OFDMA vs SC-FDMA

An OFDM signal is the sum of many subcarriers, each carrying its own modulated symbol. Now and then those subcarriers line up in phase and the time-domain waveform hits a sharp peak well above its average level. That gap between peak and average is the peak-to-average power ratio (PAPR), and it decides how hard you can drive a power amplifier.

An amplifier is most efficient near saturation, but a high-PAPR signal can't go there — the peaks would clip and splatter energy into neighbouring channels. So you back the amplifier off, leaving headroom for the peaks. That back-off wastes power and heat.

On the downlink that trade is acceptable. The eNB is mains-powered with a large amplifier, so OFDMA's high PAPR costs efficiency the operator can afford. On the uplink it isn't. A handset has a small battery and a tiny amplifier, and the uplink is usually the link that limits cell range. Forcing the UE to back off would shorten battery life and shrink coverage — exactly what you don't want. That single asymmetry is why LTE picked two different access schemes for the two directions.

SC-FDMA is best read as its other name: DFT-spread OFDM (DFT-s-OFDM). It runs the normal OFDM chain with one extra block at the front.

First the data symbols pass through an M-point DFT (the spreading or precoding step). The output is mapped onto the UE's allocated subcarriers — kept contiguous so the property survives — then through the standard IFFT, cyclic-prefix insertion, and transmission. Because every input symbol is smeared across all the allocated subcarriers instead of riding on just one, the resulting time-domain signal behaves like a single-carrier transmission, with the lower PAPR that comes with it.

The cost moves to the receiver, which suits the design. The eNB undoes the DFT spreading and handles the frequency-domain equalisation — extra processing that a base station can carry, and that you would not want to push onto the handset. The cyclic prefix is still there, so SC-FDMA keeps OFDM's clean handling of multipath while trimming the peaks.

5G NR didn't keep the strict OFDMA-down / SC-FDMA-up split. NR uses CP-OFDM for the downlink and, by default, for the uplink too — partly because CP-OFDM pairs more naturally with MIMO and flexible numerologies, and partly because UE amplifiers improved.

But the PAPR problem never went away for coverage-limited users, so NR keeps DFT-s-OFDM — the same DFT-spreading idea as LTE's SC-FDMA — as an optional uplink waveform. The network can switch a UE to DFT-s-OFDM on the PUSCH when it sits at the cell edge and needs every dB of uplink power, then move it back to CP-OFDM closer in, where higher throughput and multi-layer MIMO matter more. So the LTE lesson carries over: pick the high-efficiency waveform when power is plentiful, and the low-PAPR one when reach is on the line.