v*a
发帖数: 14 | 1 用 mean square error (mse)定义好坏,mse=sum((I-I').^2))/sum(I^2)或者
mse=sum(abs(F)-abs(F')).^2))/sum(F^2). I为功率,F为振幅.I'为设计值.
对设计值 1 进行优化,从功率分布看,结果大为改善,mse也减小许多.但用振幅分布表示的
mse却反而增大.这是为什么啊?
哪位帮我想想,先谢了! | s***l
发帖数: 2236 | 2 of course when using I, the phase information has been got rid of. So I is
better since the phase fluctuation completely disappeared. While for F, you
still have at least plus and minus values.
just my 2 cents.
【在 v*a 的大作中提到】
: 用 mean square error (mse)定义好坏,mse=sum((I-I').^2))/sum(I^2)或者
: mse=sum(abs(F)-abs(F')).^2))/sum(F^2). I为功率,F为振幅.I'为设计值.
: 对设计值 1 进行优化,从功率分布看,结果大为改善,mse也减小许多.但用振幅分布表示的
: mse却反而增大.这是为什么啊?
: 哪位帮我想想,先谢了!
| v*x
发帖数: 32 | 3 my two cents:
because I and F are not the same thing,
you optimize for I, you sacrifice F.
You optimize for F, you sacrifice I.
so the result is not surprizing.
【在 s***l 的大作中提到】
: of course when using I, the phase information has been got rid of. So I is
: better since the phase fluctuation completely disappeared. While for F, you
: still have at least plus and minus values.
: just my 2 cents.
|
|
