In this paper, a new deconvolution estimator design method is proposed to solve the problems caused by unknown input for a discrete time-varying linear system (DTVLS). These problems arise in cases of image restoration or geological exploration. Our proposed method modifies the state estimate error x(k)âx^(k) of traditional performance measures to the input estimate error u(k)âu^(k) to develop a deconvolution estimator via game theory in an Hâ setting. In this paper, unknown input is estimated using noisy measurements, and statistical information about the process and measurement noises is not necessarily required in an Hâ setting. The new deconvolution estimator yields an estimated input that is close to the actual input and that leads to improvements in terms of MSE and PSNR for performance.