Jacket Transform

Hung-Yi Cheng()
E-mail: woody@access.ee.ntu.edu.tw
Graduate Institute of Communication Engineering
National Taiwan University, Taipei, Taiwan, ROC

1. Abstract

---------------- 2

2. Introduction

3.

2-1. What is jacket transform

----------------- 2

2-2. Why is called jacket transform

----------------- 3

The Center Weighted Hadamard Transform

3-1. What is weighted Hadamard transform (WHT)

-----------------4

3-2. Fast algorithm for high order WHT

-----------------4

3-3. The properties for WHT

-----------------5

4. The Reverse Jacket Transform

4-1. General definition of jacket matrix

-----------------7

4-2. The properties for reverse jacket transform

------------------9

4-3. Fast algorithm for jacket matrix

-----------------10

5. Complex reverse jacket transform

5-1. The weighted Hadamard transform

------------------13

5.2 The complex weighted Hadamard transform

------------------14

5.3 The Complex reverse jacket transform

------------------15

6. The Applying of Jacket Matrices in CDMA

------------------16

7. Conclusions

------------------18

8. Reference

------------------19

1. Abstract
Recently, an interesting matrix named Jacket was proposed by Prof. Moon Ho Lee
(1989, 2000, 2001, IEEE Trans. CAS), it is extended from Walsh Hadamard (1893),
but includes both orthogonal and non-orthogonal cases. In particular, we can present
the Jacket transform and a simple factor decomposition, which is used to develop
either a fast algorithm or communications for the Jacket transform. The matrix
decomposition is of the form of the matrix products Hadamard matrices and
successively lower order coefficient matrices. This decomposition very clearly leads
to a block circular sparse matrix factorization of the Jacket matrix. The main property
of Jacket is that the inverse matrices of its elements can be obtained very easily and
have a special structure. It is useful to construct space time codes, fast algorithm and
so on.

2. Introduction
2-1 What is Jacket martrix
[

In mathematics, a jacket matrix is a square matrix

] of order n if its entries

are non-zero and real, complex, or from a finite field, and

where

is the identity matrix, and

(

where T denotes the transpose of the matrix.

In other words, the inverse of a jacket matrix is determined its element-wise or
block-wise inverse. The inverse form which is only from the entrywise inverse na d
transpose. The definition above may also be expressed as:

[ J ]mxn

J11

J
m1

1/ J11
1

[ J ]mxn 1
C
1/ J m1

J1n

J mn
1/ J1n

1/ J mn

The jacket matrix is a generalization of the Hadamard matrix, also it is

a Diagonal block-wise inverse matrix.

2-2 Why is called jacket matrix

The basic idea was motivated by the cloths of Jacket. As our two sided Jacket is
inside and outside compatible, at least two positions of a Jacket matrix are replaced by
their inverse; these elements are changed in their position and are moved, for example,
from inside of the middle circle to outside or from to inside without loss of sign.

3. The Center Weighted Hadamard Transform

3-1. What is weighted Hadamard transform (WHT)
The center weighted Hadamard transform (WHT) is defined. This transform is
similar to the Hadamard transform (HT) in that it requires only real operations. The
WHT weights the region of mid-spatial frequencies of the signal more than the HT. A
simple factonzation of the weighted Hadamard matrix is used to develop a fast
algorithm for the WHT. The matrix decomposition is of the form of the Kronecker
products of fundamental Hadamard matrices and successively lower order weighted
Hadamard matrices.
Let the Hadamard and the Weighted Hadamard matrices of order N 2k be
denoted by [ H ]N and [WH ]N .
The lowest order WH matrix is of size (4 * 4) and is defined as follows:

1 1 1 1
1 1 1 1 4 0 0
1 2 2 1

1 1 1 1 1 0 6 2

[WH ]4

1 2 2 1 4 1 1 1 1 0 2 6

1 1 1 1
1 1 1 1 0 0 0

0
0
0

And the inverse of [WH ]4

2 2 2 2
1 1 1 1 4

1 2 1 1 2 1 1 1 1 1 0
1
[WH ]4

8 2 1 1 2 16 1 1 1 1 0

2 2 2 2
1 1 1 1 0

0 0 0
3 1 0
1 3 0

0 0 4

3-2. Fast algorithm for high order WHT

As with the Hadamard matrix, a recursive relation governs the generation of high
order [WH] matrices, i.e.,
[WH ]N

[WH ]N /2 [ H ]2

(1)

Where is the Kronecker product and [ H ]2 is the lowest order Hadamard matrix as
following.

1 1
[ H ]2

1 1

Then we get the N=8

[WH ]8 [WH ]4 [ H ]2
1
1

1
1
1

1
1

1
1 1 1

2 2 1 1 1

1 1 1
1 1 1 1 1 1

1
2

1
2

1 1 1
1 2 2

1
2

1 1
2 1

1 2

2 1

2 2 1

2 2

1 1 1 1 1
1 1 1 1 1

1
1

1
1
1

1
1

1
1

3-3. The properties for WHT

In order to develop a fast general algorithm for the WHT, define a weighed
coefficient matrix [ RC ]N by
[ RC ]N

[ H ]N [WH ]N

(2)

and the [ RC ]N is sparse matrix. We can get the properties with [RC]
[ RC ]N

([ H ]N /2 [ H ]2 )([WH ]N /2 [ H ]2 )
([ H ]N /2 [WH ]N2 ) ([ H ]2 [ H ]2 )
[ RC ]N /2 2[ I ]2

the inverse for RC matrix, that is

1
[ RC ]N [ RC ]N /2 [ I ]2
2

(3)

Where [ I ]2 is 2x2 identity matrix.

Since [ H ]N1

1
[ H ]N , we find the higher order WH matrix by these fast algorithm.
N

1
[ H ]N [ RC ]N
N
[WH ]N1 N [ RC ]N1[ H ]N 1
[WH ]N

For example N=4 as below

(4)

1 1 1 1
1 1 1 1 4 0 0
1 2 2 1

1 1 1 1 1 0 6 2

[WH ]4

1 2 2 1 4 1 1 1 1 0 2 6

1 1 1 1
1 1 1 1 0 0 0
1
[ H ]4 [ RC ]4
4

0
0
0

The simple recursive relationship in (1) (2) (3) and (4) can be used to formulate a high
order [WH] as the fast algorithm.
As an example for N=8 can be represented as
[WH ]8

1
[ H ]8 [ RC ]8
8
1
( [ H ]4 [ H ]2 )([ RC ]4 2[ I ]2 )
8
1
([ RC ]4 [ H ]4 ) 2[ H ]2
8

The higher order of WH can be derived by the lower order of WH and lower order of
Hadamard matrix.
After calculation , we get the [WH ]8 from the fast algorithm.
1 1 1 1 4

1 1 1 1 1 0
[WH ]8
8 1 1 1 1 0

1 1 1 1 0
1 1 1 1 1
1 1 1 1 1

1 1 2 2 2

1 1 2 2 2

1 1 2 2 2

1 1 2 2 2
1 1 1 1 1

1 1 1 1 1

2 1
2 1
2

0
0
1 1
2

0
1 1

4
1
1
1

1
1

1
1

The figure is show the fast algorithm of WH flow.

Using the algebra of Kronecker product, it can be shown that

1
[ H ]N [ RC ]N N [ RC ]N1[ H ]N 1
N
([ H ]N /2 [ RC ]N /2 2[ H ]2 ) *

[WH ]N [WH ]N1

1
([ H ]N /2 [ RC ]N /2 1 [ H ]2 )
2
[ I ]N
Property : Products WH and inverse WH is identity matrix.

4. The Reverse Jacket Transform

4-1. Basical definition of jacket matrix
A WH matrix is a slight modification of a Hadamard matrix. The WH, however,
weights the region of mid-spatial frequencies of the signal more than the Hadamard
transform.The reverse jacket transform are a generalized conventional weighted
Hadamard transform [WH] and Hadamard transform [H].The reverse jacket transform
having geometric structure property.

Now we consider the 4*4 [WH] matrix

1 1 1 1
1 2 2 1

[WH ]4
1 2 2 1
1 1

[WH ]2

1 1 1 1 ,
1 2

The upper left 2x2 block subsampling given is

[WH ]2

. We can find some a regular

recursive geometric structure.

Compare with weighted Hadamard transform, we define several matrices for [RJ]

a
[ RJ ]2 M1 T
b
The matrix

M1

b
c

consists of three elements (a, b and c), which are all non-zero and

n
take of 2 , n = 0, 1,2. We define other matrices to get the [RJ].

b a
c b
M2
M4

c b ,
b a
Finally, we obtain

[ RJ ]4

, defined by

[ RJ ]4

M1
M T
2

b
a
a b

M 2 b c c b

M 4 b c c b

a b b a

If we assume M 1 is invertible, we define the inverse RJ matrix as follows:

signum(a * c)2
[ RJ ]2 1
det([ RJ ]2 )

1,
c b

b a , where signum( x) 0,

1,

And we also define the inverse [ RJ ]4 as follow,

1/ b 1/ a
1/ a 1/ b
1/ b 1/ c 1/ c 1/ b
1

[ RJ ]4 1
4 1/ b 1/ c 1/ c 1/ b

1/ a 1/ b 1/ b 1/ a

x0
x0
x0

We can see the elements position in the forward matrix can be replaced by its
inverse matrix, and the signs are not changed between the matrix and its inverse.

The figure is show the key idea for jacket transform.

4-2. The properties for reverse jacket transform

We denote the

[ RJ ]2k

for the reverse jacket matrix with (2k * 2k). There are two

properties to the reverse jacket matrix.

Property 1 :
[ RJ ]2k is a symmetric matrix, i.e., [ RJ ]2k [ RJ ]2k T

where T denotes transpose.

Property 2 :
If [ RJ ]2k is orthogonal matrix, we can derive the equation as following

[ RJ ]2k T [ RJ ]2k 2k [ I ]2k

But is a different case from a non-orthogonal matrix,

[RJ ]2k T [ RJ ]2k [ RJ ]2k2

Property 3 :
A matrix is called the reverse jacket matrix if

[ RJ ]2k [ H ]2k1[ RJ ]*2k [ H ]2k

, where k belongs to integer N. All its components are 2k , n = 0,1,2.
According to the properties, it is easy found to be equivalent to the conditions.
[ RJ ]2k * [ H ]2k1 [ RJ ]*2k [ H ]2k
[ RJ ]2k 1

,
1
[ RJ ]2k
det([ RJ ]2k )

where k>2

So the inverse reverse jacket transform is a function of jacket transform.

4-3. Fast algorithm for jacket matrix

Now we start to get fast algorithm for the high order of reverse jacket transform. The
fast algorithm of RJ is similar as the weighted Hadamard transform and can be
expressed recursively in terms of Kronecker product.

1 1
[ RJ ]2K 1 [ RJ ]2k [ H ]2 [ RJ ]2k
,
1 1

where k>2

For example, as [ RJ ]8 is an RJ matrix, then it is

[ H ]2
[ H ]
2
[ RJ ]8
[ H ]2

[ H ]2

[ H ]2

[ H ]2

2[ H ]2
2[ H ]2

2[ H ]2
2[ H ]2

[ H ]2

[ H ]2

[ H ]2
[ H ]2
[ H ]2

[ H ]2

In similar fashion as weighed Hadamard transform, we note that

[ RC ]2K 1 [ H ]2k 1 [ RJ ]2k 1

([ H ]2k [ H ]2 )([ RJ ]2k [ H ]2 )
[ RC ]2k 2[ I ]2
Since [ H ]2k 1

1
[ H ]2k , we have from
2k

[ RJ ]2k 1

1
[ H ]2k 1 [ RC ]2k 1
2k 1

Therefore, the fast RJ transform is simply the fast Hadamard transform followed by a
spars matrix

1
[ RC ]2k .
2k

For example, a=b=1 and c=2, the RJ is the same as WH

a b 1 1 1 1 1 2 1
M 2 [ RJ ]2

b c 1 2 2 1 1 0 3
And its inverse matrix

1 2 1 1 1 1 3 0
[ RJ ]2 1

3 1 1 6 1 1 1 2
Also, we have
1 1 1 1
1 2 2 1
,
[ RJ ]4
1 2 2 1

1 1 1 1

2 2 2 2

1 2 1 1 2
1
[ RJ ]4
8 2 1 1 2

2 2 2 2

In geometric structure, the outside of [ RJ ]4 is corresponding to inside of [ RJ ]4 1

and vice versa.

The algorithm for the fast reverse jacket transform is similar as the WH. The FRJT

can be derived by decomposing [RJ] into a product of sparse matrices given each
rows/columns with only two non-zero elements.
The fast RJ transform for N=8,
1
[ RJ ]8 [ H ]8 [ RC ]8
8
1
([ H ]4 [ H ]2 )([ RC ]4 2[ I ]2 )
8
1
([ RJ ]4 2[ H ]2 )
2

In a similar manner as the fast RJ transform, we get the inverse of [RJ]

1
[ RJ ]81 ([ H ]4 [ H ]2 )([ RC ]4 1 [ I ]2 )
2
1
[ RJ ]4 1 [ H ]2
2

Table I : The inverse reverse jacket transform (forward)

Table II : The reverse jacket transform

The table I and II show the RJ matrices have a geometric structure.

5.Complex reverse jacket transform

5-1. The weighted Hadamard transform
The Hadamard transform (WHT) and discrete are used widely in signal processing.
Variations of these two transforms called center weighted Hadamard transform
(CWHT) and complex reverse jacket transform (CRJT)

have been reported and

their applications .Now we discuss about anther from for WHT in order to derive the
complex weighted Hadamard transform.
For n 2r length real vector (a0 , a1 ,
real vector ( A0 , A1 ,

, an 1 ) the transform vector is a n-length

, An 1 ) given by

A0
a0
A

1 [ H ] a1
n 1

An 1
an 1

that is,
n 1

A j (1) j ,i ai ,

j=1 ~ n-1

i 0

Where the modulo-2 inner product <j,i> is given by

j, i jr 1ir 1 jr 2ir 2

j0i0

where denotes modulo two addition.

The inverse transform is given by
aj

1 n 1
( 1) j ,i Ai ,

n i 0

j=1 ~ n-1

5-2 The center weighted Hadamard transform

The CWHT is obtained by weighting the center portion of the transform matrix for
n 2r length real vector and is given by
n 1

Aj (1) j ,i ( )( ir 1 ir 2 ) ( jr 1 jr 2 ) ai ,

j 0,1, 2,

, n 1

i 0

where is any real number is the weight. The inverse transform is given by
n 1

a j (1) j ,i ( 1 ) (ir 1 ir 2 ) ( jr 1 jr 2 ) Ai ,

j 0,1, 2,

, n 1

i 0

We can know that for 1 the CWHT is same as the WHT. The 4x4 and 8x8 CWHT
matrices corresponding to the weight are
1 1
1
[W ]4
1

1 1

1
1
1

and

1
1

1
[W ]8
1

1
1

1

1

1
1

1
1
1
1

1
1
1

1
1

1
1

5-3 The Complex reverse jacket transform

Both the CWHT and the CRJT include the WHT as a special case. The CRJT uses
the complex number j 1 which is a fourth root of unity by replacing the weight

and this CRJT is orthogonal matrix.

The 4x4 and 8x8 CRJT are shown below as
1 1
1 1
1 j j 1

[C ]4
1 j j 1

1 1 1 1

and

1
1

1
[C ]8
1

1
1

j j

1 j

j 1

j j 1

j j

1 1 1 1

1 1

1
1
1

1
1
1

1
1

1
1

The following diagram is shown the different matrix and function with DFT,DCT
Hadamard and Jacket transform.

6. The Applying of Jacket Matrices in CDMA

The Jacket transform can be an orthogonal matrix with highly practical value for
wireless communication. The computation of the transform of a signal consists of
additions and subtractions of the signal samples. The Jacket transform is also the
non-sinusoidal orthogonal transforms. Using the orthogonal Jacket transform, we can
apply this matrix to sequence generator in code-division multiple access (CDMA)
spread-spectrum communication.
Consider the general 4x4 Jacket matrix is

[ RJ ]4

b
a
a b
b c c b

b c c b

a b b a

Its inverse matrix is

1/ b 1/ a
1/ a 1/ b
1/ b 1/ c 1/ c 1/ b
1

[ RJ ]4 1
4 1/ b 1/ c 1/ c 1/ b

1/ a 1/ b 1/ b 1/ a

and we have known that

[ RJ ]4 [ RJ ]41 [ I ]4

At the transmitter, assume we send the massage (0 0 1 1) at the same

time. Choosing the Jacket codes used to modulate the signal these CDMA
systems. Since this matrix is orthogonal and easily to inverse matrix. We can
reconstruct the massage easily with the inverse Jacket transform.
As the figure 1, the stations transmit the signal to the multiplexer which is
composed with Jacket matrix. The TX transmit the data with summing the signals
through the Jacket transform.

Figure 1: Jacket matrix as the multiplexer in CDMA system.

The signal transmitting from the TX is
0
2b 2c

2b 2c

b
a 1
a b
b c c b 1

b c c b 1

a b b a 1

Assume we received the signals without noise and the demultiplexer in CDMA
system is the inverse Jacket transform. The received signal multiplex the each low of
Jacket matrix. Because the Jacket transform is orthogonal matrix, we can get the
message perfectly as following step.

Figure 2: Jacket matrix as the demultiplexer in CDMA system.

We received the signal (0, -2b+2c, -2b-2c, 0) from TX with Jacket transform.
Through the demultiplexer, the signal became
1/ b 1/ a 0
1
1/ a 1/ b

1
1/ b 1/ c 1/ c 1/ b 2b 2c
1

1 4 1/ b 1/ c 1/ c 1/ b 2b 2c

1
1/ a 1/ b 1/ b 1/ a 0

We can reconstruct the message perfectly with the Jacket transform.

This is very simple example for using the Jacket transform.

7. Conclusions
The reserve Jacket matrix is a generalized form of the Weighted Hadamard and
Hadamard. The matrix has a recursive structure and symmetric characteristics.
The elements positions in the forward matrix can be replaced by its inverse matrix,
and the signs are not changed between the matrix and its inverse. The Hadamard
matrix is a special case of the Jacket matrix. The fast transform algorithm is the
matrix decomposition of the Hadamard matrices and succesively lower order reverse
Jacket transform. The [ RC ]N sparse matrix of [ RJ ]N leads to very clear
decomposition. Using the orthogonal Jacket transform, we can apply the matrix to
wireless communication such as CDMA.

8. Reference
[1]M.H. Lee, The Center Weighted Hadamard Transform, IEEE Trans.1989 AS-36,
(9),pp.1247-1249.
[2]S.-R.Lee and M.H.Lee, On the Reverse Jacket Matrix for Weighted Hadamard
Transform, IEEE Trans. on Circuit Syst.II, vol.45.no.1, pp.436-441,Mar.1998.
M.H. Lee, A New Reverse Jacket Transform and its Fast Algorithm, IEEE Trans.
Circuits Syst.-II , vol 47, pp.39-46, 2000.
[3] M.H. Lee and B.S. Rajan, A Generalized Reverse Jacket Transform, IEEE Trans.
Circuits Syst. II, Analog Digit. Signal Process., vol. 48 no.7 pp 684-691, 2001.
J. Hou, M.H. Lee and J.Y. Park, New Polynomial Construction of Jacket
Transform, IEICE Trans. Fundamentals, vol. E86-A no. 3, pp.652-659, 2003.
[4]W.P. Ma and M. H. Lee, Fast Reverse Jacket Transform Algorithms, Electronics
Letters, vol. 39 no. 18 , 2003.
[5]Moon Ho Lee, Ju Yong Park, and Jia Hou,Fast Jacket Transform Algorithm Based
on Simple Matrices Factorization, IEICE Trans. Fundamental, vol.E88-A, no.8,
Aug.2005.
[6]Moon Ho Lee and Jia Hou, Fast Block Inverse Jacket Transform, IEEE Signal
Processing Letters, vol.13. No.8, Aug.2006.
[7] Moon Ho Lee, and Ken.Finlayson, A Simple Element Inverse Jacket Transform
Coding, Information Theory Workshop 2005, ITW 2005, Proc. of IEEE ITW 2005,
28.Aug-1.Sept., New Zealand, also IEEE Signal Processing Letters, vol. 14 no. 5,
May 2007
[8]M. H. Lee,A new reverse jacket transform and its fast algorithm IEEE Trans.
Circuits Syst.II, vol. 47, no.1, pp.39-46, Jan.2000.