Online-Learnt and Deployed MoDem System

The telecommunications industry has seen a significant increase in the demand for high-quality transmission and processing of digital information. As a result, research efforts have been focused on developing high-performance systems for the modulation and demodulation of digital information signals. These efforts have led to exploring artificial intelligence (AI) techniques for implementing demodulation. However, traditional AI-based demodulation techniques use specific data for supervised training, which results in models that hardly adapt to changing environments. These models do not consider the time variance that can affect a realistic communication chain, where physical devices and their functionalities inevitably influence the signal. Moreover, the non-idealities that characterize the communication channel are subject to unpredictable and random temporal fluctuations. To address these issues, this manuscript proposes a novel system that implements modulation and demodulation of digital signals and can adapt the demodulation to the characteristics of the specific front-end adopted, as well as to those of the communication channel, during execution on the field. The proposed system estimates the main parameters that characterize the received signal and performs neural demodulation at the receiver. This is achieved through an online learning process that takes place at the transmitter and is sent to the receiver in real-time. This approach allows the system to adapt to the statistically varying conditions of ongoing communication, resulting in improved performance and reliability.