assessment) and only transmits if it is clear — plus occupied-channel-bandwidth rules and power limits set by regional regulators. To fit LBT's channel-access granularity, NR-U adds wideband operation split into 20 MHz LBT sub-bands and features like the Channel Occupancy Time (COT) and configured-grant enhancements. Note that n96 reaching to 7125 MHz is precisely why the FR1 ceiling was defined at 7.125 GHz rather than a round "6 GHz". For an operator, NR-U is a cheap capacity top-up: no license fee, but no guaranteed access either, so it is best used as an opportunistic downlink-heavy booster over a licensed anchor.

Summary

Hold spectrum as a hierarchy. At the top, two frequency ranges: FR1 (410 MHz–7.125 GHz, coverage + mid-band capacity, SCS 15/30/60 kHz, up to 100 MHz) and FR2 (24.25–71 GHz mmWave, SCS 60/120/480/960 kHz, up to 2 GHz) with an empty gap between them. Inside each range sit licensed operating bands (n-numbers), each with fixed UL/DL ranges and a duplex mode — FDD on low paired bands, TDD dominant in mid-band and all of FR2, plus the supplementary SUL and SDL carriers.

Inside a band you place carriers whose PRB count follows from bandwidth ÷ (12 × SCS), minus guard band — the canonical result being 273 PRB for 100 MHz at 30 kHz. Two rasters discretise placement: the fine global frequency raster (ARFCN/NREF, step ΔFGlobal) for carrier centres, and the sparse synchronization raster (GSCN) for where an SSB may sit so blind cell search stays fast.

Everything traces back to TS 38.101-1/-2 and TS 38.104. Know the range boundaries, the 273-PRB headline, the duplex logic, and the two-raster split, and the rest of the spectrum picture assembles around them.