The neural network with complex sequence input can be converted to one that accepts real features, and the real part and imaginary part of complex sequence are divided into two real features input neural networks respectively. During the training, the network learns how to combine these two real features into a complex number so that the output of a complex number sequence can be obtained.
The feasibility and effectiveness of this approach depends on the problem you are trying to solve and the neural network architecture you are using. For some problems, this transformation may result in loss of information, which may degrade the performance of the neural network. If you want to train a neural network that accepts real number characteristics, it is recommended to use loss functions of real number values, such as mean square error or cross entropy loss. When training, you need to convert the real output to the complex output, for example by reconstructing the complex output as the real and imaginary parts of the complex number respectively.
