Method and apparatus for active terminal detection using spread code learned by using deep learning in multiple communications

1 . An active terminal detection method for wireless communication between a plurality of terminals and a receiver, the active terminal detection method comprising:

providing an active terminal detection device executing a first learning network and a second learning network, wherein the first learning network is trained to generate a second signal to be transmitted to the receiver by applying a spread code to a first signal generated by each of the plurality of terminals and the second learning network is trained to determine whether each of the plurality of terminals is active;

generating, by the first learning network, the second signal corresponding to each of the plurality of terminals by combining the spread code with the first signal generated by each of the plurality of terminals; and

determining, by the second learning network, whether each of the plurality of terminals is active based on the second signal, wherein the first learning network and the second learning network form an end-to-end learning structure, and

wherein the first learning network includes a first hidden layer deriving the spread code in which a cross-correlation value of the spread code decreases as activation probability of the plurality of terminals increases.

2 . The active terminal detection method of claim 1 , wherein

the first learning network and the second learning network are trained by using an error of active terminal detection as binary cross entropy loss according to a stochastic gradient descent algorithm.

3 . The active terminal detection method of claim 1 , wherein

the first learning network is trained to reduce active terminal detection errors by using an error value, activation probability of each of the plurality of terminals, channels of the plurality of terminals and the receiver, and a number of the plurality of terminals.

4 . The active terminal detection method of claim 1 , wherein

the second learning network is trained by using batch normalization and a residual network.

5 . An active terminal detection device for wireless communication between a plurality of terminals and a receiver, the active terminal detection device comprising:

a memory storing an active terminal detection program; and

a processor configured to execute the active terminal detection program,

wherein the active terminal detection program

forms an end-to-end learning structure by using a first learning network and a second learning network, wherein the first learning network is trained to generate a second signal to be transmitted to the receiver by assigning a spread code to a first signal generated by each of the plurality of terminals and the second learning network is trained to determine whether each of the plurality of terminals is active,

generates the spread code using the end-to-end learning structure,

generates, by the first learning network, the second signal corresponding to each of the plurality of terminals by combining the spread code with the first signal generated by each of the plurality of terminals, and

determines, by the second learning network, whether each of the plurality of terminals is active based on the second signal,

wherein the first learning network includes a first hidden layer deriving the spread code in which a cross-correlation value of the spread code decreases as the activation probability of the terminal increases.

6 . The active terminal detection device of claim 5 , wherein

first learning network and the second learning network are trained by using an error of active terminal detection as binary cross entropy loss according to a stochastic gradient descent algorithm.

7 . The active terminal detection device of claim 5 , wherein

the first learning network is trained to reduce active terminal detection errors by using an error value, activation probability of each of the plurality of terminals, channels of the plurality of terminals and the receiver, and a number of the plurality of terminals.

8 . The active terminal detection device of claim 5 , wherein

the second learning network is trained by using batch normalization and a residual network.