Datasets:
Tasks:
Other
Formats:
parquet
Size:
100K - 1M
Tags:
wireless
physical-layer-security
covert-communication
low-probability-of-detection
virtual-mimo
anomaly-detection
License:
| """CP-OFDM and configurable variants. | |
| Default grid: 64-point FFT, 16-sample cyclic prefix, 4 OFDM symbols -> 4*80 = | |
| 320 samples. 48 occupied subcarriers (logical frequencies -24..-1, 1..24; DC | |
| and band edges null) -> 0.75 fractional bandwidth. | |
| Pilot patterns (both give 48 pilot REs + 144 data REs, matched resources, but a | |
| different pilot signature for Eve): | |
| * "block": symbol 0 is a full pilot symbol; symbols 1..n_sym-1 carry data | |
| (48 pilot REs, (n_sym-1)*n_occ data REs). | |
| * "comb": every ``comb_spacing``-th subcarrier is a pilot in every symbol; | |
| the rest carry data (n_sym*n_pilot_sub pilot REs, n_sym*n_data_sub data). | |
| Variants are subclasses that set the grid/pilot class attributes. DFT-s-OFDM | |
| and OTFS subclass this and override the grid-filling. | |
| """ | |
| from __future__ import annotations | |
| import numpy as np | |
| from ..pilots import pilot_matrix | |
| from .base import Waveform | |
| class OFDM(Waveform): | |
| name = "ofdm" | |
| n_fft = 64 | |
| n_cp = 16 | |
| n_occ = 48 | |
| n_sym = 4 | |
| pilot_pattern = "block" | |
| comb_spacing = 4 | |
| def __init__(self, cfg): | |
| assert cfg.n_samples == self.n_sym * (self.n_fft + self.n_cp), "grid must fill T samples" | |
| logical = np.concatenate([np.arange(-self.n_occ // 2, 0), np.arange(1, self.n_occ // 2 + 1)]) | |
| self.occ = logical % self.n_fft # ascending logical frequency | |
| if self.pilot_pattern == "comb": | |
| self.pilot_sub = np.arange(0, self.n_occ, self.comb_spacing) | |
| self.data_sub = np.array([k for k in range(self.n_occ) if k not in set(self.pilot_sub)]) | |
| self._n_pilot_re = len(self.pilot_sub) * self.n_sym | |
| assert cfg.n_data == len(self.data_sub) * self.n_sym | |
| else: # block | |
| self._n_pilot_re = self.n_occ | |
| assert cfg.n_data == (self.n_sym - 1) * self.n_occ | |
| self.pilots = pilot_matrix(cfg.pilot_scheme, cfg.n_tx, self._n_pilot_re) | |
| super().__init__(cfg) | |
| def n_pilot_re(self) -> int: | |
| return self._n_pilot_re | |
| # -- grid <-> time --------------------------------------------------------- | |
| def _grid_to_time(self, grid): | |
| """(..., M, n_sym, n_occ) -> (..., M, T)""" | |
| f = np.zeros(grid.shape[:-1] + (self.n_fft,), dtype=np.complex128) | |
| f[..., self.occ] = grid | |
| x = np.fft.ifft(f, axis=-1, norm="ortho") | |
| x = np.concatenate([x[..., -self.n_cp :], x], axis=-1) # (..., n_sym, n_fft+n_cp) | |
| return x.reshape(x.shape[:-2] + (self.cfg.n_samples,)) | |
| def _time_to_grid(self, u): | |
| """(..., M, T) -> (..., M, n_sym, n_occ)""" | |
| x = u.reshape(u.shape[:-1] + (self.n_sym, self.n_fft + self.n_cp))[..., self.n_cp :] | |
| return np.fft.fft(x, axis=-1, norm="ortho")[..., self.occ] | |
| # -- grid filling (overridden by DFT-s-OFDM / OTFS) ------------------------ | |
| def _fill_grid(self, s): | |
| if self.pilot_pattern == "comb": | |
| return self._fill_comb(s) | |
| data = s.reshape(s.shape[:-1] + (self.n_sym - 1, self.n_occ)) | |
| pil = np.broadcast_to(self.pilots[:, None, :], s.shape[:-1] + (1, self.n_occ)) | |
| return np.concatenate([pil, data], axis=-2) | |
| def _extract_data(self, grid): | |
| if self.pilot_pattern == "comb": | |
| return self._extract_comb(grid) | |
| d = grid[..., 1:, :] | |
| return d.reshape(d.shape[:-2] + (self.cfg.n_data,)) | |
| def _fill_comb(self, s): | |
| lead = s.shape[:-1] | |
| grid = np.zeros(lead + (self.n_sym, self.n_occ), dtype=np.complex128) | |
| data = s.reshape(lead + (self.n_sym, len(self.data_sub))) | |
| pil = np.broadcast_to( | |
| self.pilots.reshape(self.cfg.n_tx, self.n_sym, len(self.pilot_sub)), | |
| lead + (self.n_sym, len(self.pilot_sub)), | |
| ) | |
| grid[..., self.data_sub] = data | |
| grid[..., self.pilot_sub] = pil | |
| return grid | |
| def _extract_comb(self, grid): | |
| d = grid[..., self.data_sub] | |
| return d.reshape(d.shape[:-2] + (self.cfg.n_data,)) | |
| def _modulate_raw(self, s): | |
| return self._grid_to_time(self._fill_grid(s)) | |
| def demodulate(self, u): | |
| return self._extract_data(self._time_to_grid(np.asarray(u, dtype=np.complex128))) | |
| class OFDMComb(OFDM): | |
| name = "ofdm_comb" | |
| pilot_pattern = "comb" | |
| comb_spacing = 4 # 12 pilot + 36 data subcarriers per symbol (matched: 48 pilot, 144 data) | |
| class OFDM48(OFDM): | |
| """Smaller-FFT, more-symbol variant (T=320, 144 data, 0.75 BW). | |
| Block pilots need n_occ (=36) divisible by M, so this variant is valid only | |
| for M in {4, 6, 9, 12, 18, 36} -- not the default M=8. Not in the default | |
| registry; instantiate directly when using a compatible M. | |
| """ | |
| name = "ofdm48" | |
| n_fft = 48 | |
| n_cp = 16 | |
| n_occ = 36 | |
| n_sym = 5 | |