NR PUCCH: Why Uplink Control Needs Five Formats
PUCCH carries anything from a single HARQ bit to a full CSI report, from the cell centre to the edge. No single format spans that, which is why NR defines five — and how the network picks between them.
The uplink control channel has to carry two things that look nothing alike. At one extreme it is a single HARQ acknowledgement — one bit, needed within a slot or two, from a UE that may be anywhere in the cell. At the other it is a full channel state report: tens or hundreds of bits describing rank, precoder and quality across the band.
No single channel design covers both ends of that range. One bit sent with the machinery needed for a large report wastes almost all of it; a large report sent with the machinery sized for one bit does not decode. NR's answer is five PUCCH formats, and the way to understand them is as two independent choices rather than five options in a list.
Two axes, not five choices
The first axis is duration. A format is either short — one or two OFDM symbols, sitting at the end of a slot — or long, four to fourteen symbols.
The second is payload. A format either carries up to two bits, or more than two.
That gives four combinations, and NR defines a format for each:
| Up to 2 bits | More than 2 bits | |
|---|---|---|
| Short (1–2 symbols) | Format 0 | Format 2 |
| Long (4–14 symbols) | Format 1 | Format 3 |
Format 4 is the fifth, and it is a variant of Format 3 rather than a new point on the grid: it uses spreading so that several UEs can share the same resource. It exists for cells with many UEs each sending small reports, where giving every one its own long allocation would be wasteful.
Once you see it as a two-by-two, the formats stop being a list to memorise and become the answer to two questions the scheduler is actually asking.
The short formats carry no channel estimate of their own
Formats 0 and 1 both carry at most two bits, and the difference between them is not payload but how the receiver knows what it is looking at.
Format 0 carries no reference signal at all. The information is in which cyclic shift of a base sequence the UE transmits — the receiver correlates against the possible shifts and picks the strongest. With one or two symbols there is no room for a demodulation reference, so the signal has to be self-identifying.
That is a neat trick and it has a hard ceiling: the number of distinguishable shifts is what limits the payload to two bits. You cannot extend Format 0 to carry more by making it longer, because its whole design is built on there being nothing else in the resource.
Format 1 occupies four to fourteen symbols and alternates data and reference symbols. The extra symbols are not there to carry more information — the payload is still capped at two bits — they are there to accumulate energy. This is the format for a UE at the cell edge that needs to send one HARQ bit and cannot be heard in one symbol.
The long formats trade time for reach
Formats 2, 3 and 4 all carry real payloads with their own demodulation reference signals, and the choice between them is a coverage decision, not a capacity one.
A UE near the gNB sends its CSI report in one or two symbols with Format 2 and gets on with it. The same UE at the cell edge cannot — the energy in two symbols is not enough — so the network configures Format 3, which spreads the same report over up to fourteen symbols. If that is still not enough, the format can be repeated across slots.
Nothing about the information changed. What changed is how much energy the uplink has to put behind it, and the only currency the UE has for that is time, because its transmit power is already at the maximum the regulator and the amplifier allow.
This is the same trade that shows up in the uplink waveform choice and in preamble format selection. The uplink is power-limited in a way the downlink is not, and most uplink design decisions are ultimately about that.
What actually rides on PUCCH
Three kinds of information, and they have different deadlines:
HARQ acknowledgements are the most time-critical. A late acknowledgement stalls the process it belongs to, which is why the short formats exist at all.
Scheduling requests are a single bit meaning I have something to send. The UE has no grant, so it cannot say how much — that comes later in a buffer status report on PUSCH. The SR is periodic: the network configures occasions, and the UE transmits on the next one. A short SR period costs uplink resource on every UE whether or not they have anything to say; a long one adds latency to every uplink packet arrival. It is one of the more consequential single parameters in a latency-sensitive deployment.
Channel state reports carry CQI, PMI and RI. They are the largest payload and the least urgent — a report that arrives a slot late is still useful, where a HARQ acknowledgement is not.
Where PUCCH goes when there is PUSCH
If a UE has been granted PUSCH in the same slot, its control information does not go on PUCCH at all. It is multiplexed onto the PUSCH instead, puncturing or rate-matching around the data.
This is worth internalising, because it means PUCCH is the fallback, not the default. A UE that is actively transmitting data barely uses it. PUCCH exists for the case where the UE has control information and no data grant to carry it on — which, for a UE mostly receiving downlink traffic, is most of the time.
The parts that bite
PUCCH resources are configured, not negotiated. The UE is told which format and which resources to use through RRC, with some dynamic selection via a field in the downlink DCI. A UE that moves from the cell centre to the edge does not switch format by itself; the network has to reconfigure it. Cells that never move UEs onto long formats show cell-edge HARQ feedback loss that looks like a coverage hole.
Format 0 and the SR share a mechanism. Both are carried in the choice of cyclic shift, so a HARQ acknowledgement and a scheduling request in the same occasion are jointly encoded rather than sent separately. That is efficient and it is also a place where the payload limits bite sooner than people expect.
Long formats consume uplink symbols other UEs could use. Pushing everyone onto Format 3 for safety is a real capacity cost in a loaded cell, not a free robustness setting.
The mental model
Two axes, not five options: short or long duration, small or large payload. Format 4 is Format 3 with multiplexing.
Format 0 has no reference signal. The information is in the cyclic shift, which is exactly why it cannot carry more than two bits.
Long formats buy reach, not capacity. The payload cap does not change; the energy does.
PUCCH is the fallback. Control information rides on PUSCH whenever a UE has one.
The uplink is power-limited, so time is the currency. Nearly every PUCCH decision is a version of that sentence.
Further reading
- 3GPP TS 38.211 §6.3.2 — PUCCH formats and sequence generation
- 3GPP TS 38.213 §9 — UCI reporting procedures and resource determination
- 3GPP TS 38.331 — PUCCH-Config and the resource sets
- ShareTechnote — 5G PUCCH, for per-format resource mapping diagrams and configuration examples
Practise this on 5G6GTech
Our companion site turns these topics into flashcards and quizzes.
- Drill 5G concepts with the flashcard deck (opens on 5G6GTech in a new tab)
700 standards-aligned 5G concepts on a spaced-repetition schedule, covering RAN, the core and service-based architecture, the physical layer and 5G-Advanced.
- Test yourself in the 5G practice quiz (opens on 5G6GTech in a new tab)
Timed multiple-choice questions across the same 5G domains, with explanations for each answer.
Related Articles
NR PUSCH: The Uplink Is Not a Mirror of the Downlink
PUSCH looks like PDSCH turned around, and the differences are where the engineering is: a choice of waveform, a power-limited transmitter, and a grant that costs a round trip unless you configure one in advance.
Carrier Aggregation in 5G NR: PCell, SCells, and Why Throughput Doesn't Just Add Up
How 5G NR carrier aggregation works: component carriers, the PCell/SCell relationship, intra- and inter-band types, and why aggregate throughput rarely matches the sum of the parts.
HARQ in 5G NR: Soft Combining, Process IDs, and Flexible Timing
How HARQ combines FEC with retransmission, why soft combining beats plain ARQ, and what NR's K1/K2 timing flexibility changes for latency and URLLC.
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.