Blind Roll-Off Factor Estimation for Paired Carrier Multiple Access Satellite Signals

The downlink signal of a paired carrier multiple access (PCMA) satellite system consists of two single-carrier signals that use the same time-frequency resources and hence appear as co-channel interference w.r.t. each other. Any blind receiver for such downlink signal requires knowledge of the various signal parameters in order to enable the detection of the data of both carriers e.g. by sequence estimation schemes. Assuming linear modulation and root-raised cosine pulses, this paper focuses on the estimation of the transmit filter roll-off factors of the respective carrier signals. Their knowledge is critical for blind PCMA receivers because distortions by suboptimal receiver input filtering besides the expected strong inter-carrier interference should be avoided. In contrast to single-carrier scenarios, prior works do not offer solutions for co-channel signals as present in the PCMA scenario. Here, we derive an approximate maximum-likelihood (ML) scheme for estimation of the roll-off factors. Furthermore, we propose an artificial neural network based scheme using a relatively compact architecture comprising only few convolutional and fully-connected layers which overcomes the practical limitations of the ML scheme. The neural network is trained on a data set of simulated PCMA received signals. We evaluate and compare the performance of the different approaches via numerical simulation results for varying system parameters.