Spatial modulation (SM) is a recently introduced multiple-input and multiple-output (MIMO) scheme based on a new operation principle. In contrast to conventional spatial multiplexing (SMX) MIMO schemes, which usually activate all transmitter (TX) antennas for signal transmission, SM activates one out of all available TX antennas in each transmission interval. Changing the active TX antenna enables signal transmission in the spatial domain, parallel to the already present transmission in the in-phase and quadrature (IQ) domain. In this way, SM transmits practically two independent data streams (in the IQ and the spatial domain) utilizing only one RF chain at the TX. As a result, SM achieves significant hardware complexity and signal processing reductions compared to conventional SMX MIMO systems. Also, the use of a single RF chain at the TX provides energy efficiency gains in SM systems. ^Conventional SM works without the knowledge of channel state information at TX (CSIT), which is generally considered an advantage in wireless communications. On the other hand, many techniques, that tend to improve the performance of SM systems, require the knowledge of CSIT. One of these techniques is transmit precoding. The focus of this thesis is on the performance gains of SM systems, achieved by implementing this technique. In addition, transmit precoding can be used for design of new transmission schemes such as receive spatial modulation (RSM), whose properties are also studied in this thesis. For the sake of completeness, the precoding schemes for SM systems are analyzed considering signal propagation in the low-GHz (i.e., below 6 GHz) and in the millimeter-wave (mmWave) frequency range. For RSM with minimum mean square error (MMSE) precoding we derive an upper-bound bit error probability (BEP) expression and perform a detail error performance analysis. In comparison to RSM with zero forcing (ZF) precoding, RSM with MMSE precoding achieves better error rates and this error rate difference generally depends on different systems parameters. To obtain a deeper understanding of the properties of MMSE precoding, we introduce three different detectors for RSM systems with MMSE precoding. For those detectors we evaluate their error performance characteristics and computational complexities. In addition, the concept of RSM is investigated to the mmWave frequency range in an indoor environment that is dominated by the line-of-sight (LOS) signal components. Taking the characteristics of the LOS components, the optimal operation condition that ensures minimum error probability is derived and under this condition a simple TX hardware architecture, which uses only phase shifters for precoding, is proposed. ^Furthermore, we optimize the error performance of SM systems in mmWave indoor LOS channels by introducing a simple precoder that is purely based on phase shifting. As the precoding coefficients of this precoder are chosen from a fixed predefined codebook, only a low-rate feedback channel is needed between RX and TX. This precoder is particularly efficient in SM systems with low order IQ constellations. Finally, we propose the utilization of channel cut-off rate as a metric to design precoding schemes that optimize the mutual information and the channel capacity of SM systems. Numerical results show that cut-off rate gains achieved by using the proposed precoding schemes in SM systems are very similar to proper mutual information gains. Also, we can observe that the precoding schemes which are designed for increasing array gain in SM systems give the largest cut-off rate gain at low signal-to-noise ratios (SNRs). On the other hand, the precoding schemes that increase the minimum inter-symbol Euclidean distances in SM systems are most convenient for an implementation at medium to high SNR values.