This paper is about an efficient implementation of adaptive filtering for echo cancelers. Recently a fast converging algorithm called Robust Proportionate Normalized Least Mean Squares (RPNLMS++) against double-talk has been proposed. This paper presents a realization of an improved version of the previous RPNLMS++ adaptive filter using block structure in which the filter coefficients are adjusted one per each output block. Then, an efficient implementation of the block filtering process is proposed using Number Theoretic Transforms (NTT) which can significantly reduce the computation complexity of filter implantation on Digital Signal Processor (DSP). Analyses of convergence properties, during single and double-talk, and complexity show that the new block adaptive filter permits fast implementations while maintaining performance equivalent to that of the widely used RPNLMS++ adaptive filter.