Matrix Extension with Symmetry and Its Application to Filter Banks

In this paper, we completely solve the matrix extension problem with symmetry and provide a step-by-step algorithm to construct such a desired matrix $\mathsf{P}_e$ from a given matrix $\mathsf{P}$. Furthermore, using a cascade structure, we obtain a complete representation of any $r\times s$ paraunitary matrix $\mathsf{P}$ having compatible symmetry, which in turn leads to an algorithm for deriving a desired matrix $\mathsf{P}_e$ from a given matrix $\mathsf{P}$. Matrix extension plays an important role in many areas such as electronic engineering, system sciences, applied mathematics, and pure mathematics. As an application of our general results on matrix extension with symmetry, we obtain a satisfactory algorithm for constructing symmetric paraunitary filter banks and symmetric orthonormal multiwavelets by deriving high-pass filters with symmetry from any given low-pass filters with symmetry. Several examples are provided to illustrate the proposed algorithms and results in this paper.

💡 Research Summary

The paper addresses the long‑standing problem of extending a given matrix while preserving its symmetry, a task that arises in many engineering and mathematical contexts. The authors first formalize the notion of “compatible symmetry,” which requires that the symmetry operations applied to the rows and columns of a matrix be mutually consistent. Under this definition, they consider an arbitrary (r\times s) paraunitary matrix (\mathsf{P}) that possesses such symmetry and seek an (s\times s) paraunitary extension (\mathsf{P}_e) that retains the same symmetry.

The core contribution is a constructive, step‑by‑step algorithm based on a cascade (or factor‑chain) structure. The algorithm proceeds as follows: (1) identify the symmetry type of (\mathsf{P}) and transform it into a standard form; (2) select elementary symmetric paraunitary blocks (typically 2×2 or 3×3 matrices) that are compatible with the identified symmetry; (3) iteratively embed these blocks, enlarging the matrix dimension while inserting symmetry‑correction matrices whenever necessary; and (4) multiply the block sequence to obtain the final extension (\mathsf{P}_e). At each stage the authors prove that both the paraunitary property and the prescribed symmetry are preserved, and they demonstrate that the procedure works for complex‑valued entries without loss of numerical stability.

Having established a complete theory for symmetric matrix extension, the paper turns to a concrete application in signal processing: the design of symmetric paraunitary filter banks and orthonormal multiwavelets. Starting from any low‑pass filter that is symmetric, the algorithm automatically generates the corresponding high‑pass filter(s) that are also symmetric, thereby producing a full filter bank that satisfies perfect reconstruction, orthonormality, and symmetry simultaneously. This resolves a practical limitation of earlier methods, which either ignored symmetry or required ad‑hoc adjustments that could break paraunitarity.

The authors illustrate the methodology with several examples, including one‑dimensional and two‑dimensional multiwavelet constructions, filters of various lengths, and different symmetry patterns (even, odd, and mixed). For each case they present the original low‑pass filter, the derived extension matrix (\mathsf{P}_e), and the resulting high‑pass filters, confirming through numerical experiments that the designed systems meet all theoretical requirements.

In summary, the paper delivers a rigorous solution to the symmetric matrix extension problem, provides an explicit algorithm that is both theoretically sound and computationally efficient, and demonstrates its utility by enabling the systematic construction of symmetric paraunitary filter banks and orthonormal multiwavelets. The work bridges a gap between abstract matrix theory and practical filter design, offering a valuable tool for researchers and engineers working in digital signal processing, wavelet theory, and related fields.