Skip to main content

Table 5 Training time comparison between the investigated SAFs LSTM-based channel equalizers

From: Modified state activation functions of deep learning-based SC-FDMA channel equalization system

AF

LSTM-based CE-SD Adam

LSTM-based CE-SD RMSProp

LSTM-based CE-SD SGdm

Cross entropy (M:S)

Sum squared errors (M:S)

Cross entropy (M:S)

Sum squared errors (M:S)

Cross entropy (M:S)

Sum squared errors (M:S)

Tanh

37:46

35:55

22:19

20:13

21:18

20:21

Cloglogm

56:39

44:20

26:28

19:49

30:41

19:39

Elliott

52:22

45:18

17:21

19:40

26:30

19:57

Modified Elliott

54:16

47:22

26:05

20:10

25:41

20:45

Gaussian

57:51

45:33

35:41

24:24

25:33

26:15

Bitanh1

54:18

48:52

25:42

24:39

24:54

24:42

Bitanh2

53:02

53:31

21:57

26:06

21:27

28:42

Rootsig

47:41

47:54

22:11

17:34

20:46

23:11

Softsign

48:19

49:20

24:40

18:53

25:28

20:15

Wave

48:06

46:31

25:19

23:38

24:29

20:52

Aranda

61:36

59:54

32:36

32:01

34:17

32:22

GELU

48:44

:

21:15

:

20:50

: