E4215: Analog Filter Synthesis and Design

Frequency Transformation
Nagendra Krishnapura (nkrishnapura@mltc.com)

4 Mar. 2003

S = Σ + jΩ Prototype frequency variable

s = σ + jω Transformed frequency variable
Qm  
S
k=1 1− Zk
H0 (S) = Q   Prototype transfer function
n S
k=1 1− Pk
QM  s

k=1 1− zk
H(s) = Q   Transformed transfer function
N s
k=1 1− pk

The prototype transfer function H0 (s) has n poles, m finite zeros, and n−m zeros at infinity 1 . Ap and As are
the attenuation of the prototype filter at Ω p and Ωs . Ap = −20 log 10 |H0 (Ωp )|, As = −20 log 10 |H0 (Ωs )|.

1 Lowpass to Lowpass transformation

• Transformation
S s
↔
Ωp ωp
Ωωp
S = jΩ ↔ s = jω = j
Ωp

• Scaled poles and zeros

Zk ωp
Zk ↔
Ωp
Pk ωp
Pk ↔
Ωp

• Every real pole P results in a scaled real pole p.

• Every complex conjugate pole pair P r ± jPi results in a complex conjugate pole pair p r ± jpi . The
prototype and the transformed pole pairs have the same quality factor.

• Resulting filter has N = n poles and M = m finite zeros.

1
Usually not mentioned explicitly

1
2

2 Lowpass to Highpass transformation

• Transformation
S ωp
↔
Ωp s
Ωp ωp
S = jΩ ↔ s = −j
Ω

• “Inverted” poles and zeros

Ωp ωp
Zk ↔ z k =
Zk
Ωp ωp
Pk ↔ pk =
Pk

• Every real pole P results in a scaled real pole p.

• Every complex conjugate pole pair P r ± jPi results in a complex conjugate pole pair p r ± jpi . The
prototype and the transformed pole pairs have the same quality factor.

• Resulting filter has N = n poles. The n − m zeros at infinity move to the origin.

3 Lowpass to Bandpass transformation

• Transformation: Ω = 0 (dc) transforms to ω = ω 0 (geometric center of the passband). Every fre-
quency is transformed into two frequencies whose geometric mean is ω 0 . i.e. if there is a peak
at Ωpeak in the prototype response, the transformed response has two peaks at ω p1 and ωp2 where
ωp1 ωp2 = ω02 .

S 1 s2 + ω02
↔
Ωp ωb s
Ωp ω − ω02
2
jΩ ↔ j
ωb ω

• ωb is the bandwidth, the width of the passband |ω p1 − ωp2 |, within which the attenuation is less than
Ap .

• Every real pole P is transformed into a complex conjugate pole pair p r ± jpi .

• Every complex conjugate pole pair P r ± jPi is transformed into two complex conjugate pole pairs
pr1 ± jpi1 and pr2 ± jpi2 both of which have the same quality factor Q. The quality factor of the
transformed pole pair increases as the ratio ω 0 /ωb increases.

• Resulting filter has N = 2n poles. The order is doubled.

 3

4 Lowpass to Band elimination transformation

• Transformation: Ω = 0 (dc) transforms to ω = ω 0 (geometric center of the stopband). Every fre-
quency is transformed into two frequencies whose geometric mean is ω 0 . i.e. if there is a peak
at Ωpeak in the prototype response, the transformed response has two peaks at ω p1 and ωp2 where
ωp1 ωp2 = ω02 .

S s
↔ ωb
Ωp s2 + ω02
ω
jΩ ↔ −jΩp ωb
ω 2 − ω02

• ωb is the width of the band, |ωp1 − ωp2 |, within which the attenuation is more than A p .

• Every real pole P is transformed into a complex conjugate pole pair p r ± jpi .

• Every complex conjugate pole pair P r ± jPi is transformed into two complex conjugate pole pairs
pr1 ± jpi1 and pr2 ± jpi2 both of which have the same quality factor Q. The quality factor of the
transformed pole pair increases as the ratio ω 0 /ωb increases.

• Resulting filter has N = 2n poles. The order is doubled.

5 Using frequency transformation to synthesize filters

0
• If it is a bandpass or a band elimination filter, convert the specified frequencies ω p1,p2 0
, ωs1,s2 to
2
ωp1,p2 , ωs1,s2 which have the same geometric mean ω 0 (ωp1 ωp2 = ωs1 ωs2 = ω0 ). While doing so, the
specifications should be tightened 2 , not loosened.

• Translate the given specifications A s , Ap , ωp , ωs (or ωp1,p2 , ωs1,s2 ) to a lowpass prototype specifica-
tion As , Ap , Ωp , Ωs . The choice of Ωp or Ωs depends on the available filter tables.

• Look up3 the filter transfer function that satisfies A s , Ap , Ωp , Ωs . There are usually several types of
filters. The choice depends on complexity of the active realization or additional specs., e.g. group
delay, if present.

• If a cascade structure is being designed, factorize the transfer functions into first and second order
terms in the numerator and the denominator. Transform the prototype transfer function into the desired
transfer function. Realize each pole/pole-pair (with associated zeros) using opamp-RC or g m -C first
and second order structures.

• If a ladder structure is being designed, look up the corresponding prototype ladder structure. Trans-
form the passive structure (Fig. 1) into the desired filter. Realize the resulting structure using element
replacement or leapfrog synthesis.

• Simulate the resulting active structure with ideal components to verify the integrity of the design.
Resimulate with nonidealities and modify/improve the circuit if need be.
4

prototype lowpass highpass bandpass band elimination

ωb/ω02ΩpL
LΩp/ωp 1/LΩpωp

LΩpωb/ω02
L

1/LΩpωb
LΩp/ωb

ωb/ω02ΩpC
CΩp/ωb
CΩpωb/ω02

C CΩp/ωp 1/CΩpωp 1/CΩpωb

Figure 1: Transformation of passive elements

A filter design software package can eliminate one or more steps in the synthesis. You could get the
poles and zeros and the passive ladder structure directly from the specifications. Mathematical tools like
MATLAB can provide you poles and zeros for a variety of standard filter types. Note that unnormalized
coefficients, as provided by MATLAB, can have a very wide range 4 , especially in high order filters and can
lead to gross errors in the frequency response. In those cases, the tool can be used to design a low frequency
prototype and the resulting poles and zeros scaled up.

2
The transition bandwidth will decrease on one of the sides
3
A. I. Zverev, Handbook of Filter Synthesis, Wiley, New York, 1967. The mother of all filter tables!
4
e.g. The denominator of a second order filter with ωp = 1 Grad/s and Q = 1 is 10−18 s + 10−9 s + 1.
 5

Low pass prototype Low pass High pass Band pass Band stop
passband attenuation ≤ Ap dB
stopband attenuation ≥ As dB
passband edge(s) Ωp ωp ωp ωp1 , ωp2 ωp1 , ωp2
stopband edge(s) Ωs ωs ωs ωs1 , ωs2 ωs1 , ωs2
frequency variable S = Σ + jΩ s = σ + jω
S s ωp 1 s2 + ω02 s
Equivalence ωb 2
Ωp ωp s ωb s s + ω02
Ω ω ωp 2
1 ω − ω0 2 ω
Equivalence − −ωb 2
Ωp ωp ω ωb ω ω − ω02
√ √
parameters ω0 = ωp1 ωp2 = ωs1 ωs2
ωb = ωp2 − ωp1
passband “center” Ω=0 ω=0 ω=∞ ω = ±ω0 ω = 0, ∞
stopband “center” Ω=∞ ω=∞ ω=0 ω = 0, ∞ ω = ±ω 0
Ωp ωp ωs ωp2 − ωp1 ωs2 − ωs1
passband/stopband
Ωs ωs ωp ωs2 − ωs1 ωp2 − ωp1
L LΩp/ωp 1/LΩpωp ωb/ω02ΩpL

LΩpωb/ω02
1/LΩpωb
LΩp/ωb
inductor
C CΩp/ωp 1/CΩpωp
ωb/ω02ΩpC

CΩpωb/ω02
CΩp/ωb

1/CΩpωb
capacitor