5G NR Numerology: The One Parameter That Shapes the Whole Air Interface
Why 5G NR has five subcarrier spacings instead of LTE's one, what each value buys you, and how μ ripples through symbol, slot and frame timing.
Why this is the first thing to learn
Almost every number in the NR physical layer — symbol length, slot length, cyclic prefix, how many resource blocks fit in a carrier, how often the UE has to blind-decode — is derived from one integer, μ (mu). Learn how μ works and half of this chapter becomes arithmetic.
The mental model: gears on a bicycle
OFDM splits the band into many narrow subcarriers. The subcarrier spacing (SCS) is the distance between them, and it sets the OFDM symbol length: a subcarrier of spacing Δf needs a symbol of length 1/Δf to stay orthogonal to its neighbours.
Think of the SCS as a gear:
- Low gear (15 kHz) — long symbols (66.7 µs). Lots of "torque" against multipath: the cyclic prefix can be long, so the signal survives big cells and long echoes. But everything moves slowly: a slot is a full millisecond.
- High gear (120 kHz) — short symbols (8.3 µs). Fast: a slot is 125 µs, so latency is low, and the short symbol shrugs off the phase noise and Doppler that grow with carrier frequency. But the cyclic prefix is short, so it only works where echoes are short — small cells, mmWave.
LTE had one gear (15 kHz) because it lived in a narrow range of frequencies and cell sizes. NR runs from 410 MHz to 52.6 GHz (and beyond), from rural macro cells to indoor mmWave hot-spots. No single gear fits, so NR gives you five, and — the clever part — makes them mesh.
The mechanics
Subcarrier spacing is defined as
Δf = 2^μ × 15 kHz , μ ∈ {0, 1, 2, 3, 4}
Everything else follows (normal cyclic prefix):
| μ | SCS | Useful symbol (1/Δf) | Slot length | Slots per subframe | Slots per 10 ms frame | Where it is used |
|---|---|---|---|---|---|---|
| 0 | 15 kHz | 66.67 µs | 1 ms | 1 | 10 | FR1: low bands, LTE refarming, large cells |
| 1 | 30 kHz | 33.33 µs | 0.5 ms | 2 | 20 | FR1: the mid-band workhorse (n77/n78/n41) |
| 2 | 60 kHz | 16.67 µs | 0.25 ms | 4 | 40 | FR1 (URLLC, only SCS with extended CP) and FR2 data |
| 3 | 120 kHz | 8.33 µs | 0.125 ms | 8 | 80 | FR2: mmWave data and SSB |
| 4 | 240 kHz | 4.17 µs | 0.0625 ms | 16 | 160 | FR2: SSB only, never for data |
Two invariants make this table easy to reconstruct:
- Symbols per slot is always 14 (normal CP). So as the symbol halves, the slot halves.
- The subframe (1 ms) and frame (10 ms) never change. So the number of slots per subframe is simply 2^μ.
Which SCS is allowed where (TS 38.104 / 38.101): FR1 data uses 15, 30 or 60 kHz; FR1 SSB uses 15 or 30 kHz. FR2 data uses 60 or 120 kHz; FR2 SSB uses 120 or 240 kHz. Rel-17 added μ = 5 and 6 (480 and 960 kHz) for the 52.6–71 GHz range (FR2-2); treat those as 5G-Advanced extras.
Why powers of two — the meshing gears
The factor 2^μ is not just tidy. It guarantees that symbol boundaries line up across numerologies: every 15 kHz symbol boundary coincides with every second 30 kHz boundary, every fourth 60 kHz boundary, and so on. That is what allows:
- Mixed numerologies on one carrier (different bandwidth parts with different SCS) without the two grids drifting apart.
- A 15 kHz SSB on a 30 kHz data carrier — the UE can translate between the two grids with a fixed offset.
- Scheduling across numerologies (a PDCCH at one SCS scheduling a PDSCH at another) with clean slot alignment.
If the spacings were, say, 15 / 20 / 25 kHz, none of this would work.
What each choice really trades
| You want… | Push μ… | Because… | But… |
|---|---|---|---|
| Lower latency | up | slots and mini-slots get shorter; HARQ turns around faster | the CP shrinks, so the cell must be small |
| Bigger cells / long echoes | down | the CP (≈ 7 % of the symbol) grows with the symbol | slots get longer, latency rises |
| Robustness to phase noise and Doppler (high carrier, high speed) | up | a wider subcarrier tolerates more frequency error before inter-carrier interference bites | same CP penalty |
| Wide channel bandwidth with a manageable FFT | up | fewer subcarriers per MHz | fewer resource blocks per MHz, coarser frequency-domain scheduling |
| Fine frequency-domain scheduling granularity | down | more, narrower RBs per MHz | — |
The design rule of thumb that falls out: choose the smallest SCS whose CP still covers the delay spread and whose subcarrier still tolerates the phase noise. That lands on 15/30 kHz in FR1 and 120 kHz in FR2.
A worked example
An operator has 100 MHz in n78 (3.5 GHz). Options: 30 kHz gives 273 RBs (98.28 MHz used), 0.5 ms slots and a 2.34 µs CP — enough for typical urban delay spread. 15 kHz cannot fill a 100 MHz carrier (max 50 MHz at 15 kHz, because the FFT would exceed 4096 points), and 60 kHz would shorten the CP to 1.17 µs, too short for macro cells. So the answer is 30 kHz — which is why almost every mid-band NR network on earth runs μ = 1.
Common confusions
- "Higher SCS means more capacity." No. Capacity per MHz is essentially the same at every SCS; what changes is latency, robustness and CP overhead.
- "240 kHz is a data numerology." It is defined only for the SS/PBCH block in FR2.
- "The symbol is 66.7 µs at 15 kHz." That is the useful part. The transmitted symbol is 66.7 + 4.7 = 71.4 µs with the normal CP (see the cyclic prefix article).
- "Numerology = bandwidth part." A BWP has a numerology; the numerology is the parameter, the BWP is the container.
- "You pick μ per cell." You pick it per BWP; one cell can run several.
Spec pointers
- TS 38.211 §4.2 — supported numerologies (Table 4.2-1), cyclic prefix per μ
- TS 38.211 §4.3 — frame and slot structure per μ
- TS 38.104 §5.3 / TS 38.101-1, -2 §5.3 — allowed SCS per band and channel bandwidth
- TS 38.213 §4.1 — SSB SCS per frequency range
Recap
Δf = 2^μ × 15 kHz. Symbols per slot stay at 14, the subframe stays at 1 ms, so slot length = 1/2^μ ms. Powers of two make the grids mesh, which lets one carrier host several numerologies. Small μ for coverage, large μ for latency and high carrier frequencies; 30 kHz in FR1 and 120 kHz in FR2 are the practical sweet spots.
Interview questions
The questions below are drawn from this topic and phrased the way they tend to come up. Try each one out loud before revealing the answer.
18 questions
- The numerology is the integer μ that selects the subcarrier spacing through Δf = 2^μ × 15 kHz. It is a single parameter from which symbol duration, slot length and slots per subframe all follow.
- 15, 30, 60, 120 and 240 kHz, for μ = 0 to 4. Release 17 added 480 and 960 kHz (μ = 5 and 6) for FR2-2 above 52.6 GHz.
- LTE operated over a narrow range of frequencies and cell sizes, so 15 kHz suited everything. NR runs from sub-1 GHz macro cells to mmWave hot-spots, and no single spacing serves both.
- They are reciprocal. A subcarrier spacing of Δf requires a useful symbol length of 1/Δf to keep subcarriers orthogonal, so doubling the spacing halves the symbol.
- Fourteen with normal cyclic prefix, regardless of numerology. Twelve with extended cyclic prefix, which exists only at 60 kHz.
- The slot duration. Since the symbol halves each time μ increases by one, the slot halves too: 1 ms at 15 kHz down to 62.5 µs at 240 kHz.
- 2^μ. The subframe is always 1 ms and the frame always 10 ms, so only the slot count varies.
- So symbol boundaries align across numerologies. Every 15 kHz boundary coincides with every second 30 kHz boundary and so on, which lets different bandwidth parts use different spacings on one carrier without the grids drifting apart.
- FR1 data uses 15, 30 or 60 kHz. FR2 data uses 60 or 120 kHz.
- 15 or 30 kHz in FR1; 120 or 240 kHz in FR2. 240 kHz is SSB-only and never carries data.
- Phase noise and Doppler both scale with carrier frequency. A wider spacing means a shorter symbol and greater tolerance to that frequency error, at the cost of a shorter cyclic prefix.
- The cyclic prefix shrinks in proportion, so the tolerable delay spread falls. High spacings only work where echoes are short, which means small cells.
- No. At 120 kHz the normal cyclic prefix is roughly 0.6 µs, so a 2 µs echo would run past it and destroy orthogonality. 15 or 30 kHz would be appropriate.
- Yes, in different bandwidth parts. The power-of-two relationship keeps their symbol boundaries aligned, though guard handling between them still needs care.
- It is the crossover spacing used in both FR1 and FR2, where delay spread can still be significant relative to the shortened symbol. Extended CP buys robustness at the cost of two symbols per slot.
- The subframe at 1 ms, the radio frame at 10 ms, and the number of symbols per slot. These invariants are what make the tables easy to reconstruct from memory.
- Shorter slots mean shorter transmission opportunities and faster HARQ turnaround, so latency falls roughly in proportion to 1/2^μ. This is why URLLC favours higher spacings.
- Δf = 2^3 × 15 kHz = 120 kHz. Useful symbol = 1/120 kHz = 8.33 µs. Fourteen symbols per slot, and the subframe is 1 ms with 2^3 = 8 slots, so the slot is 125 µs.
Related Articles
5G NR Waveforms: CP-OFDM and DFT-s-OFDM, and When the Uplink Should Use Which
Why NR uses CP-OFDM everywhere but keeps DFT-s-OFDM as an uplink option, what PAPR really costs, and how the network chooses per UE.
5G NR Transmitter and Receiver Chain: Following the Bits From MAC to Antenna and Back
A block-by-block walk through the NR physical-layer transmit chain and its mirror-image receiver, explaining what each stage does and why it sits where it does.
5G NR Synchronisation Signals (PSS and SSS): How a Phone Finds a Cell From Nothing
PSS and SSS explained from the UE's point of view: what each signal gives the receiver, how the PCI is built from them, and why NR has 1008 cell identities.