Timing recovery loops are the state-of-the-art systems in telecommunication receivers with the purpose to recover the correct sampling time. They retrieve the symbol phase maximizing the Modulation Error Rate (MER) performance. In this paper, we propose a PSK symbol synchronizer based on a Reinforcement Learning method called Q-Learning. We designed an intelligent system which autonomously becomes able to minimize the symbol amplitude variance by a trial-and-error iterative process. This results in a correct timing synchronization. The experimental results show that, below a $E_{b}/N_{0}=22.5\mathbf{dB}$ channel noise level, the MER figures are lower (2 to 2.5dB) yet comparable with a Mueller and Müller loop. Above such a noise threshold, we measured 1dB of MER improvement. Moreover, the system response to impulsive, step and ramp phase perturbation was characterized in terms of number of symbols to recover a stable locked state: 65 in the first, 420 in the second and 750 in the third case.
Paper
Full text
A Q-Learning based PSK Symbol Synchronizer
Semantic Scholar · Engineering · 2019
Abstract
Timing recovery loops are the state-of-the-art systems in telecommunication receivers with the purpose to recover the correct sampling time. They retrieve the symbol phase maximizing the Modulation Error Rate (MER) performance. In this paper, we propose a PSK symbol synchronizer based on a Reinforcement Learning method called Q-Learning. We designed an intelligent system which autonomously becomes able to minimize the symbol amplitude variance by a trial-and-error iterative process. This results in a correct timing synchronization. The experimental results show that, below a $E_{b}/N_{0}=22.5\mathbf{dB}$ channel noise level, the MER figures are lower (2 to 2.5dB) yet comparable with a Mueller and Müller loop. Above such a noise threshold, we measured 1dB of MER improvement. Moreover, the system response to impulsive, step and ramp phase perturbation was characterized in terms of number of symbols to recover a stable locked state: 65 in the first, 420 in the second and 750 in the third case.