Low-Density Parity-Check code(LDPC) decoding is one of the key steps in reference signal reconstruction of passive radar. Decoding algorithms can improve data bit errors caused by noise, but with the high complexity. Combining high memory throughput and tremendous computational horsepower of GPU graphics processor, this paper proposes three types of parallel LDPC decoding algorithms suitable for GPU processing based on hard, mixed, and soft decisions. The complexity, decoding performance, and impact on radar signal processing of the three types of algorithms are described; finally, a parallel GPU implementation scheme is given, and the real-time time-consuming of the algorithms is compared. The simulation and measured results verify that the soft decision parallel algorithm is superior to other algorithms on performance and effectiveness.