What Is 5G NR? Numerology, Frequency Ranges, and the Flexible Frame
5G NR explained for engineers: scalable numerology, FR1 and FR2, the flexible slot structure, bandwidth parts, and what actually changed from LTE.
If you already know LTE, the fastest way to understand 5G NR is through one design decision: LTE fixed almost everything, NR made almost everything a parameter.
LTE gave you 15 kHz subcarrier spacing, a 1 ms subframe, and a 20 MHz maximum carrier. Those numbers were baked in. NR replaced each with a scalable family — and nearly every other NR feature follows from that choice.
NR (New Radio) is the 3GPP radio access technology for 5G, defined from Release 15 onward. It's the air interface only — not the core, not the architecture. A UE talks NR to a gNB; what happens behind the gNB is a separate question.
Numerology: the idea everything hangs off
NR defines subcarrier spacing as a scalable family indexed by µ:
Δf = 2^µ × 15 kHz
That gives five values, each with its own slot duration:
| µ | Subcarrier spacing | Slot duration | Typical use |
|---|---|---|---|
| 0 | 15 kHz | 1 ms | FR1, wide-area coverage |
| 1 | 30 kHz | 500 µs | FR1, the mainstream mid-band choice |
| 2 | 60 kHz | 250 µs | FR1/FR2, low-latency services |
| 3 | 120 kHz | 125 µs | FR2 mmWave |
| 4 | 240 kHz | 62.5 µs | FR2, synchronisation signals only |
A slot is always 14 OFDM symbols. Doubling the subcarrier spacing halves the symbol duration, which halves the slot duration. That's the entire mechanism.
The trade-off you can't escape
Wider subcarrier spacing is not simply better. It buys and costs specific things:
Wider spacing gives shorter slots — meaning lower latency — and much better resistance to Doppler and phase noise, which is why it's mandatory at mmWave. It costs you cyclic prefix duration, which shrinks proportionally, so the system tolerates less delay spread.
Narrower spacing gives a longer cyclic prefix, which survives the large delay spreads of wide-area cells and heavy multipath. It costs latency and Doppler tolerance.
That's why 15 kHz survives for low-band wide-area coverage, 30 kHz dominates mid-band deployments, and 120 kHz is the practical choice above 24 GHz. The physics picks the numerology, not the marketing.
FR1 and FR2
NR splits its spectrum into two frequency ranges with genuinely different engineering characteristics.
FR1 spans roughly 410 MHz to 7.125 GHz — the sub-7 GHz range. Channel bandwidths run from 5 to 100 MHz. This is where most commercial 5G capacity actually lives, particularly the n78 band around 3.5 GHz.
FR2 covers roughly 24.25 to 71 GHz — mmWave. Channel bandwidths run from 50 to 400 MHz per carrier.
The difference isn't just numbers. FR2 propagation is severe enough that beamforming stops being an optimisation and becomes a precondition — an FR2 cell without beamforming has essentially no usable range. FR2 also makes blockage a first-class problem: a hand, a body, or a passing vehicle can drop the link entirely, which is why beam failure recovery exists as a specified procedure.
FR1 gives you coverage and building penetration. FR2 gives you bandwidth in exchange for a much harder radio problem. Most networks use FR1 as the backbone and FR2 for dense capacity hotspots, if at all.
The flexible frame
NR keeps LTE's 10 ms radio frame and 1 ms subframe as fixed anchors. Everything below that scales with numerology.
The consequential change is at symbol level. In LTE, a subframe was downlink or uplink and that was that. NR allows each symbol in a slot to be designated downlink, uplink, or flexible.
This enables three things LTE couldn't do:
Configurable TDD patterns — an operator picks a downlink/uplink ratio matching actual traffic asymmetry, rather than choosing from seven fixed LTE configurations.
Self-contained slots — data and its acknowledgement can occupy the same slot, collapsing the HARQ round trip dramatically.
Mini-slots — transmission can start at any symbol and last 2, 4, or 7 symbols rather than waiting for a slot boundary. This is the mechanism behind NR's latency claims, and the reason URLLC is feasible at all.
A URLLC packet arriving mid-slot doesn't wait. It gets a mini-slot, and can preempt an ongoing eMBB transmission if the scheduler decides the latency budget requires it.
Bandwidth parts
This one has no LTE equivalent and it's frequently misunderstood.
A Bandwidth Part (BWP) is a contiguous subset of a carrier's resource blocks, with its own numerology, that a UE is configured to operate within. A UE can be configured with up to four BWPs per direction, but only one is active at a time.
Two problems it solves:
Device power. Monitoring a full 100 MHz carrier is expensive. A UE with light traffic can operate in a 20 MHz BWP and switch to the wide one only when there's data worth the power. RF front-end and baseband both scale down.
Numerology coexistence. Different BWPs can use different subcarrier spacings on the same carrier — a 30 kHz BWP for eMBB traffic and a 60 kHz BWP for latency-sensitive traffic, simultaneously.
BWPs are also what let a device with narrower RF capability operate on a wide carrier at all — the foundation RedCap builds on.
The reference signals did the real work
The single biggest structural change from LTE is easy to miss because it is a subtraction rather than an addition: NR has no always-on cell-specific reference signal.
LTE transmitted CRS continuously across the whole band in every subframe, whether or not anyone was connected. Devices used it for demodulation, measurement and cell detection, which made it convenient — and it meant an idle LTE cell still radiated constantly, burning power and interfering with its neighbours.
NR replaces this with signals that are transmitted only when needed and only where needed:
DM-RS is transmitted alongside a scheduled transmission and precoded identically, so the device estimates the effective channel including the precoder without ever being told what the precoder was.
CSI-RS is configured per device for channel measurement, transmitted periodically, semi-persistently, or aperiodically on demand.
SRS is the uplink sounding signal, letting the gNB estimate the uplink channel and — in TDD, via reciprocity — the downlink channel too.
PT-RS exists purely for high frequencies, tracking the phase noise that becomes significant in mmWave oscillators.
The consequence is lean carrier operation: an NR cell with no active users transmits almost nothing beyond its SSB. This is the foundation for network energy saving, and it is why NR cells can be switched to very low duty cycles in ways LTE never could.
The SSB and initial access
What remains always-on is the SS/PCH Block — primary and secondary synchronisation signals plus the PBCH — transmitted in bursts, typically every 20 ms.
In FR2 the SSB is transmitted as a burst set of up to 64 beams swept in sequence. The device measures each, identifies the best, and derives its random access resources from that index. This is how beam selection happens before any connection exists: the device tells the network which beam it heard by choosing the PRACH occasion associated with it.
That mechanism is the reason initial access in mmWave works at all, and it is a genuine architectural difference from LTE rather than a parameter change.
What the protocol stack looks like
The user-plane stack is LTE's with one addition:
SDAP is new in NR. It maps QoS Flows from the 5G Core onto Data Radio Bearers. LTE had no equivalent because LTE's QoS model was bearer-based end to end.
PDCP handles ciphering, integrity protection, header compression, reordering, and duplicate detection.
RLC provides segmentation and, in AM mode, ARQ. NR removed LTE's concatenation function — segmentation happens per transmission opportunity, which reduces processing latency.
MAC handles multiplexing, HARQ, scheduling, and random access.
PHY handles coding, modulation, and the physical channels.
The control plane adds RRC and NAS above PDCP. Note where the split lands in a disaggregated gNB: RRC, PDCP, and SDAP sit in the CU; RLC, MAC, and PHY in the DU.
Coding
NR made a clean break here. LDPC replaced turbo codes for data channels — it parallelises better at high throughput and has a lower error floor. Polar codes replaced convolutional coding for control channels, where blocks are short and reliability matters more than throughput.
Deployment: SA and NSA
NR can be deployed two ways, and the distinction matters more than people expect.
Standalone (SA) — gNB connects to a 5G Core. Full 5G feature set: network slicing, RedCap, URLLC with a 5G QoS model, service-based architecture.
Non-Standalone (NSA) — an LTE eNB anchors the control plane while the gNB adds NR capacity, both connected to the LTE EPC. Faster and cheaper to deploy, but you get NR throughput without the 5G Core features.
Most operators launched NSA and are migrating to SA. Worth being clear about what that means practically: an NSA network cannot do network slicing, because slicing lives in the 5G Core. A lot of 5G capability claims quietly assume SA.
The NSA vs SA article covers the deployment options in detail.
What actually changed from LTE
| LTE | 5G NR | |
|---|---|---|
| Subcarrier spacing | 15 kHz fixed | 15–240 kHz scalable |
| Max carrier bandwidth | 20 MHz | 100 MHz (FR1), 400 MHz (FR2) |
| Scheduling granularity | 1 ms subframe | Slot or mini-slot (2–7 symbols) |
| TDD configuration | 7 fixed patterns | Symbol-level flexibility |
| Data channel coding | Turbo | LDPC |
| Control channel coding | Convolutional | Polar |
| Frequency range | Sub-6 GHz | Sub-7 GHz and mmWave |
| Beamforming | Optional enhancement | Fundamental, mandatory in FR2 |
| Bandwidth adaptation | None | Bandwidth parts |
| Always-on signals | CRS across the carrier | Lean design, minimal always-on |
That last row deserves a mention because it's a genuine design improvement. LTE transmitted cell-specific reference signals continuously across the whole carrier whether or not anyone was connected. NR's lean carrier design minimises always-on transmission, which cuts both energy consumption and inter-cell interference. It's a quiet change with real operational impact.
Where the latency actually went
"5G is low latency" is repeated so often that the mechanism behind it gets lost. The reduction is not one change but four, and only some of them are radio changes at all.
Shorter slots. Higher numerology shrinks the slot. At 30 kHz a slot is 0.5 ms; at 120 kHz it is 0.125 ms. Every procedure timed in slots inherits the reduction.
Mini-slots. A transmission need not wait for a slot boundary. Two, four or seven symbols can be scheduled starting at any symbol, which removes the alignment delay that dominates short transfers. This matters more than raw slot length for latency-sensitive traffic.
Self-contained slots. A slot can carry downlink data and its acknowledgement, or an uplink grant and the resulting transmission. LTE's fixed 4 ms HARQ timing becomes a configurable value negotiated per transmission, and the round trip collapses.
Front-loaded DM-RS. The demodulation reference signal sits at the start of the transmission rather than being spread through it, so the receiver can begin channel estimation immediately rather than waiting for the whole slot to arrive. This is a decoding-latency change, invisible in the frame structure but significant in silicon.
Above the radio, edge deployment removes propagation and transport delay by shortening the path to the application. That is an architectural choice, not an NR feature, and it is often the largest single contributor to an advertised latency figure.
The honest summary: air-interface latency improved by roughly an order of magnitude for the traffic that can exploit mini-slots. End-to-end latency improved by whatever the deployment allows, which varies enormously.
Uplink waveform: the choice LTE did not offer
LTE used SC-FDMA in the uplink exclusively, for one reason: a low peak-to-average power ratio, which lets a handset's power amplifier operate efficiently and reach further.
NR keeps that option — DFT-s-OFDM — but adds CP-OFDM as an alternative, and lets the network switch between them per device.
The trade is straightforward. CP-OFDM supports spatial multiplexing properly and uses the same processing chain as the downlink, so it delivers higher throughput near the cell centre. DFT-s-OFDM has a PAPR several dB lower, so it reaches further at the cell edge where the amplifier is the binding constraint.
Networks therefore configure CP-OFDM for devices in good coverage and switch to DFT-s-OFDM as they move out. The device supports both; the choice is the network's, signalled in the configuration.
This is a small change with an outsized effect on real coverage maps, and it is one of the clearest examples of NR's design philosophy: where LTE picked one option for everyone, NR parameterises the choice and lets the network decide per situation.
The mental model
Numerology (µ = 0–4) scales subcarrier spacing and slot duration together. Everything else follows.
FR1 is coverage and capacity. FR2 is bandwidth at the cost of a much harder radio problem.
Symbol-level TDD flexibility plus mini-slots is where NR's latency advantage actually comes from.
Bandwidth parts decouple device capability from carrier width.
NR is the radio. SA versus NSA determines which 5G features you actually get.
NR is best understood as LTE with the constants replaced by parameters. The OFDM foundations are the same. What changed is that the system can now be configured for a 700 MHz rural cell and a 28 GHz stadium small cell using the same specification — and that flexibility is both the achievement and the source of most of the complexity.
Further reading
- 3GPP TS 38.211 — Physical channels and modulation
- 3GPP TS 38.300 — NR and NG-RAN overall description
- 3GPP TS 38.104 — Base station radio transmission and reception, FR1 and FR2 definitions
- 3GPP TS 38.331 — RRC protocol specification
Related Articles
Non-Terrestrial Networks: 3GPP Satellites, Delay Budgets, and Direct-to-Device
How 3GPP adapted NR for satellites: LEO/MEO/GEO trade-offs, the timing advance and HARQ changes long delays forced, transparent versus regenerative payloads, and what direct-to-device really delivers.
5G Interview Questions: What Gets Asked and How to Answer It
A structured 5G interview guide: the questions that actually come up across radio, RAN, core and QoS, worked troubleshooting and system-design scenarios, and a framework for answering well.
5G Scheduling Algorithms: Round Robin, Proportional Fair, and What Real Schedulers Do
How the gNB MAC scheduler decides who transmits: the classic algorithms, why Proportional Fair became the baseline, and how QoS-aware scheduling handles URLLC and eMBB together.