What phase noise actually does
A perfect oscillator spins at a constant rate. A real one jitters — its phase does a slow random walk. Over one OFDM symbol that shows up mostly as a common phase error (CPE): every subcarrier rotates by nearly the same angle. The catch is the walk accumulates across symbols, so a single estimate at the start of the slot goes stale. Crank the noise and watch the phase drift.
De-rotating the constellation
PT-RS symbols are known, so the receiver measures the per-symbol rotation from them and spins it back out — exactly as DM-RS does for the channel, but updated every few symbols to chase the drift. Toggle PT-RS and watch a smeared 64-QAM cloud tighten.
CP-OFDM pilots vs DFT-s-OFDM groups
This is the uplink technical twist most miss. Under CP-OFDM, PT-RS is a frequency-domain pilot: one subcarrier on every K-th RB, on every L-th symbol. Under DFT-s-OFDM it can't be — that would break the single-carrier property — so PT-RS is inserted in the time domain before the DFT, as groups of samples. Flip between them.
Time density L & frequency density K
The network ties the two densities to the impairment. L (every L-th symbol) comes from the MCS via thresholds ptrs-MCS1..4 — higher MCS ⇒ denser (smaller L). K (every K-th RB) comes from the scheduled bandwidth via NRB0..1 — wider band ⇒ sparser (larger K), because CPE is common across subcarriers. Move the sliders across the thresholds.
Time density L ← MCS
Frequency density K ← bandwidth
Resulting CP-OFDM pattern & overhead
PT-RS groups & samples
Under transform precoding, PT-RS is defined by a number of groups and samples per group, both set by the scheduled bandwidth (TS 38.214 Table 6.2.3.2-1). The groups are spread across the DFT block so the receiver can track phase along the single-carrier symbol.