The fifth generation of mobiles communications (5G) requires a variety of applications that the fourth generation (4G) cannot achieve, such as a higher data rate, ultra-low power consumption and short response times. In order to achieve these requirements one of the proposed physical layer waveform is the generalized frequency division multiplexing known as GFDM. In this thesis the characteristics and the feasibility of GFDM are studied. First of all, the GFMD principles are introduced. Then, the implementation of the GFDM scheme on the Vienna 5G simulator, including modulation and demodulation schemes are detailed. Hence, the Fast Fading channel object of the Vienna LTE simulator is adapted to the next generation in order to analyze the new waveform. Since the one tap equalization already implemented is not effective for the new waveform, a time domain equalization based on dividing the channel in Sub-blocks and averaging each one of them is implemented. After the development of the waveform the parameter optimization using machine learning and deep learning techniques is performed. First, different regression models that predict the BER are implemented and analyzed. Secondly, these regression models are optimized, using genetic algorithms, in order to find the optimal number of channel sub-blocks that minimize the BER. Subsequently, after applying the parameter optimization, the GFDM waveform performance i.e, BER results, PSD and the Tx signal, compared with other waveforms that are also possible candidates for the fifth generation systems as FBMC or OFDM. Finally, it is proved that GFDM is a suitable candidate for the 5G standard, and that the optimization algorithm developed has an outstanding performance and highly scalability.

Se ha implementado el esquema de la modulación GFDM (Generalized Frequency Division Multiplexing) en el simulador 5G de la TU Wien. Se ha comparado el comportamiento de la modulación GFDM con OFDM y FBMC. Se ha simulado un canal de alta movilidad, para analizar el comportamiento de GFDM en escenario de alta movilidad. Se ha desarrollado un método de ecualización conjuntamente con un algoritmo que optimiza uno de los parámetros clave de la ecualización. El algoritmo utiliza técnicas de aprendizaje autónomo ( Machine Learning)

S'ha implementat l'esquema de la modulació GFDM (Generalized Frequency Division Multiplexing) en el simulador 5G de la TU Wien. S'ha comparat el comportament d'aquesta modulació amb OFDM i FBMC. S'ha simulat un canal d'alta mobilitat, per tal d'analitzar el comportament de GFDM en escenaris d'alta mobilitat. S'ha desenvolupat un mètode d'equalització juntament amb un algoritme que optimitza un dels paràmetres clau d'aquesta equalització. L'algoritme utilitza tècniques d'aprenentatge autònom (Machine Learning)