The authors study the dependence of the shape of electromagnetic pulse received in near radiation zone of the antenna on the observation point. The paper discusses negative and positive effects of this phenomenon on the wireless impulse communication and presents a new method of information extraction form ultrawideband electromagnetic pulse, comparing it to the traditional way of signal processing. The method is based on modern deep learning technics and recurrent neural networks, namely physical long short-term memory. Moreover, the paper presents a concept of direct sequence ultrawideband (DS-UWB) impulse radio receiver based on a physical neural network. It is proposed to change the traditional way of radio signal processing and use a single neural network instead of a matched filter, a magnitude amplifier and a FPGA processor. The architecture of the physical neural network was designed with an intention to study the behavior of ultrawideband short pulse (UWB-SP) radio signal in near and far radiation zones. The applicability of the neural radio concept is proved by simulation of AWGN communication channel for multiuser environment and real time RX signal processing by the designed neural network. The paper contains the results of a numerical modeling of the radiation-reception process and illustrations of the neural network training process. The lens impulse radiation antenna is considered as radiator of transient electromagnetic field for simulation. The radiation process is modeled with the help of the antenna’s transient response obtained using the evolution approach and the superposition principle in the form of Duhamel integral. The prospects of using the proposed methodology in the problems of the Internet of Things are analyzed. The study shows that using the proposed method allows solving multipathing and multiuser problems even in near radiation zone.