Abstract:This paper proposed a novel search strategy of the nearest neighbor partition mutation, and a new codebook design algorithm via the hybrid engine combining the evolutionary programming (EP) with the estimation of distribution algorithm (EDA). The nearest neighbor partition control vector is introduced as the chromosome for implementing the nearest neighbor partition mutation to control the centroid track. Compared with other methods using GAs, the search space is relative small since the size of chromosome equals to the dimension of training vector, and easy for manipulation. In the proposed hybrid algorithm, the search direction is estimated by EDA, thereby the convergence rate is accelerated. The experimental results show the new codebook design algorithm makes better performance on improving the codebook quality.

Codebook Design Optimization Based on EDA-EP