Professional Documents
Culture Documents
Lecture 3 - Transient Response and Transforms: 3.1 Bessel-Thomson Filters
Lecture 3 - Transient Response and Transforms: 3.1 Bessel-Thomson Filters
Transforms
The filters so far considered (Butterworth, Chebyshev and elliptic) were designed
in mind; the impulse response , and
step response, may be poor. If we are concerned with preserving the signal shape,
then we may have to use a different type of filter (remember, nevertheless, that
low-pass filtering will inevitably round the corners of the waveform off and delay
it).
Consider a signal split into its Fourier components. To preserve shape, all
components must get through with the same gain factor and the same delay.
(Distortion due to frequency-dependent gain is amplitude distortion, that due to
frequency-dependent delay is phase distortion). If we low-pass filter, there is ob-
viously a gain loss at higher frequencies, hence a “rounding-off” of the shape, but
does less damage to the shape than a fast
fall-off. For all frequencies passed to have the same delay, we need a phase lag
proportional to in the pass-band.
Let! with constant; then an input leads to an output
#" $!#"% ie. a delay of for all frequencies, with the
amplitude changed from to . In fact,for
an & th-order
all-pole filter, (' )+ *
as ' , , and so we cannot have - for all . But clearly linear phase in
the passband is better than non-linear phase (as Fig. 3.1)
teristic, we make the derivatives of , with respect to , zero at
/.
These filters have a maximally-flat time delay; in order to arrive at such a charac-
, from the
30
linear passband phase non-linear passband phase
φ φ
n π /2 n π /2
amplitude ω amplitude ω
passband passband
02143 0 9: 8
;
to the &657 , for an & th-order filter. The first derivative, , is the time delay.
Thus:
<
=>-? terms in
+ )>@BA and higher C
D
=FEG? +)
D terms in and higher C
Consider &H I
. A general 3-pole (no zero) transfer function, with low-
frequency time delay , might be written:
K
L
J E
Hence E#?J? +~ SJ>W + ? A ~ KJW R
A
D B=FEg" E Q } ? E !WZG"
and D 00 EE 0 [>[>[ C
In general, for an & th order Bessel-Thomson filter:
32
&E O+ OXR Ox O~ O}
"
0 EL I "
0 E
E "
I I 0 0 E LE E . "
EL EL I LE "
EE II E . LE LE qE . I
These coefficients can also be calculated from Bessel polynomials (hence the
name of the filter).
0.5
−0.5
−2 −1.5 −1 −0.5 0
-. [
Figure 3.2: Bessel-Thompson poles (x), Butterworth (*) and Chebyshev Type I
. [ E I T. [
z. [ 0
(+). Dampings are Butterworth, , Chebyshev , and
Bessel-Thomson, .
33
3.1.2 Impulse response
0.06 1
0.05 0.8
0.04
0.6
g(t)
r(t)
0.03
0.02
τ 0.4
0.01 0.2
0 0
0 1 2 3 4 5 0 1 2 3 4 5
t t
Figure 3.3: Bessel & filter. Impulse response (left) and step response (right).
E
& : s ¡
O !Q
) )
For an th order filter,
For &¢ E ,
E E E
E . ' .
For &^ 0 ,
E E X£
RA !W + E . ' [
34
0
n=1
−100
−200
−300
φ
−400
−500
−600
n=8
−700
0 2 4 6 8 10
ω
35
3.1.4 Comparison of delays for Bessel-Thomson & Butterworth
filters
Fig. 3.5 gives a comparison of the time delays for 6th-order Bessel-Thomson and
Butterworth low-pass filters. Only with the Bessel filter will there be minimal
waveform distortion in the pass-band.
4.5 Bessel
Butterworth
3.5
3
delay (s)
2.5
1.5
0.5
−1 0 1
10 10 10
log ω
36
3.2 Analogue Frequency Transformations
We have now derived design procedures for low-pass filters using Butterworth,
Chebyshev or elliptic approximations to an ideal low-pass characteristic. The
high-pass (HP), band-pass (BP) and band-stop (BS) transfer functions can all be
obtained through well-known frequency transformations.
¥¦ E
ZEG?M§T= :
: ¨ + C Y©
is then mirrored about
U¤ in a log. plot since
Lªe¦ E
9E ?P§«= :>: ¨ + C XY©
2: : ¬|¨ 0 :L: ®b¨
(eg. response at is now the same as that at +A ).
U¤
Reminder:
U¤
= 3dB cut-off frequency for Butterworth filter
U¤ ¯
°u±
= pass-band edge frequency for Chebyshev filter
= geometric mean for elliptic filter (usually)
37
Delay characteristics
KJ ' : ± ¨Y
These do not transform in the same way as the amplitude response under the
non-linear frequency transformation given on the previous page; for ex-
ample, the low-pass phase lag of a Bessel-Thomson filter is approximately (
¶ : A : AY
constant). When this filter is transformed to a HPF, we have, instead a phase-lead
of , ie. a time advance of , which is non-uniform across the pass-band.
¸ u¥¦
¹ p
R +
Let be the 30dB cut-off frequency for the LPF.
¸ u¥¦
¹ »
º E . >
E º E R
R
E. .
p rads/sec
38
: ±¨ R
KJLL Uª ¦ E#? 0 ± È? 0 ± +
ie. :
¨ :
¨ ?
: ± ¨ R
the same as
KJLL ¥¦ except for 3 zeroes at the origin.
L
¥ ¦ E
L
ª e ¦ E
Finally, É E ? : : ¨ } and thus Ê #E ?z : : ¨ }
E
Check: For ...2Ë
Ì ªU¦ A Ç \ Í . [ . E , ie. 36dB attenua-
¯ AS@ Ä x
tion.
39
|G|
ωc
log ω
ω ωn ω log ω
l u
Figure 3.6: Lowpass to bandpass transform.
LPF BPF
"yU¤ U) ³
0
u¤ eÐ
" , 0
e
Õ
Ð Ñ
" U
³ 0 U¤ , despite what itU³looks
Note: this does not mean that
if you work through using the transform it is easy to see that
"TUÐ like!
U¤ .
In fact,
40
Phase characteristics
J y
J ? :± Y
Band-pass filters derived from low-pass prototypes using '
Î have char-
onwhich
acteristics SJ29¥¦ be ortheanti-symmetrical
are symmetrical (for amplitude) (for phase)
Z"# Z) ¥¦ complex conjugate of Z¥¦ , will have the same amplitude
about a logarithmic scale. Let low-pass transfer function;
but
opposite phase.
Ú Û Ø0 + Ñ
" + Ø 0fÚ Û Eg"v 0 Ø +
) ) )
Ø
(NB: is complex!) JÕ Ú ?
° If
)fÝÞÝ Ø
, as is often the case (narrow-band filter), poles are at
)
+
Ú )
ie. the pole pattern is reproduced at half-scale,
but with the origin transferred to
41
Bode Diagrams
From: U(1)
−20
−25
−30
−35
−45
−50
−55
−60
800
600
400
To: Y(1)
200
−200
−1 0 1
10 10 10
Frequency (rad/sec)
Figure 3.7: Band-pass filter Bode plot - amplitude (top) and phase (bottom).
¼ ½
{
¼ ½ E .~ . ¯ E . 0 E
¬
À Á
&
¼ ½ ¾ :
¨ ¼
½ } å]å]å;* [
:
å]å;*
ç = 3 meets the specification
KJLL ¥¦ +4ÄbÅ ¨Ç +4ÄWA Å ¨Ç Y ÄbÅ ¨Ç
Therefore, we have
AS@ Æ @ Æ @ Æ r
KJL
ÒW¦ KJG? )J + L ¥¦ E
EG? 0 ±
@ :
¨ Î È? 0 ± Y ±!@ :
: ¨ ÎY + ?z ± Y ±!@ :
: ¨ ÎY R
± Y : Y
J R
SJLLÒW¦ J } ?MO ~ J ~ ?PO J x ?POFè RJ R ?MO J + ?MO J ?PO
i.e. x +
A å
écêWë Ó ë ¤ R
Oè } OXR ¤R ? U¤S +
O å 0 U) ¤S x O x 0 ¤+ ? I + )
O +A 0 ¤+ ) + ? I x O ~ 0 u¤ )
) )
3.4 Conclusion
The frequency transformations, although highly non-linear, are appropriate for
transforming the standard low-pass filters (except for Bessel-Thomson filters) to
high-pass, band-pass and band-stop filters. This is because the frequency response
of the filters being transformed is an approximation to a piecewise constant char-
acteristic in the frequency bands of interest. Thus, taking the example of the ellip-
tic low-pass filter of Fig. 3.8, the non-linearity of the mapping affects the spacing
of the ripple peaks and valleys but has no effect on the amplitude of these ripples.
Therefore, the filters designed using this frequency transformation preserve the
&
equiripple character of the prototype filter. Figure 3.9 shows Butterworth Bode
plots for low-pass, high-pass and band-pass filters. All are filters.
43
(a) (b)
−10 −10
−15 −15
Phase (deg); Magnitude (dB)
−20 −20
−25 −25
0 0
−50 −50
To: Y(1)
To: Y(1)
−100 −100
−150 −150
−200 −200
−1 0 1 −1 0 1
10 10 10 10 10 10
44
(a)
From: U(1)
−75
−80
−85
1000
To: Y(1)
−1000
−1 0 1
10 10 10
(b)
From: U(1)
−75
−80
−85
0
To: Y(1)
−500
−1000
−1 0 1
10 10 10
Phase (deg); Magnitude (dB)
(c)
From: U(1)
−20
−40
−60
1000
To: Y(1)
−1000
−1 0 1
10 10 10
Frequency (rad/sec)
Figure 3.9: Butterworth filters, & . (a) lowpass, (b) highpass & (c) bandpass.
45