Title:From Channel Measurement to Training Data for PHY Layer AI Applications

Abstract:Learning-based techniques such as artificial intelligence (AI) and machine learning (ML) play an increasingly important role in the development of future communication networks. The success of a learning algorithm depends on the quality and quantity of the available training data. In the physical layer (PHY), channel information data can be obtained either through measurement campaigns or through simulations based on predefined channel models. Performing measurements can be time consuming while only gaining information about one specific position or scenario. Simulated data, on the other hand, are more generalized and reflect in most cases not a real environment but instead, a statistical approximation based on a mathematical model. This paper presents a procedure for acquiring channel data by means of fast and flexible software defined radio (SDR) based channel measurements along with a method for a parameter extraction that provides configuration input to the simulator. The procedure from the measurement to the simulated channel data is demonstrated in two exemplary propagation scenarios. It is shown, that in both cases the simulated data is in good accordance to the measurements