Multirate Digital Signal Processing

Applications to
Communication Systems
fred harris
[email protected]
Part 2

June -1, 2011

filter_ten_a, filter_ten_b, filter_ten_c

Critically Sampled
Analysis/Synthesis Filter Bank
H0 (Z)

G 0 (Z)

H1 (Z)

H0 (Z)

G 1 (Z)
4-Point
IFFT

H2 (Z)

H1 (Z)
4-Point
FFT

G 2 (Z)

H3 (Z)

H2 (Z)

G 3 (Z)

H3 (Z)

Non-Critically Sampled
Analysis/Synthesis Filter Bank
H0 (Z)

G 0 (Z)

H0 (Z)

H1 (Z)

G 1 (Z)

H1 (Z)

H2 (Z)

G 2 (Z)

H3 (Z)

H4 (Z)

H2 (Z)

G 3 (Z)
8-Point
IFFT

H3 (Z)
8-Point
FFT

G4 (Z)

H4 Z)

G 5 (Z)

H5 (Z)

f
H5 (Z)

H6 (Z)

H7 (Z)

G 6 (Z)

G 7 (Z)

H6 (Z)

H7 (Z)

Channelizer Time Response to

Tone Burst
0.2

2
channel 1

channel 0

channel 2

0.2

-0.2
-0.2

20

40

60

80

0.1

-2

20

80

20

0.1

40

60

80

-0.1

40

60

80

-0.1

40

60

80

60

80

channel 5

20

40

60

80

-0.1

20

40

channel 7

Channel Time Series

20

20

0.1

channel 6

0.1

-0.4

channel 4

-0.1

60

0.1
channel 3

-0.1

40

20

40

60

80

Channelizer Spectral Response to

Tone Burst
0

dB

channel 0

-50

-100

-100

-100

-150
-1

-150
1
-1

-0.5

0.5

-0.5

0.5

0
channel 3

dB

channel 2

-50

-150
-1

-50

-100

-100

-100

0.5

-150
-1

0.5

-0.5

0.5

-150
-1

-0.5

0.5

0
channel 6

channel 7

-50

-50

-100

-100

-150
-1

channel 5

-50

-0.5

-0.5

0
channel 4

-50

-150
-1

dB

0
channel 1

-50

-0.5

0
Frequency

0.5

-150
-1

Channelized Spectra

-0.5

0
Frequency

0.5

Reconstructed Tone Burst from

Analysis Filter Bank
Reconstructed Signal
1.5

Amplitude

1
0.5
0
-0.5
-1
-1.5
0

50

100

150
200
250
Spectrum: Reconstructed Signal

-3

-2

-1

300

350

400

dB

-50

-100

-4

0
1
Normalized Frequency (f/f BW )

1-to-16 Up-Sample in 32-Point IFFT

Circ ula r Buffer
fs

h0(n)
h 1(n)
flg= 0

16 Active
Inp ut Ports

...
flg= 0

h 16(n)

.....

flg =
1

.....

h15(n)
32-PNT
IFFT

flg =

Input
Samp le Ra te
12 Mhz
p er Cha nnel

h2(n)

h30(n)
h 31(n)
hr(n)= h(r+ nM)

16 fs

FDM

32-Path Polyphase Partition with

First Step of 1-to-16 Up-Sampling
32

H0( Z )

H0( Z )
1:16

-1

Z
1:16

....

....

....
32

H15 ( Z )

-1

H1( Z )

....

32

H1( Z )

-15

H15 ( Z )

-15

1:16

....

1:16

-30

-30

32

H30 ( Z )

-16

H16 ( Z )

....

....

-16

H16 ( Z )

H30 ( Z )

....

32

1:16

Z
1:16

32

-31

H31 ( Z )

-31

H31 ( Z )

Z
1:16

32-Path Polyphase Partition with Second

and Third Steps of 1-to-16 Up-Sampling
2

H0( Z )

H0( Z )

1:16

1:16
2

....

....

....
2

H15 ( Z )

-15

H15 ( Z )
-1

-1

H30 ( Z )

-14

-1

1:16

-1

Z
1:16

Z
1:16

-15

-1

-14

H30 ( Z )

1:16

H31 ( Z )

-15

H16 ( Z )

....

....

1:16

....

Z
1:16

1:16

H16 ( Z )

Z
1:16

....

Z
1:16

-1

H1( Z )

-1

....

H1( Z )

-1

-15

H31 ( Z )

Z
1:16

32-Path Polyphase Partition with

Final Step of 1-to-16 Up-Sampling
2

H0( Z )

Two Versions of Path

Filters in Partition

....

H1( Z )

Hk(Z )

H15 ( Z )
2

-1
Z H16 ( Z )

h(k+ 32)

h(k+ 64)

....

h(k)

h(k+ 96)

h(k+ 128)

-1

Z H(k+ 16)(Z )

-1

-1

Z H30 ( Z )
h(k+ 16)

Z H31 ( Z )

h(k+ 48)

h(k+ 80)

h(k+ 112)

h(k+ 144)

Embedded 2-to-1 Down-Sampler in 32-path Polyphase Filter

Requires Circular Buffer to Align Phase of Sinusoids
In Successive Output Blocks from IFFT
Circ ula r Buffer

IV
0

III
0

II
0

I
0

fs

flg = 1

32-PNT
IFFT
flg =

. ... .

flg= 0

flg= 0

2T
Phase Continuity

3T

4T

Single Channel Polyphase Filters

Down-Sampler and Up-Sampler
Embedded in Filter
Constant Workload Single Channel Filter

Nyquist Sample Criterion

M-to-1

x(n)

y(n)

Filter
H(Z)

fs

fs

y(nM)

fs/M

fS > 2BW
Nyquist
fS = 2BW + f
harris

f
2BW

fs

0
2BW+f

fs/M

fs

Interesting Relationship
x(n)
fs

N-Tap
Lowpass
Filter

y(n)

2BW = fs /M

M-to-1

fs

y(nM)
fs
M

....

....

f
-f s /M

Filter Length at Input Sample Rate :

N=

f s /M
0
f= (1-)fs /M

f S A(dB )
f 22

Signal 2 - Sided BW as Fraction of Output Sample Rate :

fS
; (Protect from Aliasing When Down Sampling)
M
Transition BW, f as Fraction of Output Sample Rate :
2 BW =

fS
; (Allowable Aliasing When Down Sampling)
M
Substitute in Filter Length at Input Sample Rate :
f = (1 )

N=

fS
A(dB)
A( dB )
1
=
(1 ) f S / M 22
(1 ) / M 22

Dividing both sides by M

N
1 A( dB)
=
;
M (1 ) 22

N (Ops)
N (Ops/Output)
=
M (Input/Output) M (Input)

Efficient Filtering When

Sample Rate is Large
Compared to Bandwidth

Bad Mismatch: Sample Rate Large

Compared to Bandwidth
200 Hz
0.1 dB
80 d B

-6 d B/Oc tave

f
200 Hz
20 kHz

20 kHz Inp ut
Sam p le Rate

365 Tap
FIR Filter

20 kHz Output
Sa m ple Rate

Nyquist Rate for Filter is 200 Hz+200Hz = 400 Hz or fs/50

Long Filters, High Sample Rate: Expensive!

365 Tap Protype Low Pass Filter
1
0.8
0.6
0.4
0.2
0
-0.2
-200

-150

-100

-50

50

100

150

200

Spectrum

-20
-40
-60

Sample Rate: 20.0 kHz

Pass Band: 0.0-to-0.1 kHz
Stop Band: 0.4-to-10 kHz
Stop Band Atten: 80 dB

Log Magnitude (dB)

Log Magnitude (dB)

Spectrum: Zoom to Pass-Band Ripple

0.2
0.1
0
-0.1
-0.2
-0.2

-0.1

Frequency (kHz)

0.1

0.2

-80
-100
-2

-1.5

-1

-0.5

Frequency (kHz)

0.5

1.5

Filter to Extract Low Bandwidth

Secondary Signal
Low BW
Second ary Signa l

Prima ry
Sig na l

Prima ry
Sig na l

0
-300 -100

-10,000

100 300

10,000

Reduce Sample Rate at Input to Filter:

Very Efficient Implementation!
365 Tap s
20 kHz

Polyp ha se
Low Pa ss Filter

400 Hz

50-to -1
8-tap s

20 kHz

400 Hz
8-tap

Selec t
Pa th
Weig hts

48
49

20 kHz

2
400 Hz

Coefficient
Bank

Down Sample to Reduce Sample Rate

Proportional to Bandwidth Reduction and
Up Sample to Preserve Input Sample Rate.
365 Tap s
20 kHz

365 Tap s

Polyp ha se
Low Pa ss Filter

400 Hz

20 kHz

Polyp ha se
Low Pa ss Filter

50-to -1

1-to-50
8-tap s

48
49

400 Hz

20 kHz

8-tap s

20 kHz

48
49

Efficient Polyphase Filter

20 kHz Inp ut
Sam ple Rate

365 Tap
FIR Filter

20 kHz

Selec t

20 kHz Output
Sa m ple Ra te

20 kHz

400 Hz
8-tap

8-tap

Coeffic ient
Ba nk

Coeffic ient
Ba nk

Selec t

Two Processes in Boxes: How can you

tell which is which from outside box?
(The Wet Finger Test)
White Box

20kHz

20kHz

365-Tap
Lowpass
Filter

365-ops/input

White Box

400 Hz

20kHz

8-Tap
Filter

Coeffic ient
Bank

Coeffic ient
Bank

Selec t

16-ops/input

20kHz

8-Tap
Filter

Selec t
State Mac hine

Clean-Up Filter Between

Polyphase Resampling Filters
400 Hz

20 kHz

400 Hz

8-tap

20 kHz
8-tap

16-tap
Clea n-up
Filter

Selec t

Coeffic ient
Ba nk

Coeffic ient
Selec t
Ba nk

Filter Bank
Response

f
-200

0
-100

200
100

Commercial FM Signal Structure

Composite
Stereo

L+ R

19-kHz

L-R

Pilot

X2

X2
19-kHz

38-kHz

76-kHz

SCA

Pilot Filter
Pilot
Attenuation
60 dB

Transition BW
4 kHz
L+ R

L-R

L-R
53

38
15 19 23
Sam ple Rate: 200

SCA

SCA
76

f(kHz)
100

Stereo FM Receiver
Composite
Stereo
The Diffic ult One
to Im plem ent

Low
Pass

19-kHz

Pilot
Filter

2L

Low
Pass

76-kHz

X2

X2
38-kHz

Low
Pass

2R

SCA

Extracting Pilot Signal From

Composite Stereo FM Signal
pilot
L+ R

L-R
15

38

23
19

140 Tap s
FIR
Filter

L-R

f
53

100

DLY
38kHz Pilot

19 kHz Pilot

N=

fS atten(dB)
200 60
=
= 136.6 => 140
df
22
4 22

Polyphase Pilot Extraction

j 0k2
10
2

H0 (Z )
200 kHz
Com posite
Stereo
(Real)

j1

2
10

j2

2
10

H1 (Z )
e

x(n)

H2 (Z )

y(nM,1)

Low Pa ss
Filter

H0 (Z )
c os(2

2
10 )

H1 (Z )
2
c os(410
)

p(n)
p(nM,1)

H2 (Z )
c os(18 2 )

2
j 9 10

10

..

..
H9 (Z )

2
c os( 0 10
)

.
...

.
...
e

20 kHz
Aliased to
Baseb and
Filtered Pilot
(Complex)

20 kHz
Aliased to
Baseband
Pilot
(Com plex)

20 kHz
Frequenc y
Doubled Aliased
Baseb and Pilot
(Real)

200 kHz
Double Frequenc y
38-kHz Pilot
(Real)

H9 (Z )

200 kHz
Up Sam pled
and
Translated
38-kHz Pilot
(c omp lex)

Prototype Filter with Multiple Stop

Bands and Dont-Care Bands
Impulse Response, 10-to-1 Downsample Prototyple Low Pass Filter
1

Amplitude

0.8
0.6

Batman Filter

0.4
0.2
0
-0.2

-15

-10

-5

0
Time Samples

10

15

-20

-40

-60

-80
-100

-80

-60

-40

-20

0
Frequency (kHz)

20

40

60

80

100

Input and Output Spectra from

Nyquist Zone 1 in 10 Stage Polyphase Filter
Input Spectrum: Pilot at 19 kHz

Magnitude

Nyquist Zone Centered at 20 kHz

0.8

Input Polyphase Filter Frequency Response

in First Nyquist Zone

0.6
0.4
0.2
0
0

10

15

20

25
Frequency (kHz)

30

35

40

45

50

1st Nyquist Zone Polyphase Output Spectrum: Pilot at -1 kHz

1
Magnitude

Log-Magnitude (dB)

Frequency Response of Baseband Clean-up Filter

0.8
0.6
0.4
0.2
0
-10

-8

-6

-4

-2

0
Frequency (kHz)

10

Extracted and Processed Aliased

Pilot Signal
Low-pass Output Spectrum: Pilot at -1 kHz

Magnitude

1
0.8
0.6
0.4
0.2
0
-10

-8

-6

-4

-2

0
Frequency (kHz)

10

10

Doubler Output Spectrum: Pilot at -2 kHz

Magnitude

1
0.8
0.6
0.4
0.2
0
-10

-8

-6

-4

-2

0
Frequency (kHz)

Pilot Aliased into Nyquist Zone-2 in 10Stage Polyphase Up Sampler

2nd Polyphase Output Spectrum: Pilot at 38 kHz

Magnitude

1
0.8
0.6
0.4
0.2
0
0

10

15

20

25
Frequency (kHz)

30

35

40

45

50

35

40

45

50

Time Series: Pilot and Double Frequency Pilot

Amplitude

1
0.5
0
-0.5
-1
0

10

15

20

25
Time Samples

30

Interpolators
And Interpolation

Applications
y Fixed Up-Sampler Interpolators
y Fixed Down-Sample Filters
y Reduced Cost Filtering When Large Ratio of

Sample Rate to Bandwidth

y Timing Recovery Re-Sampling of Time Series
y Timing Recovery Re-Sampling of Matched
Filter
y Clock Domain Alignment

Spectrum of Interpolator and

Periodic Spectrum of
Zero-Packed Shaping Filter
S pec trum of S haping F ilter and 1-to-32 Interpolating F ilter
0

Gain (dB)

-20
-40
-60
-80

-30

-20

-10

0
Norm aliz ed F requenc y

10

20

30

Zoom to S pec trum

0

Gain (dB)

-20
-40
-60
-80
-8

-6

-4

-2

0
2
Norm aliz ed F requenc y

Spectrum of 1-to-32 Interpolated

Shaping Filter
S pec trum of Interpolated S haping F ilter
0

Gain (dB)

-20
-40
-60
-80

-30

-20

-10

0
Norm aliz ed F requenc y

10

20

30

Zoom to S pec trum

0

Gain (dB)

-20
-40
-60
-80
-8

-6

-4

-2

0
2
Norm aliz ed F requenc y

Polyphase Partition of M-Path

Resampling Filter
N/M= 4
H0 (Z )
H1 (Z )
H2 (Z )

y(m )

....

....

x(n)

HM-1 (Z )

Efficient Hardware Implementation

of 1-to-M Polyphase Interpolator
N/M= 4
x(n)

Hr (Z )

y(m )

h(0+ nM)
h(1+ nM)
h(2+ nM)

...

.. .

h(M-1+ nM)

Interpolation Options
Initial Sa m ple Positions

Interpola ted Sam p le Positions

Initial Sa m ple Grid,
Unit dista nc e
Between Sa m ples

Sam e Rate Sam p le Grid,

Unit Dista nc e
Between Sam ples

Higher Rate Sam ple Grid ,

Less Tha n Unit Distanc e
Between Sam p les

Lower Rate Sam ple Grid,

More Tha n Unit Distanc e
Between Sam p les

M-Path, 1-to-M/Q Interpolator

Polypha se Filter
x(n)

y(m)
H(Z)

1:M

Q:1

N/M= 4
H0 (Z )
H1 (Z )
H2 (Z )

HM-1 (Z )

y(m )

....

....

x(n)

Q:1

5/3, Rational Ratio Re-Sampling

K(m+1)=[k(m)+3] modulo(5)
y5(m )= x(n+ k/5)

phs(0)

In
n
n+ 1
n+ 2

3-to-1

phs(1)

y(m )= x(n+ 3k/5)

x(n)
phs(2)
phs(3)

Out
0,3
1,4
2

phs(4)

n+ 1

n+ 3

n+ 2

0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4

x x

x x

x x

x x

x x

x x

Rational Ratio Interpolation.

Example; up 8, down 3
Input Sa m ples and availab le
1-to-8 Interpola ted Sam p les

n+ 3
n

n+ 2

n+ 1

3-to-8 Interpolated Sa mples

(up 8, down 3)
m+ 6
m

m+ 1

m+ 2

m+ 3

m+ 4

m+ 5

m+ 7

m+ 8

Interpolation To Time Position Between

Available Interpolation Points
(Arbitrary Ratio Interpolation)
Desired Sample Value

Error

Available Sam ple Value

n

n+ 1

Input
Sample

Nearest Available
Sample Position
n+ k/M

Desired Sample
Position n+ k/M+

Zero Order Hold Model of

Nearest (Left) Neighbor Interpolation

Interpolated Sam ple Values

Error
Zero-Order-Hold
Analog Levels

n+ 1

Desired Sample Position n+ k/M+

Spectrum of Up-Sampled Signal at

Input and Output of Virtual DAC
f

BW= 1
0

N
Output Sam ple Rate

2N

DAC Response
BW= 1
f

N
Output Sam ple Rate

2N

Frequency Response of DAC

at First Spectral Null
H(f)= - 1 f
N

DAC Response

f
N-0.5

N+ 0.5

1
1
1
1
b
:
f : | H ( ) |=
<2
2
2N
2N
N
(b 1)
7
, Say b = 8(bits ), N > 2 = 128
N >2

| H ( f ) |=

When signal is already 4-times oversampled

Need 32 stage up-sampler to suppress spectral artifacts to -48 dB

Shaping Filter: Time and Frequency

Response, Four Times Over Sampled
time response of shaping filter
0.3

amplitude

0.2

0.1

-0.1
-5

10

15

20

25

30

35

40

45

50

time
spectral response of shaping filter

log magnitude

-20

-40

-60
-2

-1.5

-1

-0.5

0
frequency

0.5

1.5

Time and Frequency Response of

32/6.4 Left Neighbor Interpolator
0.05
0.04
0.03
0.02
0.01
0
-0.01
0

50

100

150

200

250

0
-20
-40
-60
-80
-10

-8

-6

-4

-2

10

Time and Frequency Response of 32/6.37

Left Neighbor Interpolator
0.05
0.04
0.03
0.02
0.01
0
-0.01
0

50

100

150

200

250

0
-20
-40
-60
-80
-10

-8

-6

-4

-2

10

Prototype Interpolator Length for 8bit data, initially Over Sampled by 2.

f= 1

f
-2 -1.5

-4

-0.5 0 0.5

1.5 2

f= 1

DC

DC

DC

DC
f

-4

-2 -1.5

-0.5 0 0.5

1.5 2

To Obtain 128 Over Sample, M=64, N=(128/1)(66/22)=384

N/M=6: Need 64 6-tap filters in Polyphase Interpolator

Prototype Interpolator Length for 8bit data, initially Over Sampled by 4.

f= 3

f
-1.5

-4

-8

3.5 4

-0.5 0 0.5

f= 3

DC

DC

DC

DC
f

-8

-0.5 0 0.5

-4 -3.5

3.5 4

To Obtain 128 Over Sample, M=32, N=(128/3)(66/22)=128

N/M=4: Need 32 4-tap filters in Polyphase Interpolator

Address Control:
Modulo Accumulator
TOUT

Inp ut
Sam ples

TIN

Outp ut
Sam ples

Inp ut Tim e ind ex n

n
n+ 2
n+ 1
.... 8 9 0 1 2 3 4 5 6 .... 9 0 1 2 3 4 5 6 .... 9 0 1 2 Polyp hase index k

On Overflow,
x(n)
Insert New Input

d -acc = M

Polyp hase
Weig hts

TOut
f
= M In
TIn
fOut

...

Fractional Offset: d-acc

Filter

y(m )

k(m )

(m )
ac c (m )
Mod (M)
d-a cc

-1

Int(--)

Fractional Part
(For later use)

Two Neighbor,
Linear Slope Interpolator
Right Available
Interpolated
Sample Value

Desired
Sample
Value

Left Available
Interpolated
Sample Value
Linear
Interpolator

Input
Sample
value

Interpolated
Sample Value

n+ (k+ 1)/M
n+ 1
n+ k/M
Right
Available
Input
Left Available Desired Sample
Interpolation
Sample Interpolation
Position k+
Sample Position
value Sample Position

Equivalent Interpolating Kernel

x(k+ )
x(k) TRI(k)

k-1
M

x(k+ 1)
x(k+ 1) TRI(k+ 1)

x(k)

k
M

k+ 1
M

k+ 2
M

Spectrum of Up-Sampled Signal at

Input and Output of Virtual Linear
Interpolator
BW= 1
0

f
N
Outp ut Sa mp le Rate

2N

Rep ea ted
Spec tral Zeros

Tria ngle Sp ec tral

Resp onse
BW= 1

f
0

N
Outp ut Sa mp le Rate

2N

Frequency Response at First

Spectral Null of Linear Interpolator
H(f)=

[ N1 f]

Tria ngle
Response
f
N-0.5

N+ 0.5

2
2
2
1
1
1
1
1
b
b / 2
| H ( f ) |= f : | H ( ) |=
<2
:
<2 :
2
N
2N
2N
2N
(b / 21)
7
, Say b = 16(bits ), N > 2 = 128
N >2
When signal is already 4-times oversampled
Need 32 stage up-sampler to suppress spectral artifacts by -96 dB

Estimate y(n+k/M) & y(n+k/M)

With 3 Arms of Polyphase Filter
.

PHS-(k-2)

PHS-(k-1)

y(n+ k/M)

y(n)
PHS-(k)
y(n+ k/M)
PHS-(k+ 1)
PHS-(k+ 2)

.
Estimate y(n+k/M) & y(n+k/M)
With 2 Polyphase Filters
y(n+ k/M)

Polyphase
Matched
Filter

....

y(n+ k/ M)
k
Polyphase
Derivative
Matched
Filter

....

y(n)

.
y(n+k/M) & y(n+k/M)
With 2
Efficient Polyphase Filters
y(n+ k/ M)

y(n)
1-Stage Filter

Polyphase
Matched Filter
Coeffic ients
.

Coeffic ient
Selection

y(n+ k/M)

1-Stage Filter
Polyphase
Derivative
Matched Filter
Coeffic ients

Interpolation with Polyphase Low-pass

Filter and Polyphase Derivative Filter
for Local Slope Correction
y(n+ k/M)
x(n)

x(n)

y(n+ k/M+ /M) =

Filter

y(n+ k/M)+ y(n+ k/M)

hk(n)

Derivative Polyphase Filter

k(m )

dh=conv(h,[1 0 -1]*M/2

y(n+ k/M)

x(n)

dh=dh(2:length(dh)-1);

Filter

dhk(n)
k(m )
ac c (m)
Mod (M)
d-a cc

-1

(m )
Int(--)

Input Shaping Filter at

4-Samples per Symbol

Spectra of 64/10.49 Interpolated Signal

Interpolated Shaping Filter
1
0.8
0.6
0.4
0.2
0
-0.2
0

200

400

600

800

1000

1200

Frequency Response

Log magnitude (dB)

0
-20
-40
-60
-80
-100
-120

-20

-15

-10

-5

Normalized Frequency (f/fsym )

10

15

20

Signal Conditioning and Processing

Half
Band
Filter

Half
Band
Filter

2
2

160 MHz

Interp
Bank
Phase
Ac c um ulator

DDS

24-Path
Polyphase
Filter

24-PNT
FFT

12-to-1

DDS

Spectral Centers 1.7 MHz Separation

Channel BW: 1.7 MHz
Channels Span 30 MHz ( 17 Channels)
24-Channel Channelizer: 24*1.7=40.8MHz
12-to-1 Down Sample in Channelizer
Output Sample Rate; 3.4 MHz/Channel

Wide Dynamic Range Resampler

16 Channels

Interpolate
Filter
2

3.4
MHz

...

ADC

3.4
MHz

3.4
MHz

40.8
MHz

40.8
MHz

81.6
MHz

Circ ular Buffer

163.2
MHz

160
MHz

Spectra from 24-channel Channelizer

at 3.4 MHz Sample Rate
0

-50

-50

-50

-50

-100

-100

-150

-100

-150
-1.5

-1

-0.5

0.5

1.5

-100

-150
-1

-150

-1

-50

-50

-50

-50

-100

-100

-150

-100

-150
-1.5

-1

-0.5

0.5

1.5

-1

-50

-50

-50

-50

-100

-100

-100

-150

-150

-150

-150

-1.5

-1

-0.5

0.5

-1

1.5

-1

-50

-50

-50

-50

-100

-100

-100

-150

-150

-150

-150

-1

-0.5

0.5

1.5

-1

-1

-50

-50

-50

-50

-100

-100

-150

-100

-150
-1.5

-1

-0.5

0.5

1.5

-1

-50

-50

-50

-50

-100

-150
-1

-0.5

0.5

1.5

-1

-1

-1

-150

-150
-1

-100

-100

-150
-1.5

-1

-150

0
-100

-100

-150
-1

-100
-1.5

-1
0

-100

-150

-100

-150
-1

-1

-1

Time Series from 24-channel

Channelizer at 3.4 MHz Sample Rate
2

-2
0
2

200

-2
0
2

200

-2
0
2

200

-2
0
2

-2
0
2

200

-2
0
2

200

-2
0
2

200

-2
0
2

-2
0
2

200

-2
0
2

200

-2
0
2

200

-2
0
2

-2
0
2

200

-2
0
2

200

-2
0
2

200

-2
0
2

-2
0
2

200

-2
0
2

200

-2
0
2

200

-2
0
2

-2
0

200

-2
0

200

-2
0

200

-2
0

50

50

50

50

50

50

100

100

100

100

100

100

150

150

150

150

150

150

50

50

50

50

50

50

100

100

100

100

100

100

150

150

150

150

150

150

50

50

50

50

50

50

100

100

100

100

100

100

150

150

150

150

150

150

50

100

150

200

50

100

150

200

50

100

150

200

50

100

150

200

50

100

150

200

50

100

150

200

Equipment Bay: 192-Stereo FM Modulators

Conversation with Client!

y How big a room will we need to house the DSP

version of this Transceiver?

y Answer: I think it will fit on one chip.
y Response: Dont be Absurd, You Cant Pack a
Room into a Single Chip!
y Results: 48-Analog Devices Blackfin Processors
to Demodulate 192 MP3 Stereo Channels.
y 1 Virtex V-4 for 192 Digital Stereo FM
Modulators and 256 Channel Channelizer @ 293
kHz Bandwidth per channel. (60% of Chip)

Prototype Analog Stereo FM Modulator

LPF
14 kHz
3.2 MHz
Right
L- R

d bx
Enc ode

50- sec
Pre-emph

VCO
32 kHz

LPF
14 kHz
3.2 MHz

SCA

d bx
Enc ode

50- sec
Pre-emph

LPF
7.5 kHz

100

..

75- sec
Pre-emph

L+ R

BPF
15-50 kHz

IF
Output

40

..

Left

VCO
80 kHz

BPF
60-90 kHz

DSP Based Stereo FM Modulator

Left

Right

(L+ R)

(L-R)

dbx
Enc oder

IIR

IIR

75-usec
Pre-em ph

LPF
14-kHz

IIR
50-usec
Pre-emph

LPF
14-kHz

IIR

IIR

50-usec
Pre-em ph

LPF
14-kHz

48-to-293
Arb itrary
Re-Sam ple

IIR
48-to-293
Arb itrary
Re-Sample
KACC

SCA

dbx
Enc oder

48-to-293
Arb itrary
Re-Sam ple
KACC

Satellite Cloc k Dom ain

CORDIC
DDS FM-MOD
&
Up-Converter
32 kHz
CORDIC
DDS FM-MOD
&
Up-Converter

IIR

Gain

BPF
35-kHz
Gain
IIR
BPF
30-kHz

32 kHz
Transceiver Cloc k Dom ain

Gain

Gain

128 Pa th Polypha se Filte r

11-Tap s Pe r Path

75.008 Mhz

225.024 Mhz
1-to-3
Up-Sa mp le

128 Pa th Polypha se Filte r

11-Tap s Pe r Path

256 Channels Ad de r

Ha lf Ba nd Pha se Shift

Even
Samp les

128 Po int FFT

Od d
Samp les

128 Po int FFT

256 Channels 1:2 Up _Samp ler

256 Channel Channelizer for 50-MHz

Digital IF Sampled at 225.024 MHz

Radix-2 Butterfly of two 128-Point FFTs

DDS
50 MHz

Qua ntize

DAC

225.024 Mhz

New Directions
in
Channelized Receivers

M-Channel Channelizer
Resampled M-Path Narrowband Filter
with Rotators Replaced by M-Point IFFT

Armstrong to Tuned RF with Alias

Down Conversion to Polyphase Receiver
-j n
e k

Digital
Low-Pass

Digital
Band-Pass

H(Z)

H(Ze
M-to-1

-j k

)
M-to-1

M-Path Digital -j 2 rk
e M
Polyphase

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!

Channelizer Parameters
Center frequencies, hence channel spacing, and

the number of paths in filter partition

defined by length M of IFFT.
Channel bandwidth and spectral characteristics,
in-band ripple, out-of band attenuation,
and transition BW defined by prototype
low-pass filter in polyphase partition.
Channelizer output sample rate determined by
input commutator span of P inputs per Mpoint IFFT output.
Three Parameters are independent and adjustable.

Two Channelizer BW Options

Channelizer for High Quality Spectrum Analyzer
Crossover
Transition BW
BW

f
Channel Spac ing
Channelizer for High Quality FDM Receiver
Transition BW
Channel BW

f
Channel Spac ing

Overlapped Channel BW and

Output Sample Rate Options

The Winner!

InterestingVariationsofChannelizer
Fast Channelizer:
Time Series and Spectrum From Same Channelizer
Fast Channelizer:
Spectrum and Time Series from same Channelizer
Variable Bandwidth Filter:
Fast Convolution, Efficient, Low Workload
Multiple Bandwidth Channelizer
Arbitrary Channel Spacing Channelizer

Block Diagram of Parallel Processor Spectral

Sniffer Steered Digital Drop Receiver
720 Channel Channelizer
720
Path
PolyPhase
Filter

fS=90 MHz

8192 Bin Spec trum Analyzer

8192
Path

fBW=11 kHz
fS =11 kHz

32,768
Point
4-Fold
Folded
Window

8192
Point

16-Sets:
Channel
Phase
Rotator
Vectors

Spectral
Mask &
Channel
Selec t
Log 10

IFFT

.2

| |

Ensemble
Average

Channel
Processing
and
Multiplexing

fBW=125kHz
fS=500kHz

16 Digital Drop Receivers, Brute Force

1440 Ta p

fs= 500 kHz

Low Pa ss
1

2
2
fc

180:1 fBW= 125 kHz

DDS
1440 Ta p

fs= 500 kHz

Low Pa ss
2

2
2

180:1

fBW= 125 kHz

DDS

fc

16-Copies of
Same Filter

..........

..........

fs= 90 MHz

1440 Ta p

fs= 500 kHz

Low Pa ss
16

2
2
fc

180:1

fBW= 125 kHz

DDS

Resample in Single Polyphase Filter, use

Rotators to Extract 16 Specific Aliases
720-p a th
720
Polypha se
Comp le x
Coe ffic ients
720
Ro ta tors

Samp le
Data
Buffer
2

1
720

179
180

359

1 Polyphase
Filter
Down
Samples
And Services
all Channels

360

539
540
719

Ro ta tors
2

720

Ro ta tors

......

180-to -1

1440 Ta p s

fs= 90 MHz

16-Sets of
Complex Rotators
720 Mult & Add
per Output

2
720

Ro ta tors
16

46,080 Multiplies
At 500 kHz Rate

Replace 16-Sets of Rotators with

Single 720 Point IFFT
720-p ath
Polypha se
Coe ffic ients
720

Samp le
Data
Buffer

1
720

179
180

2
720

359
360

1440 Ta p s

180-to -1

3
720
Point
IFFT

539
540
719

......

fs= 90 MHz

A 720 point IFFT,

Prime Factors: 5,8,9
Implemented with
Winograd Transform
Workload is
2,400 Multiplies.
5.2% of workload to
compute 16 Outputs

720
720

Three Options for Digital Drop Receiver

16 Channel Channelizer
16 Sets
1440 Tap
Low Pa ss
2

2
2
DDS

fc

58,000 Multiplies at
500 kHz Rate

16 Channel Channelizer
720
Path
PolyPhase
Filter

16-Sets:
Channel
Phase
Rotator
Vec tors

49,000 Multiplies at
500 kHz Rate

720 Channel Channelizer

720
Path

720
Point

PolyPhase
Filter

IFFT

5,300 Multiplies at
500 kHz Rate

Spectrum Analyzer:
Polyphase Partition of Overlapped Window
8,192 Point IFFT
160,000 Multiplies
Per Transform
At 11-KHz Rate

y(nM,0)

H0( Z )

y(nM,1)

H1( Z )
x(n)

y(nM,2)

H2( Z )

M-Point
IFFT

....

....

y(nM,k)

32,768 Point Window

65,000 Multiplies
Per Transform
At 11-KHz Rate

y(nM,M-2)

HM-2( Z )

y(nM,M-1)

HM-1( Z )

Window and IFFT

193,000 Multiplies
Per Transform
At 11-KHz Rate

We have accounted for the Two Major Blocks:

Spectrum Analyzer and Channelizer!
720 Channel Channelizer
720
Path

fS=90 MHz

PolyPhase
Filter

8192 Bin Spec trum Analyzer

8192
Path

fBW=11 kHz
fS =11 kHz

32,768
Point
4-Fold
Folded
Window

8192
Point

16-Sets:
Channel
Phase
Rotator
Vectors

Spectral
Mask &
Channel
Selec t
Log 10

IFFT

.2

| |

Ensemble
Average

Channel
Processing
and
Multiplexing

fBW =125 kHz

fS =500 kHz

Block Diagram of Cascade Channelizer and

Spectrum Analyzers
720 Channel Channelizer
720
Path

720
Point

PolyPhase
Filter

IFFT

Channel
Processing
and
Multiplexing

Spectral
Mask &
Channel
Selec t
Windo w

IFFT

Log 10

180 45-Bin
Spectrum Analyzers
Windo w

IFFT

Windo w

IFFT

Windo w

IFFT

.2

| |

Ensemble
Average

M-to-2 Down-Sampled
M-path Polyphase Channelizer

M-1

State Eng ine

........

M/2
M/2+ 1

M-Point IFFT

M/2-1

Circular Output Buffer

....

M-Path Polyphase Filter

M-Path Input Data Buffer

FDM

M-2
M-1

TDM

Spectrum and Zoom Detail of Input Test Signal

Course Spectrum and Zoom Detail: Power Output from

180 Channelizer Filters

Time Series from 60 Channels of 180 Path Input

Channelizer Simulation

Spectrum from 60 Channels of 180 Path Input Channelizer Simulation

Power Spectra from Selected Channels

Closing Comments
(on this Topic)
y Full Channelizers are Surprisingly

Efficient and Inexpensive

y Dont Waste Bandwidth Reduction and
Sample Rate Reduction Offered by
Channelizer
y Perform Spectrum Analysis at Output of
Channelizer Rather than at Input
y Noise Figure Improvement due to
Spectrum Analysis of Decoupled Channels