Doppler in 5G NR: shift, spread, and the mobility problem.

Two different things: shift and spread.

Doppler does two distinct things to a signal, and conflating them is where confusion starts.

The shift is the familiar one: a transmitter and receiver closing at velocity v see the carrier moved by f_d = (v/c)·f_c. It is a single, bulk carrier-frequency offset (CFO) — a bias. You can estimate it and subtract it. The spread is subtler: in a real channel the signal arrives by many paths, each with a different relative velocity, so a single transmitted tone comes back as a little band of frequencies, roughly 2·f_d,max wide. That spread is a measure of how quickly the channel is changing — its coherence time — and it cannot be subtracted away, only tracked.

Why OFDM is sensitive to it.

5G is OFDM, and OFDM leans entirely on its subcarriers being orthogonal — each one peaks exactly where its neighbours are zero. A frequency shift slides every subcarrier off that alignment, so each one starts picking up energy from the others. That leakage is inter-carrier interference (ICI), and it rises with the ratio of Doppler to subcarrier spacing. As long as f_d ≪ SCS the system barely notices; when the Doppler becomes an appreciable fraction of the spacing, the orthogonality that OFDM depends on quietly erodes.

Why 5G cares more than LTE did.

Two things pushed Doppler up the priority list. First, 5G runs at much higher carrier frequencies — mid-band at 3.5 GHz and mmWave at 28 GHz and above — and Doppler scales directly with carrier frequency, so the same speed produces far more shift. A car at 120 km/h sees a few hundred hertz at 3.5 GHz but several kilohertz at 28 GHz.

Second, 5G deliberately took on the hard mobility cases: high-speed rail at 350–500 km/h, and non-terrestrial networks. NTN is the extreme — a low-earth-orbit satellite moves at roughly 7.5 km/s relative to the ground, producing Doppler of tens of kilohertz that changes continuously as the satellite crosses the sky. That is a different regime from anything terrestrial, and it dominates the NTN physical-layer design.

The defences.

There is no single fix; 5G layers several.

Wider subcarrier spacing. Because ICI scales with f_d / SCS, choosing a higher numerology (30, 60 or 120 kHz) makes a given Doppler a smaller fraction of the spacing. This is the tightest link between Doppler and the frame structure. Denser DM-RS. More demodulation pilots in time let the receiver re-estimate a fast-changing channel before it goes stale. PT-RS. The phase-tracking reference signal follows phase drift and residual offset, and earns its keep at mmWave where both are worst. CFO loops. The receiver estimates the bulk shift — from the sync signals and pilots — and corrects it continuously. NTN pre-compensation. For satellites the shift is too large and too predictable to leave to the receiver, so the UE and gNB pre-correct it using the satellite’s ephemeris and the UE’s own GNSS position before transmitting.

The numerology trade-off.

Reaching for wider subcarrier spacing to beat Doppler is not free. A wider spacing means a shorter symbol and a shorter cyclic prefix, which tolerates less delay spread and can cost coverage; it also raises overhead. So the numerology choice is a balance between the Doppler you must survive and the delay spread and range you need — which is exactly why NR made subcarrier spacing scalable in the first place rather than fixing it as LTE did.

What it costs to build.

Doppler handling is receiver DSP. A frequency-synchronisation path has to acquire and then track the offset within the sync budget; the channel estimator has to interpolate across pilots that are now spaced for a fast channel; the PT-RS path adds phase tracking; and for NTN the modem has to run a pre-compensation calculation against ephemeris and GNSS in real time. Get the loops and the pilot processing right and a link holds at 500 km/h or to a satellite; get them wrong and it falls over long before the headline data rate is the problem.