POLYPHASE FIR FILTER

IMPLEMENTATION

Presented by-
Sunil Kumar
PHD Scholar
PHD/EC/10006/2019
ECE Department, BIT Mesra
 OUTLINE
 Introduction
 Multirate processing
 Downsampling
 Upsampling
 Polyphase implementation of FIR
filters decimation case
 Polyphase implementation of FIR
filters interpolation case
 INTRODUCTION
 Polyphase filters are becoming a very
important component in the design of
various filter structures due to the fact
that it reduces the cost and complexity
of the filter by doing the process of
decimation prior to filtering which
reduces the multiplications per input
sample
 Polyphase filtering is a computationally
efficient structure for applying resampling
and filtering to a signal. Most digital filters
can be applied in a polyphase format, and
it is also possible to create efficient
resampling filterbank
 Multirate Processing
 Involves two actions on the digital signal:
 Downsampling: resampling downwards the
digital signal in the digital domain.

Fe Fe/M Retain one sample over M and discard

M
the M-1 others, every M samples.

 Upsampling: resampling upwards the digital

signal in the digital domain.

Fe LFe Insert L-1 zeros between each sample

L

Down sampling decreases the sampling rate by discarding samples

(decimation).Up sampling adds extra samples (interpolation).
 Downsampling
 x( n) if n  mM ,
x(n) M y(m) y( m )  
0 else.

 M 1 k
1
y( m )  x ( mM )   x ( n) ( n  mM ) 
1 j 2
 x(n)e
n
x( n)  M

m   M M k 1

Folding term
 Downsampling
 Anti-aliasing Filter
fc : (Fe/M)/2
x(n) H(z) M y(m)

Fe Fe/M

 Noble identity for decimation

H(zM) M M H(z)

The anti-aliasing filter removes frequencies above the

Nyquist frequency for the new sampling rate.
 Upsampling
 x( m ) if n  mL ,
x(m) L y(n) y ( n)  
0 else.

   
 
Y (z)   y( n) z
n  
n
  x( m ) z
m  
L m
 X zL
 Upsampling
 Interpolating Filter

fC : (Fe/L)/2
x(m) L H(z) y(n)

Fe LFe

 Noble identity for upsampling

L H(zL) H(z) L
Polyphase Implementation of FIR Filters
Decimation Case 1 of 4
H(z) M E(zM) M

mM 1
H z    h( n ) z n

n0

Let n=lM+k

 
M 1 N / M 1
H z    z Ek z
k M with E k z    hlM  k z l

k 0 l 0
 Polyphase Implementation of FIR Filters
Decimation Case 2 of 4
Processing load (MAC/s)
H(z) M
N

E0(zM)
z-1
E1(zM) MTe Time
z-1
• M-1 filter evaluation over M

EM-1(zM) M are discarded.

• N filter length

Fe Fe/M
 Polyphase Implementation of FIR Filters
Decimation Case 3 of 4
 Using noble identity

Processing load (MAC/s)

M E0(z)
z-1 N
M E1(z)

z-1
M EM-1(z)
MTe Time

Fe Fe/M
• No more useless computations, but one sampling period over M, CPU is
burdned with N MAC/s.
 Polyphase Implementation of FIR Filters
Decimation Case 4 of 4
 Equivalent commutator model

Processing load (MAC/s)

E0(z)

E1(z)

N/M

EM-1(z)
MTe Time

Fe Fe/M
• Commutator runs at Fe,. At each input sample only one component is computed and accu-
mulated with the result of the previous one. The result is output when the last component
is reached and accumulator is reset. This spreads the processing load over MTe.
Polyphase Implementation of FIR Filters
Interpolation Case 1 of 5
L H(z) L R(zL)

lL1
H z    h( n) z  n
n0

Let n=mL+L-1-k

 
L 1 N / L 1
H z    z  ( L 1 k )
Rk z L with Rk z    hmL  L  1  k z m

k 0 m 0
Polyphase Implementation of FIR Filters
Interpolation Case 2 of 5
Processing load (MAC/s)
L H(z)
N

L R0(zL)
z-1

R1(zL) Te/L Time

z-1 • L-1 multiplications by 0 over L

For each filter evaluation.
RM-1(zL)
• N filter length.

Fe LFe
Polyphase Implementation of FIR Filters
Interpolation Case 3 of 5
 Using noble identity

R0(z) L • At each output sampling instant,

z-1
only one component is non zero

R1(z) L

z-1
RM-1(z) L

Fe LFe
Polyphase Implementation of FIR Filters
Interpolation Case 4 of 5
 Equivalent commutator model

R0(z) Processing load (MAC/s)

R1(z)

N/L

RM-1(z)
Te/L Time
Fe LFe
• For each output sampling instant one polyphase component is computed.
When we reach again the first component (M-1) a new input sample is inputed
in the delay line of each polyphase component.
Polyphase Implementation of FIR Filters
Interpolation Case 5 of 5
 Linear Periodically Varying Time system

z-1 z-1

hL-1 h2L-1 h3L-1 z-1 z-1

z-1 z-1

h1 hL+1 h2L+1 h0 hL h2L

h1 hL+1 h2L+1

z-1 z-1 hL-1 h2L-1 h3L-1

h0 hL h2L
 Polyphase Decomposition
The Decomposition
 Consider an arbitrary sequence {x[n]}
with a z-transform X(z) given by
 n
X ( z)   n   x[ n ] z
 We can rewrite X(z) as
M 1  k
X ( z)   M
k 0 z X k ( z )
where
 n  n
X k ( z)   n   xk [n] z  n   x[Mn  k ] z
0  k  M 1
18
 Polyphase Decomposition

 The subsequences {xk [n]} are called the

polyphase components of the parent
sequence {x[n]}

 The functions X k (z ) , given by the

z-transforms of {xk [n]} , are called the
polyphase components of X(z)

19
 Polyphase Decomposition
 The relation between the subsequences{xk [n]}
and the original sequence {x[n]} are given by
 In matrix form we can write
xk [n]  x[Mn  k ], 0  k  M  1

 X0(zM ) 
 
X ( z )  1  M
z 1 .... z ( M 1)  X1.( z ) 
 .. 
X . M 
20  M 1 ( z )
 Polyphase Decomposition

 A multirate structural interpretation of

the polyphase decomposition is given
below

21
 Polyphase Decomposition
 The polyphase decomposition of an IIR
transfer function H(z) = P(z)/D(z) is not that
straight forward
 One way to arrive at an M-branch polyphase
decomposition of H(z) is to express it in the
form P '( z ) / D '( z M ) by multiplying P(z) and
D(z) with an appropriately chosen polynomial
and then apply an M-branch polyphase
decomposition to P '( z )

22
 FIR Filter Structures Based on
Polyphase Decomposition

 We shall demonstrate later that a

parallel realization of an FIR transfer
function H(z) based on the polyphase
decomposition can often result in
computationally efficient multirate
structures
 Consider the M-branch Type I
polyphase decomposition of H(z):

M 1  k
H ( z)   k 0 z Ek ( z M)
23
 FIR Filter Structures Based on
Polyphase Decomposition

 A direct realization of H(z) based on the Type

I polyphase decomposition is shown below

24
 FIR Filter Structures Based on
Polyphase Decomposition

 The transpose of the Type I polyphase

FIR filter structure is indicated below

25
 Conclusion
• The multirate operations of decimation and
interpolation that were introduced in the previous
section will now be used to decompose any system
function H(z) into its polyphase representation

• polyphase implementation of an FIR filter h whose

output was downsampled by the factor N . The
derivation was based on commuting the downsampler
with the FIR summer. We now derive the polyphase
representation of a filter of any length algebraically by
splitting the impulse response h into N polyphase
components.
THANK YOU