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
