Realizations of Discrete-Time
Systems
Direct Form II
For a discrete-time system described by
N M
∑a k y [n −k]= ∑bk x[n −k]
k=0 k=0
the transfer function is of the form
Y (z) b0 + b1z−1 +L + bN z−N b0zN + b1zN −1 +L + bN
H (z )= = −N
=
X (z) a0 + a1z +L + aN z
−1
a0zN + a1zN −1 +L + aN
⎛ Here the order of the numerator and denominator are both ⎞
⎜ indicated as N. If the order of the numerator is actually less ⎟
⎜ ⎟
⎜⎝ than N , some of the b’s will be zero. But a must not be zero.⎟
⎠
0
Direct Form II
H (z )= H 1 (z)H 2 (z)
where Y1 (z) 1
H1 (z )= =
X (z) a0zN + a1zN −1 +L + aN
Y (z)
H 2 (z )= = b0zN + b1zN −1 +L + bN
Y1 (z)
Rearranging H1 (z ),
zN Y1 (z)= (1/a0 ){X (z)−⎡⎣a1zN −1 Y1 (z)+L + aN Y1 (z)⎤⎦}
Direct Form II
N −1
z N Y1 (z )= (1/a0 ){X (z)−⎡a
⎣1 z Y1 (z)+L + aN Y1 (z)⎤
⎦}
Direct Form II
Direct Form II
For the special case of FIR filters.
(The number of delays has been
changed to M - 1 to conform to
conventions in the DSP literature.)
M −1
h [n ]= ∑ bkδ [n −k]
k=0
This type of filter has M - 1 finite
zeros and M - 1 poles at z = 0.
Direct Form II
One desirable characteristic of an FIR filter is that it can have
linear phase in its passband.
The impulse response is
h [n ]= h[0]δ [n]+ h[1]δ [n −1]+L + h[M −1]δ ⎡
⎣n −(M −1)⎤
⎦
and itsz transform is
H (z)= h[0]+ h[1]z−1 +L + h[M −1]z−(M −1)
and itsfrequency response is
H (ejΩ )= h[0]+ h[1]e−jΩ +L + h[M −1]e−j(M −1)Ω
Direct Form II
The response y [n ] to an excitation x[n] is
y[n]= b0 x[n]+ b1 x[n −1]+L + bM −1 x ⎡⎣n −(M −1)⎤ ⎦.
LetM be even and letthe filtercoefficientsbe chosen such that
h[0]= h[M −1] , h[1]= h[M −2] , L , h[M /2−1]= h[M /2]
Direct Form II
Then the frequency response is
−j(M −1)Ω −jΩ −j(M −2)Ω
⎧
⎪h[0 ]+ h[0]e + h [1 ]e + h[1]e +L ⎫
⎪
H (e )= ⎨
jΩ
−j(M /2−1)Ω −j(M /2)Ω ⎬
⎪
⎩ + h[M /2−1 ] e + h[ M /2−1 ]e ⎪
⎭
or
⎧h[0](ej((M −1)/2)Ω + e−j((M −1)/2)Ω ) ⎫
⎛ M −1⎞ ⎪ ⎪
⎝ 2 ⎟⎠ ⎪ ⎪
−j⎜ Ω
j (M −3)/2)Ω −j M −3 /2 Ω
jΩ
H (e )= e ⎨ + h[1](e ( + e (( ) ) )+L ⎬
⎪ −jΩ jΩ
⎪
⎪
⎩ + h[M /2−1 ]( e + e ) ⎪
⎭
or
⎛ M −1⎞
⎧ ⎛⎛ M −1⎞ ⎞ ⎛⎛ M −3⎞ ⎞ ⎫
−j⎜
⎝ 2 ⎟⎠
Ω ⎪h[0]cos⎜⎜ ⎟ Ω⎟ + h[1]cos⎜⎜ ⎟ Ω⎟ +L ⎪
H (ejΩ )= 2e ⎨ ⎝⎝ 2 ⎠ ⎠ ⎝⎝ 2 ⎠ ⎠ ⎬
⎪ + h[M /2−1]cos(Ω ) ⎪
⎩ ⎭
Direct Form II
In the frequency response
⎛ M −1⎞
⎧ ⎛⎛ M −1⎞ ⎞ ⎛⎛ M −3⎞ ⎞ ⎫
−j⎜
⎝ 2 ⎟⎠
Ω ⎪h[0]cos⎜⎜ ⎟ Ω⎟ + h[1]cos⎜⎜ ⎟ Ω⎟ +L ⎪
H (e jΩ )= 2e ⎨ ⎝⎝ 2 ⎠ ⎠ ⎝⎝ 2 ⎠ ⎠ ⎬
⎪ + h[M /2−1]cos(Ω ) ⎪
⎩ ⎭
−j M −1 /2 Ω
there isa factore (( ) ) w hich hasa linearphase and the restofthe
frequency response ispurely real.
Direct Form II
The recursion relation is
y [n ]= b0 (x[n]+ x ⎡
⎣n −(M −1)⎤
⎦)
+b1 (x[n −1]+ x ⎡
⎣n −(M −2)⎤
⎦)+L
+bM /2−1 (x ⎡
⎣n −(M /2−1)+ x[M /2]⎤ ⎦)
w hich can be realized in thisform
thatreducesthe num berofm ultiplications
by half.
Direct Form II
It can be shown that symmetric or anti-symmetric, even or odd
impulse responses yield linear phase shift in the frequency response.
Cascade Realization
Direct Form II is no the only form of realization. There are
several other forms. Two other important forms are the
cascade form and the parallel form.
The cascade form is realized by first factoring the transfer function
z−z1 z−z2 z−zM 1 1 1
H (z )= A L L
z− p1 z− p2 z− pM z− pM +1 z− pM +2 z− pN
Each individual factor is realized as a small Direct Form II
subsystem and the subsystems are then cascaded.
Parallel Realization
The parallel form is realized by first
expressing the transfer function
in partial-fraction form
K1 K2 KN
H (z )= + +L +
z− p1 z− p2 z− pN
Each individual term is realized
as a small Direct Form II subsystem
and the subsystems are then
paralleled.
Complex Poles and Zeros
In either the cascade or parallel realization, the first-order
subsystems may have complex poles and/or zeros. In such a case
two first-order subsystems should be combined into one second-
order subsystem to avoid the problem of complex coefficients in the
first-order subsystems. Also, for reasons we will soon see, it is
common to do cascade and parallel realizations with second-order
subsystems even when the poles and/or zeros are real.
1st Order vs 2nd Order
In the case of FIR filters the second-order subsystems take
this form.
2 delays
3 multiplications
2 additions
Compare this two two first-order cascaded stages.
2 delays
4 multiplications
2 additions
1st Order vs 2nd Order
If the FIR filter has linear phase, a fourth-order structure reduces
number of multiplications even further compared with cascaded
first-order or cascaded second-order subsystems.