Abstract: The frequency domain adaptive filter (FDAF) has been widely used in practice due to its low complexity and good convergence performance. The choice of the fixed step size should be balanced between several factors, for instance, the initial convergence, the steady-state misadjustment, the tracking capability and the robustness to noise interference. The optimal step-size control of the unconstrained FDAF algorithm is studied in this paper. The convergence performance of the unconstrained FDAF is analyzed, which is then adopted to derive the optimal step size and estimate the misalignment factor. Computer simulations in system identification and echo cancellation show that the proposed algorithm can achieve both faster convergence and lower steady-state misadjustment than the fix stepsize version, and it is free of double-talk detector owing to the robustness to near-end interference.