EDITOR: B. SOMANATHAN
NAIR
1.
INTRODUCTION
In
this blog, we explain the step-by-step design procedure of designing a Butterworth
filter in accordance with the given specifications. As stated in a previous
blog, the step-by-step design of Butterworth digital filters begins with the
design of its analog counterpart. From the analog design steps, we find
expression for H(s). From H(s) we find H(z) either by direct
conversion or with the help of a suitable transformation. Then using H(z)
we construct the desired filter structure.
For the design of digital filters,
we use the same set of equations that we have used for the design of analog
filters, with some modifications incorporated in them. We know that digital
signals are obtained from analog signals by means of analog-to-digital
conversion. For A/D conversion, analog signals are to be sampled by means of sampling
pulses. Hence, sampling frequency must be included into the specifications.
To incorporate sampling, we modify the specifications
suitably, as illustrated in the example given below.
Example 1: Design a Butterworth digital low-pass filter for the
following specifications:
1. Pass-band gain required : 0.89
2. Frequency up to which pass-
band gain must remain more
or
less steady, f1 : 25 Hz.
3. Amount of
attenuation required : 0.215
4. Frequency
from which
the
attenuation must start, f2 : 75
Hz
5. Sampling frequency, fs : 300 Hz
In the specifications given above,
the only modification we have made is to add the sampling-frequency term fS. This is the frequency
with which analog-to-digital conversion is done. According to sampling theorem,
we must have
fs ³ 2 fm (1)
where
fm is the maximum
frequency in the analog input signal to be processed. In our example, we have
been given fs = 300 Hz.
This value is much higher than the required minimum criterion given in (1). If
it is not specified in a given design problem, then we can assume a suitable
value for fs based on (1).
We now proceed with our design.
STEP 1: NORMALIZATION OF
FREQUENCIES
Before
starting the actual procedure for the digital filter design, it is quite common
to normalize the given frequencies
for the convenience of using small values of w’s. Similar techniques may be used in analog designs
also. It may be noted that if normalized frequencies are initially used, then
to get the final results denormalization
must be done.
To normalize a given frequency, we
use the equation
ωn = 2πf/fs (2)
where
ωn = normalized angular
frequency, f is either f1 or f2, and fs = sampling frequency. Note that the normalized frequency
is a mere number and has no units attached to it.
In certain
cases, frequencies will be specified in terms of angular frequencies (i.e. in
terms of ω’s). In such cases, we
still proceed in the same way as given above for normalizing the frequencies.
This is because, by definition, angular frequency
ω = 2πf (3)
Then
ω1 = 2πf1 (4)
ω2
= 2πf2 (5)
Now,
using Eq. (10.65), we get the normalized frequencies
ω1 = ωn1 = 2πf1/fs = 2πx25/300 = π/6 (6)
ω2
= ωn2 = 2πf2/fs= 2πx75/300 = π/2 (7)
STEP 2: DETERMINATION OF n AND
wC
Given
H1(ω) = 0.89
We
substitute w1 = p/6 in the Butterworth polynomial to yield
(0.89)2
= 1/[1+(π/6ωc)2n] (8)
Similarly, at w2 = p/2 in the
Butterworth polynomial, we have
(0.215)2 =
1/[1+(π/2ωc)2n] (9)
Solving (8)
and (9) simultaneously, we get n =
1.98. Choosing the next higher integer, we find the order of the filter
n = 2 (10)
To
find ωc, we use Eq.
(10.74). Substituting for n = 2 in (9)
and solving, we get
ωc
= 0.737 (11)
STEPS 2 AND 3: DETERMINATION POLES AND VALID POLES
We
have already seen that the poles of a second-order filter lie at 90o
with respect to each other on the circle of radius ωc, and that the valid poles are
B1
= ωc(‒0.707+j0.707) = 0.737(‒0.707+j0.707) = ‒0.521+j0.521 (12)
The second
pole B2 is the complex
conjugate of B1, and
therefore
B2 = ‒0.521‒j0.521 (13)
STEP 4: FINDING THE
EXPRESSION FOR H(S) IN THE ANALOG DOMAIN
By using the poles, we can now write
the expression for H(s) as:
H(s)
= ωc2/(s2 + 2δωcs+ ωc2) = ωc2/(s+a+jb)(s+a‒jb)
= 0.7372/(s+0.521+j0.521)(s+0.521‒j0.521)
= 0.543/(s+0.521)2+0.5212) (14)
Taking
the inverse Laplace transform of (14), we
obtain
h(t) = 1.042e‒0.521t sin
(0.521t) (15)
By taking the z transform of h(t), the
transfer function of desired digital filter will be obtained.
No comments:
Post a Comment