End-to-end DVB-S2X system design with deep learning-based channel estimation over satellite fading channels

Digital Video Broadcasting – Satellite Second generation extension (DVB-S2X) has been introduced with a relatively higher number of modulation schemes and code rates (MODCODs) to satisfy the demand for high data rates and qualified broad�casting services. However, the atmospheric impairments are considered a serious problem in satellite communication in tropical regions, which are mostly character�ized by heavy precipitation, especially at high frequencies. For these reasons, the design of satellite fading channels for tropical regions becomes an urgent necessity not only to study the effect of heavy fading caused by these impairments on the performance of such a satellite system but also to find solutions to enhance the performance of the DVB-S2X system in these heavy fading channels. In this thesis, the contribution can be divided into four main parts: In the first part, the end-to-end DVB-S2X system with most of its MOCODs and two frame sizes were introduced. Monte Carlo simulation is used to implement the sys�tem model with two scenarios; the Additive White Gaussian Noise (AWGN) channel is used in the first scenario to validate the DVB-S2X system by comparing the re�sults with the European Telecommunications Standards Institute (ETSI) standard. In the second scenario, the system is evaluated with a Rician channel which represents the real channel for satellite transmission. Comparisons in bit error rates have been made between those two models to observe the impact for Shannon channel capacity and spectral efficiency for different (MODCODs). Moreover, the study improves the assessment level of DVB-S2X system performance with different types of channels and MODCODs. The atmospheric impairments on the Ka-band satellite channel are considered in the channel design, especially the rainfall effect, which is the most effective atmospheric impairment that degrades the system performance. For this reason, two rainy fading channels are designed in the second part of this thesis, one for the tropical region termed as (Tropical channel) and the other for the temperate region termed as (Tem�perate channel), using real rain data from these two areas. In the third part, the first full design of the DVB-S2X system with multi-user�multiple-input-single-output (MU-MISO DVB-S2X), with most of its modulation and coding schemes (MODCODs), over rainy fading channels is presented. The proposed model mitigates the fade in heavy fading channels by utilizing zero-forcing beamforming (ZFBF) and semi-orthogonal user selection (SUS) techniques. Besides, the user scheduling influence on the bit error rate (BER) performance of the MU-MISO DVB-S2X system is tested and compared with the conventional MISO DVB-S2X system. Simulation results show that the proposed system can achieve a significant improvement in terms of BER performance with at least 20 dB for 128 amplitude and phase-shift keying (128APSK) MODCOD over the tropical channel and 14 dB for 32APSK MODCOD over temperate channel when the number of users is six. The BER performance is more improved when the number of users increased to 20. The enhancement in error rates proves that the MU-MISO DVB-S2X system with scheduling can be the key solution for DVB-S2X system performance degrada�tion in fading channels, especially rainy fading channels. In the fourth part a deep learning (DL) algorithm of channel estimation for two fad�ing channel models, Tropical and Temperate in the satellite communication system is presented. The Normalized Mean Square Error (NMSE) and the BER perfor�mances for different DVB-S2X system MODCODs are investigated and the results for these algorithms are compared with the conventional Minimum Mean Square Er�ror (MMSE) and Least Square (LS) channel estimation techniques. Two DL-based channel estimators are proposed termed as (DLBLSTM) and (DLGRU ). The channel estimation results indicate that the adopted DL architectures are more robust than conventional techniques when fewer training pilots are used for both fading channels. Although the conventional algorithm, MMSE, outperforms the pro�posed algorithms when the number of pilots increased but it is not applicable in real transmission as it is required prior knowledge about the channel statistic which is not the case with DL-based estimators which rely only on the pilots. For example, when the number of pilots p = 37, the NMSE performance for the MMSE estima�tor is 5.147×10−4 for the normal frame. Whereas, the DLBLSTM estimator gives slightly lower performance than the MMSE with 7.216×10−4 . The DLGRU estima�tor achieves 8.849×10−4 which is the worst performance among all estimators. In addition, the complexity of the proposed schemes is lower than those of competi�tive algorithms. Finally, we can conclude that DL still has potential although more efficient architectures are required.

Text FK 2021 94 UPMIR.pdf Download (1MB)

Actions (login required)

View Item