An M-point DFT before the IFFT
Ordinary OFDM puts one symbol on each subcarrier, so the IFFT sums many independent symbols — occasionally they all align and create a huge peak. Transform precoding first runs the M data symbols through an M-point DFT and maps the result to M contiguous subcarriers. Now every time-domain sample is essentially one spread symbol — the peaks vanish.
Symbols
d(0)…d(M−1)
M-point DFT
spread
Subcarrier map
M contiguous
IFFT
N-point
+ CP
transmit
Time-domain envelope
Same data, two waveforms. CP-OFDM swings wildly (high PAPR); DFT-s-OFDM stays flat (low PAPR), so the PA can be driven closer to saturation without clipping — that is the coverage gain.
π/2-BPSK
For the deepest coverage, transform precoding pairs with π/2-BPSK: every other symbol is rotated by 90°, which avoids transitions through the origin. No zero-crossings means an even flatter envelope and the lowest PAPR of any NR modulation. Toggle the rotation.