Rapid and accurate motion control of industrial robots has a vital demand on the precise load torque estimation. To achieve highly robust motion control against dynamic torque, our previous research had introduced the Kalman-filter based instantaneous state observer (ISOB) to attain the instantaneous load torque estimation of a robot arm. This paper proposes the load torque estimation by ISOB based on singular spectrum analysis (SSA) to improve the torque estimation performance. The proposed method achieves highly precise load torque estimation, especially in impulsive responses, with effective noise suppression. The proposed method is targeted to realize the high performance load acceleration control for industrial robots. The validity of the proposed method is verified by numerical simulation results based on the speed-control experimental data of a robot arm.