The present invention relates to a method for encoding a vector for transmission from a transmitter to a receiver, comprising a step of selecting a first vector in a predefined M-dimensional vector codebook, and at least one refinement step wherein an error vector between the random vector and the first vector is quantised by means of selecting a further vector from a further vector codebook with dimensionality reduced by one relative to the previous step, wherein the M-dimensional vector codebook and the further predefined vector codebook are known to both the transmitter and receiver.