How can I do 1D convolution on a 880*1 feature with filter size 45*1 with stride of 42?

I have a 885*1 feature and I would like to do a convolution with 45*1 with stride 42.

I want to get a 21*1 feature as a result.

If input layer(885*1) is named A, what sholud be the code?

Answer 1

[Matt J]

You can create a weight matrix that computes the convolution using func2mat on the File Exchange

https://www.mathworks.com/matlabcentral/fileexchange/44669-func2mat-convert-linear-function-to-matrix

For example,

>> k=rand(45,1); 
>> C=func2mat(@(z) conv(z,k,'valid'),rand(885,1)); 
>> C=C(1:42:end,:); 

>> whos C
  Name       Size             Bytes  Class     Attributes

  C         21x885            22208  double    sparse    

If you wish, you can verify that C*x gives the equivalent convolution as follows,

>> x=rand(885,1); 

>> y1=conv(x,k,'valid'); y1=y1(1:42:end);

>> y2=C*x;

>> norm(y1-y2)

ans =

   1.2809e-14