PDSCH — Downlink Shared Channel in LTE 4G

Introduction

PDSCH is the LTE downlink data pipe. When a UE is browsing, streaming, or downloading, the bits are arriving on PDSCH; when it reads system information, is paged, or completes random access, those messages also arrive on PDSCH. It is a shared channel — its time-frequency resources are not owned by any UE but handed out subframe by subframe by the eNB scheduler — which is exactly what makes bursty mobile traffic efficient: an idle UE consumes nothing.

PDSCH sits at the physical layer, carrying the DL-SCH transport channel that MAC hands down. It is used the entire time a UE is camped on or connected to a cell, and it is one of the very first channels a UE must be able to decode: SIB1, paging and the random-access response all ride PDSCH before the UE has any dedicated configuration. Because of that, PDSCH must be decodable both from broadcast information alone (the fallback path) and, once connected, in its full high-throughput MIMO form.

The crucial property to hold onto is that PDSCH is not self-describing. It carries no header telling you which resource blocks it occupies or how it was coded. All of that lives in a matching DCI on the PDCCH, and the RNTI that de-scrambles that DCI's CRC also fixes what the PDSCH means. Understanding PDSCH is therefore inseparable from understanding the grant that points at it.

Why PDSCH works the way it does

Three design choices follow directly from "shared, dynamically scheduled, and not self-describing." First, the resources must be described externally: the DCI supplies frequency allocation, MCS, HARQ process, NDI, RV and precoding, because the shared channel cannot describe itself. Second, the modulation-and-coding must be adaptive: because the scheduler picks a fresh MCS every subframe from CQI feedback, the same UE can be served with QPSK at the cell edge and 64QAM in the centre, maximising throughput per unit of spectrum. Third, decoding must be robustly retransmittable: HARQ with soft combining lets a failed transport block be recovered by adding coding gain rather than starting over.

What PDSCH carries

PDSCH is the physical channel that carries the Downlink Shared Channel (DL-SCH). Most of the time that means user-plane data, but the same channel is reused for several essential system messages that simply travel as DL-SCH payload. There is no separate "broadcast data channel" and no separate "paging channel" on the air — the logical BCCH, PCCH and CCCH are all mapped onto DL-SCH and delivered by PDSCH.

Because SIB1, paging and RAR all ride PDSCH, a UE must be able to receive PDSCH before it has any dedicated configuration at all. Those transmissions therefore use the compact fallback format DCI 1A in the common search space, scrambled by SI-RNTI, P-RNTI or RA-RNTI as appropriate, with the simplest allocation rules and QPSK or low-order modulation, so that any UE can decode them from broadcast information alone. SIB1 itself is scheduled dynamically with SI-RNTI and its scheduling of the other SIBs is described in SystemInformationBlockType1.

Where PDSCH sits in the subframe

An LTE downlink subframe is 1 ms and, with the normal cyclic prefix, has 14 OFDM symbols split into two 0.5 ms slots. The first 1–3 OFDM symbols form the control region, which carries PCFICH, PHICH and PDCCH spread across the whole system bandwidth. Everything after the control region is the PDSCH region, and PDSCH lives only in the resource blocks that your DCI allocates.

The PDSCH starting symbol is therefore not fixed — it is one symbol past the control region, and the size of that region is signalled by the Control Format Indicator (CFI) carried on PCFICH in every subframe. The UE decodes PCFICH first, learns whether the control region is 1, 2 or 3 symbols (up to 4 for the narrowest 1.4 MHz bandwidth), and only then knows where PDSCH begins. For cross-carrier scheduling in carrier aggregation the start is instead taken from the RRC parameter pdsch-Start, since the scheduling PDCCH sits on a different carrier.

Cell-specific reference signals (CRS) are scattered through every resource block for channel estimation, and PDSCH is mapped around them. The mapping also steps around the PSS/SSS and PBCH, which occupy the centre six resource blocks in subframes 0 and 5, and around any UE-specific DMRS when those are configured. The figure below shows one resource-block pair with all three regions.

Scheduled by DCI

A PDSCH is scheduled by a downlink-assignment DCI carried on the PDCCH (see PDCCH & DCI). The format chosen depends on the transmission mode:

• DCI 1A — compact fallback format; single transport block, always decodable with minimal configuration, used for SI, paging and RAR as well as normal single-layer data. It always uses resource-allocation Type 2.
• DCI 1 — single-codeword scheduling with the more flexible resource-allocation types (Type 0 / Type 1).
• DCI 1B / 1D — single-codeword with closed-loop precoding information (rank-1 precoding, multi-user hints).
• DCI 2, 2A, 2B, 2C, 2D — the "2x" family for spatial multiplexing, able to carry two transport blocks (two codewords) plus precoding and antenna-port information.

Whichever format is used, the DCI supplies what the shared channel cannot describe about itself: the frequency-domain resource allocation, the MCS index (per codeword), the HARQ process number (3 bits in FDD, up to 4 in TDD), the new-data indicator (NDI, 1 bit), the redundancy version (RV, 2 bits, i.e. RV 0/1/2/3), the transmit-power control command for PUCCH (2 bits), and — for the 2x formats — precoding and layer information. The UE combines these fields to know exactly which resource elements to demodulate and how to interpret them. The DCI's CRC is scrambled by the UE's C-RNTI (or SPS-C-RNTI for semi-persistent scheduling), which is how the UE knows the grant — and hence the PDSCH — is its own.

Downlink resource-allocation types

LTE frequency-domain allocation is built on the Resource Block Group (RBG): a set of P consecutive RBs treated as one schedulable unit, where P (1, 2, 3 or 4) grows with the channel bandwidth. The DCI's resource-allocation field is interpreted according to one of three types — covered in more depth on the Resource Allocation page.

Type 0 gives coarse but flexible non-contiguous placement, useful for frequency-selective scheduling across the band. Type 1 trades some addressing range for RB-level precision inside one RBG subset. Type 2 is the most compact: a single Resource Indicator Value (RIV) jointly encodes the starting RB (RB_start) and the length (L_CRBs), and the fallback DCI 1A always uses it. In its distributed flavour the virtual RBs are spread across the band so that even a small allocation gains frequency diversity; in the localized flavour the virtual RBs map straight through to physical RBs.

Transport block size determination

Once the UE knows how many RBs it has and which MCS was signalled, it looks up the transport block size (TBS). This is a deliberately two-step table lookup defined in TS 36.213, and getting the two steps straight is the whole trick.

In Step 1, the 5-bit MCS index (0–28 for the normal table) maps to a modulation order and a separate I_TBS. Note that I_TBS is not the same number as the MCS: several MCS values share an I_TBS at the boundary between modulations (for example MCS 9 and MCS 10 both give I_TBS 9, and MCS 16 and MCS 17 both give I_TBS 15), because the boundary MCS switches to a higher modulation while keeping the same effective code rate. In Step 2, that I_TBS together with the number of allocated PRBs (N_PRB, from 1 up to N_RB^DL) indexes a large two-dimensional table that reads off the exact TBS. This is why the same MCS yields a bigger transport block whenever more RBs are granted — the code rate is held roughly constant while the payload scales with bandwidth.

The reserved indices MCS 29, 30 and 31 are used only for retransmissions: no new I_TBS is looked up, the modulation is implied, and the TBS is inherited from the initial transmission of that HARQ process. When 256QAM is configured (Rel-12), a different pair of tables is used — Table 7.1.7.1-1A for the MCS mapping and an extended TBS table — with I_TBS reaching up to 33 to expose the extra spectral efficiency.

The peak TBS at I_TBS 26 scales cleanly with bandwidth, which is where LTE's headline single-stream rates come from:

Two codewords / spatial multiplexing. PDSCH can carry one or two transport blocks per subframe. A single codeword serves single-layer transmission and transmit diversity; a second codeword is added only for spatial multiplexing with rank greater than one. Crucially, the TBS lookup is performed independently per codeword: each codeword carries its own MCS in the DCI, so each runs its own Step 1 and Step 2, and each is HARQ-managed separately. That is why a 2×2 MIMO UE can peak at twice the single-codeword TBS, and why the eNB can send one codeword with 64QAM and the other with 16QAM if the two spatial layers see different quality. The number of layers (up to four in Rel-8/9, eight in Rel-10) maps onto the one or two codewords by the fixed codeword-to-layer rule in TS 36.211.

Modulation, scrambling, layers and RE mapping

After the transport block is channel-coded and rate-matched, PDSCH goes through a fixed physical-layer chain: scrambling → modulation → layer mapping → precoding → RE mapping (TS 36.211 clause 6.3).

Scrambling. The coded bits are scrambled by a length-31 Gold sequence whose initial value depends on the cell, the RNTI and the codeword, so neighbouring cells and different UEs randomise differently and inter-cell interference looks like noise. The scrambling initialiser is:

Here n_RNTI is the UE's RNTI, q is the codeword index (0 or 1), n_s is the slot number and N_ID_cell is the physical cell ID (0–503). Because the RNTI is part of the seed, another UE cannot descramble your PDSCH even if it decoded the same resource elements.

Layer mapping and precoding are driven entirely by the transmission mode (see Transmission Modes). The one or two codewords are mapped onto 1–8 layers, and the layers are precoded onto the physical antenna ports. In codebook-based modes the DCI carries a precoding-matrix indicator; in non-codebook modes (TM7–TM10) the precoder is transparent to the UE, applied at the eNB and estimated implicitly through the reference signals.

RE mapping. The complex symbols for each antenna port are mapped to resource elements in the allocated RBs, in order of increasing subcarrier first and then increasing symbol, starting at the PDSCH start symbol. The mapping skips any RE used by CRS, UE-specific DMRS, PSS/SSS, PBCH, and the control region — those resource elements are simply not counted, which is exactly why the number of usable REs, and hence the achievable code rate, depends on how many reference-signal ports and control symbols are configured. The table below sketches the RE budget for one RB-pair.

CRS-based vs DMRS-based demodulation

How you demodulate PDSCH depends on the transmission mode (TM). There are two families, and the split lands at TM7.

In the classic modes TM1–TM6 (and TM8's fallback paths) the UE equalises the channel using the always-present cell-specific reference signals (CRS). CRS are broadcast to the whole cell on 1, 2 or 4 antenna ports; every UE sees them, uses them for channel estimation, and must know the eNB's precoder (from the DCI) to reconstruct the effective channel per layer.

In the beamforming and advanced-MIMO modes TM7–TM10, PDSCH is instead demodulated from UE-specific DMRS (demodulation reference signals, historically "UE-specific RS"). These DMRS are transmitted through the same precoder as the data, so the UE estimates the effective (precoded) channel directly and never needs to be told the precoder at all. This is what makes non-codebook beamforming and TM9/TM10 multi-layer transmission practical: the eNB can point energy however it likes, and the UE just follows its own DMRS. TM10 extends this to coordinated multipoint (CoMP) with a configurable quasi-co-location and rate-matching behaviour.

HARQ for PDSCH

PDSCH decoding is protected by Hybrid ARQ. The scheduling DCI names the HARQ process number, toggles the NDI to signal a new transmission versus a retransmission, and sets the RV to select which coded bits are sent. After decoding, the UE reports ACK/NACK on PUCCH (or multiplexed on PUSCH when it has an uplink grant in the same subframe). A NACK triggers a retransmission that the UE soft-combines with the earlier attempt in its soft buffer (Chase combining when the RV repeats, incremental redundancy when it changes), so each retransmission adds coding gain rather than starting over. See the dedicated HARQ page for the buffer and timing details.

Downlink HARQ in LTE is asynchronous and adaptive: the eNB may retransmit in any later subframe (asynchronous) and may change the resource allocation, MCS and RV each time (adaptive), because every retransmission comes with its own DCI. This contrasts with the uplink, where HARQ is synchronous. There are up to 8 HARQ processes per component carrier in FDD (more in TDD, depending on the UL/DL configuration), running as parallel stop-and-wait channels so the pipe stays full while individual processes wait for feedback.

Summary

PDSCH carries the DL-SCH — user data plus SIBs, paging, RAR and Msg4 — on shared, dynamically scheduled resources that no UE owns. It is never self-describing: a matching downlink-assignment DCI (fallback DCI 1A, the 1-family, or the spatial-multiplexing 2-family) supplies the frequency allocation (Type 0/1/2), the per-codeword MCS, the HARQ process, NDI and RV, and any precoding, and the RNTI on that DCI's CRC both proves ownership and fixes meaning.

The starting symbol comes from the PCFICH CFI; the transport block size comes from a two-step lookup (MCS → Q_m, I_TBS, then I_TBS × N_PRB → TBS in bits), run independently per codeword. The physical chain is scramble (seeded by RNTI, codeword, slot and cell ID) → modulate (QPSK/16QAM/64QAM, 256QAM from Rel-12) → layer-map → precode → RE-map around CRS/DMRS/PSS/SSS/PBCH/control. Demodulation is CRS-based in TM1–TM6 and DMRS-based in TM7–TM10, and reliability comes from up to 8 asynchronous, adaptive HARQ processes with soft combining. NR keeps the MCS-driven, HARQ-protected shared-channel idea but drops CRS in favour of always-DMRS demodulation and swaps turbo coding for LDPC.