Everything in the 5G air interface is built on one object: the OFDM symbol, laid out on a grid of time and frequency. Understand how bits become a symbol, how symbols stack into slots and frames, and how numerology scales the whole thing — and the rest of the physical layer stops being a wall of acronyms.
An OFDM transmitter turns a stream of bits into a time-domain waveform in a fixed pipeline, and every stage has a clear job.
1 — Bit to symbol. Bits are grouped and mapped to complex points — amplitude and phase — using QPSK, 16QAM, 64QAM or 256QAM. More points per symbol carries more bits, but needs a cleaner channel. 2 — Resource grid. Those points are placed on a two-dimensional grid: subcarriers up the frequency axis, OFDM symbols along time. 3 — IFFT. This is the actual conversion: an inverse FFT turns the parallel frequency-domain subcarriers into one combined time-domain signal. 4 — Cyclic prefix and filtering. A copy of the symbol’s tail is prepended as a guard interval, then the signal is filtered or windowed before the DAC and the RF chain.
Radio signals arrive by many paths — direct, and bounced off buildings and terrain — so a symbol smears into the one after it. That is inter-symbol interference (ISI). The cyclic prefix (CP) is a copy of the end of the symbol glued to its front, forming a guard interval: as long as the worst echo arrives within the CP, the receiver can discard the guard and recover a clean symbol. It costs a little airtime, and it is why delay spread — how long the echoes last — sets a floor on symbol length.
NR nests time in a fixed hierarchy. A radio frame is 10 ms, split into ten subframes of 1 ms each. A subframe holds one or more slots, and every slot carries 14 OFDM symbols (with the normal cyclic prefix). The subframe is the anchor that stays 1 ms no matter what; the number of slots inside it is what changes.
LTE fixed the subcarrier spacing at 15 kHz. NR makes it scalable through the numerology μ: the spacing is 15 kHz × 2μ, and because symbol duration is inversely proportional to spacing, each step doubles the subcarrier spacing and halves the symbol — so twice as many slots fit in the same 1 ms subframe.
| μ | Subcarrier spacing | Useful symbol Tu | Slots / subframe | Main use |
|---|---|---|---|---|
| 0 | 15 kHz | ~66.7 µs | 1 | FR1 sub-6 GHz, standard coverage |
| 1 | 30 kHz | ~33.3 µs | 2 | FR1 mid-band — the workhorse |
| 2 | 60 kHz | ~16.7 µs | 4 | FR1 / FR2, URLLC low latency |
| 3 | 120 kHz | ~8.33 µs | 8 | FR2 mmWave, high throughput |
| 4 | 240 kHz | ~4.17 µs | 16 | FR2, SSB / synchronisation only |
The point is one framework serving very different needs. Wide spacing (short symbols) means low latency and tolerance of the phase noise and Doppler of mmWave; narrow spacing means longer symbols, better coverage and less overhead. The scheduler picks the numerology that fits the band and the service.
Uplink and downlink have to be kept apart, and there are two ways to do it. FDD (frequency-division duplex) uses paired spectrum: a downlink band and an uplink band that both run continuously, at the same time, on different frequencies. It is simple and gives symmetric, low-latency links, but it needs a paired allocation and a duplexer.
TDD (time-division duplex) uses a single, unpaired band and splits it in time — some symbols or slots are downlink, some uplink, some flexible. It fits the single large channels available in mid-band and mmWave, and it lets the network shift the ratio toward whichever direction has traffic. Most 5G deployments are TDD, and it is a large part of why NR made HARQ asynchronous and scheduling dynamic — a fixed timing relationship cannot survive a slot whose direction can change.
The frame structure is not just theory — it sets the clock the whole modem runs against. The FFT size and rate follow from the numerology and bandwidth; the slot length is the latency budget the scheduler, HARQ and decoder all have to finish inside; and TDD switching points drive the RF front-end timing. Choose a numerology and you have chosen the sample rates, the buffer sizes and the timing margins the baseband has to hit, every slot, without fail.
The basic unit of the 5G air interface: a set of data modulated onto many closely spaced subcarriers and converted to a time-domain waveform by an IFFT, with a cyclic prefix added as a guard. A slot carries 14 of them under the normal cyclic prefix.
A scaling factor μ that sets the subcarrier spacing at 15 kHz × 2^μ (15, 30, 60, 120, 240 kHz). Higher numerology means wider spacing, shorter symbols, lower latency and more slots per millisecond — used for mmWave and URLLC; lower numerology favours coverage.
To absorb multipath. A copy of the symbol’s tail is prepended as a guard interval, so echoes that arrive within that guard do not bleed into the next symbol. It removes inter-symbol interference at the cost of a little airtime.
FDD separates uplink and downlink by frequency, using paired bands that both run continuously. TDD uses one unpaired band and separates the directions in time. Most 5G is TDD, which suits the large single channels in mid-band and mmWave and lets the uplink/downlink ratio follow the traffic.
2^μ. A subframe is always 1 ms, so numerology 0 (15 kHz) has 1 slot, numerology 1 (30 kHz) has 2, numerology 2 has 4, and so on — each step doubles the slot count as the symbol duration halves.
Fourteen, with the normal cyclic prefix. (An extended cyclic prefix, used only at 60 kHz spacing, gives twelve.) This holds regardless of numerology — what changes with numerology is how long each symbol, and therefore each slot, lasts.
What rides on the grid: PDSCH, PUSCH and more.
KNOWLEDGEWhy the time-split needs asynchronous HARQ.
KNOWLEDGEThe FEC that protects each block.
KNOWLEDGEThe synchronisation signals on the grid.
TECHNOLOGYThe OFDM chain in silicon.
KNOWLEDGEWhere TDD and FDD bands come from.
FFT rate, slot latency budget and TDD switching timing all fall out of the numerology you pick. Tell us the band and the service and we will tell you what the baseband has to hit.