Convex Design of Higher Order Allpass Transformed DCT Filter Banks with Perfect and Near-Perfect Reconstruction

Löllmann H, Kaup A (2026)

Publication Language: English

Publication Type: Conference contribution, Conference Contribution

Publication year: 2026

Pages Range: 1-5

Event location: Bruges, Belgium BE

Abstract

A new numerical design for higher order allpass transformed discrete cosine transform (DCT) analysis-synthesis filter banks with non-critical subsampling is presented. The allpass transformation of higher order allows a flexible adjustment of the non-uniform time-frequency resolution of the filter bank, which can be used, e.g., to construct auditory filter banks. Moreover, an allpass transformed DCT filter bank provides real-valued subband signals and can be implemented efficiently by a polyphase network, where its signal delay is much lower than for comparable tree-structured filter banks. In contrast to existing designs, the coefficients of the FIR synthesis filter bank are determined by a convex least-squares-error (LSE) optimization by which an allpass transformed DCT filter bank with perfect reconstruction (PR) is achieved. The presented convex LSE design provides a global optimum and can be solved rather efficiently. Moreover, it is shown how a relaxation of the PR constraint by limiting the aliasing distortions allows to improve the frequency selectivity of the synthesis filters, which is of interest for applications where small reconstruction errors are tolerable.

Authors with CRIS profile

Heinrich Löllmann Professur für Signalverarbeitung Andre Kaup Lehrstuhl für Multimediakommunikation und Signalverarbeitung (LMS)

How to cite

APA:

Löllmann, H., & Kaup, A. (2026). Convex Design of Higher Order Allpass Transformed DCT Filter Banks with Perfect and Near-Perfect Reconstruction. In Proceedings of the European Signal Processing Conference (EUSIPCO) (pp. 1-5). Bruges, Belgium, BE.

MLA:

Löllmann, Heinrich, and André Kaup. "Convex Design of Higher Order Allpass Transformed DCT Filter Banks with Perfect and Near-Perfect Reconstruction." Proceedings of the European Signal Processing Conference (EUSIPCO), Bruges, Belgium 2026. 1-5.

BibTeX: Download