Specific Emitter Identification with Limited Labelled Signals Based on Variational Autoencoder Embedded in Information-Maximising Generative Adversarial Network and Gradient Penalty

Algorithm 1

Training procedure of VAE-InfoGAN-GP.

Input:

 : Raw RF signal data;

 : Latent code;

 : Latent vector;

Output:

 : Parameters of sub-networks E, G, D, Q

1. Perform bispectrum analysis on RF signal data and obtain bispectrum as RFFs representation data ;

2. Initialize network parameters of encoder, generator, discriminator and auxiliary classifier;

3. Fix the network parameters of encoder and generator, generate real and fake RFFs representation data based on the RFFs representation data , latent code and latent vector :

4. Calculate the loss function of discriminator:

5. Network parameters of discriminator are updated according to the loss function:

6. Calculate the loss function of auxiliary classifier:

7. Network parameters of auxiliary classifier are updated according to the loss function:

8. Fix network parameters of discriminator and auxiliary classifier, and train encoder and generator;

9. Calculate the loss function of generator:

10. Network parameters of generator are updated according to the loss function:

11. Calculate the loss function of encoder:

12. Network parameters of encoder are updated according to the loss function:

13. Repeat steps (2)-(12) until the network converges, and then auxiliary classifier is used to conduct semi-supervised SEI;