Array API Standard Support: `signal`#

This page explains some caveats of the signal module and provides (currently incomplete) tables about the CPU, GPU and JIT support.

Caveats#

JAX and CuPy provide alternative implementations for some signal functions. When such a function is called, a decorator decides which implementation to use by inspecting the xp parameter.

Hence, there can be, especially during CI testing, discrepancies in behavior between the default NumPy-based implementation and the JAX and CuPy backends. Skipping the incompatible backends in unit tests, as described in the Adding tests section, is the currently recommended workaround.

The functions are decorated by the code in file scipy/signal/_support_alternative_backends.py:

import functools
import types
from scipy._lib._array_api import (
    is_cupy, is_jax, scipy_namespace_for, SCIPY_ARRAY_API, xp_capabilities
)

from ._signal_api import *   # noqa: F403
from . import _signal_api
from . import _delegators
__all__ = _signal_api.__all__


MODULE_NAME = 'signal'

# jax.scipy.signal has only partial coverage of scipy.signal, so we keep the list
# of functions we can delegate to JAX
# https://jax.readthedocs.io/en/latest/jax.scipy.html
JAX_SIGNAL_FUNCS = [
    'fftconvolve', 'convolve', 'convolve2d', 'correlate', 'correlate2d',
    'csd', 'detrend', 'istft', 'welch'
]

# some cupyx.scipy.signal functions are incompatible with their scipy counterparts
CUPY_BLACKLIST = [
    'abcd_normalize', 'bessel', 'besselap', 'envelope', 'get_window', 'lfilter_zi',
    'sosfilt_zi', 'remez',
]

# freqz_sos is a sosfreqz rename, and cupy does not have the new name yet (in v13.x)
CUPY_RENAMES = {'freqz_sos': 'sosfreqz'}


def delegate_xp(delegator, module_name):
    def inner(func):
        @functools.wraps(func)
        def wrapper(*args, **kwds):
            try:
                xp = delegator(*args, **kwds)
            except TypeError:
                # object arrays
                if func.__name__ == "tf2ss":
                    import numpy as np
                    xp = np
                else:
                    raise

            # try delegating to a cupyx/jax namesake
            if is_cupy(xp) and func.__name__ not in CUPY_BLACKLIST:
                func_name = CUPY_RENAMES.get(func.__name__, func.__name__)

                # https://github.com/cupy/cupy/issues/8336
                import importlib
                cupyx_module = importlib.import_module(f"cupyx.scipy.{module_name}")
                cupyx_func = getattr(cupyx_module, func_name)
                kwds.pop('xp', None)
                return cupyx_func(*args, **kwds)
            elif is_jax(xp) and func.__name__ in JAX_SIGNAL_FUNCS:
                spx = scipy_namespace_for(xp)
                jax_module = getattr(spx, module_name)
                jax_func = getattr(jax_module, func.__name__)
                kwds.pop('xp', None)
                return jax_func(*args, **kwds)
            else:
                # the original function
                return func(*args, **kwds)
        return wrapper
    return inner


# Although most of these functions currently exist in CuPy and some in JAX,
# there are no alternative backend tests for any of them in the current
# test suite. Each will be documented as np_only until tests are added.
untested = {
    "argrelextrema",
    "argrelmax",
    "argrelmin",
    "band_stop_obj",
    "bode",
    "check_NOLA",
    "chirp",
    "coherence",
    "csd",
    "czt",
    "czt_points",
    "dbode",
    "dfreqresp",
    "dlsim",
    "dstep",
    "find_peaks",
    "find_peaks_cwt",
    "freqresp",
    "gausspulse",
    "iirdesign", # There's no reason this shouldn't work. It just needs tests.
    "istft",
    "lombscargle",
    "lsim",
    "max_len_seq",
    "peak_prominences",
    "peak_widths",
    "periodogram",
    "place_poles",
    "sawtooth",
    "sepfir2d",
    "ss2tf",
    "ss2zpk",
    "step",
    "sweep_poly",
    "symiirorder1",
    "symiirorder2",
    "tf2ss",
    "unit_impulse",
    "welch",
    "zoom_fft",
    "zpk2ss",
}


def get_default_capabilities(func_name, delegator):
    if delegator is None or func_name in untested:
        return xp_capabilities(np_only=True)
    return xp_capabilities()

bilinear_extra_note = \
    """CuPy does not accept complex inputs.

    """

uses_choose_conv_extra_note = \
    """CuPy does not support inputs with ``ndim>1`` when ``method="auto"``
    but does support higher dimensional arrays for ``method="direct"``
    and ``method="fft"``.

    """

resample_poly_extra_note = \
    """CuPy only supports ``padtype="constant"``.

    """

upfirdn_extra_note = \
    """CuPy only supports ``mode="constant"`` and ``cval=0.0``.

    """

xord_extra_note = \
    """The ``torch`` backend on GPU does not support the case where
    `wp` and `ws` specify a Bandstop filter.

    """

convolve2d_extra_note = \
    """The JAX backend only supports ``boundary="fill"`` and ``fillvalue=0``.

    """

zpk2tf_extra_note = \
    """The CuPy and JAX backends both support only 1d input.

    """

abcd_normalize_extra_note = \
    """The result dtype when all array inputs are of integer dtype is the
    backend's current default floating point dtype.

    """

capabilities_overrides = {
    "abcd_normalize": xp_capabilities(extra_note=abcd_normalize_extra_note),
    "bessel": xp_capabilities(cpu_only=True, jax_jit=False, allow_dask_compute=True),
    "bilinear": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                jax_jit=False, allow_dask_compute=True,
                                reason="Uses np.polynomial.Polynomial",
                                extra_note=bilinear_extra_note),
    "bilinear_zpk": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                    jax_jit=False, allow_dask_compute=True),
    "butter": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                              allow_dask_compute=True),
    "buttord": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                               jax_jit=False, allow_dask_compute=True,
                               extra_note=xord_extra_note),
    "cheb1ord": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                jax_jit=False, allow_dask_compute=True,
                                extra_note=xord_extra_note),
    "cheb2ord": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                jax_jit=False, allow_dask_compute=True,
                                extra_note=xord_extra_note),
    "cheby1": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                              allow_dask_compute=True),

    "cheby2": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                              allow_dask_compute=True),
    "cont2discrete": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "convolve": xp_capabilities(cpu_only=True, exceptions=["cupy", "jax.numpy"],
                                allow_dask_compute=True,
                                extra_note=uses_choose_conv_extra_note),
    "convolve2d": xp_capabilities(cpu_only=True, exceptions=["cupy", "jax.numpy"],
                                  allow_dask_compute=True,
                                  extra_note=convolve2d_extra_note),
    "correlate": xp_capabilities(cpu_only=True, exceptions=["cupy", "jax.numpy"],
                                 allow_dask_compute=True,
                                 extra_note=uses_choose_conv_extra_note),
    "correlate2d": xp_capabilities(cpu_only=True, exceptions=["cupy", "jax.numpy"],
                                   allow_dask_compute=True,
                                   extra_note=convolve2d_extra_note),
    "correlation_lags": xp_capabilities(out_of_scope=True),
    "cspline1d": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                 jax_jit=False, allow_dask_compute=True),
    "cspline1d_eval": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                      jax_jit=False, allow_dask_compute=True),
    "cspline2d": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                 jax_jit=False, allow_dask_compute=True),
    "deconvolve": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                  jax_jit=False, allow_dask_compute=True),
    "decimate": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "detrend": xp_capabilities(cpu_only=True, exceptions=["cupy", "jax.numpy"],
                               allow_dask_compute=True),
    "dimpulse": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "dlti": xp_capabilities(np_only=True,
                            reason="works in CuPy but delegation isn't set up yet"),
    "ellip": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                             allow_dask_compute=True,
                             reason="scipy.special.ellipk"),
    "ellipord": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                jax_jit=False, allow_dask_compute=True,
                                reason="scipy.special.ellipk"),
    "findfreqs": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 jax_jit=False, allow_dask_compute=True),
    "firls": xp_capabilities(cpu_only=True, allow_dask_compute=True, jax_jit=False,
                             reason="lstsq"),
    "firwin": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                              jax_jit=False, allow_dask_compute=True),
    "firwin2": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                               jax_jit=False, allow_dask_compute=True,
                               reason="firwin uses np.interp"),
    "fftconvolve": xp_capabilities(cpu_only=True,
                                   exceptions=["cupy", "jax.numpy", "torch"]),
    "freqs": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                             jax_jit=False, allow_dask_compute=True),
    "freqs_zpk": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 jax_jit=False, allow_dask_compute=True),
    "freqz": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                             jax_jit=False, allow_dask_compute=True),
    "freqz_sos": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 jax_jit=False, allow_dask_compute=True),
    "group_delay": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                   jax_jit=False, allow_dask_compute=True),
    "invres": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "invresz": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "iircomb": xp_capabilities(xfail_backends=[("jax.numpy", "inaccurate")]),
    "iirfilter": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 jax_jit=False, allow_dask_compute=True),
    "kaiser_atten": xp_capabilities(
        out_of_scope=True, reason="scalars in, scalars out"
    ),
    "kaiser_beta": xp_capabilities(out_of_scope=True, reason="scalars in, scalars out"),
    "kaiserord": xp_capabilities(out_of_scope=True, reason="scalars in, scalars out"),
    "lfilter": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                               allow_dask_compute=True, jax_jit=False),
    "lfilter_zi": xp_capabilities(cpu_only=True, allow_dask_compute=True,
                                  jax_jit=False),
    "lfiltic": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                               allow_dask_compute=True),
    "lp2bp": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                             allow_dask_compute=True, jax_jit=False),
    "lp2bp_zpk": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 allow_dask_compute=True, jax_jit=False),
    "lp2bs": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                             allow_dask_compute=True, jax_jit=False),
    "lp2bs_zpk": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 allow_dask_compute=True, jax_jit=False),
    "lp2lp": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                             allow_dask_compute=True, jax_jit=False),
    "lp2lp_zpk": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 allow_dask_compute=True, jax_jit=False),
    "lp2hp": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                             allow_dask_compute=True, jax_jit=False),
    "lp2hp_zpk": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 allow_dask_compute=True, jax_jit=False),
    "lti": xp_capabilities(np_only=True,
                            reason="works in CuPy but delegation isn't set up yet"),
    "medfilt": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                               allow_dask_compute=True, jax_jit=False,
                               reason="uses scipy.ndimage.rank_filter"),
    "medfilt2d": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                 allow_dask_compute=True, jax_jit=False,
                                 reason="c extension module"),
    "minimum_phase": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                     allow_dask_compute=True, jax_jit=False),
    "normalize": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                 jax_jit=False, allow_dask_compute=True),
    "oaconvolve": xp_capabilities(
        cpu_only=True, exceptions=["cupy", "torch"],
        xfail_backends=[("dask.array", "wrong answer")],
    ),
    "order_filter": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                    allow_dask_compute=True, jax_jit=False,
                                    reason="uses scipy.ndimage.rank_filter"),
    "qspline1d": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                 jax_jit=False, allow_dask_compute=True),
    "qspline1d_eval": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                      jax_jit=False, allow_dask_compute=True),
    "qspline2d": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "remez": xp_capabilities(cpu_only=True, allow_dask_compute=True, jax_jit=False),
    "resample": xp_capabilities(
        cpu_only=True, exceptions=["cupy"],
        skip_backends=[
            ("dask.array", "XXX something in dask"),
            ("jax.numpy", "XXX: immutable arrays"),
        ]
    ),
    "resample_poly": xp_capabilities(
        cpu_only=True, exceptions=["cupy"],
        jax_jit=False, skip_backends=[("dask.array", "XXX something in dask")],
        extra_note=resample_poly_extra_note,
    ),
    "residue": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "residuez": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "savgol_filter": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                     jax_jit=False,
                                     reason="convolve1d is cpu-only"),
    "sawtooth": xp_capabilities(jax_jit=False,
                                skip_backends=[("dask.array", "dask tests fail")]),
    "sepfir2d": xp_capabilities(np_only=True),
    "sos2zpk": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                               allow_dask_compute=True),
    "sos2tf": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                              allow_dask_compute=True),
    "sosfilt": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                               allow_dask_compute=True),
    "sosfilt_zi": xp_capabilities(cpu_only=True, allow_dask_compute=True,
                                  jax_jit=False),
    "sosfiltfilt": xp_capabilities(
        cpu_only=True, exceptions=["cupy"],
        skip_backends=[
            (
                "dask.array",
                "sosfiltfilt directly sets shape attributes on arrays"
                " which dask doesn't like"
            ),
            ("torch", "negative strides"),
            ("jax.numpy", "sosfilt works in-place"),
        ],
    ),
    "sosfreqz": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                jax_jit=False, allow_dask_compute=True),
    "spline_filter": xp_capabilities(cpu_only=True, exceptions=["cupy"],
                                     jax_jit=False, allow_dask_compute=True),
    "tf2sos": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                              allow_dask_compute=True),
    "tf2zpk": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                              allow_dask_compute=True),
    "unique_roots": xp_capabilities(np_only=True, exceptions=["cupy"]),
    "upfirdn": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                               allow_dask_compute=True,
                               reason="Cython implementation",
                               extra_note=upfirdn_extra_note),
    "vectorstrength": xp_capabilities(cpu_only=True, exceptions=["cupy", "torch"],
                                      allow_dask_compute=True, jax_jit=False),
    "wiener": xp_capabilities(cpu_only=True, exceptions=["cupy", "jax.numpy"],
                              allow_dask_compute=True, jax_jit=False,
                              reason="uses scipy.signal.correlate"),
    "zpk2sos": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                               allow_dask_compute=True),
    "zpk2tf": xp_capabilities(cpu_only=True, exceptions=["cupy"], jax_jit=False,
                              allow_dask_compute=True,
                              extra_note=zpk2tf_extra_note),
    "spectrogram": xp_capabilities(out_of_scope=True),  # legacy
    "stft": xp_capabilities(out_of_scope=True),  # legacy
    "istft": xp_capabilities(out_of_scope=True),  # legacy
    "check_COLA": xp_capabilities(out_of_scope=True),  # legacy
}


# ### decorate ###
for obj_name in _signal_api.__all__:
    bare_obj = getattr(_signal_api, obj_name)
    delegator = getattr(_delegators, obj_name + "_signature", None)

    if SCIPY_ARRAY_API and delegator is not None:
        f = delegate_xp(delegator, MODULE_NAME)(bare_obj)
    else:
        f = bare_obj

    if not isinstance(f, types.ModuleType):
        capabilities = capabilities_overrides.get(
            obj_name, get_default_capabilities(obj_name, delegator)
        )
        f = capabilities(f)  # pyrefly:ignore[not-callable]

    # add the decorated function to the namespace, to be imported in __init__.py
    vars()[obj_name] = f

Note that a function will only be decorated if the environment variable SCIPY_ARRAY_API is set and its signature is listed in the file scipy/signal/_delegators.py. E.g., for firwin, the signature function looks like this:

def firwin_signature(numtaps, cutoff, *args, **kwds):
    if isinstance(cutoff, int | float):
        xp = np_compat
    else:
        xp = array_namespace(cutoff)
    return xp

Support on CPU#

Legend

✔️ = supported

✖ = unsupported

N/A = out-of-scope

function/class	torch	jax	dask
CZT	✖	✖	✖
ShortTimeFFT	✖	✖	✖
StateSpace	✖	✖	✖
TransferFunction	✖	✖	✖
ZerosPolesGain	✖	✖	✖
ZoomFFT	✖	✖	✖
abcd_normalize	✔️	✔️	✔️
argrelextrema	✖	✖	✖
argrelmax	✖	✖	✖
argrelmin	✖	✖	✖
band_stop_obj	✖	✖	✖
bessel	✔️	✔️	✔️
besselap	✔️	✔️	✔️
bilinear	✔️	✔️	✔️
bilinear_zpk	✔️	✔️	✔️
bode	✖	✖	✖
buttap	✔️	✔️	✔️
butter	✔️	✔️	✔️
buttord	✔️	✔️	✔️
cheb1ap	✔️	✔️	✔️
cheb1ord	✔️	✔️	✔️
cheb2ap	✔️	✔️	✔️
cheb2ord	✔️	✔️	✔️
cheby1	✔️	✔️	✔️
cheby2	✔️	✔️	✔️
check_COLA	N/A	N/A	N/A
check_NOLA	✖	✖	✖
chirp	✖	✖	✖
choose_conv_method	✔️	✔️	✔️
closest_STFT_dual_window	✖	✖	✖
coherence	✖	✖	✖
cont2discrete	✖	✖	✖
convolve	✔️	✔️	✔️
convolve2d	✔️	✔️	✔️
correlate	✔️	✔️	✔️
correlate2d	✔️	✔️	✔️
correlation_lags	N/A	N/A	N/A
csd	✖	✖	✖
cspline1d	✔️	✔️	✔️
cspline1d_eval	✔️	✔️	✔️
cspline2d	✔️	✔️	✔️
czt	✖	✖	✖
czt_points	✖	✖	✖
dbode	✖	✖	✖
decimate	✖	✖	✖
deconvolve	✔️	✔️	✔️
detrend	✔️	✔️	✔️
dfreqresp	✖	✖	✖
dimpulse	✖	✖	✖
dlsim	✖	✖	✖
dlti	✖	✖	✖
dstep	✖	✖	✖
ellip	✔️	✔️	✔️
ellipap	✔️	✔️	✔️
ellipord	✔️	✔️	✔️
envelope	✔️	✔️	✔️
fftconvolve	✔️	✔️	✔️
filtfilt	✔️	✔️	✔️
find_peaks	✖	✖	✖
find_peaks_cwt	✖	✖	✖
findfreqs	✔️	✔️	✔️
firls	✔️	✔️	✔️
firwin	✔️	✔️	✔️
firwin2	✔️	✔️	✔️
firwin_2d	✖	✖	✖
freqresp	✖	✖	✖
freqs	✔️	✔️	✔️
freqs_zpk	✔️	✔️	✔️
freqz	✔️	✔️	✔️
freqz_sos	✔️	✔️	✔️
freqz_zpk	✔️	✔️	✔️
gammatone	✔️	✔️	✔️
gauss_spline	✔️	✔️	✔️
gausspulse	✖	✖	✖
get_window	✔️	✔️	✔️
group_delay	✔️	✔️	✔️
hilbert	✔️	✔️	✔️
hilbert2	✔️	✔️	✔️
iircomb	✔️	✖	✔️
iirdesign	✖	✖	✖
iirfilter	✔️	✔️	✔️
iirnotch	✔️	✔️	✔️
iirpeak	✔️	✔️	✔️
impulse	✔️	✔️	✔️
invres	✖	✖	✖
invresz	✖	✖	✖
istft	N/A	N/A	N/A
kaiser_atten	N/A	N/A	N/A
kaiser_beta	N/A	N/A	N/A
kaiserord	N/A	N/A	N/A
lfilter	✔️	✔️	✔️
lfilter_zi	✔️	✔️	✔️
lfiltic	✔️	✔️	✔️
lombscargle	✖	✖	✖
lp2bp	✔️	✔️	✔️
lp2bp_zpk	✔️	✔️	✔️
lp2bs	✔️	✔️	✔️
lp2bs_zpk	✔️	✔️	✔️
lp2hp	✔️	✔️	✔️
lp2hp_zpk	✔️	✔️	✔️
lp2lp	✔️	✔️	✔️
lp2lp_zpk	✔️	✔️	✔️
lsim	✖	✖	✖
lti	✖	✖	✖
max_len_seq	✖	✖	✖
medfilt	✔️	✔️	✔️
medfilt2d	✔️	✔️	✔️
minimum_phase	✔️	✔️	✔️
normalize	✔️	✔️	✔️
oaconvolve	✔️	✔️	✖
order_filter	✔️	✔️	✔️
peak_prominences	✖	✖	✖
peak_widths	✖	✖	✖
periodogram	✖	✖	✖
place_poles	✖	✖	✖
qspline1d	✔️	✔️	✔️
qspline1d_eval	✔️	✔️	✔️
qspline2d	✖	✖	✖
remez	✔️	✔️	✔️
resample	✔️	✖	✖
resample_poly	✔️	✔️	✖
residue	✖	✖	✖
residuez	✖	✖	✖
savgol_coeffs	✔️	✔️	✔️
savgol_filter	✔️	✔️	✔️
sawtooth	✔️	✔️	✖
sepfir2d	✖	✖	✖
sos2tf	✔️	✔️	✔️
sos2zpk	✔️	✔️	✔️
sosfilt	✔️	✔️	✔️
sosfilt_zi	✔️	✔️	✔️
sosfiltfilt	✖	✖	✖
sosfreqz	✔️	✔️	✔️
spectrogram	N/A	N/A	N/A
spline_filter	✔️	✔️	✔️
square	✔️	✔️	✔️
ss2tf	✖	✖	✖
ss2zpk	✖	✖	✖
step	✖	✖	✖
stft	N/A	N/A	N/A
sweep_poly	✖	✖	✖
symiirorder1	✖	✖	✖
symiirorder2	✖	✖	✖
tf2sos	✔️	✔️	✔️
tf2ss	✖	✖	✖
tf2zpk	✔️	✔️	✔️
unique_roots	✖	✖	✖
unit_impulse	✖	✖	✖
upfirdn	✔️	✔️	✔️
vectorstrength	✔️	✔️	✔️
welch	✖	✖	✖
whittaker_henderson	✖	✖	✖
wiener	✔️	✔️	✔️
zoom_fft	✖	✖	✖
zpk2sos	✔️	✔️	✔️
zpk2ss	✖	✖	✖
zpk2tf	✔️	✔️	✔️

Support on GPU#

Legend

✔️ = supported

✖ = unsupported

N/A = out-of-scope

function/class	cupy	torch	jax
CZT	✖	✖	✖
ShortTimeFFT	✖	✖	✖
StateSpace	✖	✖	✖
TransferFunction	✖	✖	✖
ZerosPolesGain	✖	✖	✖
ZoomFFT	✖	✖	✖
abcd_normalize	✔️	✔️	✔️
argrelextrema	✖	✖	✖
argrelmax	✖	✖	✖
argrelmin	✖	✖	✖
band_stop_obj	✖	✖	✖
bessel	✖	✖	✖
besselap	✔️	✔️	✔️
bilinear	✔️	✖	✖
bilinear_zpk	✔️	✔️	✖
bode	✖	✖	✖
buttap	✔️	✔️	✔️
butter	✔️	✖	✖
buttord	✔️	✔️	✖
cheb1ap	✔️	✔️	✔️
cheb1ord	✔️	✔️	✖
cheb2ap	✔️	✔️	✔️
cheb2ord	✔️	✔️	✖
cheby1	✔️	✖	✖
cheby2	✔️	✖	✖
check_COLA	N/A	N/A	N/A
check_NOLA	✖	✖	✖
chirp	✖	✖	✖
choose_conv_method	✔️	✔️	✔️
closest_STFT_dual_window	✖	✖	✖
coherence	✖	✖	✖
cont2discrete	✔️	✖	✖
convolve	✔️	✖	✔️
convolve2d	✔️	✖	✔️
correlate	✔️	✖	✔️
correlate2d	✔️	✖	✔️
correlation_lags	N/A	N/A	N/A
csd	✖	✖	✖
cspline1d	✔️	✖	✖
cspline1d_eval	✔️	✖	✖
cspline2d	✔️	✖	✖
czt	✖	✖	✖
czt_points	✖	✖	✖
dbode	✖	✖	✖
decimate	✔️	✖	✖
deconvolve	✔️	✖	✖
detrend	✔️	✖	✔️
dfreqresp	✖	✖	✖
dimpulse	✔️	✖	✖
dlsim	✖	✖	✖
dlti	✖	✖	✖
dstep	✖	✖	✖
ellip	✔️	✖	✖
ellipap	✔️	✔️	✔️
ellipord	✔️	✖	✖
envelope	✔️	✔️	✔️
fftconvolve	✔️	✔️	✔️
filtfilt	✔️	✔️	✔️
find_peaks	✖	✖	✖
find_peaks_cwt	✖	✖	✖
findfreqs	✔️	✔️	✖
firls	✖	✖	✖
firwin	✔️	✔️	✖
firwin2	✔️	✖	✖
firwin_2d	✖	✖	✖
freqresp	✖	✖	✖
freqs	✔️	✔️	✖
freqs_zpk	✔️	✔️	✖
freqz	✔️	✔️	✖
freqz_sos	✔️	✔️	✖
freqz_zpk	✔️	✔️	✔️
gammatone	✔️	✔️	✔️
gauss_spline	✔️	✔️	✔️
gausspulse	✖	✖	✖
get_window	✔️	✔️	✔️
group_delay	✔️	✖	✖
hilbert	✔️	✔️	✔️
hilbert2	✔️	✔️	✔️
iircomb	✔️	✔️	✖
iirdesign	✖	✖	✖
iirfilter	✔️	✔️	✖
iirnotch	✔️	✔️	✔️
iirpeak	✔️	✔️	✔️
impulse	✔️	✔️	✔️
invres	✔️	✖	✖
invresz	✔️	✖	✖
istft	N/A	N/A	N/A
kaiser_atten	N/A	N/A	N/A
kaiser_beta	N/A	N/A	N/A
kaiserord	N/A	N/A	N/A
lfilter	✔️	✖	✖
lfilter_zi	✖	✖	✖
lfiltic	✔️	✖	✖
lombscargle	✖	✖	✖
lp2bp	✔️	✔️	✖
lp2bp_zpk	✔️	✔️	✖
lp2bs	✔️	✔️	✖
lp2bs_zpk	✔️	✔️	✖
lp2hp	✔️	✔️	✖
lp2hp_zpk	✔️	✔️	✖
lp2lp	✔️	✔️	✖
lp2lp_zpk	✔️	✔️	✖
lsim	✖	✖	✖
lti	✖	✖	✖
max_len_seq	✖	✖	✖
medfilt	✔️	✖	✖
medfilt2d	✔️	✖	✖
minimum_phase	✔️	✔️	✖
normalize	✔️	✔️	✖
oaconvolve	✔️	✔️	✖
order_filter	✔️	✖	✖
peak_prominences	✖	✖	✖
peak_widths	✖	✖	✖
periodogram	✖	✖	✖
place_poles	✖	✖	✖
qspline1d	✔️	✖	✖
qspline1d_eval	✔️	✖	✖
qspline2d	✔️	✖	✖
remez	✖	✖	✖
resample	✔️	✖	✖
resample_poly	✔️	✖	✖
residue	✔️	✖	✖
residuez	✔️	✖	✖
savgol_coeffs	✔️	✔️	✔️
savgol_filter	✔️	✖	✖
sawtooth	✔️	✔️	✔️
sepfir2d	✖	✖	✖
sos2tf	✔️	✖	✖
sos2zpk	✔️	✖	✖
sosfilt	✔️	✖	✖
sosfilt_zi	✖	✖	✖
sosfiltfilt	✔️	✖	✖
sosfreqz	✔️	✔️	✖
spectrogram	N/A	N/A	N/A
spline_filter	✔️	✖	✖
square	✔️	✔️	✔️
ss2tf	✖	✖	✖
ss2zpk	✖	✖	✖
step	✖	✖	✖
stft	N/A	N/A	N/A
sweep_poly	✖	✖	✖
symiirorder1	✖	✖	✖
symiirorder2	✖	✖	✖
tf2sos	✔️	✖	✖
tf2ss	✖	✖	✖
tf2zpk	✔️	✖	✖
unique_roots	✔️	✖	✖
unit_impulse	✖	✖	✖
upfirdn	✔️	✖	✖
vectorstrength	✔️	✔️	✖
welch	✖	✖	✖
whittaker_henderson	✖	✖	✖
wiener	✔️	✖	✔️
zoom_fft	✖	✖	✖
zpk2sos	✔️	✖	✖
zpk2ss	✖	✖	✖
zpk2tf	✔️	✖	✖

Support with JIT#

Legend

✔️ = supported

✖ = unsupported

N/A = out-of-scope

function/class	jax
CZT	✖
ShortTimeFFT	✖
StateSpace	✖
TransferFunction	✖
ZerosPolesGain	✖
ZoomFFT	✖
abcd_normalize	✔️
argrelextrema	✖
argrelmax	✖
argrelmin	✖
band_stop_obj	✖
bessel	✖
besselap	✔️
bilinear	✖
bilinear_zpk	✖
bode	✖
buttap	✔️
butter	✖
buttord	✖
cheb1ap	✔️
cheb1ord	✖
cheb2ap	✔️
cheb2ord	✖
cheby1	✖
cheby2	✖
check_COLA	N/A
check_NOLA	✖
chirp	✖
choose_conv_method	✔️
closest_STFT_dual_window	✖
coherence	✖
cont2discrete	✖
convolve	✔️
convolve2d	✔️
correlate	✔️
correlate2d	✔️
correlation_lags	N/A
csd	✖
cspline1d	✖
cspline1d_eval	✖
cspline2d	✖
czt	✖
czt_points	✖
dbode	✖
decimate	✖
deconvolve	✖
detrend	✔️
dfreqresp	✖
dimpulse	✖
dlsim	✖
dlti	✖
dstep	✖
ellip	✖
ellipap	✔️
ellipord	✖
envelope	✔️
fftconvolve	✔️
filtfilt	✔️
find_peaks	✖
find_peaks_cwt	✖
findfreqs	✖
firls	✖
firwin	✖
firwin2	✖
firwin_2d	✖
freqresp	✖
freqs	✖
freqs_zpk	✖
freqz	✖
freqz_sos	✖
freqz_zpk	✔️
gammatone	✔️
gauss_spline	✔️
gausspulse	✖
get_window	✔️
group_delay	✖
hilbert	✔️
hilbert2	✔️
iircomb	✖
iirdesign	✖
iirfilter	✖
iirnotch	✔️
iirpeak	✔️
impulse	✔️
invres	✖
invresz	✖
istft	N/A
kaiser_atten	N/A
kaiser_beta	N/A
kaiserord	N/A
lfilter	✖
lfilter_zi	✖
lfiltic	✔️
lombscargle	✖
lp2bp	✖
lp2bp_zpk	✖
lp2bs	✖
lp2bs_zpk	✖
lp2hp	✖
lp2hp_zpk	✖
lp2lp	✖
lp2lp_zpk	✖
lsim	✖
lti	✖
max_len_seq	✖
medfilt	✖
medfilt2d	✖
minimum_phase	✖
normalize	✖
oaconvolve	✔️
order_filter	✖
peak_prominences	✖
peak_widths	✖
periodogram	✖
place_poles	✖
qspline1d	✖
qspline1d_eval	✖
qspline2d	✖
remez	✖
resample	✖
resample_poly	✖
residue	✖
residuez	✖
savgol_coeffs	✔️
savgol_filter	✖
sawtooth	✖
sepfir2d	✖
sos2tf	✖
sos2zpk	✖
sosfilt	✖
sosfilt_zi	✖
sosfiltfilt	✖
sosfreqz	✖
spectrogram	N/A
spline_filter	✖
square	✔️
ss2tf	✖
ss2zpk	✖
step	✖
stft	N/A
sweep_poly	✖
symiirorder1	✖
symiirorder2	✖
tf2sos	✖
tf2ss	✖
tf2zpk	✖
unique_roots	✖
unit_impulse	✖
upfirdn	✖
vectorstrength	✖
welch	✖
whittaker_henderson	✖
wiener	✖
zoom_fft	✖
zpk2sos	✖
zpk2ss	✖
zpk2tf	✖

Array API Standard Support: signal#

Caveats#

Support on CPU#

Support on GPU#

Support with JIT#

Array API Standard Support: `signal`#