RealKernelProcessor<T>.Convolve Method

public override void Convolve(
	int signalLength,
	ReadOnlySpan<T> signal,
	int signalStride,
	int resultLength,
	Span<T> result,
	int resultStride
)

Public Overrides Sub Convolve ( 
	signalLength As Integer,
	signal As ReadOnlySpan(Of T),
	signalStride As Integer,
	resultLength As Integer,
	result As Span(Of T),
	resultStride As Integer
)

public:
virtual void Convolve(
	int signalLength, 
	ReadOnlySpan<T> signal, 
	int signalStride, 
	int resultLength, 
	Span<T> result, 
	int resultStride
) override

abstract Convolve : 
        signalLength : int * 
        signal : ReadOnlySpan<'T> * 
        signalStride : int * 
        resultLength : int * 
        result : Span<'T> * 
        resultStride : int -> unit 
override Convolve : 
        signalLength : int * 
        signal : ReadOnlySpan<'T> * 
        signalStride : int * 
        resultLength : int * 
        result : Span<'T> * 
        resultStride : int -> unit 

Parameters

signalLength  Int32

The length of the input signal.

signal  ReadOnlySpan<T>

A read-only span containing the real input signal.

signalStride  Int32

The stride between successive elements in the signal. Use 1 for contiguous data.

resultLength  Int32

The expected length of the result (should match GetOutputLength(Int32)).

result  Span<T>

A span to store the convolution result. Must have capacity for at least resultLength elements.

resultStride  Int32

The stride between successive elements in the result. Use 1 for contiguous output.

Real convolution computes: y[n] = Σ_m x[n - (m - a)] · h[m], where a is the anchor index. This is the standard discrete convolution operation used in digital signal processing and filtering.

This method automatically selects between direct (time-domain) and FFT (frequency-domain) algorithms based on the ProcessingMethod setting and signal/kernel sizes. The direct method is typically faster for small kernels, while FFT is more efficient for large kernels.

Comparison with other libraries:

• NumPy np.convolve(signal, kernel, mode='full'|'same'|'valid'): Use FullLength, SameAsSignal, or NoPadding respectively.
• SciPy scipy.signal.convolve: Similar to NumPy with additional boundary modes via SignalPadding.
• OpenCV cv2.filter2D: Typically uses Centered with Replicate.

Reference