Deep Learning Based Communication: an Adversarial Approach
Ref: CISTER-TR-190605 Publication Date: 27 to 28, Jun, 2019
Deep Learning Based Communication: an Adversarial ApproachRef: CISTER-TR-190605 Publication Date: 27 to 28, Jun, 2019
Deep learning based communication using autoencoder have revolutionized the design of physical layer in wireless communication. In this paper, we propose an adversarial autoencoder to mitigate vulnerability of autoencoder against adversarial attacks. Results confirm the effectiveness of adversarial training by reducing block error rate (BLER) from 90 percent to 56 percent.
Poster presented in 3rd Doctoral Congress in Engineering (DCE 2019).