compute the finite linear convolution of two arrays of samples
Convolution( A, B, options )
Array, Vector, or list of real or complex numeric sample values for first signal
Array, Vector, or list of real or complex numeric sample values for second signal
algorithm : keyword auto, direct, or fft for the algorithm to use for computation; default is direct
container : Array or Vector, predefined container for holding result
shape : keyword full, same, or valid for the shape of the convolution; default is full
subtype : keyword Array, Vector, Vector[column], or Vector[row] for the subtype of the output; default is Array
The Convolution(A, B) command computes the full convolution of the signals A and B of length M and N respectively, storing the result in a container C of length M+N−1 and having datatype float or complex, which is then returned.
The full convolution is defined by the formula
for each k from 1 to M+N−1, with Aj=0 for M<j and Bj=0 for N<j.
For all choices of shape, the full convolution of size P=M+N−1 is computed. When shape=same, the full convolution is trimmed on both sides so that the result is of length Q=M. Note that when the number of elements to be trimmed is odd, one more element will be trimmed from the left side than the right. When shape=valid, the final convolution will be found in a similar manner to the shape=same case, but the value of the size Q will be M when N=0, and max⁡M−N+1,0 otherwise. The valid convolution effectively discards the elements which involve padded zeros for the signals.
Before the code performing the computation runs, A and B are converted to datatype float (if the values are all real-valued) or complex (if all the values are complex-valued, but not all real-valued) if they do not have that datatype already. For this reason, it is most efficient if A and B have one of these datatypes beforehand.
If either A or B is an rtable that is not a 1-D Array, it is accepted by the command and converted to an Array. Should this not be possible, an error will be thrown.
If the container=C option is provided, then the results are put into C and C is returned. With this option, no additional memory is allocated to store the result. The container must be a 1-D rtable of appropriate size having datatype float (for two real signals) or complex (for one or two complex signals).
The algorithm=name option can be used to specify the algorithm used for computing the convolution. Supported algorithms:
auto - automatically choose the fastest algorithm based on input.
direct - use direct convolution formula for computation. This is the default.
fft - use an algorithm based on the Fast Fourier Transform (FFT). This is a much faster algorithm than the direct formula for large samples, but numerical roundoff can cause significant numerical artifacts, especially when the result has a large dynamic range.
The SignalProcessing[Convolution] command is thread-safe as of Maple 17.
For more information on thread safety, see index/threadsafe.
a ≔ Array⁡1,2,3,'datatype'='float'8
b ≔ Array⁡1,−1,1,−1,'datatype'='float'8
c ≔ Array⁡1..numelems⁡a+numelems⁡b−1,'datatype'='float'8:
A ≔ Vectorrow⁡2−I,0,5+3⁢I,0,4⁢I
B ≔ Vectorrow⁡−7,3+10⁢I,9−2⁢I,1
C1 ≔ Convolution⁡A,B,'algorithm'='fft'
C2 ≔ `~`round⁡C1
A ≔ Vector'row'⁡2,3,5,7,11
B ≔ Vector'row'⁡13,17,19,23
The SignalProcessing[Convolution] command was introduced in Maple 17.
For more information on Maple 17 changes, see Updates in Maple 17.
The A and B parameters were updated in Maple 2020.
The algorithm option was introduced in Maple 2020.
For more information on Maple 2020 changes, see Updates in Maple 2020.
The SignalProcessing[Convolution] command was updated in Maple 2023.
The shape and subtype options were introduced in Maple 2023.
For more information on Maple 2023 changes, see Updates in Maple 2023.
Download Help Document
What kind of issue would you like to report? (Optional)