A new approach to multipulse LPC coder design

Two new ideas for the design of multipulse excited LPC coders are presented in this paper. The first idea relates to improving the extraction of the all-pole filter parameters by taking into account the error weighting function. This function, which takes advantage of the noise masking properties of the ear, is ignored in the conventional covariance and autocorrelation methods. The new approach leads to an iterative algorithm, in which the first iteration is essentially the covariance method. Each new iteration involves estimation of the residual and increases a likelihood function, taking into account the error weighting function. The second idea relates to improving the derivation of the excitation parameters. A recursive algorithm for the optimal choice of the i-th pulse location, given the previously obtained (i-l) pulse locations, is presented. The low complexity of the new algorithm enables its combination with a tree search algorithm, thus giving a solution which is better than the solution of the algorithms reported earlier , when the error weighting function is used