The design question 3GPP actually faced
LTE’s uplink was a single, non-negotiable decision: every PUSCH transmission uses SC-FDMA. 5G NR does not make that same trade-off unconditionally — it exposes the choice as a per-configuration parameter. That’s a more interesting design decision than “5G moved to OFDM,” and it’s worth understanding why 3GPP declined to simply carry SC-FDMA forward.
The short version: LTE’s uplink was designed around one dominant constraint — UE power efficiency. 5G NR’s uplink has to serve simultaneously as a power-limited link (cell-edge devices, IoT, mmWave-constrained UEs) and a capacity-maximizing link (massive MIMO, wide bandwidths, dense scheduling). No single waveform optimizes for both. So NR standardized two, and pushed the selection decision to the network.
Where the PAPR problem actually comes from
An OFDM signal combines many independently modulated, phase-aligned subcarriers. Occasionally, these subcarriers align constructively, causing instantaneous peaks that far exceed the average envelope—resulting in a high peak-to-average power ratio. For a base station with a robust PA and fixed power budget, managing this is straightforward. However, for battery-powered user equipment (UE), it’s more challenging: the PA must stay linear across the full peak range, requiring it to back off from saturation. This backoff, known as Maximum Power Reduction, directly lowers the UE’s effective transmit power, not just its efficiency.
LTE’s response and the reasons NR did not adopt it entirely.
LTE addressed this by DFT-precoding the modulation symbols before OFDM modulation, collapsing the multi-carrier envelope into something closer to single-carrier behavior — hence SC-FDMA, equivalently DFT-s-OFDM. It solved the PAPR problem cleanly, at a structural cost: DFT-precoded symbols are spread across the entire allocation, which constrains frequency-selective scheduling and — critically for 5G — is incompatible with multi-layer spatial multiplexing beyond a single stream.
That constraint was tolerable in LTE, where uplink MIMO ambitions were modest. It’s a real limitation in NR, where uplink is expected to support up to 4-layer (and in Release 18, up to 8-layer) spatial multiplexing for high-end devices, flexible numerology, mini-slot scheduling, and non-contiguous resource allocation. Locking every UE into a single-carrier-like waveform would have foreclosed exactly the capacity gains 5G was built to deliver.
3GPP’s resolution wasn’t to pick a winner — it was to standardize both and make the choice configurable.
The mechanism: transformPrecoder
The selection is carried in PUSCH-Config as transformPrecoder ENUMERATED {enabled, disabled} (TS 38.331):
disabled→ CP-OFDM. No DFT stage. Full scheduling flexibility, full MIMO layer support, but the UE bears OFDM’s native PAPR and the associated backoff.enabled→ DFT-s-OFDM. The processing chain gains one stage — DFT spreading before resource mapping — which restores the low-PAPR, near-single-carrier envelope at the cost of being limited to one antenna port and one spatial layer.
This is an RRC-configured property of the PUSCH, not a per-transmission or per-symbol choice. A given configured PUSCH uses one waveform for its lifetime; changing it requires reconfiguration. Release 18 introduces dynamic waveform switching — allowing the network to shift a UE between DFT-s-OFDM and CP-OFDM without a full RRC reconfiguration round-trip — aimed specifically at coverage-limited scenarios where the operating point needs to track channel conditions faster than RRC signaling allows. Public evidence of large-scale commercial use of this specific capability is still limited as of this writing.
What the trade-off actually buys you
Read as a decision matrix rather than a spec table: this is 3GPP encoding a capacity-vs-coverage trade-off directly into the waveform layer, rather than leaving it entirely to power control and MCS selection. That’s a meaningfully different design philosophy from LTE’s uplink, where the waveform was fixed, and every other lever (power control, scheduling, HARQ) had to compensate for a single, non-negotiable PAPR profile.
Where this actually lands in commercial networks
Most operational 5G NR deployments today default to CP-OFDM, for a straightforward reason: current commercial priorities — massive MIMO gains, dense urban capacity, flexible scheduling — are the exact axes CP-OFDM is built to serve, and mid-band macro geometries in most launched networks don’t yet demand the coverage extension DFT-s-OFDM offers. DFT-s-OFDM remains standardized and available precisely for the cases where that calculus flips: rural/large-cell deployments, power-constrained IoT-class devices, and mmWave uplink, where the coverage penalty of high PAPR is the binding constraint rather than a tolerable cost.
Three claims worth retiring
“5G moved away from SC-FDMA.” It didn’t — DFT-s-OFDM is SC-FDMA’s direct successor in the NR spec, available via transformPrecoder: enabled.
“5G always uses OFDM.” True for the downlink only. PUSCH is genuinely dual-waveform; PDSCH has no transform-precoding option at all.
“A UE freely switches waveforms mid-transmission.” No — waveform is a property of the configured PUSCH, changed only through (re)configuration, with Release 18’s dynamic switching narrowing that gap but not eliminating the concept of a configured state.
The takeaway
NR’s dual-waveform uplink is best read as an acknowledgment that “5G” isn’t one deployment profile — it’s simultaneously an urban capacity network and a coverage-extension network, often on the same carrier serving devices with very different power constraints. Rather than force one waveform to serve both cases adequately, 3GPP pushed the trade-off down to a configurable parameter and let the network — not the standard — decide where each UE sits on the capacity/coverage curve.