In order to effectively improve the communication quality in the extremely-low frequency (ELF) communication, a whole model of analog circuits and transform domain algorithm is constructed. Analog circuits include a pair of magnetic antennas, an amplifier and a group of filters. The distributed capacitance of the magnetic antenna is effectively reduced by the segmented winding method. Analog circuits used to amplify and filter received signal are designed. Besides, a magnetic sensor with high sensitivity is produced. The Karhunen-Loève transform (KLT) algorithm applied to the field of interference suppression is deduced in detail. The transform successfully passes the received signal along the basis vector in sub-band, but the interference signal along the vector is attenuated. Therefore, the problem of an optimal filter is converted into the solution of transform factor for each sub-band. Then the relationship between the KLT transform and the time domain algorithm in the interference suppression problem is given. Based on the KLT algorithm, Fourier transform (FT) that makes the correlation matrices of the received signal diagonalized approximately is applied to the interference suppression algorithm. Based on the deduction results, the final optimal filter expressions are basically the same as the KLT algorithm. Finally, the experiments are carried out by using the simulated signal and real collected data in the laboratory, respectively. The schematic diagram of the real collected device is presented. The experimental result shows that, no matter the analog signal or the real collected data, the proposed algorithm can effectively suppress the interference. For the simulation, the performance of KLT algorithm is basically same as that of FT algorithm, but KLT algorithm is obviously better than FT algorithm for real collected data.