create a complex array with fake imaginary part?

Hi, everyone,

I want to create a complex array but with real values. I tried things like b=(complex(a)), but it turns out that imaginary part does not exist. Then I pass the array to a mex function, and the mex runs a mxGetPi and fails because the imaginary part is not allocated, then it crashes.

Is there a way to fake a complex array in matlab with imaginary part allocated?

Thanks

Dehuan

Answer 1

[Matt J]

The only thing that comes to mind is to add a hopefully negligible non-zero imaginary part to one of the elements,

    a(k)=a(k)+eps(a(k))*i;

Note that only one imaginary element in the array needs to be non-zero in order for mxGetPi to see an imaginary part. You might be able to be strategic about your choice of k, depending on what your application is doing.

[Dehuan]

I tried this, actually, but when I cast the array to integer, then the small imaginary part disappears. Is there's a way to solve/hack it?

[John D'Errico]

There is no complex integer type in MATLAB. So pass in the real integer part to your code, and add a zero imaginary part there. What is the problem? If you will sometimes have a non-zero imaginary part, then separate the real and imaginary parts in MATLAB, and pass them in separately, then recombine them in C.

[Matt J]

There are complex integer types, e.g.,

    >> a=uint8(1+1i); whos a

      Name      Size            Bytes  Class    Attributes

      a         1x1                 2  uint8    complex

but setting the imaginary part to zero will make MATLAB strip away the complex Attribute. I guess you would indeed have to add the imaginary array inside the MEX file.

[Matt J]

Dehuan Commented:

The C function was expecting interleaved real/imaginary in int16 format.

And the mex wrapper is written in such a way that it accepts complex int16s and interleave the real and imaginary part in the mex wrapper.

Although I understand the best way to do it is to pass double complexes to mex and cast and interleave there, or cast and interleave in matlab before passing data to mex(personally I prefer the latter), I'm seeking a way to to use that mex function with some real-only data without changing the mex code.

[Matt J]

I'm seeking a way to to use that mex function with some real-only data without changing the mex code.

I don't think you'll find it, unless the operation has special properties like linearity ( see my other answer ).

I'm not sure why you're reluctant to modify the mex code, however. It should just require a few lines at the beginning of the code to add a zero imaginary part that mxGetPi can see.

[Dehuan]

The reason is as simple as "One can not make his/her boss to change his/her code because he/she want to use it in a hacked/ugly way."

[Matt J]

"One can not make his/her boss to change his/her code because he/she want to use it in a hacked/ugly way."

Your boss' code is already ugly if it can't handle strictly real input data. Imagine an fft() routine that could only handle a signal with a non-zero imaginary part.

I would find a diplomatic way to tell him/her that, and offer to upgrade the code as we've described.

Answer 2

[Matt J]

This might be wishful thinking, but if the operation f() to be performed by the MEX is linear, you could leverage that,

   f(a) = f(a+b*i)+f(a-b*i)

where b is any fake non-zero imaginary data that you wish to add.

Answer 3

[Matt J]

You could try creating your own mex function that appends a zero imaginary part to a real int16 input. If you use mxCreateNumericArray with the ComplexFlag argument set to true, I think there's a chance that it will let you allocate a zero imaginary part.

You would need to use the conversion tool in the very last step, just before the input needs to be fed to your boss' mex code.