The optimization of physical layer security scheme in the multiple input multiple output (MIMO) fullduplex two-way communication system is studied

where legitimate nodes send confidential information and null space artificial noise while receiving information. For the case where the eavesdropping's channel state information is only statistically known and there is residual self-interference

the closed-form expression of the average achievable secrecy sum rate is first derived

then its lower bound is further given. Aiming at maximizing the lower bound

the power allocation factors of the artificial noise and the information signal

along with the power allocation matrices of the information signal are jointly optimized.An iterative method is used. In each iteration

we first fix the former and optimize the latter

then fix the latter and optimize the former. The DC (Difference of Concave/Convex

DC) programming and the genetic algorithm are used to optimize the information’s power allocation matrices as well as the power allocation factors of the information signal and artificial noise respectively. The simulation results prove that the theoretical derivation is correct

and the optimization algorithm can effectively improve the average secrecy sum rate.