The following sections guide you through the process of calculating
the channel latencies required for perfect wavelet reconstruction.
This example uses the
ex_wavelets model, but
you can apply the process to perform perfect wavelet reconstruction
in any model. To open the example model, type
ex_wavelets at the MATLAB® command
You must have a Wavelet
Toolbox™ product license to run the
Before you can begin calculating the latencies required for perfect wavelet reconstruction, you must know the types of filters being used in your model.
The Dyadic Analysis Filter Bank and the Dyadic Synthesis Filter
Bank blocks in the
ex_wavelets model have the
Filter order [synthesis/analysis] =
Number of levels =
Tree structure =
Once you know the types of filters being used by the Dyadic
Analysis and Dyadic Synthesis Filter Bank blocks, you need to calculate
the group delay of those filters. To do so, you can use the Signal
Before you can use
fvtool, you must first
reconstruct the filters in the MATLAB workspace. To do so, type
the following code at the MATLAB command line:
[Lo_D, Hi_D, Lo_R, Hi_R] = wfilters('bior3.5')
the low- and high-pass filters used by the Dyadic Analysis Filter
Bank block, and
the low- and high-pass filters used by the Dyadic Synthesis Filter
After you construct the filters in the MATLAB workspace,
you can use
fvtool to determine the group delay
of the filters. To analyze the low-pass biorthogonal filter used by
the Dyadic Analysis Filter Bank block, you must do the following:
fvtool(Lo_D) at the MATLAB command
line to launch the Filter Visualization Tool.
When the Filter Visualization Tool opens, click the Group delay response button () on the toolbar, or select Group Delay Response from the Analysis menu.
Based on the Filter Visualization Tool's analysis, you can see that the group delay of the Dyadic Analysis Filter Bank block's low-pass biorthogonal filter is 5.5.
Repeat this procedure to analyze the group delay of each of
the filters in your model. This section does not show the results
for each filter in the
ex_wavelets model because
all wavelet filters in this particular example have the same group
To determine the delay introduced by the analysis and synthesis filter bank system, you must reconstruct the tree structures of the Dyadic Analysis Filter Bank and the Dyadic Synthesis Filter Bank blocks. To learn more about constructing tree structures for the Dyadic Analysis Filter Bank and Dyadic Synthesis Filter Bank blocks, see the following sections of the DSP System Toolbox™ User's Guide:
Because the filter blocks in the
use biorthogonal filters with three levels and an asymmetric tree
structure, the filter bank system appears as shown in the following
The extra delay values of M and N on paths 3 and 4 in the previous figure ensure that the total delay on each of the four filter paths is identical.
Now that you have reconstructed the filter bank system, you can calculate the delay on each filter path. To do so, use the following Noble identities:
You can apply the Noble identities by summing the delay on each signal path from right to left. The first Noble identity indicates that moving a delay of 1 before a downsample of 2 is equivalent to multiplying that delay value by 2. Similarly, the second Noble identity indicates that moving a delay of 2 before an upsample of 2 is equivalent to dividing that delay value by 2.
fvtool analysis in step 1 found that
both the low- and high-pass filters of the analysis filter bank have
the same group delay (F0 = F1 =
5.5). Thus, you can use F to represent the group
delay of the analysis filter bank. Similarly, the group delay of the
low- and high-pass filters of the synthesis filter bank is the same
so you can use G to represent the group delay of
the synthesis filter bank.
The following figure shows the filter bank system with the intermediate delay sums displayed below each path.
You can see from the previous figure that the signal delays on paths 1 and 2 are identical: 7(F+G). Because each path of the filter bank system has identical delay, you can equate the delay equations for paths 3 and 4 with the delay equation for paths 1 and 2. After constructing these equations, you can solve for M and N, respectively:
fvtool analysis in step 1 found the group
delay of each biorthogonal wavelet filter in this model to be 5.5
samples. Therefore, F = 5.5 and G =
5.5. By inserting these values into the two previous
equations, you get M = 11 and N =
33. Because the total delay on each filter path
must be the same, you can find the overall delay of the filter bank
system by inserting F = 5.5 and G =
5.5 into the delay equation for any of the four
filter paths. Inserting the values of F and G into
7(F+G) yields an overall delay
of 77 samples for the filter bank system of the
Now that you know the latencies required for perfect wavelet
reconstruction, you can incorporate those delay values into the model.
ex_wavelets model has already been updated
with the correct delay values (M = 11, N =
33, Overall = 77), so it is ready to run.
After you run the model, examine the reconstruction error in the Difference scope. To further examine any particular areas of interest, use the zoom tools available on the toolbar of the scope window or from the View menu.
 Strang, G. and Nguyen, T. Wavelets and Filter Banks. Wellesley, MA: Wellesley-Cambridge Press, 1996.