PDCCH & DCI Formats in 5G NR

Introduction

DCI is the control payload that makes dynamic scheduling possible in NR, and the PDCCH is the physical channel that carries it. Every time the gNB wants to send you downlink data, grant you uplink resources, page you, adjust your power, or tell you the slot format, it packs the instruction into a DCI of a specific format, protects it, and transmits it on the PDCCH inside a region the UE is already monitoring (see CORESET & Search Space).

DCI is used at every stage of the UE lifecycle. During initial and random access it carries SIB and RAR scheduling with robust fallback formats. Once connected, it delivers the steady stream of unicast grants that move data on PDSCH and PUSCH, plus a family of group-common commands (slot format, pre-emption, power control, wake-up). The formats are defined in TS 38.212; how the UE monitors and interprets them is in TS 38.213.

It matters because the data channels carry no self-describing header — without a correctly decoded DCI, the resources underneath are unusable. Control reliability therefore gates everything: this is why DCI is fixed at the most robust modulation, protected by a long CRC, and coded so it can be found by brute-force blind decoding rather than any explicit pointer.

Why DCI on PDCCH is needed

Concretely, the shared channels PDSCH and PUSCH are pure data pipes with no embedded header describing their own contents. Something external must tell the UE which PRBs and symbols the data occupies, which modulation and code rate were used, which HARQ process it belongs to, and when to act. DCI is that "something." And because control must be understood even before dedicated configuration exists — during access, paging and random access — it is engineered for robustness first: a fixed low-order modulation, a strong CRC, and short-blocklength coding that degrades gracefully. If the control is lost, the data underneath is lost with it, no matter how well the data was coded.

PDCCH carries DCI

The PDCCH is the channel; the DCI is the payload it carries. When the gNB decides to schedule you, it builds a DCI message, protects it, and transmits it on the PDCCH inside a region of time and frequency the UE is already watching. The UE blindly decodes candidate locations, and if a decode succeeds and passes its CRC check, it acts on the grant.

Three processing choices define the physical DCI. First, the CRC attached to the DCI is scrambled by an RNTI (Radio Network Temporary Identifier) — this is how addressing works without spending bits on an explicit UE ID. Second, the channel coding is Polar coding, the same family used for the Broadcast Channel. Third, the modulation is fixed at QPSK: a DCI is never higher-order QAM-modulated, because robustness matters far more than spectral efficiency for control. Get the control wrong and the data underneath is lost regardless of how well it was coded.

PDCCH physical structure: REG, CCE, and aggregation levels

A PDCCH is not sprayed arbitrarily across the grid. It is tiled from small, standardised units so the UE knows exactly which resource elements to combine when it attempts a decode. Two units matter, and one derived quantity governs robustness.

• REG (Resource Element Group) = 1 resource block × 1 OFDM symbol = 12 resource elements. Of those 12 REs, 3 carry the PDCCH-DMRS (the demodulation reference signal, at subcarriers 1, 5, 9) and the remaining 9 carry coded control data.
• CCE (Control Channel Element) = 6 REGs = 72 REs, of which 54 carry data and 18 carry DMRS. The CCE is the currency of PDCCH allocation — everything is counted in CCEs.
• Aggregation Level (AL) = the number of CCEs used for one PDCCH transmission. NR supports AL 1, 2, 4, 8, and 16.

The PDCCH-DMRS is what makes the control channel demodulable at all. Because control is decoded before any data, the UE cannot rely on data-region reference signals; instead each PDCCH candidate carries its own DMRS, scrambled by pdcch-DMRS-ScramblingID (0–65535, defaulting to the physical cell ID when absent), so the receiver can estimate the channel across exactly the CCEs it is combining. Whether that estimate can be shared across REGs depends on precoderGranularity.

The point of aggregation levels is link adaptation for control. A DCI payload is a fixed number of bits; spreading it across more CCEs simply lowers the effective code rate and adds redundancy. A UE near the cell centre with strong SINR decodes reliably at AL 1 or AL 2 (spending few CCEs, leaving control capacity for other UEs). A cell-edge UE in poor SINR is given AL 8 or AL 16, buying reliability at the cost of consuming a large slice of the CORESET. The gNB estimates the downlink quality (often from CSI or from the UL SINR) and picks the AL per transmission — there is no explicit AL field in the DCI, because the AL is implicit in which candidate the UE successfully decodes.

Polar coding, the 24-bit CRC, and RNTI scrambling

The journey from DCI bits to PDCCH symbols is short but exact. The gNB first appends a 24-bit CRC to the DCI payload. It then scrambles (XORs) that CRC with an RNTI. The combined block is Polar-encoded (with interleaving and rate-matching to fit the chosen aggregation level), scrambled, QPSK-modulated, and mapped to the CCEs of a candidate.

The UE never sees an explicit "this is for UE #7" field. Instead, at each monitoring occasion it takes each candidate, Polar-decodes it, and then de-scrambles the recovered CRC with the RNTI(s) it currently owns. If the CRC checks out against, say, the UE's C-RNTI, the DCI is for that UE and carries the meaning associated with that RNTI. If no de-scrambling passes, the candidate was not for this UE and is discarded. This is elegant because it does double duty: the RNTI both addresses the UE and selects the interpretation — the very same DCI format decoded with SI-RNTI means "system information scheduling," while decoded with C-RNTI it means "a dynamic unicast grant for you."

Polar coding is used because it performs well at the short block lengths typical of control (tens to low hundreds of bits) and degrades gracefully. The 24-bit CRC is long enough to keep the false-alarm rate negligible (on the order of 2^-24 per hypothesis) even though the UE runs many blind decodes per slot across multiple RNTIs and aggregation levels. RNTI values themselves are 16-bit (0x0001–0xFFEF for assignable identities such as C-RNTI, with the top of the range reserved for fixed system RNTIs).

The full DCI format catalogue

DCI comes in a fixed catalogue of formats, grouped by job: uplink grants (0_x), downlink assignments (1_x), and a family of group-common control messages (2_x). A UL grant schedules the PUSCH; a DL assignment schedules the PDSCH. The 2_x formats are broadcast-style commands read by a group of UEs, each addressed by a shared RNTI.

Fallback (x_0) vs non-fallback (x_1) vs compact (x_2)

Within each scheduling direction the three variants trade robustness, richness, and latency. Understanding the split explains why NR carries three DL formats and three UL formats rather than one.

Formats 0_0 and 1_0 are the fallback DCIs: compact, with a size that is fixed and predictable from minimal configuration, and monitored primarily in the Common Search Space (CSS). They are robust and always understood, which is exactly what you want during initial access, random access, paging, and any moment the UE and gNB might disagree about configuration. The SI-RNTI, RA-RNTI, P-RNTI, and TC-RNTI procedures all rely on these fallback formats because they must be decodable before dedicated RRC configuration exists.

Formats 0_1 and 1_1 are the non-fallback DCIs: larger, feature-rich, and monitored in the UE's UE-specific Search Space (USS). They unlock the advanced features a fully configured connected UE needs — multi-layer MIMO with antenna-port and TCI signalling, bandwidth-part switching, carrier-aggregation fields, ZP/rate-matching control, and richer HARQ handling.

Formats 0_2 and 1_2 are the compact / configurable variants introduced in Release 16 for URLLC. Many of their field widths are RRC-configurable, so the network can shrink the DCI to the minimum bits a low-latency service needs, improving reliability (fewer bits at a given AL) and enabling faster, smaller-footprint scheduling.

Field-by-field: DCI 1_1 (downlink assignment)

The non-fallback DL format is where the real scheduling detail lives. Many field widths depend on RRC configuration, so the bit counts below are typical rather than fixed. The time-domain field is an index into a configured PDSCH TDRA table; the selected row expands to the slot offset K0 and the SLIV (start symbol and length). Note that MCS/NDI/RV appear per transport block when two-codeword MIMO is configured.

Field-by-field: DCI 0_1 (uplink grant)

The non-fallback UL format mirrors much of 1_1 but points at a PUSCH and adds the fields needed to control uplink transmission — the SRS-based beam/precoder selection, codebook precoding, and CSI/beta-offset triggering. Its time-domain field indexes the PUSCH TDRA table, yielding the slot offset K2 and the SLIV. There is no PUCCH/HARQ-timing field, because the UE is transmitting data, not feeding back ACKs.

Summary

The PDCCH is the physical control channel and the DCI is the bit-field it carries. A DCI is built from REGs (12 REs, 3 DMRS) grouped six-to-a-CCE, aggregated over 1/2/4/8/16 CCEs to trade capacity for robustness, protected by a 24-bit CRC that is XOR-scrambled with an RNTI, Polar-coded, and QPSK-modulated. The RNTI does double duty — it both addresses the UE and selects the interpretation — and there is no explicit AL field, so the UE learns both which grant is its and how big it is only by a CRC that unmasks cleanly.

The format catalogue splits by job (0_x uplink, 1_x downlink, 2_x group-common) and by richness (fallback / non-fallback / compact). Fallback formats keep access and paging decodable before configuration exists; the non-fallback 1_1/0_1 carry the full feature set of a connected UE, with FDRA/TDRA pointing indirectly at resources and TDRA rows yielding K0/K2 while K1 comes from the feedback-timing field. Keeping the per-cell DCI-size budget (at most 4, at most 3 with C-RNTI) in mind is what keeps the whole scheme within the UE's blind-decode limits.

Quick Q&A