Computationally Efficient AI for Extending DSP Functions in Wireless Communications and Sensing

The ability to support communications and sensing in congested RF environments is an important and growing need for commercial and military applications. In particular there is a need for overcoming (at reasonable cost) the challenges posed by fast changing and complex non-stationary channels and very busy spectrum, which include many users including non-cooperative ones. Standard DSP methods that process signals in the temporal and spatial domains are often limited in their ability to efficiently adapt to such environments with limited resources. We will describe a general framework for using physics-infused and prior-informed AI based methods that provide a cost-effective replacement for traditional DSP functions while dramatically improving the ability to operate in such environments. We will show results of using these methods in the HF band and their ability to overcome challenges posed by the ionospheric channel and by strong interference. In some important cases, we achieve over 30dB of improvement in signal to interference levels compared with current methods. In addition, we will describe an FPGA based implementation of this AI architecture, which offers a low-cost embedded solution.