The power line is originally designed for power transmission instead of communication. The power line channel contains many power devices. The electronic components with switching functions can easily introduce high energy impulsive noise. Used as a communication system, such impulsive noise can seriously degrade the data transmission efficiency of the entire system. According to the research, the periodic impulsive noise with cyclostationary characteristic is the most threatening interference in narrowband power line channel, and its energy is usually dozens of times higher than background noise. In this thesis, we refer to the physical layer specification of IEEE P1901.2 standard as the simulation platform and propose a robust receiver architecture for narrowband power line communication. The frequency shift filter (FRESH filter) is used to estimate the cyclostationary impulsive noise and eliminate it. The fact that an orthogonal frequency-division multiplexing (OFDM) signal has cyclostationary characteristics due to the cyclic prefix (CP) can be used to reconstruct the received signal interfered by the impulsive noise. However, the cyclic frequency of impulsive noise is not a constant, and such cyclic frequency error can greatly degrade the performance of the FRESH filter. Therefore, we also propose an adaptive FRESH filter algorithm for cyclic frequency error compensation in this thesis. The simulation results show that the algorithm proposed by us can effectively estimate and compensate the cyclic frequency error, eliminate the cyclostationary impulsive noise, and hence improve the system performance.