| Paper ID | SPCOM-8.3 |
| Paper Title |
MITIGATING CLIPPING DISTORTION IN OFDM USING DEEP RESIDUAL LEARNING |
| Authors |
Muhammad Shahmeer Omar, Xiaoli Ma, Georgia Institute of Technology, United States |
| Session | SPCOM-8: Deep learning for communications |
| Location | Gather.Town |
| Session Time: | Friday, 11 June, 14:00 - 14:45 |
| Presentation Time: | Friday, 11 June, 14:00 - 14:45 |
| Presentation |
Poster
|
| Topic |
Signal Processing for Communications and Networking: [SPC-ML] Machine Learning for Communications |
| IEEE Xplore Open Preview |
Click here to view in IEEE Xplore |
| Virtual Presentation |
Click here to watch in the Virtual Conference |
| Abstract |
The high peak-to-average power ratio (PAPR) of orthogonal frequency division multiplexing (OFDM) transmissions is susceptible to non-linear distortion caused by power amplifier (PA) saturation. In this work, we propose a novel technique, using residual neural networks and soft clipping, to deterministically limit the peak amplitude of the signal, thus lowering its PAPR and circumventing PA distortion. We show that the proposed solution is capable of significantly reducing PAPR and in-band distortion, while obeying a spectral mask. Furthermore, we show that the neural network is able to generalize for a range of peak amplitudes, thus eliminating the need to re-train the network when the requirement changes. |
