Finite Wordlength Effects

• Finite register lengths and A/D converters

cause errors in:-
(i) Input quantisation.
(ii) Coefficient (or multiplier)
quantisation
(iii) Products of multiplication truncated
or rounded due to machine length

1 Professor A G Constantinides
  

Finite Wordlength Effects

• Quantisation
Output
eo (k )

Q
ei (k )
Input

Q Q
  ei ,o (k ) 
2 2

2 Professor A G Constantinides
 Finite Wordlength Effects
• The pdf for e using rounding
1
Q

Q Q

2 2

Q 2
• Noise power 2 2
   e p (e).de  E{e } 2

Q 2
or 2
2 Q
 
12
3 Professor A G Constantinides
 Finite Wordlength Effects
• Let input signal be sinusoidal of unity
amplitude. Then total signal power P  1
  2
b
• If b bits used for binary then Q  2 2
2  2b
 so that   2 3
•  Hence 2 3  2b
P   .2
  2
or SNR  1.8  6b dB
4 Professor A G Constantinides
  

Finite Wordlength Effects

• Consider a simple example of finite
precision on the coefficients a,b of second
order system with poles e  j

1
H ( z)  1 2
1  az  bz
1
H ( z)  1 2 2
1  2  cos .z   .z

• where a  2  cos b 2

5 Professor A G Constantinides
  

Finite Wordlength Effects

bit pattern 2  cos ,  2 
000 0 0
• 001 0.125 0.354
010 0.25 0.5
011 0.375 0.611
100 0.5 0.707
101 0.625 0.791
110 0.75 0.866
111 0.875 0.935
6 1.0 1.0 1.0Professor A G Constantinides
 Finite Wordlength Effects
• Finite wordlength computations

INPUT OUTPUT

7 Professor A G Constantinides
 Limit-cycles; "Effective Pole"
Model; Deadband
• Observe that for H ( z )  1
(1  b1 z 1  b2 z 2 )
• instability occurs when b2  1
• i.e. poles are
• (i) either on unit circle when complex
• (ii) or one real pole is outside unit
circle.
• Instability under the "effective pole" model
8
is considered as follows Professor A G Constantinides
  

Finite Wordlength Effects

• In the time domain with H ( z )  Y ( z )
X ( z)
•
y (n)  x(n)  b1 y (n  1)  b2 y (n  2)

• With b2  1 for instability we have

Q b2 y (n  2) indistinguishable from y (n  2)
• Where Q  is quantisation

9 Professor A G Constantinides
  

Finite Wordlength Effects

• With rounding, therefore we have
b2 y (n  2)  0.5 y ( n  2)
are indistinguishable (for integers)
or b2 y (n  2)  0.5  y (n  2)
• Hence  0.5
y ( n  2) 
1  b2
• With both positive and negative numbers
 0.5
y ( n  2) 
10
1  b2 Professor A G Constantinides
  

Finite Wordlength Effects

 0.5
• The range of integers
1  b2

constitutes a set of integers that cannot be

individually distinguished as separate or from the
asymptotic system behaviour.
• The band of integers  0.5 0.5 
  ,  
 1  b2 1  b2 
is known as the "deadband".
• In the second order system, under rounding, the
output assumes a cyclic set of values of the
deadband. This is a limit-cycle.
11 Professor A G Constantinides
  

Finite Wordlength Effects

• Consider the transfer function
G( z)  1
(1  b1 z 1  b2 z  2 )
yk  xk  b1 yk 1  b2 yk  2

• if poles are complex then impulse response

is given
hk by

k
hk  .sin (k  1) 
sin 
12 Professor A G Constantinides
 Finite Wordlength Effects
1   b1 
• Where   b2   cos  
 2 b2 
• If b2  1 then the response is sinusiodal
with frequency
1 1   b1 
  cos  2 
T  
• Thus product quantisation causes instability
implying an "effective “ b2  1 .

13 Professor A G Constantinides
 Finite Wordlength Effects
• Consider infinite precision computations for
yk  xk  yk 1  0.9 yk  2 x0  10
xk  0 ; k  0
10

-2

-4

-6

-8

-10
-10 -5 0 5 10

14 Professor A G Constantinides
 Finite Wordlength Effects
• Now the same operation with integer
precision
10

-2

-4

-6

-8

-10
-10 -5 0 5 10

15 Professor A G Constantinides
 Finite Wordlength Effects

• Notice that with infinite precision the

response converges to the origin

• With finite precision the reponse does not

converge to the origin but assumes
cyclically a set of values –the Limit Cycle

16 Professor A G Constantinides
 Finite Wordlength Effects

• FIR filters

17 Professor A G Constantinides
 Finite Wordlength Effects
• Assume  e1 (k ) , e2 (k ) ….. are not
correlated, random processes etc.
2 2  2 Q 2
 0i   e  hi (k )  2

k 0 e 12
Hence total output noise power
2 2 2 22b  2 k sin 2  (k  1) 
0   01   02  2.  .
12 k 0 sin 2 
b
• Where Q  2 and
sin (k  1) 
k
h1 (k )  h2 ( k )   . ; k 0
sin 
18 Professor A G Constantinides
 Finite Wordlength Effects

• ie
 2b
2 2 1  2
1 
0  .
6 1   1   4  2  2 cos 2 
 2

19 Professor A G Constantinides
 Finite Wordlength Effects

A(n) B(n+1)
• For FFT
B(n) -
B(n+1)
W(n)

A( n  1)  A( n)  W ( n).B ( n)
B ( n  1)  A( n)  W ( n).B ( n)
A(n)
A(n+1)
B(n+1)
B(n) B(n)W(n)

20 Professor A G Constantinides
 Finite Wordlength Effects
• FFT
2 2
A(n  1)  B (n  1)  2
2 2
A(n  1)  2 A(n)
A(n)  2 A(n)
• AVERAGE GROWTH: 1/2 BIT/PASS

21 Professor A G Constantinides
 Finite Wordlength Effects
IMAG 1.0
• FFT
-1.0 1.0
REAL

-1.0
Ax ( n  1)  Ax ( n)  Bx ( n)C ( n)  B y ( n) S ( n)
Ax ( n  1)  Ax ( n)  Bx ( n) C ( n)  B y ( n) S ( n)
Ax ( n  1)
 1.0  C ( n)  S ( n)  2.414....
Ax ( n)

• PEAK GROWTH: 1.21.. BITS/PASS

22 Professor A G Constantinides
 Finite Wordlength Effects
• Linear modelling of product quantisation
x(n) ~
x ( n)
Q 

• Modelled as

x(n) + ~
x ( n)  x ( n)  q ( n)
q(n)

23 Professor A G Constantinides
 Finite Wordlength Effects
• For rounding operations q(n) is uniform
distributed between  Q2 , Q2
and where Q is
the quantisation step (i.e. in a wordlength of
bits with sign magnitude representation or
mod 2, Q  2 b ).
• A discrete-time system with quantisation at
the output of each multiplier may be
considered as a multi-input linear system

24 Professor A G Constantinides
 Finite Wordlength Effects
• FFT
q1 (n)...q2 ( n)...q p (n)

 x(n) h(n)  y (n)

• Then
 p
y ( n)   x ( r ).h( n  r )    q ( r ).h ( n  r ) 

r 0  1
 r 0 
• where h (n) is the impulse response of the
system from  the output of the multiplier
to y(n).
25 Professor A G Constantinides
 Finite Wordlength Effects
• For zero input i.e. x(n)  0, n we can write
p 
y ( n)   qˆ .  h (n  r )
 1 r 0

• where q̂ is the maximum of q (r ) ,  , r

which is not more than Q
2
Q p
• ie y (n)  .   h (n  r ) 
2  1n0 

26 Professor A G Constantinides
 Finite Wordlength Effects
• However
 
 h (n)   h(n)
n 0 n 0

• And hence
pQ 
y ( n)  .  h( n)
2 n 0
• ie we can estimate the maximum swing at
the output from the system parameters and
quantisation level
27 Professor A G Constantinides