I have two functions x(t) and y(t), and say t = seq(1,10,1). I want to get z(t), that is a convolution of x(t) and y(t) using FFT. Here is what I have done: I generated a vector vx from x(t) and vy from y(t) and used the function 'convolve' from state package (that uses the Fast Fourier Transform to compute the several kinds of convolutions of two sequences): conv
you are now clear that both are producing different results.
the question now is which is best.
the answer could be it depends on the number of computations each takes to perform.
you use tic and toc functions in your program. at least you come to know how much time each one is consuming. and accordingly you come to conclusion.
thanking you
I suppose that the difference between the calculation of convolution by means of FFT and the function convolveolve ( conv in MatLab) takes place. The point is that the conv function in Matlfb carries out the usual convolution of the sliding average type (similar to polynomial multiplication), but the Fourier-based convolution is cyclical (with the moving last elements of the sliding vector to its beginning). In order for the cyclic convolution to coincide with the usual one, it is necessary to padding the vectors vx and vy with zeros by the length of the vectors themselves (length of the t).
From a theoretical point of view, convolving vx (10 samples length) and vy (also 10 samples length) should provide an output signal of length length(vx) + length(vy) -1 = 19 samples.
Using matlab functions, you can do
1- conv(vx,vz) that performs the complete convolution. The length of the result is length(vx)+ length(vy)-1=19 samples
2- filter(vx,[1],vy) that gives an output of the same length than vy (i.e. 10 samples)
Pascal Scalart
the following is the description of the inbuilt convolve function (conv):
c = conv(a,b) outputs the convolution of input vectors a and b, at least one of which must be a fi object.
c = conv(a,b,'shape') returns a subsection of the convolution, as specified by the shape parameter:
full — Returns the full convolution. This option is the default shape.
same — Returns the central part of the convolution that is the same size as input vector a.
valid — Returns only those parts of the convolution that the function computes without zero-padded edges. In this case, the length of output vector c is max(length(a)-max(0,length(b)-1),
0).
kindly look in to the description before using the inbuilt function.
it is better to write our own program if the task is simple as here
when we convolve two sequence with L and M as the lengths, the convoluted sequence is having the length of L+M-1.
when you use the inbuilt function you are not getting the above mentioned length. it means both are not correct.
perhaps the inbuilt function is performing the circular convolution.
you can as well use the inbuilt function for sure. but you have to perform zero padding before using the function.
thanking you
Thank you all for your answers. It seems that convolve in R uses FFT for efficiency and if set to default arguments, it performs a circular convolution (i.e. concolve(vx,vy)). But I am still confused which is the best convolution way to use, regular convolution using the Matlab function 'conv' that produces 19 samples (length(vx)+ length(vy)-1) or convolve function to get the same length of x(t) and y(t) i.e 10? Thank you
you are now clear that both are producing different results.
the question now is which is best.
the answer could be it depends on the number of computations each takes to perform.
you use tic and toc functions in your program. at least you come to know how much time each one is consuming. and accordingly you come to conclusion.
thanking you
- Badges
- Science topic