dspk.ops.resample

Sample-rate changes on the device, with scipy.signal.decimate / resample_poly semantics.

decimate	Lowers the sample rate by an integer factor, behind an anti-alias filter.
resample_poly	Changes the sample rate by up / down with a polyphase FIR filter.

Functions

decimate

decimate(
    data,
    q: int,
    *,
    ftype: str = 'iir',
    order: int = DECIMATE_DEFAULT_ORDER,
    ripple_db: float = DECIMATE_DEFAULT_RIPPLE_DB,
    taps_per_factor: int = DECIMATE_FIR_TAPS_PER_FACTOR,
    runtime: str | None = None,
) -> np.ndarray

Lowers the sample rate by an integer factor, behind an anti-alias filter.

As scipy.signal.decimate(x, q, ftype=ftype, zero_phase=True).

Parameters

• data : array_like
  Signal (channels, samples). float32, int8, int16, uint16 and int32 arrays go to the device as they are and are converted there; other dtypes and lists become float32 first.
• q : int
  Factor: every q-th sample is kept, ceil(samples / q) samples out.
• ftype : {"iir", "fir"}, default 'iir'
"iir": a zero-phase Chebyshev I of order and ripple_db, cut at DECIMATE_IIR_CUTOFF of the new Nyquist. "fir": a Hamming firwin of taps_per_factor * q + 1 taps, applied as resample_poly does.
• order : int, default DECIMATE_DEFAULT_ORDER (8)
  Order of the IIR filter.
• ripple_db : float, default DECIMATE_DEFAULT_RIPPLE_DB (0.05)
  Passband ripple of the IIR filter, in dB.
• taps_per_factor : int, default DECIMATE_FIR_TAPS_PER_FACTOR (20)
  FIR taps per unit of q.
• runtime : str, default None
  A runtime from runtime.available ("cuda", "hip", "wgpu", "cpu") for this call. By default: the one chosen with runtime.set, else DSPK_RUNTIME, else the first available (GPUs first).

Returns

• np.ndarray — (channels, ceil(samples / q)), float32.

Raises

• ValueError: For q below 1, an unknown ftype, or data not 2-D.
• RuntimeError: When the runtime has no device.

Notes

With "iir", samples near the ends can differ slightly from scipy’s, which pads the edges differently; the interior matches.

Examples

>>> import numpy as np
>>> import dspk
>>> x = np.random.default_rng(0).standard_normal((4, 30_000)).astype(np.float32)
>>> lfp = dspk.ops.resample.decimate(x, 30)
>>> lfp.shape
(4, 1000)

resample_poly

resample_poly(
    data,
    up: int,
    down: int,
    *,
    beta: float = RESAMPLE_POLY_BETA,
    edge: str = RESAMPLE_POLY_DEFAULT_EDGE,
    runtime: str | None = None,
) -> np.ndarray

Changes the sample rate by up / down with a polyphase FIR filter.

As scipy.signal.resample_poly: up-sample by up, low-pass with a Kaiser-windowed FIR, down-sample by down.

Parameters

• data : array_like
  Signal (channels, samples). float32, int8, int16, uint16 and int32 arrays go to the device as they are and are converted there; other dtypes and lists become float32 first.
• up : int
  Up-sampling factor.
• down : int
  Down-sampling factor.
• beta : float, default RESAMPLE_POLY_BETA (5.0)
  Kaiser window shape (larger: less ripple, wider transition).
• edge : {"zeros", "odd", "reflect", "nearest"}, default RESAMPLE_POLY_DEFAULT_EDGE (‘zeros’)
  What the samples past each end are taken to be.
• runtime : str, default None
  A runtime from runtime.available ("cuda", "hip", "wgpu", "cpu") for this call. By default: the one chosen with runtime.set, else DSPK_RUNTIME, else the first available (GPUs first).

Returns

• np.ndarray — (channels, ceil(samples * up / down)), float32.

Raises

• ValueError: For up or down below 1, an unknown edge, or data not 2-D.
• RuntimeError: When the runtime has no device.

Examples

>>> import numpy as np
>>> import dspk
>>> x = np.random.default_rng(0).standard_normal((4, 30_000)).astype(np.float32)
>>> y = dspk.ops.resample.resample_poly(x, 2, 3)
>>> y.shape
(4, 20000)

Constants

Name	Type	Value	Description
DECIMATE_DEFAULT_ORDER	int	8
DECIMATE_DEFAULT_RIPPLE_DB	float	0.05
DECIMATE_FIR_TAPS_PER_FACTOR	int	20
DECIMATE_IIR_CUTOFF	float	0.8
RESAMPLE_POLY_BETA	float	5.0
RESAMPLE_POLY_DEFAULT_EDGE	str	'zeros'
RESAMPLE_POLY_HALF_LEN_PER_RATE	int	10