Multirate Digital Signal Processing

Applications to
Communication Systems
fred harris
[email protected]
Part 1

June -1, 2011

What the Customer Expects

What the Customer Will Pay

When The Customer Wants It

MORE

MORE

MO RE

MORE

MORE

MORE

MORE

MORE

MORE
MORE

MORE
MO RE
MOR E

MORE

MO RE
MORE
MORE

MORE

MO RE
MORE
MOR E

MORE

MORE

MORE

MORE

NE
XTW
EEK
TO
MO
R RO
W
TH
IS
A FT
ERN
OO
N

MORE

RE

MORE

MO RE
MORE

MORE

M
O
R
E

MORE

MORE

Size Customer Wants

Its all done with

Computer Chips

We each own a
Billion Transistors
We have an amazing wealth of resources
at our disposal! Just How big is a Billion?
A stack of a billion bank notes would be
76.2 kilometers High.
A billion seconds ago was 32.5 years ago.

We each own a
1,000,000,000 Transistors

By way of Comparison,
the Eiffel Tower Contains 18,084 Parts
It is Fastened together by 2.5 Million Rivets

The world manufactures more

transistors than
it grows grains of rice.

How big is a billion grains of rice?

y 8mm x 2mm x 2mm (Long Grain)
y 1-billion grains of rice
y 8 Meters x 2 Meters x 2 Meters
y Or 32 Cubic Meters
y Or a cube 3.2 Meters on a side
y It weighs 24,000 kg (26.6 tons)
y Market price, $1,000/ton; $26,600
y CLS-350 Mercedes Benz weighs 2,200 kg

Gordon_Moore_ISSCC-02-10-03

A Billion Transistors costs $20.00

0.00000001

Adam @ Home
Brian Basset

A review of useful things

you should have learned in
grade school about
sampled data filters

Parameters of
Sampled Data
Low Pass Filter
fS-Sample Rate
f1-Passband Edge
f2-Stopband Edge
1-Passband Ripple
2-Stopband Ripple

Estimate of Filter Length N

N = function (fs , f1 , f2 , 1 , 2 )
f
s K ( f , , ), f = f f
1 2
2 1
f
for Filters Designed with Windows, 1 = 2 ,
N = K( 2 ,f)

-20 *log10 ( 2 ) fS A(dB) fS

fS

=
f
22
f
22 f

Filter Length as Function

of Filter Specifications
N = K(1 , 2 ,f)

fS
f

1+1

H(f)
1

11

f
0

f2

f1
1

E(f)
0

f1

fS /2
2

f2

2
fS/2

Fixed Length Filter: Attenuation Proportional to

(Transition BW)/(Bin Width), A(dB)22 f/(fs/N)

Fixed Transition BW: Attenuation Proportional to

(Transition BW)/(Bin Width), A(dB)22 f/(fs/N)

Its not what you dont know

that gets you in trouble!
Its what you know for sure
to be true that just aint so!
Samuel Clemens

Spectral Resolution
Gaussian Window
Gaussian Window and Spectrum, Maximum Level Sidelobe -60 dB
1
Amplitude

0.8
0.6
0.4
0.2
0

-0.5

-0.4

-0.3

-0.2

-0.1
0
0.1
Normalized Time Interval

0.2

0.3

0.4

0.5

Zoom to Main Lobe

0
-20
-20

(dB)

Attenuation (dB)

-40

-40

-60

-60

-80
-0.1

-80
-0.5

-0.4

-0.3

-0.2

-0.1

0
0.1
Normalized Frequency

-0.05

0
Frequency

0.05

0.1

0.3

0.4

0.5

0.2

Spectral Resolution
Kaiser-Bessel Window
Kaiser Window and Spectrum, Maximum Level Sidelobe -60 dB
1
Amplitude

0.8
0.6
0.4
0.2
0

-0.5

-0.4

-0.3

-0.2

-0.1
0
0.1
Normalized Time Interval

0.2

Spectrum

0.4

0.5

Zoom to Main Lobe

0

-20
-20

(dB)

Attenuation (dB)

0.3

-40

-40

-60

-60

-80
-0.1

-80
-0.5

-0.4

-0.3

-0.2

-0.1

0
0.1
Normalized Frequency

-0.05

0.2

0
Frequency

0.05

0.1

0.3

0.4

0.5

Spectral Resolution, Remez Minimum

BW Window with -6-dB/Oct. Side Lobes
Remez Window and Spectrum, -6 dB/Octave, Maximum Level Sidelobe -60 dB
1

Amplitude

0.8
0.6
0.4
0.2
0
-0.5

-0.4

-0.3

-0.2

-0.1

0
Normalized Time

0.1

0.2

Spectrum

0.4

0.5

Zoom to Main Lobe

0

-20
-20

dB

Attenuation (dB)

0.3

-40

-40

-60

-60

-80
-0.1

-80
-0.5

-0.4

-0.3

-0.2

-0.1

0
0.1
Normalized Frequency

-0.05

0
Frequency

0.2

0.3

0.05

0.1

0.4

0.5

Spectral Resolution, Remez Minimum BW

Window with 0-dB/Oct. Side Lobes
Remez Window and Spectrum, 0-dB/Octave, Maximum Level Sidelobe -60 dB
1
Equivalent to Taylor or Tchebyshev Weights
Amplitude

0.8
0.6
0.4
0.2
0
-0.5

-0.4

-0.3

-0.2

-0.1

0
0.1
Normalized Time
Spectrum

0.2

0.3

0.4

0.5

Zoom to Main Lobe

0
-20

dB

Attenuation (dB)

-20
-40

-40

-60

-60

-80
-0.1

-80
-0.5

-0.4

-0.3

-0.2

-0.1

0
0.1
Normalized Frequency

Frequency

-0.05

0.2

0.05

0.1

0.3

0.4

0.5

Narrower Bandwidth FIR Filter

are Longer Filters

Constant
Form Factor

System Consideration:
Maintain Equal Delays in Each
Signal Processing Path
f

Use filters with

Odd Number of Taps

Narrow
Narrow Band
FilterBand Filter
Delay T1 Delay T1

Cross
Cross
Sig nal
Sig nal
Proc essingProc essing
e Band Filter Delay Line
Wid e BandWid
Filter
T1-T2
Delay T2 Delay T2

Right! Wrong!

Large Ratio of
Sample-Rate to Transition Bandwidth
Prima ry
Sig na l

Filter

Low Band wid th

Second ary Signa l

Prima ry
Sig na l
f

-100 0 100

-10,000
-300

10,000

300

f Sample Atten(dB ) 20, 000 80

=
= 365
fTransition
22
200 22

A Major Motivation for Multirate Signal Processing: Reduce Filter Length

We can not do that!
Increase Transition BW!
Decrease Out of Band Attenuation! We can not do that!
Running out of Options; there is only one left!
Reduce Sample Rate! We can do that! Yes Indeed, we can do that!

In-Band Ripple; Source of Preand Post Echoes

x(t)

y(t)

h(t)
H()

X()

x(t-TD)
0.5 x(t-TD+ Tp)
y(t)

x(t)

TD-TP

0.5 x(t-TD-Tp)

TD

TD+ TP

Effect of In Band Ripple-I

Impulse Response
1
Amplitude

0.8
0.6
0.4
0.2
0
-0.2
-30

-20

-10

10

20

30

20 Log10(Magnitude) (dB)

Frequency Response
Filter Pass-Band Ripple

0
1
dB

-20
0

-40
-1
-60

-0.5

-100
-4

-3

-2

-1

0
Normalized Frequency (f/f

0.5

Frequency

-80
2

BW

Effect of In Band Ripple-II

Amplitude

1
Input Signal to Filter
0.5
0
-72

-64

Amplitude

-56

-48

-40

-32

-24

-16

-8

16

24

32

40

48

56

64

72

-16

-8

16

24

32

40

48

56

64

72

-8
0
8
Time Samples

16

24

32

40

48

56

64

72

Output Signal From Filter

(Time Aligned with Input Signal)

0.5
0
-72

-56

-48

-40

-32

-24

Difference Between Input and Time Aligned Output

0.06
Amplitude

-64

0.04
0.02
0

-0.02
-72

-64

-56

-48

-40

-32

-24

-16

Recursive Filter:
Non-Uniform and Equalized Phase
Impulse Response

Phase Equalized Impulse Response

0.2
Amplitude

Amplitude

0.2
0.1
0
-0.1
0

20

40

60

80

0.1
0
-0.1
0

100

20

40

Attenuation (dB)

0
-50
-100
-4

-3

-2

-1

Samples

Group Delay

20
0
-4

-3

-2

-1

0
1
Frequency

80

100

0
-50

-100
-4

0
2

-2

Group Delay

In-Band
14
12
10
8
6
-0.5

40

60

Magnitude Response

0.5
3

4
In-Band

60
Samples

Attenuation (dB)

Magnitude Response

50
40
45

20

-0.5
0
-4

-2

0
Frequency

0.5

Effect of In Band Ripple-III

Amplitude

Input Pulse to Non Uniform Phase Recursive Filter

0.5
0
-72

-64

Amplitude

-48

-40

-32

-24

-16

-8

16

24

32

40

48

56

64

72

-16

-8

16

24

32

40

48

56

64

72

-8
0
8
Time Samples

16

24

32

40

48

56

64

72

Output Pulse From Filter

(Time Aligned with Input Pulse)

0.5
0
-72

0.05
Amplitude

-56

-64

-56

-48

-40

-32

-24

Difference Between Input and Time Aligned Output Pulses

-0.05
-72

-64

-56

-48

-40

-32

-24

-16

Effect of In Band Ripple-IV

Amplitude

Input Pulse to Phase Equalized Recursive Filter

0.5
0
-96 -88 -80 -72 -64 -56 -48 -40 -32 -24 -16

Amplitude

-8

16

24

32

40

48

56

64

72

80

88

96

-8

16

24

32

40

48

56

64

72

80

88

96

-8
0
8
Time Samples

16

24

32

40

48

56

64

72

80

88

96

Output Pulse From Filter

(Time Aligned with Input Pulse)

0.5
0
-96 -88 -80 -72 -64 -56 -48 -40 -32 -24 -16

Amplitude

0.02
0
-0.02

Difference Between Input and Time Aligned Output Pulses

-96 -88 -80 -72 -64 -56 -48 -40 -32 -24 -16

When Filter
Reduces Bandwidth
System Should
Reduce Sample Rate
M-to-1
h(n)
x(n)

y(n)

y(m )

Re-Sampling!
Does That Mean
We Didnt Do it Right
the First time?

Out-of-Band Side Lobe Levels

and Sample Rate Reduction
Filter Spectrum at Input Sample Rate

Attenuation (dB)

0
-20
-40
-60
-10

-8

-6

-4

-2

0
Frequency (MHz)

Filter Spectrum at Output Sample Rate

Attenuation (dB)

0
Increased In-band and out-of-band Noise
and interference due to Sum of Aliased Sidelobes

-20

Aliased Sidelobes

Sum of Aliased Sidelobes

-40
-60
-1

-0.5

0
Frequency (MHz)

0.5

10

Dont Use Equal Ripple Stop Band!

1/f Stopband Rolloff Remez Filter
0

-20

-20

-40

-40

dB

dB

Equal Ripple Remez Filter

0

-60

-60

-80

-80

-100
0

-100
0

10

20
Frequency

30

40

10

Zoom to Passband

30

40

Zoom to Passband
0.05

dB

0.05

dB

20
Frequency

-0.05

-0.05

-10

-5

0
Frequency

10

-10

-5

0
Frequency

10

Stop Band Ripple Controlled by

Penalty Function in Remez
T(f)

T(f)

f 1/f3 f 2/f3

remezfrf

W(f)

myfrf

W(f)

w(2)

f 1/f3 f 2/f3

1+1

H(f)

1
w(2)

f 1/f3 f 2/f3

1+1

H(f)
2

f
0

w(1)

w(1)
0

f 1/f3 f 2/f3

f 1/f3 f 2/f3

f
0

f 1/f3 f 2/f3

Relating Frequency Domain Specifications

to Time Domain Parameters
H(f)

BW

f
0
h(t)

-fS

fS
1
fS
t
1
BW

N=

Num ber of Sam ples

From Pea k to first Zero

fS
K(1 ,2 )
f

f
1/BW
= S
1/fS
BW

Common Down Sample Filters

0
1

2
M-2
M-1

POLYPHASE FILTER
0

DYADIC HALF-BAND FILTER

M:1

z -1

z -1

z -1

- -1
z

HOGENAUER FILTER (CIC)

- -1
z

- -1
z

Dual LTI Filters Have Same

Transfer Function
b0

x(n)

b0

y(n)
a1

b1

-1

y(n)

x(n)
b1

-1

a1

a2

-1

b2

b2

-1

a2

Dual Graphs:
Replace Nodes with Summing Junctions,
Replace Summing Junctions with Nodes,
Reverse Direction of Arrows.

Dual LTV Filters Perform Opposite

Function

M-2

M-2

M-1

M-1

M-to-1 Down Sampler

1-to-M Up Sampler

Linear Time Varying (LTV) Filters or

Periodically Time Varying (PTV)
Dual Filters Perform the Opposite Function
x(m)

-1
Z

-1
Z

-1
Z

-1
Z

h(0)

h(3)

h(6)

h(9)

h(1)

h(4)

h(7)

h(10)

h(13)

h(2)

h(5)

h(8)

h(11)

h(14)

-1
Z

-1
Z

-1
Z

-1
Z

y(3n)

h(12)

h(12)

h(9)

h(13)

h(10)

h(14)

h(11)

h(3)

h(0)

h(7)

h(4)

h(1)

h(8)

h(5)

h(2)

h(6)

x(n)
y(n)

3-to-1 Up-Sampling Filter

1-to-3 Down-Sampling Filter

PTV Multiply-Accumulate Filter

(Programmers Filter)

PTV Partial Sum Accumulate Filter

(ASIC Designers Filter)

Up Sample Filters
0
1

2
M-2
M-1

POLYPHASE FILTER
0

DYADIC HALF-BAND FILTER

1:M

- -1
z

- -1
z

- -1
z

z -1

HOGENAUER FILTER (CIC)

z -1

z -1

Motivation For Using

Multirate Filters

Processing Task:
Obtain Digital Samples of Complex
Envelope Residing at Frequency fC
Analog

Digital

Rec eiver
Multi-Channel
FDM
Input Signal

Single-Channel
Base banded
Output Signal

Input Spectrum

Selec ted
Narrow Band
Signal

f
fC

See!

Uta h

DSP Insertion in Communication Systems

Instinctive First Response:
Copy Legacy Analog Prototype
y We should avoid this approach!!
y If we dont, we emulate an analog design!
y That is not the reason we invoke and apply DSP!
y DSP is inserted to improve performance and reduce

cost!
y Analog prototype systems incorporate design
compromises appropriate for the time they were made!
y We dont want to perpetuate those compromises!
y We have access to tools and resources not available to
past designers!

First Generation DSP Receiver

1

f
2

f
3

f
4

f
e -j0t

Analog
Signal Processing

Low Pass
Filter

fs

Signal Conditioning for DSP Receiver

1

f
2

f
3

....

....

f
-fs/2

fs/2
Low Pass
Filter

fs

Duplicate Analog Processing in DSP

3

....

....

f
-fs/2

fs/2

....

....

f
-fs/2

fs/2

....

-fs

-fs/2

fs/2

....

fs

....
f

Ignoring
Good Advice!

-fs/M

fs/M

e -j0n

Low Pass
Filter

fs/M

Low Pass
Filter

fs

M:1

Fundamental Operation
Select Frequency,
Limit Bandwidth,
Select Sample Rate

s(n)

s(t)

s(n) e

-j0n
LOWPASS
FILTER

ADC
e
CLK

r(nM)

r(n)

LO

-j0n

h(n)

M:1

....

Spectral Description
Fundamental Operation
CHANNEL OF INTEREST

INPUT ANALOG FILTER RESPONSE

-fs/2

f
fs/2
TRANSLATED SPECTRUM

-fs/2
OUTPUT DIGITAL FILTER RESPONSE

fs/2
FILTERED SPECTRUM

0
-fs/2
fs/2
SPECTRAL REPLICATES ATDOWN-SAMPLED RATE

....

....
-fs/M 0

f
fs/M

Signal and Filter are at Different Frequencies

Which One to Move??
CHANNEL OF INTEREST

0
FILTER RESPONSE

0
TRANSLATED SPECTRUM

Second
Option

TRANSLATED FILTER

First
Option

Down Sample Complex Digital IF

3

....

....

f
-fs/2

fs/2

....

....

~
-fs

-fs/2

fs/2

....
-fs/M

fs/M

....

fs

....

....
f
-fs/M
Low Pass
Filter

fs/M

e j 0n

e -jM 0n
fs/M

Ba nd Pa ss
Filter

fs

M:1

Fundamental Operation Modified

s(n)

s(t)
ADC

r(n) e

r(nM)

BANDPASS
FILTER
h(n) e

CLK

j0n r(n)

j0n

e
LO

-j0n

M:1

Equivalency Theorem
r (n) = s (n)e j0 n * h(k )
= s (n k )e j0 ( n k ) h(k )
k

= e j0 n s (n k )h(k )e j0 k
k

= e j0 n {s (n) * h(n)e j0 n }

Signal Flow Description

of Equivalency Theorem
c os(nk)

c os(nk)
h(n)

x(n)

y(n)
h(n)

-sin(nk)

x(n)

h(n) c os(nk)

y(n)

h(n) sin(n k)

-sin(nk)

Reorder Translate and Resample

s(n)

s(t)

r(n) e

j0n

r(nM) e

j0Mn
r(nM)

BANDPASS
FILTER

ADC

h(n) e

M:1

j0n

CLK

LO

SPECTRAL DESCRIPTION
REORDERED FUNDAMENTAL OPERATION
CHANNEL OF INTEREST

INPUT ANALOG FILTER RESPONSE

-fs/2
FILTERED SPECTRUM

-fs/2

f
fs/2
TRANSLATED FILTER

0
fs/2
ALIASED REPLICATES AT DOWN-SAMPLED RATE

....

....

-fs/M 0 fs/M
TRANSLATED REPLICATES ATDOWN-SAMPLED RATE

....

....
-fs/M 0

f
fs/M

-j0Mn

Successive Transformations to turn Sampled Data

Version of Edwin Armstrongs Heterodyne Receiver
to Tuned Radio Frequency (TRF) Receiver
to Aliased TRF Receiver.
-j n
e k

Digital
Band-Pass

Digital
Low-Pass

H(Z)

H(Ze
M-to-1

Armstrong

Nyquist

Digital
Band-Pass

H(Ze

-j k

-j kn

Equivalency Theorem

M-to-1

Digital
Band-Pass

-j Mkn

-j k

H(Ze

-j k

)
M-to-1

M-to-1

M k = k 2

Any Multiple of
Output Sample Rate
Aliases to Baseband

2
or k = k
M

Coefficient Assignment of Polyphase Partition

For M-to-1 resample start at Index r and Increment by M
For 3-to-1 resample start at index r and increment by 3
C0 C1 C2 C3 C4

C0

C1

C2

0
0

Extract Delays To First

Non-Zero Coefficient

C3

C 5 C 6 C 7 C 8 C 9 C 10 C 11

0 C6

0 C7

C9

0 C4

C5

0 C8

0 C 10

0
0
C 11

C0

C3

0 C6

C9

C1

0 C4

0 C7

0 C 10

C2

C5

0 C8

C 11

This mapping from 1-D to 2-D is used by Cooley-Tukey

FFT. Polyphase Filters and CT-FFT are kissing cousins!

Polyphase Partition of Low Pass Filter

1-Path to M-Path Transformation
N 1

H ( Z ) = h( n) Z

y(n) M:1

x(n)

n=0

y(n) M:1

x(n)

H0( Z )

M 1 N 1

H ( Z ) = h(r + nM ) Z ( r + nM )

-1

H1( Z )

-2

H2( Z )

r =0 n =0

r =0

n =0

....

H ( Z ) = Z r h( r + nM ) Z nM

-(M-2)

M-Path Partition Supports M-to-1 Down Sample

Also Supports Rational Ratio
M-to-Q and M-to-Q/P Down Sample!

-(M-1)

....

N 1

....

M 1

HM-2( Z )

HM-1( Z )

Polyphase Partition of Band Pass Filter

1-Path to M-Path Transformation
Modulation Theorem of Z-Transform
N 1

G ( Z ) = h( n) e
n=0

j k n

N 1

= h(n) (e jk Z ) n = H (e jk Z )
n =0

M 1 N 1

G ( Z ) = h(r + nM ) e jk ( r + nM ) Z ( r + nM )
r =0 n=0
M 1

M k = k 2
2
or k = k
M

y(nM)

H( Z)

N 1

G ( Z ) = e jk Z r h( r + nM )e jk nM Z nM
r =0

M 1

G(Z ) = e
r =0

n =0

2
j
k
M

N 1

Z r h(r + nM ) Z nM
n =0

y(nM)

Polyphase Band Pass Filter and

M-to-1 Resampler
e

x(n)

j 2
k0
M

y(n) M:1

y(nM)

H0( Z )
e

j 2
k1
M
M

-1

H1( Z )

j 2
k2
M
M

-2

H2( Z )

....

....

....

j 2
k(M-2)
M

-(M-2)

HM-2( Z )

Z
e
-(M-1)

j 2
k(M-1)
M
M

HM-1( Z )

Noble Identity: Commute M-units of Delay

followed by M-to-1 Down Sample
M Delays

M Delays

Input Cloc k, T
1 Delay

1 Delay

Output Clock, MT
M-Units of Delay at Input Rate Same as 1-Unit of Delay at Output Rate

M:1
Z

M:1

-M

Z
M:1

-M

H(Z )

-1

M:1
-1

H(Z )

Apply Noble Identity to

Polyphase Partition
M:1

x(n)

j 2
k0
M

y(nM,k)
H0( Z )

M:1

j 2 k1

We Reduce Sample Rate

M-to-1 Prior to Reducing Bandwidth

-1

H1( Z )

M:1

j 2
k2
M

-2

(Nyquist is Raising His Eyebrows!)

H2( Z )

M:1

....

....

....

j 2
k(M-2)
M

-(M-2)

HM-2 ( Z )

Z
M:1

j 2 k(M-1)

M-fold Aliasing!
M-Unknowns!
M-Paths supply M-Equations
We can the separate Aliases!

-(M-1)

HM-1( Z )

We Intentionally Alias the Spectrum.

(Were you Paying Attention
in school when they discussed the
importance of anti-aliasing filters?)

Move Phase Spinners to Output of

Polyphase Filter Paths
M:1

x(n)

j 2 k0
M

y(nM,k)

H0( Z )
M:1
-1

M:1
-2

j 2
k2
M

j 2
k(M-2)
M

j 2
k(M-1)
M

H2( Z )

M:1
-(M-2)

....

....

....

HM-2( Z )

Z
M:1
-(M-1)

H1( Z )

j 2 k1

HM-1( Z )

Want Phase Spinners as far away from resampler as possible

Polyphase Partition with Commutator

Replacing the r Delays in the r-th Path
e

x(n)

j 2
k0
M

y(nM,k)

H0( Z )
e

j 2 k1

Note: We dont assign

Phase Spinners to Select
Desired Center Frequency
Till after Down Sampling
And Path Processing

H1( Z )
e

j 2
k2
M

j 2
k(M-2)
M

j 2
k(M-1)
M

....

....

H2( Z )

HM-2( Z )

HM-1( Z )

This Means that

The Processing for every Channel
is the same till the Phase Spinner
No longer LTI, Filter now has
M-Different Impulse Responses!
Now LTV or PTV Filter.

Armstrong to Tuned RF with Alias

Down Conversion to Polyphase Receiver
-j n
e k

Digital
Band-Pass

Digital
Low-Pass

H(Z)

H(Ze
M-to-1

M-Path Digital -j 2 rk
e M
Polyphase

-j k

)
M-to-1

H(Z)
r
M-to-1

Rather than selecting center frequency at input and reduce sample rate at
output, we reverse the order, reduce sample rate at input and select center
frequency at output. We perform arithmetic operations at low output rate
rather than at high input rate!

Down Sample 6-to-1

n-1

n-2

n-3

n-4

n-5

n-6

n-7

n-8

n-9 n-10 n-11 n-12 n-13 n-14

n-1

n-2

n-3

n-4

n-5

n-6

n-7

n-8

n+ 6 n+ 5 n+ 4 n+ 3 n+ 2 n+ 1

n-1

n-2

n-3

n-4

n-5

n-6

n-7

n-8

Polyphase Partition 1-D filter

becomes 2-D M-Path Filter

n-8 n-14
n-9 n-15
n-10 n-16
n-11 n-17
n-6 n-12

n-1 n-7 n-13

n-2
n-3
n-4
n-5

n-6 n-12

n+ 6 n

n-6 n-12

n-1 n-7 n-13

n+ 5 n-1 n-7 n-13

n-2 n-8 n-14

n+ 4 n-2 n-8 n-14

n-3 n-9 n-15

n+ 3 n-3 n-9 n-15

n-4 n-10 n-16

n+ 2 n-4 n-10 n-16

n-5 n-11 n-17

n+ 1 n-5 n-11 n-17

Reorder Filter and Resample

Hmm... this is very
good stuff....

s(n)

s(t)
ADC

LOWPASS FILTER
POLYPHASE PARTITION
r(nM,k)

h(0+ nM)

...

...

h(1+ nM)
CLK

r(nM)

.....

.....

h(r+ nM)

h(M-1+ nM)
PHASE ROTATORS
ALIASED HETERODYNE

j M rk

BANDPASS FILTER
POLYPHASE PARTITION

Phase and Gain Response

(3-Versions of Filter)
Prototype Filter,
Polyphase Filter Prior to Resampling,
Polyphase Filter after Resampling

Impulse Response
and Frequency Response of
Prototype Low Pass FIR Filter
Impulse Response, Prototype Lowpass Filter
1

Amplitude

0.8
0.6
0.4
0.2
0
-0.2
0

10

12

14

16

18

Normalized Time (1/BW)

Spectrum
Log Mag (dB)

0
0.2

Zoom to Passband

-20
0.1
0

-40

-0.1
-60

-0.2
-0.4

-80
-100
-3

-2

-1

-0.2

0.2

0.4

0.6

Normalized Frequency

Impulse Response of 6-Path Polyphase

Partition Prior to 6-to-1 Resampling

Frequency Response of 6-Path Polyphase

Partition Prior to 6-to-1 Resampling

Phase Response of 6-Path Polyphase

Partition Prior to 6-to-1 Resampling

Overlay Phase Response of 6-Path Polyphase

Partition Prior to 6-to-1 Resampling
-2
Nyquist
Zone
-2/3
Phase
Shifts

-1
Nyquist
Zone
-2/6
Phase
Shifts

0
Nyquist
Zone

Phase
Aligned

+1
Nyquist
Zone
2/6
Phase
Shifts

+2
Nyquist
Zone
2/3
Phase
Shifts

De-Trended Overlay Phase Response:

6-Path Partition Prior to
6-to-1 Resampling

3-D Paddle-Wheel Phase Profiles,

6-Path Partition Prior to 6-to-1 Resampling

Overlay 3-D Paddle-Wheel Phase Profiles,

6-Path Partition Prior to 6-to-1 Resampling

Overlay 3-D Paddle-Wheel Phase Profiles,

Showing Phase Shifts in +1 Nyquist Zone

Overlay 3-D Paddle-Wheel Phase Profiles,

Phase Shifted to Align Phases
in +1 Nyquist Zone

PolyChanDemo

Polyphase Partition
3

....

....

f
-fs/2
4

fs

Polyphase
h 0(n)

fs/M

Polyphase
h 1(n)

....

e -j0

fs/M

....

Low Pass
Filter

fs/2

fs/M

Polyphase
h M-2(n)

fs/M

Polyphase
hM-1(n)

5
6

e -j1

e -j(M-2)

f
6

e -j(M-1)

Polyphase Spectral Components

Phase and Gain Response of

Polyphase Filter Paths After
Resampling

Down Sampled Time series input to M-Path

Filter have offset Time Origins: M-Path
Filter Aligns All M-Paths to same Origin
x0(6n)

12

x1(6n)

-1

11

x2(6n)

-2

10

x3(6n)

-3

x4(6n)

-4
x5(6n)

-5

Polyphase Filter Arms: All-Pass Filters:

Different Time Delays
Impulse Response of Ten Polyphase Filters
6

0
0

10

Normalized Time

12

14

16

Polyphase Filter Phase Shift and

Phase Slope (Group Delay)
Spectral Phase Response

Spectral Group Delay Response

-7

-0.5

-7.2

-1

-7.4

Group Delay (d/d) (Samples)

-2

-2.5

-3

-3.5

-7.6

-7.8

-8

-8.2

-8.4

-4

-8.6

-4.5

-8.8

-5

-9
0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

Normalized Frequency (f/fs)

Normalized Frequency (f/fs)

Phase Profiles of 10-Path

Polyphase Filter
Spectral Phase Response of 10 Polyphase Filters (Prior to REsampling)
0

-5

-10

Phase Shift (/2)

Phase Shift (/2)

-1.5

-15

-20

-25

-30

-35

-40

-45

0.05

0.1

0.15

0.2

0.25

0.3

Normalized Frequency (f/fs)

0.35

0.4

0.45

0.5

0.5

Eight Phases of Eight-Path FIR Filter

phase slopes of eight path linear phase fir filter
0

-2

-4

-6

-8

-10

-12

-14

-16

-18

-20

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Eight Phases of Eight-Path IIR Filter

phase slopes of eight path non-linear phase iir filter
0

-2

-4

-6

-8

-10

-12

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Eight Phases of Eight-Path IIR Filter

phase slopes of eight path non-linear phase iir filter
0

-2

-4

-6

-8

-10

-12

-14

-16

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Single and Multi-Channel Channelizer

Polyphase M-Path Filter and M-Point FFT
Non Maximally Decimated Filter Banks
Arbitrary Resampling Embedded in
Polyphase Channelizer

Single Channel Armstrong and

Multirate Aliased Polyphase Receiver
e

j nk 2
M

Standard DDC
y(n,k)

x(n)

y(nM,k)

H(Z )
1

2
2-Polypha se
Filters

Polyphase DDC

y (nM)

x(n)

x(n)

1 M-to-1

j nk 2
M

y(nM,k)

H r (Z )

M-to-1

1-Polypha se
Filter

Dual Channel Armstrong and

Multirate Aliased Polyphase Receiver
e

j nk1 2
M

Standard DDC
y(n,k1)

x(n)

y(nM,k1)

H(Z )
1

2
j nk2 2
M

M-to-1

2-Polypha se
Filters
y(n,k2)

y(nM,k2)

H(Z )
1

2-Polypha se
Filters

Polyphase DDC
y (nM)

x(n)

x(n)

1 M-to-1

H r (Z )

2
M

e
M

j nk1

y(nM,k1)

1-Polypha se
Filter

M-to-1

j nk2 2
M

y(nM,k2)

M-Channel Polyphase Channelizer:

M-path Filter and M-point FFT
Polyphase
Partition
h0(n)
h1(n)
h2(n)

.....

fs

.....

FDM

.....

M-PNT
FFT

h3(n)

TDM

hM-2(n)
hM-1(n)
hr(n)= h(r+ nM)

Spectral Overlap Options for

M-path Polyphase Filter Bank
fBW

f
f
fBW

f
f
fBW

f
f

Non-Maximally Decimated Polyphase Filter Bank

Non-Maxim ally Dec im ated Polyp hase Filter
0

M-Pa th
Pa rtitioned
Filter
M Coeffic ients M

2
Dim ension
Polypha se
Filter
Circular
Input Buffer

fS

........

P-1
P

h(n,r:s)

r, row
s, sta te

M-1

Circ ular Output Buffer

1
2

Center
Frequenc ies

Spectral Centers from FFT,

Spectral Bandwidth from prototype Filter,
Sample Rate from Commutator, Circular Buffers
and State Machine

Deliver 5-Inputs
to 6-paths of
Polyphase filter:

192 kHz Channel Spacing

128 kHz Symbol Rate
256 kHz Sample Rate

Filter
Outp ut

Filter Da ta
Reg isters

Pha se
Alig ne d

-1

-1

-2

-2

-3

-3

-4

-4

Pha se
Alig ne d

State-0

State-1
Filter
Outp ut

Filter Da ta
Reg isters

Shift Data in
Serpentine Shift
Through Filter
addresses.
Circularly Shift
Output Buffer to
phase align filter
time origin with FFT
time origin

P fS
QM

fS
K
M

Sp ec tral Sha pe and Band width

Filter Da ta Filter
Reg isters Outp ut

Sta te Mac hine

Output
Sa m ple
Rate

M M-Point
FFT

Pha se
Alig ne d

Filter Da ta
Reg isters

Filter
Outp ut

10

-2

15

-3

-3

14

-4

-4

13

12

11

-1

-1

10

State-2
Filter
Outp ut

14

19

13

18

12

17

11

16

10

15

-2

State-3

Filter Da ta
Reg isters
20

Pha se
Alig ne d

Filter
Outp ut

Filter Da ta
Reg isters
25

19

13

24

18

12

23

17

11

-1

-1

22

16

10

-2

-2

21

15

-3

-3

20

14

-4

State-4

Pha se
Alig ne d

-4

State-5

Pha se
Alig ne d