Deep Generative Modeling of RF Communication Signals: A Study Using GANs for Synthesizing OFDM-QAM Waveforms

Deep Generative Modeling of RF Communication Signals: A Study Using GANs for Synthesizing OFDM-QAM Waveforms

In most cases, the recordings of radio frequency (RF) emissions from commercial communication hardware, carried out in realistic environments, are required against others undertakings like design and evaluation of the implementations of various practices of spectrum sharing technologies sha, training and testing spectrum sensing algorithms and more also for interference testing. Unfortunately, the nature of such data collections is time consuming, expensive and confounded with data-sharing restrictions presents enormous hurdles that constrain the data set availability. Further to that, modeling real-world behavior of RF emissions empirically from first principles is often very hard because one does not have full knowledge concerning system parameters and implementation details as well as the complexity of system dynamics which one wants to characterize is often not quite easy. Therefore, there exists a gap for adaptive and mechanistic framework in that data from existing datasets can be re-used in order to create more of the same class of waveforms. One development in machine-learning that shows promise in this area is deep generative modeling using generative adversarial networks (GAN) for unsupervised learning. Up to now, not much research has been done to look into GANs for RF communication signals. Here we report the first detailed analysis of the fidelity of generated baseline baseband OFDM-QAM signals, created by GANs. In the current work, we built upon previous GAN methods and constructed two new GAN architectures, testing their effectiveness on simulated datasets with known ground truth.

More specifically, we examined how the model performance varies with the incremental increase in complexity of the data set over several OFDM parameters and fading channels conditions. These results presented here inform the feasibility of use cases and lay the groundwork for further research exploring deep generative models of RF communication signals

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Read More