Highly Nonlinear Approximations for Sparse Signal Representation

• Project Overview
• Tutorial [.ps | .pdf]
• Source code
• Examples
• Publications
• EIFS
• SCEIFS
• Acknowledgements

• Home
• maths@Aston
• EPSRC Home

Updating the oblique projector $\hat{E}_{{\cal{V}}_k {\cal{W}^\bot}}$ to $\hat{E}_{{\cal{V}}_{k+1}{\cal{W}^\bot}}$

We assume that $\hat{E}_{{\cal{V}}_k {\cal{W}^\bot}}$ is known and write it in the explicit form

$$\hat{E}_{{\cal{V}}_k {\cal{W}^\bot}}=\sum_{i=1}^k v_i \langle {w}^k_i , \cdot \rangle .$$ (17)

In order to inductively construct the duals ${w}^{k+1}_i, i=1,\ldots,k+1$ we have to discriminate two possibilities

i)

${\cal{V}}_{k+1}={\mbox{\rm {span}}}\{v_i\}_{i=1}^{k+1}= {\mbox{\rm {span}}}\{v_i\}_{i=1}^{k}={\cal{V}}_k$, i.e., $v_{k+1} \in {\cal{V}}_k.$

ii)

${\cal{V}}_{k+1}={\mbox{\rm {span}}}\{v_i\}_{i=1}^{k+1} \supset {\mbox{\rm {span}}}\{v_i\}_{i=1}^{k}={\cal{V}}_k$, i.e. $v_{k+1} \notin {\cal{V}}_k.$

Case i)

Proposition 4   Let $v_{k+1} \in {\cal{V}}_k$ and vectors $w_i^k$ in (17) be given. For an arbitrary vector $y_{k+1}\in{\cal{H}}$ the dual vectors $w_i^{k+1}$ computed as

$$w_i^{k+1}= w_i^k- \langle u_{k+1}, w_i^k\rangle y_{k+1}$$ (18)

for $i=1,\ldots,k$ and $w^{k+1}_{k+1}=y_{k+1}$ produce the identical oblique projector as the dual vectors $w_i^k, i=1,\ldots,k$.

Case ii)

Proposition 5   Let vector $v_{k+1} \notin {\cal{V}}_k$ and vectors $w_i^k, i=1,\ldots,k$ in (17) be given. Thus the dual vectors $w_i^{k+1}$ computed as

$$w_i^{k+1}=w_i^k- w_{k+1}^{k+1} \langle u_{k+1}, w_i^k\rangle ,$$ (19)

where $w_{k+1}^{k+1}= \frac{\gamma_{k+1}}{\Vert\gamma_{k+1}\Vert^2}$ with $\gamma_{k+1}= u_{k+1} - \hat{P}_{{\cal{W}}_k} u_{k+1}$, provide us with the oblique projector $\hat{E}_{{\cal{V}}_{k+1}{\cal{W}^\bot}}$.

The proof these propositions are given in [10]. The codes for updating the dual vectors are FrInsert.m and FrInsertBlock.m.

Property 2   If vectors $\{v_i\}_{i=1}^k$ are linearly independent they are also biorthogonal to the dual vectors arising inductively from the recursive equation (19).

The proof of this property is in [10].

Remark 3   If vectors $\{v_i\}_{i=1}^k$ are not linearly independent the oblique projector $\hat{E}_{{\cal{V}}_k {\cal{W}^\bot}}$ is not unique. Indeed, if $\{w_i^k\}_{i=1}^k$ are dual vectors giving rise to $\hat{E}_{{\cal{V}}_k {\cal{W}^\bot}}$ then one can construct infinitely many duals as:

$$\tilde {y}_i= w_i^k+ y_i - \sum_{j=1}^k y_j \langle v_j , w_i^k\rangle \quad i=1,\ldots,k,$$ (20)

where $y_i, i=1,\ldots,k$ are arbitrary vectors in ${\cal{H}}$ (see [10]).

This project is supported by EPSRC (EP/D062632/1)