Many challenges are introduced with the evolution from the traditional electrical grid to the new smart grid. One of these main challenges is the need for a reliable communication network to exchange information and control the loads of each end user in real time. The most attractive access part solution for this is the use of Power Line Communication (PLC) technology that uses the grid as the transmission medium. The access PLC systems represent the area between the transformer to the end users in the same network, where the power line channel is made of the interconnection of various multi-conductor cables. A well designed PLC system in the access network can smoothly bridge the transmission grid and the customers together, and play a key role in the realization of smart grid. This thesis starts with the development of an accurate power line channel simulation model based on the multi-conductor transmission line theory. This model is able to predict the end-to-end transfer function for any user with any two conductors in the network, on the basis of the multi-conductor cable structure and the network topology. Furthermore, the model can easily evaluate the crosstalk between different pairs of transmission line when there are parallel transmissions in the cable. Due to unmatched impedance at the users premises and branches, the obtained transfer functions are strongly frequency-selective and thus results in another research issue of loading algorithm. With obtained noise density, the PLC channels capacity is analyzed. Calculations of the channel capacity provide encouraging results for PLC in access network. Besides the existing approaches of eliminating the impulsive noise in PLC environment, we consider the frame length as a possible parameter which can be adjusted accordingly to minimize the influence from impulsive noise. With a full knowledge of the PLC channel, the issue of resource allocation in a frequency-selective channel was addressed in the light of the water-filling theory. The algorithms aim to minimize the bit error ratio with given power and bit rate are considered as the objective is to improve the reliability of the PLC system. The improved Fischer-Huber algorithm and improved H-H algorithm appeared much flexible and easy to implement because they considered the big variation of the channels attenuation in the bandwidth of interest, a specific character of PLC environment. In the fourth and final part, we move up to MAC layer. After comparing three different popular MAC layer protocols, the most suitable protocol for reliable applications such as power grid monitoring and control is suggested. A fast retransmission policy was proposed to further advance the performance.