Transferable Adversarial Attacks against Automatic Modulation Classifier in Wireless Communications

Algorithm 1

AL-BIM.

Input: A classifier truncated model ; original signal sample to be attacked ; target signal sample ; clean signal sample ; Loss Function .

Parameter: perturbation size =0.001, number of iterations T.

Output:the adversarial sample that satisfy

.

.

put into ,obtain feature ;

put into ,obtain feature ;

put into ,obtain feature ;

  t=0 to T-1

put into ,obtain feature

Obtain the gradient ,

where ;

calculate the accumulated gradient,renew the :

update the with gradient method

 end for