Sign in

username:

password:



Not a member?

Search Online Books



Search tips

Free Online Books

Chapters

Physical Audio Signal Processing
    Artificial Reverberation
       FDN Reverberation
          Choice of Lossless Feedback Matrix
             Householder Feedback Matrix

Chapter Contents:

Search Physical Audio Signal Processing

  

Book Index | Global Index

Would you like to be notified by email when Julius Orion Smith III publishes a new entry into his blog?

  

Householder Feedback Matrix

One choice of lossless feedback matrix $ \mathbf{A}_N$ for FDNs, especially nice in the $ 4\times4$ case, is a specific Householder reflection proposed by Jot [220]:

$\displaystyle \mathbf{A}_N = \mathbf{I}_N - \frac{2}{N}\underline{u}_N\underline{u}_N^T \protect$ (3.4)

where $ \underline{u}_N^T = [1, 1, \dots, 1]$ can be interpreted as the specific vector about which the input vector is reflected in $ N$-dimensional space (followed by a sign inversion). More generally, the identity matrix $ \mathbf{I}_N$ can be replaced by any $ N\times N$ permutation matrix [156, p. 126].

It is interesting to note that when $ N$ is a power of 2, no multiplies are required [439]. For other $ N$, only one multiply is required (by $ 2/N$).

Another interesting property of the Householder reflection $ \mathbf{A}_N$ given by Eq.$ \,$(2.4) (and its permuted forms) is that an $ N\times N$ matrix-times-vector operation may be carried out with only $ 2N-1$ additions (by first forming $ \underline{u}_N^T$ times the input vector, applying the scale factor $ 2/N$, and subtracting the result from the input vector).

An example implementation of a Householder FDN for $ N=3$ is shown in Fig.2.9. As observed by Jot [156, p. 216], this computation is equivalent to $ N$ parallel feedback comb filters with one new feedback path from the output to the input through a gain of $ -2/N$.

Figure 2.9: FDN using a Householder-reflection feedback matrix.
\begin{figure}\input fig/householder1.pstex_t \end{figure}

An unusual feature of the Householder feedback matrix $ A_N$ is that for $ N\neq 2$, all entries in the matrix are nonzero. This means every delay line feeds back to every other delay line, thereby helping to maximize echo density as soon as possible.

Furthermore, for $ N=4$, the matrix entries all have the same magnitude:

$\displaystyle \mathbf{A}_4 = \frac{1}{2} \left[\begin{array}{rrrr} 1 & -1 & -1 ... ... -1 & 1 & -1 & -1\\ -1 & -1 & 1 & -1\\ -1 & -1 & -1 & 1 \end{array}\right]. $

Only the $ N=4$ case is ``balanced'' in this way. For larger $ N$, the diagonal becomes larger than the off-diagonal elements, and as $ N$ becomes very large, the FDN approaches a bank of decoupled parallel comb filters.

Due to the elegant balance of the $ N=4$ Householder feedback matrix, Jot [219] proposes an $ N=16$ FDN based on an embedding of $ N=4$ feedback matrices:

$\displaystyle \mathbf{A}_{16} = \frac{1}{2} \left[\begin{array}{rrrr} \mathbf{A... ...\mathbf{A}_4 & -\mathbf{A}_4 & -\mathbf{A}_4 & \mathbf{A}_4 \end{array}\right] $

Another method is to replace each of the four delay lines in an FDN(4) by a Gerzon vector allpass (see §