Skip to content

Python Resample API

Continuously-variable polyphase resampler — three implementations all backed by the native C library, accepting and returning complex64 NumPy arrays with state preserved across calls.

Source: python/dsp/doppler/resample/__init__.py

resample

doppler.resample — continuously-variable polyphase resampler.

Three implementations are provided:

  • :class:Resampler — Kaiser polyphase table resampler backed by the C library (dp_resamp_cf32). Default choice: fast, cache-resident for small banks.

  • :class:ResamplerDpmfs — Dual Phase Modified Farrow Structure resampler (dp_resamp_dpmfs). 608-byte coefficient footprint; ideal for multi-channel pipelines.

  • :class:HalfbandDecimator — dedicated 2:1 halfband decimator (dp_hbdecim_cf32). Exploits halfband symmetry to halve the multiply count vs a general 2-phase polyphase decimator.

All classes accept complex64 NumPy arrays and return complex64 NumPy arrays. Internal state is preserved across calls — suitable for streaming pipelines.

Typical usage

import numpy as np from doppler.resample import Resampler from doppler.polyphase import kaiser_prototype _, bank = kaiser_prototype() r = Resampler(bank, rate=2.0) x = np.ones(64, dtype=np.complex64) y = r.execute(x) y.dtype dtype('complex64') len(y) > 0 True

The reference Python implementation lives in :mod:doppler.resample.reference.

Resampler

Kaiser polyphase table resampler for cf32 IQ samples.

Wraps dp_resamp_cf32_t from the C library. Supports continuously-variable interpolation and decimation at any rational or irrational rate.

Parameters:

Name Type Description Default
bank np.ndarray, shape (L, N), dtype=float32

Polyphase coefficient bank. Row-major: bank[p, k] is the k-th tap of polyphase branch p. L must be a power of two. Pass the second element of the tuple returned by :func:doppler.polyphase.kaiser_prototype.

 required
rate float

Output-to-input sample-rate ratio fs_out / fs_in. rate > 1 → interpolation; rate < 1 → decimation.

 required

Examples:

>>> import numpy as np
>>> from doppler.polyphase import kaiser_prototype
>>> from doppler.resample import Resampler
>>> _, bank = kaiser_prototype(passband=0.4, stopband=0.6)
>>> r = Resampler(bank, rate=1.5)
>>> x = np.zeros(256, dtype=np.complex64)
>>> x[0] = 1.0          # impulse
>>> y = r.execute(x)
>>> y.dtype
dtype('complex64')
>>> len(y) >= 384 - 20  # approx ceil(256 * 1.5)
True

rate property 

rate: float

Output-to-input sample-rate ratio.

num_phases property 

num_phases: int

Number of polyphase branches (L).

num_taps property 

num_taps: int

Taps per polyphase branch (N).

execute 

execute(x: ndarray) -> np.ndarray

Resample a block of cf32 samples.

State is preserved across calls — suitable for streaming.

Parameters:

Name Type Description Default
x np.ndarray, dtype=complex64

Input samples.

 required

Returns:

Type Description
np.ndarray, dtype=complex64

Output samples. Length ≈ len(x) * rate; exact count depends on the internal phase accumulator state.

Examples:

>>> import numpy as np
>>> from doppler.polyphase import kaiser_prototype
>>> from doppler.resample import Resampler
>>> _, bank = kaiser_prototype()
>>> r = Resampler(bank, rate=0.5)
>>> y = r.execute(np.ones(128, dtype=np.complex64))
>>> len(y) <= 70      # roughly half the input
True

reset 

reset() -> None

Zero the sample history and reset the phase accumulator.

ResamplerDpmfs

DPMFS polyphase resampler for cf32 IQ samples.

Wraps dp_resamp_dpmfs_t from the C library. Uses a compact polynomial coefficient bank (608 bytes for M=3, N=19) that fits in L1 cache, eliminating table-lookup cache misses in multi-channel pipelines.

Parameters:

Name Type Description Default
coeffs DPMFSCoeffs

Polynomial coefficients from :func:doppler.polyphase.fit_dpmfs. Must expose a .c attribute of shape (2, M+1, N), dtype float32. coeffs.c[j, m, k] is the k-th tap coefficient for polynomial order m and phase half j (0 = μ ∈ [0, 0.5), 1 = μ ∈ [0.5, 1.0)).

 required
rate float

Output-to-input sample-rate ratio fs_out / fs_in.

 required

Examples:

>>> import numpy as np
>>> from doppler.polyphase import kaiser_prototype, fit_dpmfs
>>> from doppler.resample import ResamplerDpmfs
>>> _, bank = kaiser_prototype()
>>> coeffs = fit_dpmfs(bank, M=3)
>>> r = ResamplerDpmfs(coeffs, rate=1.5)
>>> x = np.zeros(256, dtype=np.complex64)
>>> x[0] = 1.0
>>> y = r.execute(x)
>>> y.dtype
dtype('complex64')
>>> len(y) >= 384 - 20
True

rate property 

rate: float

Output-to-input sample-rate ratio.

num_taps property 

num_taps: int

Taps per branch (N).

poly_order property 

poly_order: int

Polynomial order (M).

execute 

execute(x: ndarray) -> np.ndarray

Resample a block of cf32 samples.

Parameters:

Name Type Description Default
x np.ndarray, dtype=complex64

Input samples.

 required

Returns:

Type Description
np.ndarray, dtype=complex64

Output samples.

Examples:

>>> import numpy as np
>>> from doppler.polyphase import kaiser_prototype, fit_dpmfs
>>> from doppler.resample import ResamplerDpmfs
>>> _, bank = kaiser_prototype()
>>> r = ResamplerDpmfs(fit_dpmfs(bank, M=3), rate=0.5)
>>> y = r.execute(np.ones(128, dtype=np.complex64))
>>> len(y) <= 70
True

reset 

reset() -> None

Zero the sample history and reset the phase accumulator.

HalfbandDecimator

Halfband 2:1 decimator for cf32 IQ samples.

Wraps dp_hbdecim_cf32_t from the C library. Exploits halfband filter symmetry to reduce the multiply count to N/2 + 2 per output sample (vs N for a general 2-phase polyphase decimator).

Parameters:

Name Type Description Default
bank np.ndarray, shape (2, N), dtype=float32

Polyphase bank from :func:doppler.polyphase.kaiser_prototype called with phases=2. One row is the symmetric FIR branch; the other is a pure-delay branch (unit gain at the centre tap). This class detects and selects the correct row automatically.

 required

Examples:

>>> import numpy as np
>>> from doppler.polyphase import kaiser_prototype
>>> from doppler.resample import HalfbandDecimator
>>> _, bank = kaiser_prototype(phases=2)
>>> r = HalfbandDecimator(bank)
>>> x = np.zeros(256, dtype=np.complex64)
>>> x[0] = 1.0          # impulse
>>> y = r.execute(x)
>>> y.dtype
dtype('complex64')
>>> len(y) == 128
True

rate property 

rate: float

Decimation rate (always 0.5).

num_taps property 

num_taps: int

FIR branch length (N).

execute 

execute(x: ndarray) -> np.ndarray

Decimate a block of cf32 samples by 2.

State is preserved across calls — suitable for streaming. Odd-length blocks are handled transparently: the dangling even sample is buffered and consumed on the next call.

Parameters:

Name Type Description Default
x np.ndarray, dtype=complex64

Input samples.

 required

Returns:

Type Description
np.ndarray, dtype=complex64

Output samples. Length == len(x) // 2 for even-length input (or one less when the previous call left a pending sample).

Examples:

>>> import numpy as np
>>> from doppler.polyphase import kaiser_prototype
>>> from doppler.resample import HalfbandDecimator
>>> _, bank = kaiser_prototype(phases=2)
>>> r = HalfbandDecimator(bank)
>>> y = r.execute(np.ones(128, dtype=np.complex64))
>>> len(y) == 64
True

reset 

reset() -> None

Zero the sample history and clear any pending sample.