Main Content

Design vector quantizer using Vector Quantizer Design Tool (VQDTool)

Quantizers

`dspquant2`

Double-click on the Vector Quantizer Design block to start VQDTool, a GUI that allows
you to design and implement a vector quantizer. You can also start VQDTool by typing
`vqdtool`

at the MATLAB^{®} command prompt. Based on your specifications, VQDTool iteratively
calculates the codebook values that minimize the mean squared error between the training
set and the codebook until the stopping criteria for the design process is satisfied.
The block uses the resulting codebook values to implement your vector quantizer.

For the **Training Set** parameter, enter a
*k*-by-*M* matrix of values you want to use to
train the quantizer codebook. The variable k, where $$k\ge 1$$, is the length of each training vector. It also represents the
dimension of your quantizer. The variable *M*, where $$M\ge 2$$, is the number of training vectors. This data can be created using a
MATLAB function, such as the default value `randn(10,1000)`

, or
it can be any variable defined in the MATLAB workspace.

You have two choices for the **Source of initial codebook**
parameter. Select `Auto-generate`

to have the block choose the
values of the initial codebook. In this case, the block picks *N*
random training vectors as the initial codebook, where *N* is the
**Number of levels** parameter and $$N\ge 2$$. When you select `User defined`

, enter the
initial codebook values in the **Initial codebook** field. The initial
codebook matrix must have the same number of rows as the training set. Each column of
the codebook is a codeword, and your codebook must have at least two codewords.

For the given training set and initial codebook, the block performs an iterative
process, using the Generalized Lloyd Algorithm (GLA), to design a final codebook. For
each iteration of the GLA, the block first associates each training vector with its
nearest codeword by calculating the distortion. You can specify one of the two possible
methods for calculating distortion using the **Distortion measure**
parameter.

When you select `Squared error`

for the **Distortion
measure** parameter, the block finds the nearest codeword by calculating
the squared error (unweighted). Consider the codebook $$CB=\left[\begin{array}{cccc}C{W}_{1}& C{W}_{2}& \mathrm{...}& C{W}_{N}\end{array}\right]$$. This codebook has *N* codewords; each codeword has
*k* elements. The *i*-th codeword is defined as $$C{W}_{i}=\left[\begin{array}{cccc}{a}_{1i}& {a}_{2i}& \mathrm{...}& {a}_{ki}\end{array}\right]$$. The training set has *M* columns and is defined as $$U=\left[\begin{array}{cccc}{U}_{1}& {U}_{2}& \mathrm{...}& {U}_{M}\end{array}\right]$$, where the *p*-th training vector is $${U}_{p}={\left[\begin{array}{cccc}{u}_{1p}& {u}_{2p}& \mathrm{...}& {u}_{kp}\end{array}\right]}^{\prime}$$. The squared error (unweighted) is calculated using the equation

$$D={\displaystyle \sum _{j=1}^{k}{\left({a}_{ji}-{u}_{jp}\right)}^{2}}$$

When you select `Weighted squared error`

for
the **Distortion measure** parameter, enter a vector or matrix for the
**Weighting factor** parameter. When the weighting factor is a
vector, its length must be equal to the number of rows in the training set. This
weighting factor is used for each training vector. When the weighting factor is a
matrix, it must be the same size as the training set matrix. The block finds the nearest
codeword by calculating the weighted squared error. If the weighting factor for the
*p*-th column of the training vector, $${U}_{p}$$, is defined as $$Wp={\left[\begin{array}{cccc}{w}_{1p}& {w}_{2p}& \mathrm{...}& {w}_{kp}\end{array}\right]}^{\prime}$$, then the weighted squared error is defined by the equation

$$D={\displaystyle \sum _{j=1}^{k}{w}_{jp}{\left({a}_{ji}-{u}_{jp}\right)}^{2}}$$

Once the block has associated all the training vectors with their nearest codeword vectors, the block calculates the mean squared error for the codebook and checks to see if the stopping criteria for the process has been satisfied.

The two possible options for the **Stopping criteria** parameter are
`Relative threshold`

and ```
Maximum
iteration
```

. When you want the design process to stop when the
fractional drop in the squared error is below a certain value, select
`Relative threshold`

. Then, type the maximum acceptable
fractional drop in the **Relative threshold** field. The fraction drop
in the squared error is defined as

$$\frac{\text{erroratpreviousiteration}-\text{erroratcurrentiteration}}{\text{erroratpreviousiteration}}$$

When you want the design process to stop after a certain number of
iterations, choose `Maximum iteration`

. Then, enter the maximum
number of iterations you want the block to perform in the **Maximum
iteration** field. For **Stopping criteria**, you can also
choose `Whichever comes first`

and enter **Relative
threshold **and** Maximum iteration **values. The block
stops iterating as soon as one of these conditions is satisfied.

When a training vector has the same distortion for two different codeword vectors, the
algorithm uses the **Tie-breaking rule** parameter to determine which
codeword vector the training vector is associated with. When you want the training
vector to be associated with the lower indexed codeword, select ```
Lower
indexed codeword
```

. To associate the training vector with the higher
indexed codeword, select `Higher indexed codeword`

.

With each iteration, the block updates the codeword values in order to minimize the
distortion. The **Codebook update method** parameter defines the way
the block calculates these new codebook values.

**Note**

If, for the **Distortion measure** parameter, you choose
`Squared error`

, the **Codebook update
method** parameter is set to `Mean`

.

If, for the **Distortion measure** parameter, you choose
`Weighted squared error`

and you choose
`Mean`

for the **Codebook update method**
parameter, the new codeword vector is found as follows. Suppose there are three training
vectors associated with one codeword vector. The training vectors are

$$T{S}_{1}=\left[\begin{array}{c}1\\ 2\end{array}\right],\text{}T{S}_{3}=\left[\begin{array}{c}10\\ 12\end{array}\right],\text{and}T{S}_{7}=\left[\begin{array}{c}11\\ 12\end{array}\right].$$

The new codeword vector is calculated as $$C{W}_{new}=\left[\begin{array}{c}\frac{1+10+11}{3}\\ \frac{2+12+12}{3}\end{array}\right]$$

where the denominator is the number of training vectors associated with this codeword.
If, for the **Codebook update method** parameter, you choose
`Centroid`

and you specify the weighting factors $${W}_{1}=\left[\begin{array}{c}0.1\\ 0.2\end{array}\right]$$, $${W}_{3}=\left[\begin{array}{c}1\\ 0.6\end{array}\right]$$, and $${W}_{7}=\left[\begin{array}{c}0.3\\ 0.4\end{array}\right]$$, the new codeword vector is calculated as

$$C{W}_{new}=\left[\begin{array}{c}\frac{\left(0.1\right)\left(1\right)+\left(1\right)\left(10\right)+\left(0.3\right)\left(11\right)}{0.1+1+0.3}\\ \frac{\left(0.2\right)\left(2\right)+\left(0.6\right)\left(12\right)+\left(0.4\right)\left(12\right)}{0.2+0.6+0.4}\end{array}\right]$$

Click **Design and Plot** to design the quantizer with the
parameter values specified on the left side of the GUI. The performance curve and the
entropy of the quantizer are updated and displayed in the figures on the right side of
the GUI.

**Note**

You must click **Design and Plot** to apply any changes you
make to the parameter values in the VQDTool dialog box.

The following is an example of how the block calculates the entropy of the quantizer
at each iteration. Suppose you have a codebook with four codewords and a training set
with 200 training vectors. Also suppose that, at the *i*-th iteration,
40 training vectors are associated with the first codeword, 60 training vectors are
associated with the second codeword, 20 training vectors are associated with the third
codeword, and 80 training vectors are associated with the fourth codeword. The
probability that a training vector is associated with the first codeword is $$\frac{40}{200}$$. The probabilities that training vectors are associated with the
second, third, and fourth codewords are $$\frac{60}{200}$$, $$\frac{20}{200}$$, and $$\frac{80}{200}$$, respectively. The GUI uses these probabilities to calculate the
entropy according to the equation

$$H={\displaystyle \sum _{i=1}^{N}-{p}_{i}{\mathrm{log}}_{2}{p}_{i}}$$

where *N* is the number of codewords. Based on these probabilities,
the GUI calculates the entropy of the quantizer at the *i*-th iteration
as

$$\begin{array}{l}H=-\left(\frac{40}{200}{\mathrm{log}}_{2}\frac{40}{200}+\frac{60}{200}{\mathrm{log}}_{2}\frac{60}{200}+\frac{20}{200}{\mathrm{log}}_{2}\frac{20}{200}+\frac{80}{200}{\mathrm{log}}_{2}\frac{80}{200}\right)\\ H=1.8464\end{array}$$

VQDTool can export parameter values that correspond to the figures displayed in the
GUI. Click the **Export Outputs** button, or press
**Ctrl+E**, to export the **Final Codebook**,
**Mean Square Error**, and **Entropy** values to
the workspace, a text file, or a MAT-file.

In the **Model** section of the GUI, specify the destination of
the block that will contain the parameters of your quantizer. For
**Destination**, select `Current model`

to
create a block with your parameters in the model you most recently selected. Type
`gcs`

in the MATLAB Command Window to display the name of your current model. Select
`New model`

to create a block in a new model file.

From the **Block type** list, select
`Encoder`

to design a Vector Quantizer Encoder block.
Select `Decoder`

to design a Vector Quantizer Decoder block.
Select `Both`

to design a Vector Quantizer Encoder block and a
Vector Quantizer Decoder block.

In the **Encoder block name** field, enter a name for the Vector
Quantizer Encoder block. In the **Decoder block name** field, enter a
name for the Vector Quantizer Decoder block. When you have a Vector Quantizer Encoder
and/or Decoder block in your destination model with the same name, select the
**Overwrite target block** check box to replace the block's
parameters with the current parameters. When you do not select this check box, a new
Vector Quantizer Encoder and/or Decoder block is created in your destination
model.

Click **Generate Model**. VQDTool uses the parameters that
correspond to the current plots to set the parameters of the Vector Quantizer Encoder
and/or Decoder blocks.

**Training Set**Enter the samples of the signal you would like to quantize. This data set can be a MATLAB function or a variable defined in the MATLAB workspace. The typical length of this data vector is

`1e5`

.**Source of initial codebook**Select

`Auto-generate`

to have the block choose the initial codebook values. Choose`User defined`

to enter your own initial codebook values.**Number of levels**Enter the number of codeword vectors,

*N*, in your codebook matrix, where*N*≥ 2.**Initial codebook**Enter your initial codebook values. From the

**Source of initial codebook**list, select`User defined`

in order to activate this parameter. The codebook must have the same number of rows as the training set. You must provide at least two codeword vectors.**Distortion measure**When you select

`Squared error`

, the block finds the nearest codeword by calculating the squared error (unweighted). When you select`Weighted squared error`

, the block finds the nearest codeword by calculating the weighted squared error.**Weighting factor**Enter a vector or matrix. The block uses these values to compute the weighted squared error. When the weighting factor is a vector, its length must be equal to the number of rows in the training set. This weighting factor is used for each training vector. When the weighting factor is a matrix, it must be the same size as the training set matrix. The individual weighting factors cannot be negative. The weighting factor vector or matrix cannot contain all zeros.

**Stopping criteria**Choose

`Relative threshold`

to enter the maximum acceptable fractional drop in the squared quantization error. Choose`Maximum iteration`

to specify the number of iterations at which to stop. Choose`Whichever comes first`

and the block stops the iteration process as soon as the relative threshold or maximum iteration value is attained.**Relative threshold**This parameter is available when you choose

`Relative threshold`

or`Whichever comes first`

for the**Stopping criteria**parameter. Enter the value that is the maximum acceptable fractional drop in the squared quantization error.**Maximum iteration**This parameter is available when you choose

`Maximum iteration`

or`Whichever comes first`

for the**Stopping criteria**parameter. Enter the maximum number of iterations you want the block to perform.**Tie-breaking rules**When a training vector has the same distortion for two different codeword vectors, select

`Lower indexed codeword`

to associate the training vector with the lower indexed codeword. Select`Higher indexed codeword`

to associate the training vector with the lower indexed codeword.**Codebook update method**When you choose

`Mean`

, the new codeword vector is calculated by taking the average of all the training vector values that were associated with the original codeword vector. When you choose`Centroid`

, the block calculates the new codeword vector by taking the weighted average of all the training vector values that were associated with the original codeword vector Note that if, for the**Distortion measure**parameter, you choose`Squared error`

, the**Codebook update method**parameter is set to`Mean`

.**Destination**Choose

`Current model`

to create a Vector Quantizer block in the model you most recently selected. Type`gcs`

in the MATLAB Command Window to display the name of your current model. Choose`New model`

to create a block in a new model file.**Block type**Select

`Encoder`

to design a Vector Quantizer Encoder block. Select`Decoder`

to design a Vector Quantizer Decoder block. Select`Both`

to design a Vector Quantizer Encoder block and a Vector Quantizer Decoder block.**Encoder block name**Enter a name for the Vector Quantizer Encoder block.

**Decoder block name**Enter a name for the Vector Quantizer Decoder block.

**Overwrite target block**When you do not select this check box and a Vector Quantizer Encoder and/or Decoder block with the same block name exists in the destination model, a new Vector Quantizer Encoder and/or Decoder block is created in the destination model. When you select this check box and a Vector Quantizer Encoder and/or Decoder block with the same block name exists in the destination model, the parameters of these blocks are overwritten by new parameters.

**Generate Model**Click this button and VQDTool uses the parameters that correspond to the current plots to set the parameters of the Vector Quantizer Encoder and/or Decoder blocks.

**Design and Plot**Click this button to design a quantizer using the parameters on the left side of the GUI and to update the performance curve and entropy plots on the right side of the GUI.

You must click

**Design and Plot**to apply any changes you make to the parameter values in the VQDTool GUI.**Export Outputs**Click this button, or press

**Ctrl+E**, to export the**Final Codebook**,**Mean Squared Error**, and**Entropy**values to the workspace, a text file, or a MAT-file.

Double-precision floating point

[1] Gersho, A. and R. Gray. *Vector Quantization and Signal
Compression*. Boston: Kluwer Academic Publishers, 1992.