Behavior of minimum output energy algorithm for

digital VSB signals

Youngsun Ha Wonzoo Chung

Division of Computer and Communications Engineering, Division of Computer and Communications Engineering,
Korea University Korea University
Seoul, Korea Seoul,Korea
Email: [email protected] Email: [email protected]

Abstract- In this paper we show that the widely used blind simulations for general cases, showed that complex MOE-IIR
adaptive minimum output energy algorithm fails to converge to equalizers failed to converge to the desired solution.
the desired solutions when complex valued IIR equalization
structure is used for digital VSB signals. We present analytic In Section II, we formally defme the complex IIR equal­
counter example for a simplified two tap channel case showing izers for digital VSB with adaptation rules, and describes the
ill convergence of MOE algorithm. statistical properties of the digital VSB signals. In Section III,
we present a cost function analysis of the complex MOE-IIR
Index Terms- Minimum output energy algorithm, Adaptive equalizers for two-tap channels. We show the non­
blind equalizer, vestigial sideband, VSB, infinite impulse convergence of the complex MOE-IIR equalizer.
response filter, constant modulus algorithm.
II. COMPLEX IIR EQUALIZERS FOR VSB

I. INTRODUCTION Consider a complex IIR equalizer for the digital VSB

Blind adaptive equalization is preferred in wireless data described in Figure 1. Here, {vk} is a VSB signal sequence
broadcasting systems to help cope with the increasing demand Source
Noise

for high data rates. Linear blind (adaptive) equalization has

been intensively studied and widely used in industry [1], [2],
[3], [4]. More recently, the blind infmite impulse response
s',

(IIR) equalization scheme has been adopted by industry to

deal with severe multipath channels. Blind IIR equalizers are
used in the acquisition stage to produce a reasonable signal
quality to drive decision-directed (DD) least mean squares
Fig. I. IIR Equalizer for VSB with a complex recursive filter
(LMS) algorithms for decision feedback equalizers (DFE) [5],
[6], [7]. generated from an independent and identically distributed

In practice, the most popular adaptation algorithms for

(i.i.d) M-Ievel PAM sequence vk = Sk + jSk with second

blind IIR equalizers are the minimum output energy (MOE) moment m2 and fourth moment m4, where
algorithm [8], [3]. Although their stability and global
convergence issues have not been thoroughly resolved, [9], (1)
[10], [11], [12], [13] the MOE-IIR equalizers are currently
considered to be the most robust and efficient methods for
The quadrature component Skis generated from {S I } through
ensuring a reliable signal quality from a cold start-up.
a Hilbert transformation filter h ([15]),
For digital vestigial sideband (VSB) signals, especially in
American high defmition digital television receiver systems
00

[14], the equalization structure consisting of a complex valued Sn = L>kh[n -k] (2)
k=-<»
IIR equalizer followed by a DFE plays a critical role in

{�
recovering received 8- signals distorted by long and severe
multipath channels. We investigated the MOE for blind for odd n
adaptation of IIR equalizer coefficients. The closed-form
�n l= (3)
analysis for two-tap channels, which was supported by for even n

978-1-61284-857-0/11/$26.00 <92011 IEEE

Then, we have the following statistical properties: III. ANAL YSYS OF THE COMPLEX MOE-IIR

i) E{SmSI}= m2o(m - /) , Unfortunately for MOE algorithms, (6) and (7) do not
have global minima at the desired location 8 = 0 .
ii) E{SmSI}= m2o(m -I), and Property 1 For C = 0 and d =rj
iii) E{SmSI}= -E{SISm}= m2h[/ - m ] . 2m 4
JMOE(f,d) =--22 1--tan (
- (r) 1 ) (10)
Property ii) comes from the fact that the magnitude response 1- r 1!
of the Hilbert transformation in the frequency domain is flat
[15]. Using the above properties gives: Proof The equalizer output can be now written as

E{vm VI}= 0 and

00

iv)
Yk =Vk - rjYk-l = L(-rjrVk-n (11)
n=O
v) E{vmVI"}= 2m2o(m -I) + j 2m2h[m -I].
The Output energy is
Let rk be a signal received through a discrete time multipath
channel model C = [CO,C1,...CNc-1] , i.e.,
2 2
Nc-1 Assuming steady state, we have E IYkl =E IYk_11 and
rk = LCnVk-n (4) hence
n=O
Let f = [fo,.t;,...fNrl] and d =[do,dp...dNrl] be a
complex-valued transversal filter and a complex-valued recur­
Notice that
sive filter, respectively. The equalizer output, denoted by Yk'
00

is given by
E(V;Yk_l) = L(-rjrE(V;Vk_1_n)
Nrl Nd-l n=O (14)
Yk = Lf r -
n k-n LdnYk-n . (5)
2
00

n=O n=O =-2 j m2 L(-r rh[2n+ 1]

A suboptimal cost-function can be used to find the
n=O
transversal filter and the recursive filter coefficients and
approximately minimizing the mean square error without the
help of a training sequence. The MOE algorithm [8]
minimizes the output energy of the equalizer with a norm (15)
2
constraint on the feedforward equalizer IIfl1 =1, From the geometrical sum

(6)
(16)
For VSB signals, we can consider the following variant of
the MOE algorithm using only the real component
We have
2
JMOER (f,d) = E{Re(Yk) },
r (-r2)n
(7)
1
= tan- (r)
00

L (17)
The update equations of these algorithms have different error n=O 2n + 1
terms
Finally,
(8)
2
2m 4 1
JMOE (d ) =E IYk1 =--22 (1--tan - (r)) (18)
eMOER = Re(Yk)· (9) 1-r
( )
1!

Analyzing the cost functions (6), and (7) reveals the behave­ d 4m2r 4 tan-1 (r)
-JMOE(f,d) = 2 2 1--
ior of these algorithms. However, due to the recursive nature
of the cost functions, it is extremely difficult to obtain closed
� (1-r ) 1!
(19)
form expressions. In the following section, we will investigate 2m2
the recursive portion of the cost functions for the two-tap
channel.
When the above derivative at r = 0 is not zero. 00

Re(E(Sk_IYk_2)) =Re L(-rjrE(Sk_ISk_2_n)

(!!...dr .JMOE(C,d) ) r=O =-1 (20)
n=O
2
00
(26)

=m2 L(-r r h[2n+ I]

n=O
,
- - - - - - _,-
·
- ,
- -- - -� - - -- -_ .., . -- - - _ ..,.
----- - ---- ,
- - - - - -' - - - - - - _,- -
.
,
. -
Consequently,
· .
· .

-� -- ---- : - - - - - -:--- ---- : -- -- -- �- - - - - - -� ------ : - - --- -: --- - - - ----

· .
-1 - - - -

-� -- --- : ----- --: --- --: ------�-- - - - - -� -- --- : ----- --: -- --- -- -- --
,
,
,
,
.
,
,
,
,
,,, ,
,
,
,,, ,
,
,
,
,
,
-2 - - - - -

, ,,
= 2m2 Lr( _r2 )nh[2n + I]
,
, , , ,
, , , 00

-� -- - --; - - - -- -: - - - -; -- -- --�- - - - - - -� - -- --; - - - - - : ------ ------i-

, , , , ,

-3
(27)
- - - -- - ---
, , ,
"
, , "
,, , ,

n=O
,
"
� -4 -r------;--------; ------r------i-------r------;-------l
,
----- - - ---- -

, ,
4m
= 2 tan-I(r)
-� -- ----! ----- --; -- -- - -� - - -- -- �- -- -- --� ------! ----- -;----- - - ----
, , , , , ,

-�
"'C , • • , • , , •
-5 - - -

--: ------; ---- -- -:--- -- -- , -- --- -,- -------: ---- - -; -- -- -- -:- - - -- - -- ----
f : : : : : : : :
__

-6 -
1r
,
,
,
,
,
, ,, ,
, ,. ,. ,
,, , , ,

-� -- ---- : --- ---:- ------ : ------ �--- -- --� ------ : -------:- -----
· . . .
,
-7
, ,
-
,

,,
,

,
, , ,
and, furthermore,
,,.
-� -- : - - - - : - - -; - - -- -- �- - - - - -� - - - : - - -- :
· . , . ,
, , ,
, , , ,

-
· , , , , ,
-8

2r3 Re(E(sk_IYk_2)) 2 m2r 2 � >(-r 2) nh[ 2 n + 1]

---- - - -- -- - - - - - - - ------ 00

-� -- - -: ------ -: - ---: - - - - -- �- - - - - -� -- -- -: ---- -- : - - - - - - - - - - -�

, , ,
. ,
. , ,. , ,
·
,
.
, , . .
, , ,
,
, ,
=
-9 -- - -- - - --

, n=O
(28)
, , , , , , ,

Fig. 2. The ill convergence of the MOE-IIR algorithm at d = O,X � 0

Hence, assuming steady state
Hence, JMOE(d) d =0 . does not have a minimum at
E(Re(Yk)2)
Furthermore, the same result holds for JMOER (d) :

Property 2 = --
1
I- r 4 (
(1+r2)m2 - m2 (1+r2)itan-l (r)
1r
) (29)

1 m2 4 _I
JMOER(d) =-JMOE(d) =--2 I--tan (r)
2 I-r 1r
( ) (21)
m
= _2_
1- r 2
4
I- -tan
1r
_I
(
(r) )
Proof For the real component of Yk we have the follow­ 1
ing recursive equation:
Therefore, we have JMOER (r) ="2JMOE'
(22)
-,:
10 � -- - --f----- --- - -- �- -- ---f---- - ,:------� - -----f, -- : -- ---� -- -- -
- ----- -

The Output energy of the real component is

, I I I I , I

j
• I • , , , I • • •

9 f
----- i j
------------ f i
------ j------------ f i j
------ ------------ ------

2
E(Re(Yk)2)= E(sk ) +r2E(si-I)
· ..
. .. . . ..
. .. .

8 j - --- -� ------i- -- -- -j --- --- � --------i ----j - - - - - -i --------i ---- j ------

· . . . .

2 j ----� - - ---.: -----j. - -----� - - - .:- -----j ------t. - -.:- -----j - -

+r4 E(Re(Yk_2) )+ 2rE(SkSk_l)
· . , . ,

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

(23)
6 i -----f-- --:------i - ---f------: -----i---- -f------:---- -i------

- 2r2 Re(E(SkYk_2)) - 2r3 Re(E(Sk_IYk_2))

.. .. . .. ..
- -
-- - - - - -
.� · .
. . . ,

4.
· . . . .

5 �- ----�- - ---i------�------� ------i------� ------� -----i------� ------

j- ----�------: -----j------� --:------j------ ; --- :------j- -
-
� ·
.. .. . .
. .. . . . .

� 4
· .
- ----
.
---
.
- ---
. .

By using the following relation f

3 j-- ---!-------i -----j------!-------i -----j------t-------i -----j------
E(SkSk_l) = -m2h[/] (24)
2 �. -..- r·····.;... -.. � ... -. -� -.-...;..... - � ..... � ......;-..... �..-.-.

:------�-- ---� -- - -: -----� -- - ; ---------� ---- -

-

· ... .. . .
. .. .. . . ,

�- - : -
· . . .

we have 1 - - - - - ---- - - - -
: - -- --- -

· .. . . . .. . .

o �.--.-.�.-----�.-� -..�...... : - . . -.:. .

-
: : : :
00

Re(E(skYk_2)) =Re �)-rjr j E(SkSk_2_n)

.

-1

n=O (25)
2
00

0, x 2! 0
=-rm2 L(-r r h[2n+3] Fig. 3. The ill convergence of the MOE-IIR algorithm at d =

n=O
and
We confirmed MOE-IIR equalizers fail to converge to the
desired location for VSB signals.

IV. CONCLUSION
We showed the ill convergence of MOE algorithm for
blind IIR equalizers with complex recursive fIlters for VSB
signals .A cost function analysis for a two-tap channel
revealed that the MOE algorithm was not suitable for adapting
the complex valued IIR equalizers.

REFERENCES

[1] L. Tong, G. xu, and T. Kaliath, "Blind channel identification and

equalization using second-order statistics: A time-domain approach," IEEE
Trans. Iriformation Theory, vo1.40, pp.340-349

[2]. Y. Sato, "A method of self-recovering equalization for multilevel

amplitude modulation systems," IEEE Trans. Communication, pp.679-682,
June 1975.

[3]. D. N. Godard, "Self-recovering equalization and carrier tracking in a two­

dimensional data communication system," IEEE Trans. Communication,
pp.1867-1875,1980.

[4] J. R. Treichler and B. Agee, "A new approach to multipath correction of

constant modulus signals," IEEE Trans. Sign£ll Processing, Apr. 1983.

[5] T. Endres et. a1., "Carrier independent blind initialization of a DFE," IEEE
Sign£ll Processing Workshop on Sign£ll Processing Advanced in Wireless
Communications, Annapolis, MD, pp. 239-242, May 1999.

[6] J.Labat, O. Macchi, and C.Laot, "Adaptive decision feedback equalization:

Can you skip the training period?" IEEE. Trans. Communication, pp.921-930,
July 1998.

[7] W. Chung, "A dfe structure using quadrature components of 8-vsb

signals," IEEE. Transactions on Broadcasting, vo1.54, no. 3, pp. 394-400,
2008.

[8] S. Verdu, B. Anderson, and K. R. , "Blind equalization without gain

identification," IEEE Trans. Iriformation Theory, pp. 292-297, 1993.

[9] O.Macchi" Adaptive Processing. New York: John Wiley \& Sons Inc.,
1998.

[10] P. Regalia, Adaptive IIR Filtering in Sign£ll Processing and Control.

CRC,I994.

[11] C. R. Johnson, Jr. , "Adaptive iir filtering: Current results and open
issues," IEEE Trans. on Information Theory, pp. 237--250, Mar. 1984.

[12] K. J. Astrom and T. Soderstrom, "Uniquness of the maximum likelihood

estimates of the parameters of an arma model," IEEE Trans. Automatic
Control, pp. 769-773, Dec. 1974.

[13] L.Ljung, T. soderstrom, and I. Gustavsson, "Counterexamples to general

convergence of a commonly used recursive identification method," IEEE
Trans. Automatic Control, pp. 643,652, Dec. 1975,

[14] Henderson and 1. G. N. et. a1., "Atsc dtv receiver implementation,"

Proceedings of the IEEE, pp. 119-147, Jan. 2006.

[15] A. V. Oppenheim and R. W. Shafer, Discrete-Time Sign£ll Processing.

Prentice Hall, 1998.