Abstract: With the modern time series analysis method in the time domain and based on the autoregressive moving average (ARMA) innovation model and white noise estimators, new asymptotically stable multichannel optimal deconvolution filters in the ARMA innovation filter form and in the Wiener filter form are presented respectively. They avoid the solution of the Diophantine equations, and can handle the deconvolution filtering, smoothing and prediction problems in a unified framework. The relationship between the ARMA innovation filters and Wiener deconvolution filters is also given. A simulation example shows their effectiveness.