Filter sampled signals
Apply a finite impulse response or infinite impulse response operation to sampled data.
SciPy · Module
scipy.signal
scipy.signal is SciPy’s signal-processing module. It provides convolution, correlation, filtering, spectral-analysis, waveform-generation, and related signal-processing functions.
How it works
The module groups algorithms for filtering, convolution, spectral analysis, waveform generation, and continuous- and discrete-time signal processing. Its functions operate mainly on NumPy arrays and include direct, FFT-based, and overlap-add convolution options.
Syntax
import scipy.signalParameters
| Name | Type | Description |
|---|---|---|
in1 | array_like | First input array for convolution, correlation, or another selected signal-processing operation. |
in2 | array_like | Second input array for convolution, correlation, or another selected signal-processing operation. |
mode | str, optional | Controls the requested convolution output region. |
method | str, optional | Selects the convolution algorithm when supported. |
Returns
The module's functions return results specific to the selected operation; convolution functions return an array containing the computed convolution.
Key points
- The module is the signal-processing namespace within SciPy.
- signal.convolve supports convolution for N-dimensional arrays.
- signal.correlate computes cross-correlation for N-dimensional arrays.
- signal.fftconvolve performs convolution using the Fast Fourier Transform.
- signal.oaconvolve provides overlap-add convolution for N-dimensional arrays.
- signal.choose_conv_method helps select a fast convolution or correlation implementation.
Common mistakes
- FFT-based convolution can be preferable for large inputs, but it is a different algorithm from direct convolution and should be selected deliberately.
- Do not assume one convolution method is always fastest; use the method-selection utility when performance matters.
Example: Convolve two arrays
import numpy as np
from scipy import signal
x = np.array([1.0, 2.0, 3.0])
h = np.array([1.0, 1.0])
y = signal.convolve(x, h, mode="full")
print(y)Convolves two one-dimensional NumPy arrays and prints every output sample.
When to use it
Convolve or correlate data
Combine a signal with an impulse response or detect similarity between arrays.
Analyze and create waveforms
Estimate frequency content or generate frequency-swept test signals.
Work with signal systems
Model or analyze continuous- and discrete-time systems.
Interview questions
QWhat is scipy.signal used for?
scipy.signal is SciPy's signal-processing module. It exposes tools for convolution, correlation, filtering, spectral analysis, waveform generation, and system analysis.
Follow-up: How would you choose a convolution strategy? Compare direct convolution with FFT-based and overlap-add implementations, considering input sizes and performance.
QWhat does signal.convolve do?
signal.convolve computes the convolution of two N-dimensional arrays. Its documented interface accepts the input arrays plus optional mode and method arguments.
Follow-up: What does the mode argument control? Explain how full, same, and valid output regions differ.
QWhich convolution tools does the module provide?
The module provides several convolution implementations, including convolve, fftconvolve, and oaconvolve. It also provides choose_conv_method to find a fast convolution or correlation method.
Follow-up: When might overlap-add be useful? It is suited to efficiently processing long signals, especially when one input can be treated as a filter.
Related
convolvecorrelatefftconvolveoaconvolveconvolve2dcorrelate2dsepfir2dchoose_conv_method