🅼 oscfar

Functions

  1. 🅵 oscfar.do_os_cfar

🅵 oscfar.do_os_cfar 

def do_os_cfar(data: np.ndarray, guard_cells, train_cells, rank_k, threshold_factor, averaging, min_dist, min_snr, baseline, smoothing = 'mean', lowess_frac = 0.01, clustering = False, clustering_cutoff = 0.7):

Perform OS-CFAR detection and post-processing on 2D array data.

This function applies Ordered Statistics CFAR (OS-CFAR) detection to a 2D input array (e.g., a spectrogram), typically by summing along one axis to produce a 1D time series. It then applies smoothing, peak detection, SNR filtering, baseline removal, and optional clustering.

Parameters:

  • data (np.ndarray): 2D input array (e.g., frequency x time), assumed to be linear power.
  • guard_cells (int): Number of guard cells on each side of the cell under test (CUT).
  • train_cells (int): Number of training cells on each side of the CUT.
  • rank_k (float): Fraction (0-1) of training cells to use for the OS-CFAR rank (e.g., 0.75 for 75th percentile).
  • threshold_factor (float): Scaling factor (alpha) for the OS-CFAR threshold.
  • averaging (int): Window size for initial smoothing (mean or median).
  • min_dist (int): Minimum distance (in samples) between detected peaks.
  • min_snr (float): Minimum SNR required for a peak to be kept.
  • baseline (float): Threshold factor for removing peaks close to the baseline.
  • smoothing (str): Smoothing method for the time series (‘mean’, ‘median’, ‘lowess’, or None). Default is ‘mean’.
  • lowess_frac (float): Fraction of data for LOWESS smoothing if used. Default is 0.01.
  • clustering (int or bool): If int > 0, applies DBSCAN clustering to peaks with this as min_samples. Default is False.
  • clustering_cutoff (float): Cutoff value for clustering. Default is 0.7.

Returns:

  • Peaks: Peaks object containing detected peak indices and the OS-CFAR threshold array.

Raises:

  • TypeError: If input data is not a 2D NumPy array.