A new method for frequency detection by linear prediction using total least squares

This work is concerned with the use of the total least squares criterion in forward-backward linear prediction optimization, yielding a new frequency estimation method, named TLS-FBLP. A detailed comparison of our method to Tufts and Kumaresan's (1982) modified FBLP, revealed that TLS-FBLP is better than modified FBLP at low prediction orders, as well as they are similar for high orders.