Professional Documents
Culture Documents
0
8
arccos
2
1
2
0
0
H A
t y
f
filt
( )
s
s
s
t
f
f
f
= =
|
\
|
= =
1
2
2
1
2
1
max 0
=
= =
(
|
|
\
|
= =
0
0
0
1
0
0
0 0 0
0
0
2 cos 2 2 cos 2
2
, 2 cos 2 cos
( ) ( ) ( )
( ) ( ) [ ]
( ) ( ) ( ) [ ] ( ) { } ( ) ( ) 1 1 2 cos 1 2 cos 2
2 4 cos 2 cos 2
1
0 0
0
0 0 0
1 2
2
2
0
+ + +
+ =
=
n n n n F n n F
D
A
F n F F
D
A
n y n y n y
q q q q
N
N
q q mix
Slide 16 CSIR 2006 www.csir.co.za
Step 3: Low-Pass Filtering
Finite Impulse Response (FIR) digital filter
where
( ) ( )
( ) ( ) ( ) ( ) ( ) [ ] ( ) 2 2 cos 2 cos 0 2
0 0 0 0 0
1 2
2
2
0
0
n F H F n F F H F H
D
A
k n y c n y
q LPF LPF LPF
N
N
k
mix k filt
+
+
+ =
=
4
, 2 1
4
, 2
0 0 0
0 0 0
s
s
f
f F F
f
f F F
> =
\
|
0 2
arccos
2
1 2
1 2 2
0
2
LPF
N
filt
N
out
H A
D n y
n y
out
+
+
>>
+
= =
( ) ( ) ( ) ( ) ( ) ( ) n y 0
2
q
= = =
n y n y n y n y n y
q q q q mix
( ) ( ) ( ) ( ) ( ) [ ] ( ) n F H n F F H H
D
A
n y
q LPF LPF LPF
N
filt
+ + + =
0 0 0
1 2
2
2
0
2 cos 0 2
Slide 22 CSIR 2006 www.csir.co.za
Amplitude Insensitive DIFM
Divide basic DIFM filter output with amplitude estimation
Inverse cosine lookup table yield frequency estimation
Advantages
Amplitude estimation exactly aligned with frequency estimation
No external calibration or alignment required
Time-domain multiplex hardware
( ) ( )
( )
( )
( ) ( ) ( )
( )
0
0 0 0
0
0
2 cos
2 2 cos
0
2 cos
F
n F H F n F
H
F H
F n y
q LPF
LPF
LPF
div
+
( )
s
lt
f
f
F n y
0
0
2 2 =
Example Implementation
Slide 24 CSIR 2006 www.csir.co.za
SWIFT500 Digital DRFM Module with
Built-In Amplitude Insensitive DIFM
1.2 GSPS , 500 MHz IBW
16x demultiplexing
Stratix 1S30 with 96 9x9 multipliers
Slide 25 CSIR 2006 www.csir.co.za
SWIFT500 Digital DRFM Module with
Built-In Amplitude Insensitive DIFM
y
q
(n)
y
q
(n-1)
y
q
(n-15)
y
q
(n-2)
InputCast0
InputCast1
InputCast2
InputCast15
UnitDelay0
UnitDelay1
UnitDelay2
UnitDelay15
UnitDelay16
UnitDelay17
UnitDelay18
UnitDelay24
y'
q
(n-32)
y'
q
(n-33)
y'
q
(n-40)
y'
q
(n-31)
y'
q
(n-18)
y'
q
(n-17)
y'
q
(n-16)
y'
q
(n-2)
y'
q
(n-1)
y'
q
(n)
y'
q
(n-15)
U
n
i
t
D
e
l
a
y
8
MuxMult24
14
c
0
22
c
1
38
c
2
57
c
3
81
c
4
109
c
5
138
c
6
168
c
7
196
c
8
221
c
9
240
c
10
252
c
11
255
c
12
252
c
13
240
c
14
221
c
15
196
c
16
168
c
17
138
c
18
109
c
19
81
c
20
57
c
21
38
c
22
22
c
23
14
c
24
y
' q
(
n
-
2
4
)
y
'
q
(
n
-
2
5
)
Flip-Flop
MuxMult1
MuxMult0
Adder
DeMux1
Select
0 1
DeMux2
Select
0 1
UnitDelay25
UnitDelay26
Inverse
Lookup
Table
Multiply
Arccos
Lookup
Table
y
out
y
1mm
y
2mm
y
24mm
y
a
UnitDelay25
y
dm1
y
dm2
y
m
Mux#A
Mux#B
Select
0
1
Select
0
1 y'
q
(n-#)
y'
q
(n-#-1)
y'
q
(n-#-16)
Select In
Multiply
#A
Multiply
#B
Coeff In #
Y
#mm
(m+d
1
)
Slide 26 CSIR 2006 www.csir.co.za
SWIFT500 Digital DRFM Module with
Built-In Amplitude Insensitive DIFM
Key specifications
9-bit multiplication
24th order low-pass FIR filter with Chebyshev windowing
Cut-off frequency of 100 MHz and 48 dB side-lobe suppression
Frequency response 50 MHz to 550 MHz
Time-multiplexed resources to estimate amplitude and frequency
Division implemented in a two-step process
Inversion of denominator using lookup table (12-bit x 12-bit)
Multiplication of numerator with inversed denominator
12-bit by 10-bit inverse cosine lookup table
Simulation Results
Slide 28 CSIR 2006 www.csir.co.za
Monochromatic Input Signal With Additive
Coloured Noise
Slide 29 CSIR 2006 www.csir.co.za
Analysis of DIFM Accuracy
Slide 30 CSIR 2006 www.csir.co.za
Key Performance Specifications
High signal-to-noise ratios
Mean deviation less than 2 MHz
Absolute error less than 6 MHz across bandwidth
Absolute error less than 2 MHz in > 300 MHz bandwidth
RMS error less than 3 MHz across bandwidth
RMS error less than 1 MHz in > 300 MHz bandwidth
Low signal-to-noise ratios
Bias in frequency estimation
Due to bias in amplitude estimation
Reduced by implementing higher order FIR filter (longer latency)
Latency (processing time)
13 FGPA clock cycles (173.33 ns)
Throughput rate
2 FGPA clock cycles (37.5 MHz)
Experimental Verification
Slide 32 CSIR 2006 www.csir.co.za
Quantitative Laboratory Experiments
Conclusions
Slide 34 CSIR 2006 www.csir.co.za
Conclusions
Viable, shared aperture, frequency estimation technique
Implemented efficiently in current commercial hardware
Results comparable to existing analog techniques
Flexibility and ability to be optimised for the specific
requirements
Real-time changing the filter coefficients
Insensitive to temperature
Does not require periodic calibration to maintain accuracy
Operationally superior to its analog counterparts
South African provisional patent application 2006/00946,
2006-02-01