This work is part of the innovative "Active Generator" (AG) project. AG is a concept that suggests a new arrangement of the turbine-generator line of a high power utility (a few hundred of MW) in order to de-synchronize the rotation speed of the turbine-generator group from the fixed grid frequency (50 Hz or 60 Hz). This de-synchronization has essentially two advantages. First, the variable speed of the group enables the operation of the turbine at its best available efficiency in function of the delivered power. Second, the de-synchronization allows to eliminate the gearbox between the turbine and the generator without losing the important degree of freedom in the choice of optimal nominal rotation speed of the turbine. The latter advantage is particularly interesting for high power utilities, whose prime mover is a gas turbine, because for this power range the gearbox constitutes a heavy burden. The de-synchronization is realized with a static frequency converter which is a power electronics circuit composed of silicon power devices. The converter must ensure the same nominal frequency ratio than the gearbox it replaces, which can go above 50%. For such ratio the converter must be inserted between the stator windings of the generator and the grid. There are numerous different frequency converters. Some of them are available as industrial products and others are still in a development state. Not all of these different frequency converters are well adapted to high power applications. In the AG literature, a few recommendations suggest to use a low frequency commutation sequence, combined with a high number of input phases. The high number of input phases ensures a sufficient resolution of the converter's output voltage. Compared to others, this sequence is supposed to decrease the commutation losses of the converter, avoid the usual overdesign of the nominal power of the generator, and, finally, does not require the converter to include bulky intermediary DC storage components (capacitor or inductor). This sequence is a variant of the "Cosine Waveform Crossing" (CWC) method used for Naturally Commutated Cyclo-converters (NCC) and is named slowCWC. However, up till now, there is no converter that is able to run properly with this sequence. Thus a new converter is needed. This PhD work introduces a new converter that is able to fulfill the slowCWC sequence. It is derived from a slight modification of an existing topology (NCC) and is called "gate-commutated Polyphased Matrix Converter" (PPMC). It is a direct frequency converter with a high number of input phases, generally greater than twenty, and a matrix structure of the valves that allows to connect each of the three output phases to each of the generator (input) phases. The valves are bi-directional in voltage and current and are transistor-based to achieve the turn-off capability required by the commutation sequence. The PPMC requires to add protection circuits across each generator stator winding. These circuits protect the stator windings from overvoltages which appear during some forced commutations. In its first part this PhD work uses an analytical approach and the results are expressed in a per unit system that is also adequate to describe the electrical machines. In this first part, it is about the development of design rules for the components of the protection circuits. In addition the energy losses linked to these circuits are evaluated. Those losses strongly depend on the commutation type, which is itself influenced by the presence of the protection circuits. The expression of the duration of natural commutations in the per unit system is also developed in this first part and it constitutes a key parameter in the determination of the commutation type. These theoretical developments are illustrated with numerical simulations. In its second part this PhD work presents the realization of a small-scale experimental set-up with reduced power (1 kW) but a high input phase number (27). The aim of the experimental set-up is to implement and experiment in real-time the command and control algorithm of the PPMC as well as to verify the theoretical predictions developed in the first part. The results of those developments lead to the quantitative assessment of the efficiency of the PPMC. Besides the key parameters that can help to improve this efficiency are pointed out. In certain cases the efficiency of the PPMC is acceptable under the condition that a generator parameter (its leakage reactance) remains under a given limit. This work ends with a list of suggestions for future works related to the improvement of the PPMC and related to the AG project.