The weaknesses of the single trees - the tree-structured wavelet transform and binary segmentation was first addressed in [HKRV93]. The authors proposed a double tree which combines signal binary segmentation and wavelet packets. The double tree consists of the complete hierarchy of dyadic segmentations and the full wavelet packet decomposition of each segment. The double tree is illustrated in Figure 8(a). The diagram shows that the double tree does not segment any of the frequency F nodes. In other words, the outputs of the filter banks are not segmented spatially. This limits the performance of the double tree. For example, since frequency decomposition provides much of the energy compaction, branching from the root node towards the right, S, reduces the potential number of frequency iterations for the entire signal.