Effectiveness Study of Code-Aided and Non-Code-Aided ML-Based Feedback Phase Synchronizers

This paper investigates the effectiveness of a (non-)code-aided ML-based FB phase synchronizer at the low operating signal-to-noise ratio of capacity-approaching codes. We show that the performance of the code-aided synchronizer is very close to that of a data-aided synchronizer that knows all data symbols in advance. This illustrates the optimality of the code-aided synchronizer. For the non-code-aided and the data-aided synchronizer, the linearized mean square phase error (MSPE) is evaluated analytically in the case of a first order loop. We demonstrate that, the MSPE of the non-code-aided synchronizer equals that of the data-aided synchronizer when the carrier phase is essentially constant and the loop filter gain is the same for both synchronizers, but that the non-code-aided synchronizer (as compared to the data-aided synchronizer) yields a larger MSPE due to phase fluctuations. This proves that code-aided FB phase estimation outperforms non-code-aided FB phase estimation when that the phase to be estimated is time-varying.