WIT-Waveform Independent Tunable Channel Model for sub-Terahertz Communication

Ultra-broadband communication in emerging spectrum, like the sub-Terahertz (THz) and THz band is envisioned as one of the leading technologies to meet the exponentially growing data rate requirements of future wireless communication networks. However, there is a lack of availability of differentiable channel models for ultra-broadband links for such less explored spectra. Furthermore, the channel and hardware impairments are indistinguishable from each other making the problem complex to model in closed form. In this work, we propose a Generative Adversarial Network (GAN) based channel model that captures the non-linearity in both channel and the underlying hardware and can be used in an end-to-end optimization framework. The proposed model exploits the adversarial learning properties to approximate the stochastic channel distribution while the domain knowledge is used to maintain a information theoretic relevance of the generated channel model by guaranteeing high mutual information. The channel model is developed without any dependence on the baseband signal, which is essential for embracing any new modulation or pre-equalization techniques at the transmitter side. Our proposed model is tunable to various sub-THz frequencies, ultra-broad bandwidths and channel power. Results from comprehensive over-the-air experiments in 140 GHz and 240 GHz frequencies show high model accuracy of up to 93% for different types of waveforms and provides tunability using domain knowledge.