We propose and experimentally demonstrate a universal and blind polarization-state tracking scheme for M-quadrature amplitude modulation(QAM) signals based on a decision-free radius-directed linear Kalman filter(RD-LKF).The polarization tracking performance is investigated through simulation and experiments for both quadrature phase-shift keying(QPSK) and 16-QAM signals.The influence of the filter parameters on the static polarization demultiplexing performance and dynamic tracking capability are discussed via simulation.The optimization strategy of the filter parameters in modulation-format-independent scenarios is proposed and simulations are carried out to evaluate the polarization demultiplexing penalty for QPSK,16 QAM,and hybrid QPSK/16 QAM signals.Finally,the proposed decision-free RD-LKF is experimentally compared to the respective algorithms of constant modulus algorithm and multimodulus algorithm for QPSK and 16-QAM and the advantage of ultrafast polarization tracking ability is confirmed. We propose and experimentally demonstrate a universal and blind polarization-state tracking scheme for M-quadrature amplitude modulation (QAM) signals based on a decision-free radius-directed linear Kalman filter (RD-LKF). The polarization tracking performance is investigated through simulation and experiments for both quadrature phase-shift keying (QPSK) and 16-QAM signals. The influence of the filter parameters on the static polarization demultiplexing performance and dynamic tracking capability are discussed by simulation. The optimization strategy of the filter parameters in modulation-format -independent scenarios is proposed and simulations are carried out to evaluate the polarization demultiplexing penalty for QPSK, 16 QAM, and hybrid QPSK / 16 QAM signals. Finally, the proposed decision-free RD-LKF is experimentally compared to the respective algorithms of constant modulus algorithm and multimodulus algorithm for QPSK and 16-QAM and the advantage of ultrafast polarization tracking ability is confirmed.