<s>
Decorrelation	B-Algorithm
is	O
a	O
general	O
term	O
for	O
any	O
process	O
that	O
is	O
used	O
to	O
reduce	O
autocorrelation	O
within	O
a	O
signal	O
,	O
or	O
cross-correlation	O
within	O
a	O
set	O
of	O
signals	O
,	O
while	O
preserving	O
other	O
aspects	O
of	O
the	O
signal	O
.	O
</s>
<s>
A	O
frequently	O
used	O
method	O
of	O
decorrelation	B-Algorithm
is	O
the	O
use	O
of	O
a	O
matched	O
linear	B-Architecture
filter	O
to	O
reduce	O
the	O
autocorrelation	O
of	O
a	O
signal	O
as	O
far	O
as	O
possible	O
.	O
</s>
<s>
Since	O
the	O
minimum	O
possible	O
autocorrelation	O
for	O
a	O
given	O
signal	O
energy	O
is	O
achieved	O
by	O
equalising	O
the	O
power	O
spectrum	O
of	O
the	O
signal	O
to	O
be	O
similar	O
to	O
that	O
of	O
a	O
white	O
noise	O
signal	O
,	O
this	O
is	O
often	O
referred	O
to	O
as	O
signal	B-Algorithm
whitening	I-Algorithm
.	O
</s>
<s>
Although	O
most	O
decorrelation	B-Algorithm
algorithms	O
are	O
linear	B-Architecture
,	O
non-linear	O
decorrelation	B-Algorithm
algorithms	O
also	O
exist	O
.	O
</s>
<s>
Many	O
data	O
compression	O
algorithms	O
incorporate	O
a	O
decorrelation	B-Algorithm
stage	O
.	O
</s>
<s>
For	O
example	O
,	O
many	O
transform	B-Algorithm
coders	I-Algorithm
first	O
apply	O
a	O
fixed	O
linear	B-Architecture
transformation	I-Architecture
that	O
would	O
,	O
on	O
average	O
,	O
have	O
the	O
effect	O
of	O
decorrelating	O
a	O
typical	O
signal	O
of	O
the	O
class	O
to	O
be	O
coded	O
,	O
prior	O
to	O
any	O
later	O
processing	O
.	O
</s>
<s>
This	O
is	O
typically	O
a	O
Karhunen	O
–	O
Loève	O
transform	O
,	O
or	O
a	O
simplified	O
approximation	O
such	O
as	O
the	O
discrete	B-General_Concept
cosine	I-General_Concept
transform	I-General_Concept
.	O
</s>
<s>
By	O
comparison	O
,	O
sub-band	B-Algorithm
coders	I-Algorithm
do	O
not	O
generally	O
have	O
an	O
explicit	O
decorrelation	B-Algorithm
step	O
,	O
but	O
instead	O
exploit	O
the	O
already-existing	O
reduced	O
correlation	O
within	O
each	O
of	O
the	O
sub-bands	B-Algorithm
of	O
the	O
signal	O
,	O
due	O
to	O
the	O
relative	O
flatness	O
of	O
each	O
sub-band	B-Algorithm
of	O
the	O
power	O
spectrum	O
in	O
many	O
classes	O
of	O
signals	O
.	O
</s>
<s>
Linear	B-Algorithm
predictive	I-Algorithm
coders	I-Algorithm
can	O
be	O
modelled	O
as	O
an	O
attempt	O
to	O
decorrelate	O
signals	O
by	O
subtracting	O
the	O
best	O
possible	O
linear	B-Architecture
prediction	O
from	O
the	O
input	O
signal	O
,	O
leaving	O
a	O
whitened	O
residual	O
signal	O
.	O
</s>
<s>
Decorrelation	B-Algorithm
techniques	O
can	O
also	O
be	O
used	O
for	O
many	O
other	O
purposes	O
,	O
such	O
as	O
reducing	O
crosstalk	O
in	O
a	O
multi-channel	O
signal	O
,	O
or	O
in	O
the	O
design	O
of	O
echo	O
cancellers	O
.	O
</s>
<s>
In	O
image	B-Algorithm
processing	I-Algorithm
decorrelation	B-Algorithm
techniques	O
can	O
be	O
used	O
to	O
enhance	O
or	O
stretch	O
,	O
colour	O
differences	O
found	O
in	O
each	O
pixel	B-Algorithm
of	O
an	O
image	O
.	O
</s>
<s>
This	O
is	O
generally	O
termed	O
as	O
'	O
decorrelation	B-Algorithm
stretching	O
 '	O
.	O
</s>
<s>
The	O
concept	O
of	O
decorrelation	B-Algorithm
can	O
be	O
applied	O
in	O
many	O
other	O
fields	O
.	O
</s>
<s>
In	O
neuroscience	O
,	O
decorrelation	B-Algorithm
is	O
used	O
in	O
the	O
analysis	O
of	O
the	O
neural	B-Architecture
networks	I-Architecture
in	O
the	O
human	O
visual	O
system	O
.	O
</s>
<s>
In	O
cryptography	O
,	O
it	O
is	O
used	O
in	O
cipher	O
design	O
(	O
see	O
Decorrelation	B-General_Concept
theory	I-General_Concept
)	O
and	O
in	O
the	O
design	O
of	O
hardware	O
random	O
number	O
generators	O
.	O
</s>
