Intensive rollout of smart grids (SG) creates a special demand on communication networks. Especially, they have to supply sufficient quality of service (QoS) to a big number of customers. Therefore the required QoS becomes more demanding (e.g., minimum throughput have to be increased) and in meantime the cost of the network installation and management should be reduced. Power line communication (PLC) becomes a challenging technology inside SGs. It uses already available power cables and therefore reduces the installation costs of the network. In particular, the usage of PLC is beneficial for power utilities, which appear to be the owners of this communication medium. Nevertheless, PLC remains an emerging technology. Achievable QoS for particular PLC network scenarios are hard to predict and still need to be studied. It is reasoned by the fact, that the PLC environment (cable structure, electrical devices, etc.) induces special effects on the informational signal propagation over the communication medium. These effects may result in considerable impairments of the PLC signal. Nevertheless, with a usage of a cognitive mechanism a big part of them can be avoided or even taken as an advantage. This paper overviews the influence of the channel instability on the PLC QoS. A data link layer (DLL) of a PLC protocol can implement a channel prediction mechanism, which is beneficial for multiple PLC applications. A short description on this mechanism is also introduced in this paper.