A Deep Learning-Aided Detection Method for FTN-Based NOMA

Algorithm

1: Gradient descent training algorithm of neural network.

	Input: y: training data; x: training labels; α: the learning rate;
	Output: : the connection weight matrix between layer and layer
	: the bias matrix between layer and layer
	: output of layer l
(1)	Set (matrix/vector of zeros) for all l.
(2)	For i = 1 to m,
	(a) Use backpropagation to compute the partial derivatives: and .
	(i) Perform a feedforward pass, computing the activations for layers 1 to and using the equation defining the forward propagation steps.
	(ii) For the layer , set .
	(iii) For the layer set
	.
	(iv) Compute the desired partial derivatives:
	(b) Set .
	(c) Set .
(3)	Update the parameters:
	.
	.