Supervised Contrastive Learning-Based Modulation Classification of Underwater Acoustic Communication

Algorithm 1

Two-stage training of supervised contrastive learning.

Input: Encoder training: batch size 32, initial learning rate =52, epoch E=100, =0.07

 Classifier network training: batch size 128, initial learning rate = 13, epoch E=100

Output: Backbone network parameter , The Classifier network

//Encoder training

1: for epoch =1:E

2: sample a batch of data, update as described in Section

3: Backbone encodes m into F.

4: calculates loss (10)

5: update with

6: end for

//Classifier network training

7: for epoch = 1:Edo

8: Freeze encoder parameter, update as described in Section

9: Classifier network decodes F into result

10: calculates loss (8)

11: update with

12: end for

//Finish training

Return the parameter ,