4D matrix multiplication

2 次查看（过去 30 天）
Kamran
Kamran 2021-10-15
评论： Kamran 2021-10-20
I do the following in 4 loops and it takes ages to complete. Is there a way this code could be made more efficeint, without using parallel processing toolbox?
'steer' is a 136x101x101x16 matrix
'R' is a 136x16x16 matrix
'pow' and 'F' are 101x101 matrices.
pow = zeros(grdpts_y, grdpts_x); %grdpts_y, grdpts_x = 101
for l=1:nf %nf = 136
F = zeros(grdpts_y,grdpts_x);
for i=1:grdpts_x
for j=1:grdpts_y
F(i,j) = F(i,j) + 1./(squeeze(steer(l,i,j,:))'*squeeze(R(l,:,:))*squeeze(steer(l,i,j,:)));
end
end
F = F.*conj(F);
pow = pow + F;
end
Thanks in advance,
Kamran

请先登录，再回答此问题。

采纳的回答

Matt J
Matt J 2021-10-15
编辑：Matt J 2021-10-18
steer=reshape( permute(steer,[2,3,4,1]),101^2,[],136 );
R=permute(R,[2,3,1]);
F=1./sum( pagemtimes(conj(steer),R).*steer, 2);
F=reshape( abs(F).^2 ,101,101,[]);
pow=sum(F,3);
  10 个评论
显示 8更早的评论
Matt J
Matt J 2021-10-19
编辑：Matt J 2021-10-19
In your new version, F will always be real, non-negative, so I don't know why you would still be computing conj(F).
steer=reshape( permute(steer,[2,3,4,1]),101^2,[],136 );
Vec_n=cell(1,nf);
for l=1:nf
[Vec, Val] = eig(squeeze(R(l,:,:)));
[Val Seq] = sort(max(Val));
Vec_s = Vec(:,Seq(nstat ,nstat));
Vec_n{l}= Vec(:,Seq(1:nstat-1));
end
Vec_n=cat(3,Vec_n{:});
F=1./sum( abs(pagemtimes(conj(steer),Vec_n)).^2, 2);
F=reshape( abs(F).^2 ,101,101,[]);
pow=sum(F,3);
Kamran
Kamran 2021-10-20
Thank you very much. You are of course right. Thanks again for the prompt help.

请先登录，再进行评论。

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Particle & Nuclear Physics 的更多信息

标签

产品


版本

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by