AN-642: Coupling a Single-Ended Clock Source to the Differential Clock Input of

AN-642
APPLICATION NOTE
One Technology Way • P.O. Box 9106 • Norwood, MA 02062-9106 • Tel
T : 781/329-4700 • Fax: 781/326-8703 • www.analog.com
Coupling a Single-Ended Clock Source to the Differential Clock Input
of Third-Generation TxDAC® and TxDAC+® Products
By Doug Mercer, Steve Reine, and David Carr
INTRODUCTION
The third-generationTxDAC andTxDAC+ families* feature
a flexible differential clock input. The differential clock
inputs CLK+ and CLK– can be driven from a variety of singleended and differential clock sources.Transformer coupling
is useful in many single-ended-to-differential applications.
However, a magnetically coupled transformer may not be
convenient in some situations. A single-ended clock signal
may be coupled to the clock inputs in a variety of ways.
The CLK+ input can be driven while a suitable dc threshold
voltage is applied to CLK– as shown in Figure 1. Resistors R1
and R2, along with capacitor C3, generate a dc level equal
to CLKVDD/2 for the CLK– terminal. The clock source must
be unipolar and swing nearly rail-to-rail. Although simple,
this configuration does not provide the proper termination
impedance for the source and has almost no immunity to
noise on either the power supply or ground.
CLKVDD
RSERIES
CLK+
R1
1k
50
SOURCE
C3
0.1F
R2
1k
common side of the coupling capacitor C2 is connected
could adversely inject noise onto the CLK– input. Ideally,
it should be connected at the same common point as the
clock source (see Figure 2).
R2
1k
RTERM
R4
1k
CLK–
C2
1nF
CLKCOM
Figure 2. Resistive Matching Interface
The two 1 nF input coupling capacitors, C1 and C2, set the
high-pass corner frequency of the network at approximately
5 MHz. The high-pass corner frequency can be set higher
or lower according to the equation
f3dB =
where C =
REV. 0
R3
1k
50
C3
0.1F
CLKCOM
A somewhat better alternative is a simple broadband
resistive matching network (Figure 2). As shown in Figure 1,
resistors R1 and R2 and Capacitor C3 create the dc bias
point at CLKVDD/2. Resistors R3 and R4 provide dc bias for
the differential inputs CLK+ and CLK–. A termination resistor,
RTERM (50 ) provides the required source termination, while
capacitors C1 and C2 provide dc blocking.The termination
resistor should preferably be placed directly across the
input pins, CLK+ and CLK–, where it serves to lower the
possible deleterious effects of dc offset voltages for smaller
clock signal amplitudes. At low frequencies, this may not
be quite as attractive since it necessitates the use of larger
coupling capacitors. Unlike the circuit in Figure 1, the input
source can be bipolar, and there is a least some rejection of
supply and ground noise by virtue of the large ratio of R3
and R4 to the 50  RTERM. However, the way in which the
CLK+
R1
1k
50
SOURCE
CLK–
Figure 1. Single-Ended Interface
CLKVDD
C1
1nF
C1 × C 2
C1 + C 2
1
2π × C × 50
NARROW-BAND MATCHING
Transformer coupling is useful in broadband applications.
However, often in converter applications, the clock is at a
single fixed frequency. At high clock frequencies or if the
clock source is remote from the converters where it might
pick up interference along its path, it is often preferable to
use a narrow-band matching network. A narrow-band LC
match can be implemented either as a series-inductance/
shunt-capacitance or as a series-capacitance/shuntinductance. However, the concurrent requirement that
the clock inputs, CLK+ and CLK–, be ac-coupled makes
*AD9740ACP, AD9742ACP, AD9744ACP, AD9748ACP,
AD9751, AD9753, AD9755, AD9772, AD9773, AD9775,
AD9777
AN-642
a series-capacitance/shunt-inductance type match more
appropriate (see Figure 3).
The resonant frequency is defined by the equation
L2 =
R1
1k
R3
1k
C3
0.1F
R2
1k
R4
1k
When
f0 =
CLKCOM
L1
C2
L1 =
2π × f0
R S × R IN
R S × R IN
2π × f0
For f0 = 100 MHz, RS = 50 , RIN = 1 k, L1 = 356 nH.
Because L1 and L2 are in parallel, they can be combined
to give the final value for LMATCH
L MATCH =
L1 × L 2
L1 + L 2
with L1 = 356 nH and L2 = 1.01 H, LMATCH = 263 nH.
C1 and C2 can be chosen in a number of ways. First, C2 can
be set to a large value, such as 1000 pF, so that it appears
as an RF short. C1 would then be set equal to the calculated value of CMATCH. Alternatively, C1 and C2 can each
be set to twice CMATCH so that the total series capacitance
is equal to CMATCH. By making C1 and C2 slightly unequal
(i.e., select C2 to be about 10% less than C1) but keeping
their series value the same, the amplitude of the signals
on CLK+ and CLK– can be equalized so that the differential
input is driven in a more balanced manner. Any one of the
three options detailed above can be used as long as the
combined series value of C1 and C2 (i.e., (C1  C2)/(C1 + C2))
is equal to CMATCH.
CLK+
CIN
L2
1
Next solving for L1 gives:
RIN
RIN
2π L1 × C MATCH
For f0 = 100 MHz, RS = 50 , RIN = 1 k, CMATCH = 7.12 pF
Typically, the clock inputs will need to be matched to
50 . To make the matching process simpler, the input
capacitance of the CLK+ and CLK– inputs, CIN, can be temporarily removed from the calculation by adding a virtual
shunt inductor (L2, see Figure 4), which will resonate with
CIN at the desired frequency. This inductor will be factored
into the determination of the matching inductor LMATCH later
(Figure 3). This allows the main calculation to be based
on a simple resistive-to-resistive match between the 50 
source and the 1 k dc bias resistors.
50
SOURCE
1
C MATCH =
A NARROW-BAND LC MATCHING EXAMPLE AT 100 MHz
While numerous software programs are available that
allow the values of matching components to be easily calculated, a clear understanding of the calculations
involved is valuable. A frequency value of 100 MHz has
been used for this exercise because of the effects of PC
board layout parasitics at higher frequencies. RF layout
simulation software is useful when board design at higher
frequencies is required.
1k
L1
C MATCH
Solving for CMATCH gives:
Figure 3. LC Matching Interface
C1
R S × R IN =
The input will look resistive at the frequency given by
CLK–
C2
AC GROUND
(2π × f0 )2 × C IN
With CIN temporarily out of the way, the calculation to match
the 50  source (RS) to the 1 k dc bias resistor (RIN) and
calculating values for CMATCH and L1 can take place.
CIN
LMATCH
1
For a center frequency of 100 MHz and CIN = 2.5 pF,
L2 = 1.01 H.
CLK+
50
SOURCE
L2 × C IN
Therefore
CLKVDD
C1
1
2π × f0 =
There are several advantages to a band-pass network such
as this. Some amount of voltage gain can be achieved because
the input impedance of the differential inputs is high and
almost purely capacitive.This can provide increased noise
immunity when lower amplitude clock signals are all that
is available. The component count is low: two capacitors
and one inexpensive chip inductor in addition to the components needed to create the dc bias.
CLK–
1k
AC GROUND
Figure 4. Input Matching Example
–2–
REV. 0
AN-642
In all cases, the values of CMATCH (C1, C2) and LMATCH must be
chosen from standard values.The closest standard values
are C1 = 16 pF, C2 = 15 pF, and LMATCH = 270 nH (shown in
Table I). At this point, these values now need to be installed
and measured for performance at 100 MHz. Because of
board and layout parasitics, the component values from
the above example may need to be adjusted.
Table I shows the recommended values for the inductor
and capacitors in Figure 3 for some selected RF frequencies. As previously discussed, a modification of the board
layout will produce networks that may not perform as
specified.
Further, by making these capacitors unequal, the amplitudes at CLK+ and CLK– may be equalized when driving
from a single-sided source; that is, the network also serves
as a balun. Figure 5 shows the response for a center frequency of 100 MHz; note the very high attenuation at low
frequencies. The high frequency attenuation is due to the
input capacitance of the clock input pins. The amplitude
response is plotted differentially across CLK+ and CLK– as
well as single ended for each input with respect to common. Due to the resonant nature of the matching network,
there will be a phase shift of approximately 90 at the CLK+
and CLK– inputs. This is plotted in Figure 6. Care must be
taken to account for this phase difference when adjusting
the converter data setup and hold timing.
Frequency
(MHz)
LMATCH
(nH)
C1
(pF)
C2
(pF)
61.44
470
24
22
65.00
470
24
18
76.80
390
18
16
78.00
390
18
15
78.64
390
16
16
92.16
330
13
12
100.00
270
16
15
122.88
220
11
10
130.00
180
12
11
153.60
150
10
9
156.00
150
10
8
0
157.29
120
12
12
–5
184.32
120
8
7
245.76
82
6
5
260.00
56
9
8
307.20
47
7
6
312.00
47
6
6
314.57
47
6
6
368.64
39
5
4
491.52
22
5
4
520.00
18
6
5
614.40
15
4
4
624.00
15
4
3
629.15
15
4
3
737.28
12
2
2
15
10
Table I. Recommended Values for C1, C2, and LMATCH in Figure 4
DIFFERENTIAL
RESPONSE
VIN
MAGNITUDE (dB)
5
–10
–15
–20
CLK+
SINGLE-ENDED
–25
–30
–35
CLK–
SINGLE-ENDED
–40
–45
–50
–55
10M
100M
FREQUENCY (Hz)
1G
Figure 5. Amplitude Response of 100 MHz
Matching Network
175
150
125
100
DIFFERENTIAL
RESPONSE
CLK+
SINGLE-ENDED
MAGNITUDE (dB)
75
50
25
0
APPENDIX 1
Modeling and simulating this interface network with Spice
is a useful alternative to the hand calculations shown here.
It also allows the inclusion of more parasitic effects from
such things as the package and PC board traces.The listing
that follows is a Spice level subcircuit model for the clock
input structure. Referring to the schematics in Figures 7
and 8 and in the top level of the listing, inductor L1 and
capacitors C1 and C2 determine the center frequency of the
network. The values listed are for a frequency of 100 MHz.
Resistor R1 is the driving source (V3) impedance and resistors
R2 and R3 provide the dc bias (set by voltage source V4)
for the ac-coupled inputs.
–25
–50
–75
CLK–
SINGLE-ENDED
–100
–125
–150
–175
–200
10M
100M
FREQUENCY (Hz)
1G
Figure 6. Phase Response of 100 MHz Matching
Network
REV. 0
–3–
AN-642
D1 C1_B CLK_VDD DP1 1
D2 CLK_COM C1_B DN2 1
D3 C2_B CLK_VDD DP3 1
D4 CLK_COM C4_B DN2 1
D5 C3_B CLK_VDD DP3 1
D6 C4_B CLK_VDD DP1 1
L1 L1_A CLK_P 2.1n
L2 L2_A CLK_N 2.1n
R1 C2_B C1_B 193
R2 C1_B L1_A 0.0517
R3 C4_B L2_A 0.0517
R4 C3_B C4_B 193
.ENDS CLK_INPUT
CLK_VDD
CLK_P
V3
0 AC 2
L1
270N
R2
1K
CLK_COM
R3
1K
XIPT
CLK_N
C7
5P
V4
1.65
V1
3.3
V2
0
Figure 7. SPICE Model for Input Network
CLK_VDD
CLK_P
L1
2.1N
C1
0.75P
R2
0.0517
CLK_COM
CLK_N
D1
D2
D3
C2
9F
R1
193
.model DP1 D (bv=5.5 cjo=1.17088p eg=1.106 fc=500m
ibv=608.2p is=1.299342f m=632.669m)
C3
18F
L2
2.1N
R3
0.0517
C4
0.75P
D4
.model DP3 D (bv=5.5 cjo=325.2446f eg=1.106 fc=500m
ibv=608.2p is=3.609284e-16 m=632.669m)
R4
193
D6
D5
E03631–0–4/03(0)
C6
16P
R1
50
C5
9F
.model DN2 D (bv=8.0 ibv=1.54587E-06 cjo=1.411p
m=0.3675268 is=1.759f eg=1.140)
Figure 8. SPICE Subcircuit for XlPT Block
.AC DEC 400 10E6 1E9
.PROBE
.OP
.END
LISTING 1
Clock Input Matching Network Model
*
APPENDIX 2
Spice models for chip capacitors and inductors are available from manufacturers. Figure 9 is a common one used
for inductors; Figure 10 is a common subcircuit used in
model capacitors.
V1 CLK_VDD 0 3.3
V2 CLK_COM 0 0
V3 SRC_P 0 0 AC 2
V4 CML 0 1.65
R1 SRC_P CAP_P 50
R2 CLK_P CML 1K
R3 CLK_N CML 1K
C1 CAP_P CLK_P 16p
C2 CLK_N 0 15p
L1 CLK_P CLK_N 270n
XXIPT CLK_P CLK_N CLK_COM CLK_VDD CLK_INPUT
C
R1
L
R2
RVAR
Figure 9.
Values for these models can be obtained from manufacturers’ websites, such as www.coilcraft.com. RVAR in Figure 9
is frequency dependent and relates to the skin effect and
other inductor losses.
.SUBCKT CLK_INPUT CLK_P CLK_N CLK_COM CLK_VDD
C1 CLK_VDD C1_B 0.75p
C2 CLK_VDD C2_B 9f
C3 C2_B C3_B 18f
C4 CLK_VDD C4_B 0.75p
C5 CLK_VDD C3_B 9f
RP
RS
LS
C
.
Figure 10.
© 2003 Analog Devices, Inc. All rights reserved. Trademarks and registered trademarks are the property of their respective companies.
–4–
REV. 0