How can I input complex data 3-D matrices in a convolutional neural network (using deep learning toolbox)?

I need to train a convolutional neural network (CNN) whose inputs are 3-dimensional matrices containing complex numbers, but checking the Deep Learning Toolbox documentation it seems that inputs for CNNs must be images with defined channels. How can I input this type of data into a CNN using the Deep Learning Toolbox?

Answer 1

[Gabija Marsalkaite]

MATLAB does not have explicit support for complex numbers in neural networks in 2019a (this may change in later releases). In your case I would try separating real and imaginary parts as two channels (or more if you can think of better ways to represent data) and use it as 3D images with 2 channels.

[Gabriel]

Thank you, Gabija! I will try it.