Using Filter Tables
Using Filter Tables
03-03-12 P. Mathys
1 Normalized Filters
The figure below shows the structure of a lowpass ladder filter of order n with a normalized
resistive load of RL = 1 .
RS L2 Ln
R L L
VS C1 C3 Cn
RL RL
C C 1.0 C 1.0
V
n even n odd
Note that, depending on whether the filter order n is even or odd, the last element is either
a horizontal L branch (n even) or a vertical C branch (n odd).
The values of the individual capacitors and inductors depend on the filter type, the filter
order n, the value of the source resistor RS , and, of course, the cutoff or reference frequency
of the lowpass filter (LPF), r = 2fr . The computation of these values is somewhat
complicated. To simplify the design of filters, tables for the L and C have been computed
for a normalized reference frequency of r = 1 rad/s.
For Butterworth filters of order n with RS = 1 the normalized susceptances and reactances
ai (also called immittances) are given by
(2i 1)
ai = 2 sin , i = 1, 2, . . . , n ,
2n
where i is the component index. The table below shows these values for n 7.
Order RS C1 L2 C3 L4 C5 L6 C7
a1 a2 a3 a4 a5 a6 a7
1 1.0 2.0000
2 1.0 1.4142 1.4142
3 1.0 1.0000 2.0000 1.0000
4 1.0 0.7654 1.8478 1.8478 0.7654
5 1.0 0.6180 1.6180 2.0000 1.6180 0.6180
6 1.0 0.5176 1.4142 1.9319 1.9319 1.4142 0.5176
7 1.0 0.4450 1.2470 1.8019 2.0000 1.8019 1.2470 0.4450
Example: The figure below shows the schematic of a normalized Butterworth LPF of order
n = 3.
1
RS L2 N003
1 2
VS C1 C3
RL
1 1 1
AC 1
The system function H(s) = VO (s)/VS (s), where VO is the voltage across the load resistor
RL , of this filter for general component values is
RL 1
H(s) = .
RS + RL RS RL L2 C1 C3 3 RS C1 + RL C3 2 RS RL (C1 + C3 ) + L2
s + L2 s + s+1
RS + RL RS + RL RS + RL
Note: H(s) is a generalized version of the frequency response H(j) that is computed
in phasor notation as H(j) = VO /VS . The complex variable s in H(s) is of the form
s = +j, i.e., it has a real part added to the j variable. The advantage of using s instead
of just j is that the roots of the denominator polynomial in H(s) can be computed and
displayed in the complex s-plane, which gives some additional insight for the classification of
different filter types. Another important property of H(s) is that it is the Laplace transform
of the unit impulse response h(t) of the circuit with input vS (t) and ouput vO (t). The
frequency response of a system described by h(t) H(s) is obtained by evaluating H(s) at
s = j.
With the component values shown in the schematic above, the system function of the nor-
malized (r = 1 rad/s, RS = RL = 1) Butterworth LPF of order n = 3 is
1 1 1 1
H(s) = 3 2
= 2
.
2 s + 2s + 2s + 1 2 (s + s + 1)(s + 1)
The poles of H(s) (i.e., the roots of the denominator polynomial) lie in the left half of the
complex s plane on a circle with radius r and are spaced 180 /3 = 60 apart as shown in
the plot below.
2
Poles of n=3 Butterworth Filter
0.8
0.6
0.4
0.2
Im{s}
0.2
0.4
0.6
0.8
10
20
Magnitude [dB]
30
40
50
60
2 1 0
10 10 10
f [Hz]
Note that the maximum of the magnitude is -6 dB (since RL /(RS +RL ) = 1/2) and therefore
the -3 dB frequency is the frequency at which the magnitude has decreased to -9 dB.
Butterworth filters have a maximally flat frequency response magnitude in the passband.
Another class of filters, called Chebyshev filters, achieve a steeper transition from the pass-
band of the filter to the stopband by allowing some ripple in the magnitude of the passband
3
frequency response. The immittance coefficients ci for Chebyshev filters of oder n can be
computed recursively from the immittance coefficients ai for Butterworth filters using the
two intermediate quantities
RdB /10 tanh1 (1/ 1 + )
= 10 1 , and = sinh ,
n
where RdB is the maximum passband ripple in decibels. Using , , and ai , the ci are
computed as
a1 ai ai1
c1 = , and ci = , i = 2, 3, . . . , n .
ci1 2 + sin2 [(i 1)/n])
The ci for RdB = 0.2 dB and RS = 1 are given in the following table for n 7.
Order RS C1 L2 C3 L4 C5 L6 C7
c1 c2 c3 c4 c5 c6 c7
1 1.0 0.4342
2 1.0 1.0378 0.6746
3 1.0 1.2275 1.1525 1.2275
4 1.0 1.3028 1.2844 1.9762 0.8468
5 1.0 1.3394 1.3370 2.1661 1.3370 1.3394
6 1.0 1.3598 1.3632 2.2395 1.4556 2.0974 0.8838
7 1.0 1.3723 1.3782 2.2757 1.5002 2.2757 1.3782 1.3723
RS L2 N003
1 1.153
VS C1 C3
RL
1.228 1.228 1
AC 1
4
The magnitude of the frequency response of this filter is shown next.
10
20
Magnitude [dB]
30
40
50
60
2 1 0
10 10 10
f [Hz]
The 0.2 dB ripple in the passband is only visible by zooming in on the passband as shown
below.
5.2
5.4
5.6
Magnitude [dB]
5.8
6.2
6.4
6.6
6.8
7
2 1 0
10 10 10
f [Hz]
But what is very visible in the first graph is that now the -3 dB frequency is at f3 0.2 Hz or
at 3 = 1.268 rad/sec (rather than at 1 rad/s as for the normalized Butterworth filter). By
tradition (mainly driven by mathematical considerations), the design or reference frequency
r for Chebyshev lowpass filters is the highest frequency up to which the magnitude of the
frequency response in the passband stays within the maximum ripple specification.
To be able to compare Butterworth and Chebyshev filters of order n = 3, the components
of the Chebyshev filter were scaled as shown below so that it has the same -3 dB frequency
as the Butterworth filter.
5
RS L2 N003
1 1.462
VS C1 C3
RL
1.557 1.557 1
AC 1
The graph below shows a plot of the magnitude of the frequency response of the normalized
Butterworth LPF of order 3 and the 0.2 dB ripple Chebyshev LPF of order 3 with a -3
dB frequency 3 = 1 rad/sec (corresponding to a Chebyshev design reference frequency of
r = 0.788 rad/sec).
20
Magnitude [dB]
30
40
50
60
2 1 0
10 10 10
f [Hz]
The asymptotic (for large f ) decrease of the magnitude of the frequency response is -60
dB/decade (20 n dB/decade, in general) for both filters. But since the magnitude of the fre-
quency response of the Chebyshev filter decreases more quickly initially, it gets an additional
5-6 dB attenuation in the stopband. Looking at the roots of the denominator polynomial of
the system function H(s), the main difference between Butterworth and Chebyshev filters
is that the poles of the former lie on a circle (with radius r ), whereas the poles of the latter
lie on an ellipse. This is shown in the following plot for the Butterworth and the Chebyshev
example filters considered above.
6
Poles of n=3 Butterworth (L=1) and Chebyshev (L=0.7885) Filters
Butterworth
Chebyshev
1
0.8
0.6
0.4
0.2
Im{s}
0
0.2
0.4
0.6
0.8
The maximum ripple specification RdB for Chebyshev filters determines how wide (for small
RdB ) or how narrow (for larger RdB ) the ellipse is on which the pole locations lie.
2 Frequency Scaling
An existing (linear) filter design can be scaled from one reference frequency to another by
dividing all reactive elements (inductors and capacitors) by a frequency scaling factor kf ,
defined as
desired reference frequency
kf = .
existing reference frequency
Note that kf is a dimensionless number. Thus, the frequencies in the numerator and denom-
inator must have the same dimensions, either both in radians per second or both in Hertz.
7
RS L2
+
R L
VS
C1 RL
Vo
C R
AC 1
-
8
Example: Butterworth LPF of order n = 3 with -3 dB frequency fr = 1 kHz and RS =
RL = 1 (source and load resistors still normalized). We find that r = 2 1000 = 6283.2
rad/s. Thus,
desired reference frequency 6283.2
kf = = .
existing reference frequency 1
From the Butterworth filter design table we find for n = 3 that
a1 = 1.0 , a2 = 2.0 , a3 = 1.0 ,
where a1 = old C1 , a2 = old L2 , and a3 = old C3 . Dividing by kf = 6283.2 we thus
find the new element values
C1 = 159.15 F , L2 = 318.3 H C3 = 159.15 F .
The corresponding schematic is shown below.
RS L2 N003
1 318.3
VS C1 C3
RL
159.15 159.15 1
AC 1
The magnitude of the frequency response of this filter is shown in the graph below.
10
20
Magnitude [dB]
30
40
50
60
70
2 3 4
10 10 10
f [Hz]
Clearly, the design specification of fr = 1 kHz (-3 dB frequency) is very well met.
9
3 Impedance Scaling
So far only the normalized load (and source) resistance RL = 1 has been used. In general
RL will have a different value, e.g., 50 for RF applications or RL = 600 for telephony
and audio applications. To get an idea of how to scale filter designs for general RL values,
consider again the second order LPF system function
RL 1
H(s) = .
RS + RL RS L2 C1 2 RS RL C1 + L2
s + s+1
RS + RL RS + RL
In general, the kth coefficient in the denominator (and numerator, if applicable) polynomial
of H(s) has dimension (sec/rad)k since s (especially for s = j) has dimension (rad/sec)k
and H(s) is a dimensionless ratio (of output over input voltage). Clearly, terms that are just
ratios of RS and RL , such as RL /(RS + RL ) are not changing H(s) as long as RS and RL are
scaled by the same factor. But terms such as R C or L/R, which have dimension of sec/rad,
would change H(s) if only the R part is scaled. However, R C and L/R remain constant
if all the inductor values are scaled by the same factor as the resistance values and all the
capacitance values are changed by the inverse of the factor that is used for the resistance
values. Define
desired load impedance
kz = .
existing load impedance
Then, to convert a filter design from an existing to a desired load impedance, compute new
R, L, and C values as
old C
new R = kz old R , new L = kz old L , new C = .
kz
10
RS L2 N003
600 191m
VS C1 C3
RL
265.25n 265.25n 600
AC 1
The L and C values in this filter are now much more realistic and it is fairly easy to verify in
LTspice that the filter has the designed frequency response magnitude with a -3 dB frequency
of 1 kHz.
4 Filter Transformation
Let sL = L + jL denote the complex frequency variable associated with a lowpass filter
(LPF) design. To convert a LPF to a highpass filter (HPF), replace sL by 1/s in the lowpass
system function HL (sL ), i.e.,
1 1
sL = , = at sL = jL , s = + j = = = 1/L .
s jL
Thus, the frequency response H(j) of the HPF is the same as the frequency response
of the LPF evaluated at sL = j/L . In particular, the reference frequency r of the
LPF is mapped to 1/r for the HPF. Note that, for filters with real-valued coefficients,
|H(j)| = |H(j)|, i.e., the magnitude of the frequency response has even symmetry in .
The substitution sL 7 1/s affects the C and L elements of the LPF as follows:
1 s L
7 , and L s 7 .
C sL C s
(L) (L)
That is, a capacitor Ci in the LPF becomes an inductor Li = 1/Ci in the HPF and an
(L) (L)
inductor Li in the LPF becomes a capacitor Ci = 1/Li in the HPF.
Example: Chebyshev 0.2 dB ripple HPF of order n = 5 with reference cutoff frequency 2
MHz and RL = 50 . Start from the normalized Chebyshev LPF shown below.
11
RS L2 L4
1 1.337 1.337
VS C1 C3 C5
RL
1.339 2.1661 1.339 1
AC 1
RS C2 C4
1 0.748 0.748
VS
L1 L3 L5 RL
0.747 0.462 0.747 1
AC 1
Finally, apply a frequency scaling factor of kf = 2 2 106 = 1.257 107 and an impedance
scaling factor of kz = 50 for the final denormalized HPF design shown in the schematic
below.
12
RS C2 C4
50 1.19n 1.19n
VS
L1 L3 L5 RL
2.97 1.84 2.97 50
AC 1
To convert a lowpass filter with design frequency r to a bandpass filter with center frequency
c and bandwidth 2r , replace sL by (s2 + c2 )/(2s) in the lowpass system function HL (sL ),
i.e.,
s2 + c2
q q
2
sL = = s 2sL s + c2 =0 = s = sL sL c = sL j c2 s2L .
2 2
2s
That is,pfor each value of sL , two values of s are created, one with the imaginary
p part shifted
up by c sL , and one with the imaginary part shifted down by c s2L . Setting
2 2 2
sL = jL , yields
p q q
2
2 2
s = + j = jL j c (jL ) = j L c + L 2 = = L c2 + L2 .
In particular, considering only > 0, the design frequency r of the LPF maps into
p
r + c2 + r2 c r ,
where the approximation holds for narrowband BPFs for which c r . Thus, if r is the
-3 dB frequency of the LPF, then the -3 dB bandwidth of the (narrowband) BPF is 2r .
The mapping sL 7 (s2 + c2 )/(2s) leads to the following conversion of all capacitors C in
the LPF:
1 1
1 2 =
= 2
s + c Cs c2 C
Cs C +
2s 2 2s
C 2
C =
2 c2 C
13
Similarly, sL 7 (s2 + c2 )/(2s) leads to a conversion of all inductors L in the LPF as shown
below.
= s2 + c2 Ls c2 L
Ls L = +
2s 2 2s
L 2
L 2 c2 L
=
Example: Butterworth BPF of order n = 6 with bandwidth 100 kHz, center frequency
fc = 1 MHz, and RS = RL = 50 . Start from a normalized Butterworth LPF of order n = 3
with C1 = C3 = 1 F and L2 = 2 H. Then use frequency and impedance scaling factors
100, 000
kf = 2 = 3.1416 105 , and kz = 50 ,
2
to obtain the following LPF.
RS L2
50 318.3
VS C1 C3
RL
63.66n 63.66n 50
AC 1
14
Next, use c = 2 106 in sL = (s2 + c2 )/(2s), replace all capacitors with parallel resonant
LC circuits (C 0 = C/2 and L0 = 2/(c2 C)), and all inductors with series resonant LC circuits
(L0 = L/2 and C 0 = 2/(c2 L)). This yields the following BPF.
R1 L2 C2
50 159.15 159.16p
V1 C1 C3
L1 L3 R2
31.83n 795.8n 31.83n 795.8n 50
AC 1
Example: Design of n = 4 Butterworth BPF with bandwidth 400 Hz, center frequency
fc = 4.915 MHz, and RS = RL = 120 . Start from a n = 2 Butterworth LPF with design
(-3 dB) frequency 200 Hz and use the scaling factors
RS L2 n003
120 135m
VS C1
RL
9.38 120
AC 1
15
Now apply the mapping s 7 (s2 +c2 )/(2s) with c = 2 4.915106 to the capacitor and the
inductor
to obtain a parallel combination of C1 = 4.69 F and L1 = 223.57353 pH (note that
1/ L1 C1 = 2 4, 915, 000) as replacement for the LPF capacitor and a series combination
of L2 = 67.64902 mH and C2 = 15.5 fF (note again that 1/ L2 C2 = 2 4, 915, 000) as
replacement for the LPF inductor as shown in the following schematic.
RS L2 C2 n004
VS C1
L1 RL
4.69 223.57353p 120
AC 1
A problem with this circuit is the wide range over which the component values are scattered.
A capacitor with 4.69 F can conceiveably be found, but no conventional capacitor with 15.5
fF exists. Similarly, an inductor of 67.6 mH can be found (even though obtaining a high
Q for such an inductor at 4.915 MHz would be a challenge), but 224 pH is definitely too
small to be practical. Looking closely, one can see that the series resonant circuit can be
implemented using a quartz crystal. But that still leaves the problem of how to implement
the parallel resonant circuit. Fortunately, it is possible to convert the parallel resonant circuit
into a series resonant circuit, as shown in the next section.
5 Impedance Inverter
Sometimes it is necessary to convert a series resonant circuit into a parallel resonant circuit
and vice versa. This can be done with an impedance inverter circuit, such as the T circuit
shown below.
jX jX
Zi jX ZL
jX (jX + ZL ) jX ZL (jX)2 jX ZL X2
Zi = jX + = = .
jX + jX + ZL ZL ZL
16
Example: Using an impedance inverter to convert the parallel resonant LC circuit in the
n = 4 Butterworth BPF with fc = 4.915 MHz and bandwidth 400 Hz to a series resonant
circuit that can be implemented using a quartz crystal.
2012,
c P. Mathys. Last revised: 3-21-12, PM.
17