How to train complex [64 1] matrices in deeplearning nw

I am trying to train a complex numbered matrix with a fullyconnected layer,

but I can't read the value of the imaginary axis from the input layer. Is there a way?

1) The method I thought of was dividing real and imag to learn the input layer into two

, but I can't find such a method well..

Answer 1

[Srivardhan Gadila]

You can refer to the example: Modulation Classification with Deep Learning, specifically the pretrained network and "Transform Complex Signals to Real Arrays" section.

The following code might help you:

inputSize = [64 1 2];
numSamples = 128;
numClasses = 4;

%% Generate random data for training the network.
trainData = randn([inputSize numSamples]);
trainLabels = categorical(randi([0 numClasses-1], numSamples,1));

%% Create a network.
layers = [
    imageInputLayer(inputSize,'Name','input')  
    convolution2dLayer([3 1],16,'Padding','same','Name','conv_1')
    batchNormalizationLayer('Name','BN_1')
    reluLayer('Name','relu_1')
    fullyConnectedLayer(10,'Name','fc1')
    fullyConnectedLayer(numClasses,'Name','fc2')
    softmaxLayer('Name','softmax')
    classificationLayer('Name','classOutput')];

lgraph = layerGraph(layers);

analyzeNetwork(lgraph);
%% Define training options.
options = trainingOptions('adam', ...
    'InitialLearnRate',0.005, ...
    'LearnRateSchedule','piecewise',...
    'MaxEpochs',100, ...
    'MiniBatchSize',128, ...
    'Verbose',1, ...
    'Plots','training-progress');

%% Train the network.
net = trainNetwork(trainData,trainLabels,layers,options);