PUSCH — Uplink Shared Channel in 5G NR

Introduction

PUSCH is the physical channel that carries everything the UE has to send that is not a tiny piece of control information: user-plane packets, and uplink RRC messages such as RRCSetupComplete, RRCReconfigurationComplete or measurement reports. It is defined physically in TS 38.211, its coding chain in TS 38.212, its power-control and procedural rules in TS 38.213, and its resource-mapping, MCS/TBS and precoding rules in TS 38.214.

PUSCH first appears very early in the UE lifecycle: the random-access Msg3 is a PUSCH, scheduled by the RAR uplink grant, and every subsequent scheduled uplink transmission — from a browser upload to a periodic VoIP frame — rides the same channel. Understanding it therefore matters from the first moment a UE talks to the network right through to steady-state data transfer.

Because the uplink is power- and coverage-limited, PUSCH is built with far more configurability than PDSCH: the gNB can switch the waveform, restrict the number of layers, turn on frequency hopping, choose a precoding philosophy, and steer transmit power — all per UE and per situation. This page walks through those choices, how PUSCH is laid out in time and frequency, how reference signals ride with it, how control multiplexes onto it, and how its transport block size is fixed.

Why PUSCH is shaped the way it is

Concretely, three pressures shape PUSCH. First, the amplifier: a UE power amplifier must stay linear, and a high peak-to-average-power ratio (PAPR) forces it to back off from saturation, wasting precious dB — hence the low-PAPR DFT-s-OFDM option. Second, path loss and fading: a cell-edge UE cannot spread thin power across a wide band, so it concentrates into a narrow allocation and uses hopping to escape fades. Third, battery and interference: transmitting harder than necessary drains the UE and raises noise for everyone, so tight open- plus closed-loop power control keeps each PUSCH at just the level the gNB needs to decode it. Every PUSCH feature below is one of these three pressures made concrete.

What PUSCH carries

PUSCH is the workhorse of the uplink. Its primary payload is the Uplink Shared Channel (UL-SCH): user-plane data, and uplink RRC messages such as RRCSetupComplete or measurement reports. But PUSCH is more than a data pipe. When the UE has Uplink Control Information (UCI) to send — a HARQ-ACK for a downlink transport block, or a CSI report — in a slot where it is also transmitting PUSCH, that control is normally multiplexed onto the PUSCH rather than sent separately on PUCCH. This keeps the UE to a single-carrier-friendly transmission and avoids the power split of simultaneous PUSCH+PUCCH.

The very first PUSCH a UE ever sends is special: Msg3 during random access, scheduled by the RAR uplink grant and carrying the RRC connection request. It is a stripped-down PUSCH — no configured DMRS flexibility, fixed to the initial UL BWP — but it is the same physical channel, which is why understanding PUSCH matters from the very first moment a UE talks to the network.

Waveform: CP-OFDM vs DFT-s-OFDM

PUSCH is the only NR physical channel that can use either of two waveforms, and the choice is a single RRC switch: transformPrecoder inside PUSCH-Config (for dynamic grants) or ConfiguredGrantConfig (values enabled/disabled; when absent the UE follows msg3-transformPrecoder from rach-ConfigCommon). The two options are cyclic-prefix OFDM and DFT-spread OFDM, and they sit at opposite ends of the capacity-versus-coverage trade.

CP-OFDM (transform precoding disabled) is the same multi-carrier waveform used on the downlink. Because it maps modulation symbols straight onto subcarriers, it supports true spatial multiplexing — up to four layers on PUSCH — and delivers the highest spectral efficiency when the UE has power to spare and good SINR near the cell centre. Its downside is a higher peak-to-average power ratio (PAPR): the amplifier must back off from its saturation point to stay linear, wasting power headroom.

DFT-s-OFDM (transform precoding enabled) inserts a DFT "spreading" stage before the OFDM modulator. The result is a signal that behaves almost like single-carrier transmission, with a markedly lower PAPR (typically a few dB lower, which the UE can convert into extra transmit power via a smaller Maximum Power Reduction). The amplifier can run closer to its limit without distortion, so the UE radiates more usable power — directly extending uplink coverage for cell-edge and power-limited devices. The price is that DFT-s-OFDM is restricted to a single layer (no uplink MIMO) and its modulation options include a special π/2-BPSK for the most coverage-limited cases.

Time-domain mapping, DMRS and PT-RS

Where PUSCH sits inside a slot follows the same framework as PDSCH. The DCI's time-domain resource assignment (TDRA) field indexes a row of pusch-TimeDomainAllocationList giving K2 (the DCI-to-PUSCH slot offset, integer range 0…32), a SLIV (start symbol S and length L), and the mapping type.

Mapping Type A is slot-based: the allocation is referenced to the start of the slot and the first DMRS symbol sits at a fixed position (symbol 2 or 3, set by dmrs-TypeA-Position), so it suits longer allocations that fill most of the slot. Mapping Type B is non-slot-based: it can start on almost any symbol and the first DMRS is placed on the first allocated symbol, which makes it ideal for short, latency-sensitive transmissions that must begin mid-slot.

DMRS (Demodulation Reference Signal) lets the gNB estimate the uplink channel so it can coherently demodulate PUSCH. NR defines two configuration types that differ in how reference-signal density trades against the number of orthogonal ports. Configuration Type 1 packs up to 4 ports in one front-loaded symbol (8 across two symbols) using a comb-2 + cyclic-shift structure; Configuration Type 2 gives up to 6 ports in one symbol (12 across two) using frequency-domain code division, favouring higher-order MU-MIMO at the cost of slightly coarser frequency estimation. Both start with a front-loaded DMRS, and additional DMRS symbols (dmrs-AdditionalPosition, values pos0…pos3) can be inserted deeper into the slot so the channel estimate tracks fast fading for high-mobility UEs.

PT-RS (Phase-Tracking Reference Signal) is an uplink companion to DMRS that matters mainly at high carrier frequencies (FR2) and high MCS, where oscillator phase noise rotates the constellation across the slot. PT-RS is sparse in frequency but dense in time, letting the receiver track and undo common phase error symbol by symbol. Its presence and density (PTRS-UplinkConfig) scale with the scheduled bandwidth and MCS via the RRC thresholds timeDensity and frequencyDensity. For CP-OFDM the PT-RS maps to specific subcarriers/ports; for DFT-s-OFDM it is inserted as time-domain sample groups (sampleDensity) before the DFT.

Frequency hopping

A power-limited UE at the cell edge often cannot afford to transmit across a wide band, so it concentrates its power into a narrow allocation. That narrow allocation is vulnerable to a frequency-selective fade. Frequency hopping solves this by moving the allocation between two (or more) frequency regions so the transmission samples independent parts of the channel — buying frequency diversity for free.

NR PUSCH supports two hopping modes, enabled by frequencyHopping and parameterised by frequencyHoppingOffset (one of up to four RRC-listed offsets, selected by a field in the DCI):

• Intra-slot hopping — the PUSCH transmits its first half of symbols in one frequency region and its second half in another, all inside a single slot. This is the common case and works even for a one-shot transmission.
• Inter-slot hopping — the frequency position alternates from slot to slot across a multi-slot (repetition) transmission, so each repetition lands on a different part of the band. This suits slot-aggregated coverage extension.

For configured-grant and Msg3 PUSCH the hopping offsets are similarly signalled, and when hopping is on, DMRS is transmitted in each hop so the gNB can re-estimate the channel at the new frequency. Hopping combines naturally with DFT-s-OFDM: both target exactly the cell-edge, power-limited regime.

Codebook vs non-codebook transmission

When PUSCH uses CP-OFDM with multiple transmit antennas, the UE must apply a precoder — a matrix that maps the data layers onto physical antenna ports. NR defines two philosophies for choosing it, and both lean on the Sounding Reference Signal (SRS) that the UE transmits so the gNB can measure the uplink channel. The choice is set by txConfig = codebook or nonCodebook in PUSCH-Config.

Codebook-based transmission puts the gNB in charge. The gNB measures the UE's (non-precoded) SRS — from an SRS resource set with usage = codebook — decides how many layers to use and which precoding matrix from a standardised uplink codebook fits best, and signals its decision in the uplink DCI (0_1/0_2) using three fields: the SRI (SRS Resource Indicator) selecting which SRS resource — hence which antenna ports and beam — to transmit on; the TPMI (Transmitted Precoding Matrix Indicator) selecting the precoding matrix; and precodingAndNumberOfLayers, the joint field that actually encodes rank + TPMI. The UE simply applies what it is told. Which TPMIs are usable depends on the UE's antenna coherence capability — how well it can maintain a stable relative phase across its transmit chains:

• Full coherence — the UE can control relative phase across all ports, so the whole codebook (including precoders that combine all antennas coherently) is available.
• Partial coherence — the UE can keep phase coherent within pairs of ports but not across pairs; only a subset of the codebook is allowed.
• Non-coherent — the UE cannot maintain phase across ports, so only antenna-selection precoders (one layer per antenna, no coherent combining) are permitted.

RRC advertises the UE's coherence via codebookSubset (fullyAndPartialAndNonCoherent / partialAndNonCoherent / nonCoherent), which restricts the TPMIs the gNB may signal, and maxRank caps the number of layers.

Non-codebook-based transmission flips the roles. The UE computes its own candidate precoders — typically from downlink CSI-RS measurements (the resource named by NZP-CSI-RS-ResourceId in the SRS set with usage = nonCodebook), exploiting channel reciprocity — and applies each candidate to a different SRS resource. It sounds the channel through those precoders. The gNB then picks the best combination and returns only the SRI, which points at one or more SRS resources; the UE reuses the precoder tied to each indicated SRS. There is no TPMI, because the precoder is baked into the SRS. Non-codebook supports up to 4 layers, one SRS resource selected per layer.

Scheduling: dynamic grant vs configured grant

A PUSCH transmission has to be authorised before the UE may send it. NR offers two authorisation styles that differ in overhead and latency.

Dynamic grant. The gNB sends an uplink DCI on the PDCCH, and that single DCI schedules one PUSCH. Three formats exist: DCI format 0_0 is the compact fallback grant (small, robust, monitored in both common and UE-specific search spaces, CRC scrambled by C-RNTI/TC-RNTI); DCI format 0_1 is the full-feature non-fallback grant carrying SRI, TPMI, DMRS, PT-RS, CSI-request and more (UE-specific search space, C-RNTI/CS-RNTI/MCS-C-RNTI); and DCI format 0_2 is a configurable grant designed for URLLC, where many field sizes are RRC-tunable to shrink the DCI for low latency. The DCI's TDRA field yields K2 and the SLIV, so a grant received in slot n produces a PUSCH in slot n + K2. See PDCCH & DCI.

Configured grant. For periodic or predictable traffic — VoIP, industrial sensors, URLLC — sending a DCI for every transmission is wasteful and adds latency. A configured grant pre-arranges recurring PUSCH occasions so the UE can transmit without a per-transmission DCI:

• Type 1 — fully configured by RRC in ConfiguredGrantConfig → rrc-ConfiguredUplinkGrant (time/frequency resources, periodicity, MCS, DMRS, power). It is truly grant-free: once RRC sets it up the UE simply uses the occasions, with no activation DCI at all.
• Type 2 — the parameters come from RRC, but the grant is activated and released by a DCI (DCI format 0_0/0_1) scrambled with CS-RNTI. RRC arms it; an activation DCI switches it on (and supplies the frequency/MCS), and a release DCI switches it off. This blends RRC's low overhead with DCI's flexibility.

UCI-on-PUSCH, TBS, power control and Msg3

UCI multiplexing. When HARQ-ACK or CSI must be sent in a slot that already carries PUSCH, the control is rate-matched or punctured onto the PUSCH resources rather than sent on PUCCH. HARQ-ACK is placed right after the front-loaded DMRS (so it is well protected and available early); CSI is spread across the allocation. The amount of resource each UCI type steals from data is set by beta-offsets — per-UCI scaling factors (betaOffsets, semi-static or dynamic via a DCI field) that trade UCI reliability against data throughput. CSI is split into two parts so it can be decoded incrementally: CSI part 1 (fixed size — RI, CQI, and the count of part-2 payload) and CSI part 2 (variable size, e.g. the PMI whose size depends on part 1). Each of HARQ-ACK, CSI part 1 and CSI part 2 has its own beta-offset.

Transport block size (TBS). PUSCH determines its TBS with the same chain as PDSCH: from the MCS index (0…31, looked up in the mcs-Table in force) the UE reads a modulation order Q_m and target code rate R; it counts the resource elements available for data per RB (subtracting DMRS, and any UCI/PT-RS overhead, plus the RRC xOverhead term); multiplies by allocated RBs, layers, Q_m and R to get an intermediate number of information bits; then quantises that through the standardised TBS steps and rounds so the block is byte-aligned and LDPC-friendly.

The MCS table in force (64QAM, 256QAM, or the low-spectral-efficiency table for coverage) is selected by RRC (mcs-Table) and the RNTI used (MCS-C-RNTI forces the low-SE table). Because the arithmetic is identical to the downlink, if you know PDSCH TBS you already know PUSCH TBS — only the overhead terms differ.

Power control hooks. PUSCH transmit power is set by an open-loop term — a target received power P0 (p0-NominalWithGrant plus a UE-specific offset) plus a fractional pathloss compensation factor alpha (enumerated 0, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1) — scaled by the number of allocated RBs and the MCS (via a delta-MCS term), and adjusted by a closed-loop accumulation driven by TPC commands in the uplink DCI (or group-common DCI format 2_2):

The details, including multiple power-control loops and pathloss reference RS, live on the Power Control page; the point here is that PUSCH is the channel these loops ultimately steer, capped by the UE's maximum power P-Max/PCMAX.

Msg3 PUSCH. The random-access Msg3 is a constrained PUSCH: it is scheduled by the RAR uplink grant (not a normal DCI), transmitted in the initial UL BWP, uses transform precoding per msg3-transformPrecoder, and its power is set by a dedicated Msg3 power-control formula because there is no established closed loop yet. HARQ retransmissions of Msg3 are scheduled by a DCI format 0_0 scrambled with TC-RNTI in the Type1-PDCCH common search space. It is the bridge between random access and normal scheduled uplink.

Summary

PUSCH is the uplink data channel, and almost everything about it is a response to the UE being power- and coverage-limited. You pick a waveform with transformPrecoder — CP-OFDM for up to 4 layers of capacity near the cell centre, DFT-s-OFDM for low-PAPR coverage at the edge. You place it in the slot with a TDRA row (mapping Type A or B, K2 = 0…32), let DMRS (Config Type 1 or 2, plus additional positions) and PT-RS keep the channel and phase estimates honest, and optionally turn on frequency hopping for diversity. You choose a precoding philosophy — codebook (gNB sends SRI + TPMI, bounded by the UE's coherence) or non-codebook (UE pre-codes its own SRS, gNB returns just the SRI). You authorise it dynamically with DCI 0_0/0_1/0_2 or via a configured grant (Type 1 pure-RRC, Type 2 CS-RNTI-activated). And you let it carry UCI when a control-bearing slot overlaps, size its TBS exactly as PDSCH does, and steer its power with the open- plus closed-loop formula — the whole toolkit for making a battery-powered transmitter heard at the cell edge.

Quick Q&A