Download processing_code/utils/extract_radar_data.py from phi-lab-rice/GRADE_Dataset: direct link, hf CLI and curl.
- Browser
- Download file 4.12 kB
-
https://huggingface.co/datasets/phi-lab-rice/GRADE_Dataset/resolve/main/processing_code/utils/extract_radar_data.py
- Command line
-
hf download hf://datasets/phi-lab-rice/GRADE_Dataset/processing_code/utils/extract_radar_data.py
-
curl -L -o extract_radar_data.py https://huggingface.co/datasets/phi-lab-rice/GRADE_Dataset/resolve/main/processing_code/utils/extract_radar_data.py
4.12 kB
| import numpy as np | |
| def iiqq_to_iq(iiqq: np.ndarray) -> np.ndarray: | |
| """ | |
| Convert interleaved IIQQ radar data to complex IQ format. | |
| Args: | |
| iiqq: Input array with shape [..., N] where N is divisible by 4 | |
| Returns: | |
| Complex array with shape [..., N/2] and dtype complex64 | |
| """ | |
| if iiqq.dtype == np.complex64: | |
| return iiqq | |
| shape = (*iiqq.shape[:-1], iiqq.shape[-1] // 2) | |
| iq = np.zeros(shape, dtype=np.complex64) | |
| iq[..., 0::2] = 1j * iiqq[..., 0::4] + iiqq[..., 2::4] | |
| iq[..., 1::2] = 1j * iiqq[..., 1::4] + iiqq[..., 3::4] | |
| return iq | |
| def mimo(data: np.ndarray) -> np.ndarray: | |
| """ | |
| Convert raw radar data to MIMO virtual array format. | |
| Maps (batch, doppler, tx, rx, range) -> (batch, doppler, elevation, azimuth, range). | |
| Args: | |
| data: Input array with shape [..., tx, rx, range] | |
| tx=3, rx=4 | |
| Returns: | |
| Virtual array with shape [..., 2, 8, range] | |
| """ | |
| if len(data.shape) == 5: | |
| # batch, doppler, tx, rx, range = data.shape | |
| # mimo_data = np.zeros((batch, doppler, 2, 8, range), dtype=np.complex64) | |
| # mimo_data[:, :, 0, 2:6, :] = data[:, :, 1, :, :] | |
| # mimo_data[:, :, 1, 0:4, :] = data[:, :, 0, :, :] | |
| # mimo_data[:, :, 1, 4:8, :] = data[:, :, 2, :, :] | |
| raise NotImplementedError(f"Batch operation not supported.") | |
| elif len(data.shape) == 4: | |
| doppler, tx, rx, range = data.shape | |
| mimo_data = np.zeros((doppler, 2, 8, range), dtype=np.complex64) | |
| mimo_data[:, 0, 2:6, :] = data[:, 1, :, :] | |
| mimo_data[:, 1, 0:4, :] = data[:, 0, :, :] | |
| mimo_data[:, 1, 4:8, :] = data[:, 2, :, :] | |
| else: | |
| raise ValueError(f"Expected 4D or 5D input, got shape {data.shape}") | |
| return mimo_data | |
| def fft_w_shift(data: np.ndarray, dim: int = -1, shift: bool = False) -> np.ndarray: | |
| """ | |
| Apply FFT to data with shift to center the zero frequency component. | |
| """ | |
| fft_array = np.fft.fft(data, axis=dim) | |
| if shift: | |
| fft_array = np.fft.fftshift(fft_array, axes=dim) | |
| return fft_array | |
| def range_doppler_fft(data: np.ndarray) -> np.ndarray: | |
| """ | |
| Apply FFT to data in range and doppler dimensions. | |
| """ | |
| range_fft = fft_w_shift(data, dim=-1, shift=False) | |
| doppler_fft = fft_w_shift(range_fft, dim=0, shift=True) | |
| return doppler_fft | |
| def azimuth_elevation_fft(data: np.ndarray) -> np.ndarray: | |
| """ | |
| Apply FFT to data in azimuth and elevation dimensions. | |
| """ | |
| # Axis 1 is Elevation (size 2), Axis 2 is Azimuth (size 8) | |
| elevation_fft = fft_w_shift(data, dim=1, shift=True) | |
| azimuth_fft = fft_w_shift(elevation_fft, dim=2, shift=True) | |
| return azimuth_fft | |
| def process_single_frame(iiqq: np.ndarray, no_doppler: bool = False) -> np.ndarray: | |
| """ | |
| Process a single (non-batched) frame of radar data. | |
| Args: | |
| iiqq: Raw data with shape (doppler, tx, rx, range). | |
| no_doppler: Keep only chirp 0 as a size-one Doppler axis and skip the | |
| Doppler FFT. This is intentionally applied before any Doppler | |
| transform; it is not a slice of the full Doppler FFT output. | |
| Returns: | |
| A complex cube with shape (doppler, elevation, azimuth, range). The | |
| Doppler axis has length one when ``no_doppler`` is true. | |
| """ | |
| iq_data = iiqq_to_iq(iiqq) | |
| # Maps (doppler, tx, rx, range) -> (doppler, elevation, azimuth, range). | |
| mimo_data = mimo(iq_data) | |
| if no_doppler: | |
| # Preserve only the first slow-time sample and perform the range FFT. | |
| # Keeping the leading dimension makes the output layout consistent | |
| # with the full range-Doppler processing path. | |
| range_doppler_fft_data = fft_w_shift(mimo_data[0:1], dim=-1, shift=False) | |
| else: | |
| # Apply range-doppler FFT | |
| range_doppler_fft_data = range_doppler_fft(mimo_data) | |
| # Apply azimuth-elevation FFT | |
| azimuth_elevation_fft_data = azimuth_elevation_fft(range_doppler_fft_data) | |
| return azimuth_elevation_fft_data | |