The article describes and compares two OFDM based communications schemes for reducing the effects of the combination of Narrowband Interference (NBI) and Impulsive Noise (IN), which are noise types typical in Power Line Communication (PLC). The two schemes are Modified BPSK-OFDM (called MBPSK, for brevity) and QFSK-OFDM (called QFSK, for brevity), which are non-conventional OFDM schemes. We give a description of the two schemes, showing how they are derived and also show their similarities and eventually compare their performances. Performance simulation results, in terms of bit error rate, are given to compare the systems under the effect of IN and NBI. The popular Middleton Class A model is used for modelling IN. The results show that MBPSK scheme outperforms the QFSK scheme in terms of minimum distance, and hence in terms of bit error probability when no preprocessing is performed. However, under clipping/nulling, both schemes eventually reach the bit error rate floor.

Nowadays, there is a trend to employ adaptive solutions in mobile communication. The adaptive transmission systems seem to answer the need for highly reliable communication that serves high data rates. For efficient adaptive transmission, the future Channel State Information (CSI) has to be known. The various prediction methods can be applied to estimate the future CSI. However, each method has its bottlenecks. The task is even more challenging while considering the future 5G/6G communication where the employment of sub-6 GHz and millimetre waves (mmWaves) in narrow-band, wide-band and ultra-wide-band transmission is considered. Thus, author describes the differences between sub-6 GHz/mmWave and narrow-band/wide-band/ultrawide- band channel prediction, provide a comprehensive overview of available prediction methods, discuss its performance and analyse the opportunity to use them in sub-6 GHz and mmWave systems. We select Long Short-Term Memory Recurrent Neural Network (RNN) as the most promising technique for future CSI prediction and propose optimising two of its parameters - the number of input features, which was not yet considered as an opportunity to improve the performance of CSI prediction, and the number of hidden layers.