src/covcollab/formats/otfs.py

"""OTFS over the same 48x4 time-frequency grid as OFDM.

Delay-Doppler grid: 48 delay bins x 4 Doppler bins per user.
  * Embedded pilot: a user-orthogonal code-division sequence spread over the
    reserved guard region (delay bins 0..11 x all 4 Doppler bins = 48 REs),
    using the same exp(j 2 pi m k / M) sequences as the other formats.  Spreading
    the pilot over 48 REs (rather than a single delay-Doppler impulse) keeps the
    embedded-pilot structure and per-user energy but makes the pilot
    SCRAMBLE-ABLE: a common per-RE phase decorrelates the coherent sum, defeating
    Eve's pilot-matched detector, while Bob (sharing the PRN) de-scrambles and
    estimates by correlating against the known DD sequences.
  * Data: delay bins 12..47 x 4 Doppler bins = 144 REs.  Mapping order is
    delay-major: S[m, 4*i + k] -> (delay 12+i, Doppler k).

ISFFT (unitary): DFT along delay -> subcarrier, IDFT along Doppler -> time,
then the OFDM modulator (48 occupied subcarriers of a 64-FFT, CP 16) per
time symbol.  All transforms are unitary, so the loopback inverse is exact.
"""

from __future__ import annotations

import numpy as np

from .ofdm import OFDM


class OTFS(OFDM):
    name = "otfs"
    n_delay = 48
    n_dopp = 4
    pilot_guard = 12  # delay bins 0..11 reserved for the spread pilot

    def __init__(self, cfg):
        assert cfg.n_data == (self.n_delay - self.pilot_guard) * self.n_dopp
        super().__init__(cfg)  # OFDM base sets self.pilots = pilot_sequences(M, 48)

    @property
    def n_pilot_re(self) -> int:
        return self.pilot_guard * self.n_dopp  # 48 guard-region REs

    def _fill_grid(self, s):
        m = self.cfg.n_tx
        dd = np.zeros(s.shape[:-1] + (self.n_delay, self.n_dopp), dtype=np.complex128)
        dd[..., self.pilot_guard :, :] = s.reshape(
            s.shape[:-1] + (self.n_delay - self.pilot_guard, self.n_dopp)
        )
        # spread pilot: self.pilots (M, 48) -> guard region (M, 12, 4), broadcast over batch
        pil = self.pilots.reshape(m, self.pilot_guard, self.n_dopp)
        dd[..., : self.pilot_guard, :] = np.broadcast_to(
            pil, s.shape[:-1] + (self.pilot_guard, self.n_dopp)
        )
        # ISFFT: delay -> subcarrier (DFT, axis -2), Doppler -> time (IDFT, axis -1)
        tf = np.fft.fft(np.fft.ifft(dd, axis=-1, norm="ortho"), axis=-2, norm="ortho")
        return np.swapaxes(tf, -1, -2)  # (..., M, n_sym=4, n_occ=48)

    def _extract_data(self, grid):
        tf = np.swapaxes(grid, -1, -2)  # (..., M, delay-bins-as-subcarriers 48, time 4)
        dd = np.fft.fft(np.fft.ifft(tf, axis=-2, norm="ortho"), axis=-1, norm="ortho")
        d = dd[..., self.pilot_guard :, :]
        return d.reshape(d.shape[:-2] + (self.cfg.n_data,))