Satellite communication service providers aim at exploiting their available bandwidth resources as much as possible. This is not only because there is a high demand for provided links, but also the large coverage area of a satellite beam often does not allow to separate different satellite downlink signals in the spatial domain. Consequently, many terminals need to share the same time and frequency resources and the respective leasing costs are significant. One possible approach to decrease the required bandwidth for bidirectional links between two ground stations over a satellite relay is to implement their connection using the paired carrier multiple access (PCMA) technique. Here, the satellite transponder transmits the sum of the two uplink signals as a common downlink signal and thus requires only one instead of two distinct time-frequency resource slots for the downlink transmission of the two directions of that bidirectional link. The resulting self-interference can be canceled by the respective ground stations using their own transmit signal. In contrast, a third-party ground station that receives the cochannel downlink signal, e.g. for unauthorized interception of the communication of the first two ground stations, needs to process the two interfering signals blindly and demodulate both component signals jointly. In this thesis, the acquisition of the signal is assumed to be accomplished by exploiting standard frequency domain segmentation methods and possibly prior knowledge on the existence of the particular PCMA signal. Furthermore, several joint detection schemes for the two transmit signals are available or have been proposed in the literature. To realize such detection scheme, prior estimation of the signal and channel parameters, i.e., the synchronization of the two signals, needs to be conducted. These parameters include the symbol rate, the carrier frequency offsets (CFOs), the symbol timing offsets, the carrier phases and the signal amplitudes or powers, respectively. Moreover, the knowledge of the roll-off factors of the employed root-raised-cosine (RRC) transmit pulses and the noise power or signal-to-noise ratio (SNR) is desirable for the subsequent demodulation process. As a consequence of the application scenario of the blind receiver, the estimation schemes cannot rely on exploiting known data or training sequences embedded in the received signal for inference of the parameters. Because the detection performance in such blind PCMA reception scenario is limited by interference, any joint sequence estimation algorithms usually require accurate knowledge of the channel for the computation of the signal hypotheses without introducing additional distortions. In this thesis, estimation schemes for the signal and channel parameters are designed in order to facilitate such joint sequence estimation algorithms. For most of the parameters of interest, the proposed algorithms utilize estimates of higher-order and cyclostationary statistics of the received signal from which the particular signal and channel parameters can be derived. An important requirement for separating the statistical features induced by the respective component signals is a small but measurable difference of their CFOs. Specifically, the symbol rate estimation is performed using a set of estimated cyclic correlations, i.e., the coefficients of the trigonometric series representation of the corresponding periodic autocorrelations. In contrast to the related approaches for single-carrier (SC) signals from the literature, the set of cyclic correlations proposed for PCMA signals regards the interference of the two component signals. The presented approach thus avoids the phenomenon by which the contributions of the SC signals to the statistical feature that represents the symbol rate cancel each other out. The proposed scheme for estimation of the CFOs exploits the connection of the cycle frequencies of the higher-order moments to the CFOs and detects the corresponding cycle frequencies. For this, the implemented algorithm properly models the interference scenario and adjusts the space of candidate frequencies accordingly. Two different approaches for the sequential estimation of the symbol timing offset and carrier phase of each respective component signal are proposed. One derives the parameters from the phases of the estimated cyclic moments, the other one maximizes the magnitude of the moment polyphases for inference of both parameters. The estimation of the signal and noise powers is considered for two different cases of assumptions on the reception scenario. In the first case, the roll-off factors of the RRC transmit pulses of the component signals are identical and the receiver input filter captures sufficient out-of-band noise. Here, the power imbalance of the component signals is inferred using estimates of statistics called the cyclic cumulants which correspond to the respective component signals, and the noise power is derived from an estimate of the autocorrelation of the stationary noise. In another case where the RRC transmit pulses are not identical, it is proposed to estimate the respective roll-off factors by a compact neural network-based classifier which is adjusted using machine learning techniques. The knowledge of the roll-off factors facilitates the subsequent estimation of the powers of the component signals exploiting again the cyclic cumulants. Here, several variants of using these statistical features are proposed which respectively show superior performance for specific cases of modulation schemes of the component signals. For all proposed estimators, the respectively required statistical quantities are introduced and the corresponding stochastic properties are derived for the case of the PCMA signal. The estimation performance is evaluated using numerical simulations and compared to appropriate references. The considered experiments include representative cases of parameter sets and reception scenarios which allow conclusions on the applicability of the schemes in practice. Such applicability can be established based on the presented results. However, for certain modulation schemes that require the processing of statistics of high orders, the necessary length of the signal segment that is used for estimation of the statistical features is large. This observation is characteristic for the PCMA scenario which is governed by a strong cochannel interference.
Andreas Feder (Thu,) studied this question.