Abstract: For frequency hopping/M-ary frequency shift keying (FH/MFSK) Systems, combining convolutional coding, a soft decoding algorithm is proposed by adopting M-dimensions energy information for the soft-demodulation of MFSK signals as soft input of decoder, avoiding the approximate bit logarithm likelihood ratio (LLR) calculation. The theoretical performance of the proposed algorithm is analyzed for FH/MFSK systems in the presence of partial-band noise jamming (PBNJ) over additive white Gaussian noise (AWGN) and Rayleigh fading channels. The upper bound for closed-form expressions of BER performance are derived and validated by computer simulations. The results show that the proposed algorithm obtains obvious anti PBNJ processing gain compared to the approximate bit LLR and hard decision.