Soft output Viterbi algorithm(SOVA) is a turbo decoding algorithm that is suitable for hardware implementation. But its performance is not so good as maximum a posterior probability(MAP) algorithm. So it is very important to improve its performance. The non-correlation between minimum and maximum likelihood paths in SOVA is analyzed. The metric difference of both likelihood paths is used as iterative soft information, which is not the same as the traditional SOVA. The performance of the proposed SOVA is demonstrated by the simulations. For 1024-bit frame size and 9 iterations with signal to noise ratio from 1dB to 4dB, the experimental results show that the new SOVA algorithm obtains about more 0.4dB and 0.2dB coding gains more than the traditional SOVA and Bi-SOVA algorithms at bit error rate(BER) of 1×10~ -4 , while the latency is only half of the Bi-direction SOVA decoding. So its is as good as maximum a posterior probability (MAP) algorithm. So it is very important to improve its performance. The non-correlation The metric difference of both likelihood paths is used as iterative soft information, which is not the same as the traditional SOVA. The performance of the proposed SOVA is demonstrated by the simulations. For 1 024-bit frame size and 9 iterations with signal to noise ratio from 1dB to 4dB, the experimental results show that the new SOVA algorithm is more about 0.4dB and 0.2dB encoding gains more than the traditional SOVA and Bi -SOVA algorithms at bit error rate (BER) of 1 × 10 -4, while latency is only half of the Bi-direction SOVA decoding.