Filter structures
Basic concepts
• Linear-phase FIR filters, types I-IV
• Odd/even length, symmetrical/anti-symmetrical
• Group delay
• Magnitude response |H(ejω)|
Basic concepts
• Translation of filter’s pass-band
• Modulation of filter coefficients
• Complementary filter G(z) = z-N/2 – H(z)
© R.W.
3
Frequency translation/modulation of filter coefficients at frequency k =
j2⇡!k/N
e j2⇡k/M
Basic
H (e )concepts
Hk (z) = H0 (ze ), 0 k M 1
j! j(! 2⇡k/M )
k (e= H (ze ), 0 ), k0
=HH(z) 0 MkM1 j2⇡k/M
1
k 0
j2⇡nk/M
hk [n] H
=k (e
h0j![n]e 0 k) ),0 M
) = H0 (ej(!, 2⇡k/M k 1M 1
hk [n] = h0 [n]ej2⇡nk/M , 0 k M 1
• delay
Group Zero-phase/amplitude response
• Group delay
Impossible to implement in practice, but easy to implement when
d✓(!)
causality
⌧H (!) = is relaxed , ✓(!) = arg{H(ej! )}
d! d✓(!)
⌧H (!) = , ✓(!) = arg{H(ej! )}
• Unlike magnituded! response, can be negative
Zero-phase/amplitude response of Type I
• Zero-phase response
Zero-phase/amplitude ofType
response of symmetric
I even-order N (Type I) FIR
filter
H(ej! ) = e j!N/2 H̆(!)
H(ej! ) = e j!N/2
H̆(!)
N/2
N NX
N/2
H̆(!) = h[ h[ X n]
] + 2N N cos(n!)
H̆(!)
2 = h[ n=1
] + 2 2 h[ n] cos(n!)
2 n=1
2
General zero-phase responseresponse
• Zero-phase response of Type I-IV FIR filters
General zero-phase
H(ej! ) H(e
= ej!j!N/2 j
) = eej!N/2 ej H̆(!)
H̆(!)
Polyphase
olyphase decomposition
decomposition © R.W.
1
X
1
Polyphase
decomposition
R.W.
5
Polyphase decomposition
• Consider the z-transform of sequence x[n]
• X(z) can be rewritten as
• Subsequences xk[n] are called polyphase components of x[n]
• Functions Xk(z) are called polyphase components of X(z)
R.W.
6
Type 1 polyphase decomposition
• Polyphase decomposition of
FIR filter H(z)
• The structure is used to
change filtering and down-
sampling to down-sampling
and filtering
• The number of operations
remains the same but the
filter operates at lower
frequency
Type 1 polyphase decomposition
• Transpose of the polyphase
decomposition of FIR filter
H(z)
• The structure is used to
change up-sampling and
filtering into filtering and up-
sampling
Type 2 polyphase decomposition
• Obtained by setting Ri(zM) =
EM-i(zM)
• In case of fractional
sampling rate change,
polyphase decomposition
can be used to filter at rate
Fs/M instead of LFs where Fs
refers to the original
sampling rate
Computationally efficient decimator
R.W.
10
Computationally efficient interpolator
Type I Type II
R.W.
11
Commutator representation of
interpolation and decimation with
polyphase structure
interpolation decimation
© R.W.
12
Polyphase fractional sampling/
fractional delay filter
• Polyphase structure for P/Q fractional sampling
• Stage r provides a delay equal to r/P of the input
sampling interval.
• Number of stages sets the resolution.
1:Q
h0(n)
h1(n) Commutator steps
through branches
with the increments
hP-2(n)
of Q.
hP-1(n)
Polyphase fractional sampling/
fractional delay filter
• Suppose that we want to calculate the output in the place r+d (r
+d needn’t be rational any more) between the stages r and r+1.
• Linear interpolation of filter outputs between the nearest
neighbors can be interpreted as interpolation of filter
coefficients.
Cascaded
Integrator Comb
Filters
Linear-phase digital FIR filters without multipliers
© R.W.
CIC
• The simplest low-pass FIR filter is the N-point moving
average filter or (excluding the scaling by 1/N), whose
transfer function is given by
• Another form of H(z), known as recursive running-sum filter
or boxcar filter, is given by
© R.W.
16
CIC decimator
• A realization of a factor-of-N decimator of the running
sum filter (applying noble identity) is given below
integrator comb
• Because a decimator implemented by one running-
sum filter usually does not provide enough stop-
band attenuation, the filters are cascaded giving rise
to cascaded integrator comb (CIC) filter
© R.W.
17
CIC Applications
• Down-sampling of the output of sigma-delta analog-to-digital
converter
• Typically 16:1 down-sampling CIC filter followed by a 4-path
polyphase filter or two half-band filters
© R.W.
18
1-4 stage CIC filters (without
decimation)
© R.W.
19
CIC decimator
• Filter length and down-sampling factor needn’t be equal
• The structure of a two-stage CIC decimator is shown below
• It can be easily shown that the structure corresponds to a
factor-of-R decimator with a length-RN running sum filter
© R.W.
20
CIC decimator
• Further flexibility in the design is obtained by including K
feedback paths before and K feedforward paths after the
down-sampler
• Typically, the number of sections is 3-5
• Increasing word length is a problem in practice
© R.W.
21
CIC decimator
• The corresponding transfer function becomes
• The parameters N and K can be adjusted for a given down-
sampling factor R to yield the desired out-of-band attenuation
• Adding more sections improves out-of-band attenuation but
also distorts the passband
© R.W.
22
CIC Interpolator
• CIC interpolators are usually used in the last section of the
multistage interpolator where the signal is sampled already in
a high rate
• The structure actually repeats input samples x(k) R times
(when starting from zero state)
© R.W.
23
CIC interpolator
• This feature can be used to reduce complexity in multistage
CIC interpolator
• Hold interpolator removes the comb section that requires the
most number of bits – and integrator that requires the least
number of bits
© R.W.
24
