A DCT domain quadratic predictor with normalized least mean square (NLMS) algorithm, efficient implementing structure of the reduced parameter second-order Vloterra filter (RPSOVF), is proposed to investigate nonlinear real-time multi-step prediction performance of three kinds of continuous chaotic signals. Experimental results show that the real-time ahead one-step prediction mean square errors of the DCT domain quadratic predictor proposed in this paper are at least 100 times smaller than that of RPSOVF,...