PUSCH — Uplink Shared Channel (SC-FDMA) in LTE 4G

Introduction

PUSCH is the LTE uplink data pipe. Every uploaded photo, every VoLTE packet, every RRC message the UE sends after connection setup, and the MAC control that keeps the scheduler informed all travel on PUSCH as the UL-SCH transport channel. It is a scheduled channel: unlike random access, the UE never transmits PUSCH spontaneously — it transmits only when the eNodeB has granted it uplink resources.

PUSCH is used the whole time a UE is in RRC_CONNECTED with data or signalling to send uplink. Its defining feature is the waveform: where the downlink can afford the high peak-to-average ratio of OFDMA behind a large, mains-powered eNodeB amplifier, the uplink cannot — so PUSCH uses SC-FDMA to protect the handset's coverage and battery. That single choice ripples through everything else on this page: why allocations must be contiguous, why the DFT stage exists, and why the uplink and downlink of the same standard deliberately use different waveforms.

Around the data itself, PUSCH also has to carry demodulation references (DMRS), can piggyback uplink control (UCI), runs synchronous HARQ tied to a fixed timing relationship, and is held at the right received power by closed-loop power control. Understanding PUSCH means understanding how all of those fit into one low-PAPR, contiguous, scheduled transmission.

Why PUSCH uses SC-FDMA

The concrete problem is the UE's power amplifier. It must be operated with enough back-off that even the signal's peaks stay in the amplifier's linear region; otherwise the peaks clip and spill energy into adjacent channels, violating the spectral emission mask. A high-PAPR OFDMA waveform would force several extra decibels of back-off, and every decibel of back-off is a decibel of coverage lost at the cell edge. By flattening the envelope with a DFT stage, SC-FDMA lets the handset radiate more average power for the same amplifier and the same battery drain.

What PUSCH carries and why SC-FDMA

PUSCH transports the Uplink Shared Channel (UL-SCH): user-plane uplink data and uplink RRC signalling, plus MAC control elements such as the BSR and PHR that ride inside the transport block. When the UE has control information to send in a subframe where it is already transmitting PUSCH, that Uplink Control Information (UCI) — HARQ-ACK, CQI/PMI and RI — can be multiplexed onto PUSCH rather than sent separately on PUCCH. PUSCH is a scheduled channel: the UE never transmits it spontaneously, it transmits only when the eNodeB has handed it an uplink grant.

A lower PAPR translates directly into link budget. The UE's power amplifier must be operated with enough back-off that even the signal's peaks stay in the amplifier's linear region; otherwise the peaks clip and spill energy into adjacent channels, violating the spectral emission mask. A high-PAPR OFDMA waveform would force several extra decibels of back-off, and every decibel of back-off is a decibel of coverage lost at the cell edge. By flattening the envelope, SC-FDMA lets the handset radiate more average power for the same amplifier and the same battery drain — which is why the uplink and downlink of LTE deliberately use different waveforms.

OFDMA (downlink) vs SC-FDMA (uplink)

Both waveforms use the same numerology — 15 kHz subcarriers, a 12-subcarrier resource block, normal or extended cyclic prefix, and 14 (or 12) SC-FDMA symbols per subframe. The difference is one extra processing stage on the uplink: a DFT precoder before the mapping to subcarriers. That single block changes the character of the signal completely.

Because each data symbol in SC-FDMA is spread over the whole allocation by the DFT, the instantaneous envelope stays much flatter than OFDMA, where independent subcarriers occasionally add up in phase and produce large peaks. Lower peaks mean the UE can back off its amplifier less, so it radiates more useful power for the same battery drain. The price is paid at the receiver: because information about each symbol is smeared across all subcarriers, the eNodeB must run a frequency-domain equaliser after its receive DFT to unwind the channel and de-spread the symbols. That is an easy trade — the eNodeB has the processing budget and mains power that the handset does not.

The SC-FDMA transmit chain

The single defining feature is the DFT precoding (also called transform precoding) applied to the modulation symbols before they are mapped to subcarriers. The chain is: code bits are modulated to QPSK/16QAM/64QAM symbols, an M-point DFT spreads those M symbols, the DFT outputs are mapped onto M contiguous subcarriers, an N-point IFFT builds the time-domain symbol, and a cyclic prefix (CP) is prepended before the signal goes to RF.

The M-point DFT size M equals the number of allocated subcarriers, so M is always a multiple of 12 (one resource block = 12 subcarriers). For efficient implementation LTE restricts M to values whose only prime factors are 2, 3 and 5 — which is why an uplink allocation size in resource blocks is limited to numbers of the form 2^a·3^b·5^c (so 1, 2, 3, 4, 5, 6, 8, 9, 10, 12… RBs are allowed, but 7, 11, 13 are not). The subcarrier-mapping block places the M DFT outputs onto M of the N IFFT inputs; in LTE this is always localised mapping (a contiguous run of subcarriers), not the distributed/comb mapping that early SC-FDMA studies also defined. The CP then does the same job it does on the downlink: it absorbs multipath delay spread so the channel looks circular and can be equalised with a single complex multiply per subcarrier.

Contiguous allocation, RIV and frequency hopping

The single-carrier property of SC-FDMA imposes a hard rule: a UE's PUSCH must occupy a contiguous set of resource blocks in frequency. There is no distributed or comb allocation for PUSCH as there can be on the downlink; this is uplink resource allocation Type 2, localised. So an uplink grant only needs to signal a starting RB and a length, and LTE encodes both in a single Resource Indication Value (RIV) inside the uplink grant.

The uplink grant is DCI format 0 (Release 8) — carried on the PDCCH and addressed to the UE by its C-RNTI. Beyond the RIV, DCI 0 carries the MCS and redundancy version field (5 bits), a TPC command for power control (2 bits), the new-data indicator (NDI) for HARQ (1 bit), the cyclic shift for DMRS field (3 bits), a CQI request bit, and a frequency-hopping flag (1 bit). It is padded to the same size as DCI 1A so the two share a blind-decode attempt. Later releases add DCI format 4 for uplink spatial multiplexing (MIMO) and multi-cluster allocation.

Frequency hopping is an option that lets the PUSCH move to a different part of the band between or within subframes, buying frequency diversity and interference averaging — especially valuable for slow-moving, cell-edge UEs that cannot exploit frequency-selective scheduling. LTE defines two hopping mechanisms. Type 1 hopping applies an explicit hopping offset signalled in the grant. Type 2 hopping follows a predefined, cell-specific hopping pattern derived from higher-layer parameters. Both are gated by the 1-bit hopping flag in DCI 0 and configured by the hopping parameters broadcast in SIB2 (the number of hopping sub-bands pusch-HoppingOffset and the mode). Independently of type, hopping can be intra-subframe (the two slots of a subframe use different frequencies, giving diversity within a single TTI) or inter-subframe (the frequency changes from subframe to subframe).

DMRS, n+4 timing and synchronous HARQ

DMRS (Demodulation Reference Signal) for PUSCH is transmitted on the 4th symbol of each slot (symbol index 3 with normal CP) and spans exactly the same RBs as the data it references. Sitting one full DMRS per slot, roughly in the middle, lets the eNodeB estimate the uplink channel at two points across a subframe and interpolate for the data symbols on either side. The DMRS is built from a Zadoff-Chu base sequence selected per cell, onto which a cyclic shift and an orthogonal cover code are applied; the cyclic-shift field in DCI 0 lets several UEs share the same RBs (MU-MIMO) while staying separable, because different cyclic shifts of the same base sequence are orthogonal.

The uplink grant does not schedule "now." An uplink grant (DCI 0, or a PHICH ACK/NACK) received in subframe n schedules the PUSCH transmission in subframe n+4 for FDD. This fixed four-subframe gap gives the UE time to decode the grant, assemble the transport block, run the DFT-spread OFDM chain and satisfy its timing advance. The same n+4 spacing is why uplink HARQ is synchronous: unlike the asynchronous, adaptive downlink HARQ, uplink retransmissions happen at fixed timing — for FDD, a given HARQ process's retransmission occurs exactly 8 subframes after the original, giving 8 HARQ processes in a round-trip. Because the timing is fixed, the HARQ process number is implied by the subframe number rather than signalled in the grant.

The eNodeB's ACK/NACK for a received PUSCH is carried on the PHICH (Physical HARQ Indicator Channel). A NACK on PHICH triggers a non-adaptive retransmission at the fixed time (same RBs, same MCS, next redundancy version), while sending a fresh DCI 0 instead lets the eNodeB override that with an adaptive retransmission on different resources. The NDI toggling tells the UE whether the grant is for new data or a retransmission of the current HARQ process. SRS (Sounding Reference Signal) is a separate, wideband uplink reference the UE transmits in the last SC-FDMA symbol of a designated subframe so the eNodeB can sound channel quality across frequencies it is not currently scheduling; this is what enables frequency-selective uplink scheduling and uplink timing/power tracking. When an SRS subframe collides with PUSCH, the PUSCH in that subframe is shortened (its last symbol is dropped) so the SRS can be sent — the two share the subframe rather than one blocking the other.

Multiplexing UCI onto PUSCH

When the UE has to send uplink control information in a subframe where it is already transmitting PUSCH, LTE avoids the extra PAPR of transmitting PUCCH and PUSCH at the same time (which would break the single-carrier property) by multiplexing the UCI onto the PUSCH instead. The three kinds of UCI — HARQ-ACK, CQI/PMI, and RI — are treated differently, because they differ in how catastrophic an error is and in how large they are.

The distinction between puncturing and rate-matching is the heart of it. HARQ-ACK and RI are tiny but critical: a wrong HARQ-ACK derails the downlink retransmission logic entirely. So they are punctured in — the UE maps its data first as if the UCI were not there, then overwrites the specific resource elements next to the DMRS with the ACK/RI bits. Because those symbols sit right beside the reference signal, they enjoy the best channel estimate. CQI/PMI, by contrast, is larger and less time-critical, so it is rate-matched with the data: the data rate-matching accounts for the REs the CQI will occupy, and the two are jointly coded around each other with no destructive overwrite.

How many resource elements each UCI type gets is governed by its beta offset (betaOffset-ACK-Index, betaOffset-RI-Index, betaOffset-CQI-Index), configured by RRC. A larger beta offset reserves more REs for that UCI, spending PUSCH data capacity to make the control information more robust; the eNodeB tunes these per UE according to how reliable the feedback needs to be. When the UE must send only HARQ-ACK with no data (for example acknowledging a downlink SPS release) it can be given a small PUSCH grant purely to carry the ACK.

MCS, transport block size and power control

Modulation on PUSCH is QPSK, 16QAM or 64QAM (with 256QAM added for capable UEs in later releases), selected by the MCS field in the grant according to how good the uplink channel is. The uplink reuses the same MCS and I_TBS (transport block size index) machinery as the downlink: the 5-bit MCS field maps to a modulation order and an I_TBS value, and I_TBS together with the number of allocated resource blocks (N_PRB) indexes the transport block size table in TS 36.213 to yield the exact number of information bits the UE will carry. So the grant's MCS and RIV together fully determine the transport block size.

PUSCH transmit power is set by closed-loop power control. The UE computes its transmit power from an open-loop baseline — a cell-configured target received power P0 plus a fraction alpha of the estimated path loss, scaled by the number of allocated RBs and the MCS-dependent term — and then adds a closed-loop correction accumulated from the TPC commands the eNodeB sends. Each TPC command nudges the power up or down in steps; the UE accumulates them so the eNodeB can steer the received power to its target as the channel and interference change. Two effects fall out of this. First, power scales with allocation size: a wider grant needs proportionally more power to keep the same power-spectral-density at the receiver. Second, when the UE hits its maximum transmit power P_CMAX it becomes power-limited — it cannot honour a large grant at full PSD, which is exactly the situation the PHR (Power Headroom Report) exists to warn the eNodeB about so it can shrink future grants.

Here M_PRB is the allocation size in RBs, P0 combines a cell-common and a UE-specific term (p0-NominalPUSCH + p0-UE-PUSCH), alpha (0, 0.4…1 in the alpha enum) is the fractional path-loss compensation factor, PL is the downlink path-loss estimate, and f(TPC) is the accumulated closed-loop term. The SRS transmissions described earlier feed straight into this loop: by sounding the wideband channel the eNodeB learns the path loss and frequency-selective gains it needs to pick both the MCS (and hence I_TBS) and the closed-loop power corrections intelligently, rather than blindly.

Clustered SC-FDMA and later releases

The contiguous-only rule is a Release-8 constraint, and it costs the scheduler flexibility: a UE can only be placed where a single contiguous block is free. Release 10 relaxes it with clustered SC-FDMA (multi-cluster PUSCH), which lets a single UE's PUSCH occupy up to two non-adjacent clusters of contiguous RBs, signalled through DCI format 4 (and resource allocation Type 1 for the uplink). This gives the scheduler more freedom to fit UEs around each other and to exploit frequency-selective gains, at the cost of a modest PAPR increase — two clusters raise the peak-to-average ratio compared with one, but far less than fully arbitrary OFDMA allocation would, so the coverage benefit of DFT-spreading is largely retained.

Release 10 also introduces uplink spatial multiplexing (SU-MIMO) on PUSCH — up to two transport blocks and four layers with a codebook-based precoder, again carried by DCI format 4 — and simultaneous PUCCH and PUSCH transmission for capable UEs, which loosens the original "never transmit both at once" rule when the UE's power class allows it. Even with these additions, the core identity of PUSCH holds: a DFT-spread, low-PAPR, scheduled uplink channel whose fundamentals were set to protect the handset's power budget and the cell edge.

Summary

PUSCH is LTE's scheduled uplink data channel, carrying the UL-SCH plus MAC control (BSR, PHR) and, when needed, piggybacked UCI. Its defining choice is the SC-FDMA waveform: a DFT precoding stage flattens the envelope so the handset's amplifier can radiate more average power for the same battery drain, protecting cell-edge coverage — at the cost of a hard contiguous-allocation rule and RB counts restricted to products of 2, 3 and 5. The grant (DCI format 0, addressed by C-RNTI) carries a start-and-length RIV, the MCS/RV, NDI, TPC, DMRS cyclic shift and a hopping flag.

PUSCH is demodulated from DMRS on symbol 3 of each slot, scheduled with fixed n+4 timing, and retransmitted under synchronous HARQ (a fixed 8-subframe period, 8 processes, ACK/NACK on PHICH, process number implied by the subframe). UCI rides PUSCH by puncturing the critical HARQ-ACK/RI next to the DMRS and rate-matching the larger CQI/PMI with the data, sized by RRC beta offsets. Transport block size reuses the downlink two-step MCS → I_TBS → TBS machinery, and transmit power follows an open-loop-plus-closed-loop rule capped at P_CMAX, with SRS feeding the scheduler's MCS and power decisions. Later releases add clustered SC-FDMA, uplink SU-MIMO and simultaneous PUCCH+PUSCH via DCI format 4. NR carries the low-PAPR option forward as switchable DFT-s-OFDM but defaults to CP-OFDM and drops synchronous HARQ and PHICH.