Application Note AN-8
Copyright November, 1999
Ultra-Wideband Differential Measurements Using PSPL BALUNs
James R. Andrews, Ph.D., IEEE Fellow
INTRODUCTION
Almost all high-performance test instruments, including
sampling oscilloscopes and vector network analyzers, are
single-ended, ground-referenced, 50 Ohm instruments.
There are now a growing number of electronic applications,
both digital and analog, that are using differential circuit
techniques and also balanced transmission lines. The test
instrument manufacturers have been slow to recognize this
trend. As a result, many engineers are being tasked to design
new differential circuits without the benefit of having proper
test instruments. PSPL offers ultra-wideband BALUNs which
permit differential measurements to be made over extremely
wide frequencies extending from a few kHz to many GHz.
The Model 5320A covers from 5 kHz to 11 GHz while the
Model 5315 covers from 200 kHz to 17 GHz. The Model
5320A is the most popular.
the balanced secondary consists of two identical windings
that are connected as a center-taped secondary. The center
tap is usually then connected to the common ground. Coax
connectors might now be used for all three terminals. Note
that the black dots are polarity indicators for the various
transformer windings. With the arrangement shown in Figure
2, one of the secondary outputs is in-phase with the input
and is thus labeled as the (+), or non-inverting output. The
other secondary output is out-of-phase with the input and
is thus labeled as the (-), or inverting output. There are now
three output impedances to be considered. R+ and R- are
the impedances referenced to ground seen looking into the
(+) and (-) outputs respectively. There is also a differential
impedance, Rdiff, which is the impedance seen between the
two center pins of the (+) and (-) output coax connectors.
The impedance transformation is still determined by the turns
Figure 1: BALUN Transformer
Figure 2: BALUN Transformer with Center-Tapped
Output
Historically, when engineers needed to do differential
measurements with conventional test instruments, they used
a BALanced to UNbalanced transformer, or BALUN for short.
Figure 1 shows the typical schematic diagram for a BALUN.
It consists of a simple transformer with one wire of the
primary winding being the ground terminal for the unbalanced
side. The balanced secondary winding is not connected to
the ground terminal and is thus considered to be floating
with respect to ground. Impedance transformation is also
possible if the number of wire turns on the primary and
secondary are unequal. The impedance transformation is
equal to the square of the turns ratio, N. An example of a
BALUN everyone is familiar with is the antenna transformer
supplied with every TV receiver. It is used to match 300 Ohm
flat ribbon lead to 75 Ohm coax cable. For a 300 Ohm / 75
Ohm transformation, a 2 : 1 turns ratio is required.
ratio, N, of the secondary and primary windings.
The BALUN designs shown in Figures 1 and 2 are very widely
used and are available from many manufacturers. They can
be used for Balanced to Unbalanced transformations and
also to shift impedance levels by altering the turns ratio, N.
The major limitation of these designs is bandwidth. They
are built using conventional transformer designs and
techniques. The transformer core material, number of wire
turns, etc. are dictated by the desired operating frequency.
It is difficult to design transformers, including BALUNs to
operate over more than one or two decades of bandwidth.
For typical wireless applications, ultrawide bandwidth is
not a requirement. For example, the TV antenna transformer
mentioned earlier only needs to work from 50 MHz to 800
MHz. However, for digital data, ultrawide bandwidth is a
mandatory requirement. Digital data systems, such as
Figure 2 shows another example of a BALUN. In this case SONET, Gigabit Ethernet and Fiber Channel require
PICOSECOND PULSE LABS P.O. BOX 44 BOULDER, CO 80306, USA TEL: 1.303.443.1249 FAX: 1.303.447.2236
WWW.PICOSECOND.COM
AN-3048, Revision 1, 11/99
Page 1 of 5
Application Note AN-8
Copyright November, 1999
bandwidths extending from a few kHz to many GHz. For 31 ps and the diff. balance between the (+) and (-) outputs
these data applications, the BALUN designs of Figures 1 was < 0.2 dB for times < 7 ns. Figure 4 shows the pulse
and 2 are unsatisfactory.
transmission response at 20 ps/div. of the 5320A BALUN.
The input test signal was a 10 ps risetime step function.
Figure 3: PSPL Ultra-WideBand BALUN
PSPL BALUN DESIGN
PSPL has developed an ultra-wideband BALUN design
which works over many decades of bandwidth. The PSPL
BALUN design is shown in Figure 3. It consists of a 50 Ohm,
6 dB resistive power divider, a 50 Ohm coaxial inverting 1:1
transformer, and a length of 50 Ohm coax cable. The input
signal is split into two identical signals by the 6 dB power
divider. One of these signals is then inverted (180 degree
phase shift) by the 1:1 inverting transformer. The other signal
is sent through a coax cable whose length is chosen to match
the propagation delay time of the 1:1 inverting transformer.
The input impedance is 50 . The output impedances of
both the (+) and (-) coaxial outputs are also 50 Ohm. The
differential output impedance is thus 100 Ohm. The 1:1
inverting transformer is a special PSPL proprietary design
which is a hybrid of coax cable and conventional transformer
designs. This design results in an ultra-wideband transformer
covering 6 decades. The PSPL Model 5320A BALUN has
a -3 dB bandwidth from 5 kHz to 11 GHz. Its risetime is 31
ps. The two outputs are matched in amplitude and phase to
0.2 dB and within less than 2 ps. The major limitations in
this design are the 6 dB insertion loss suffered in the 6 dB
power divider and that the impedance transformation is
limited to 2:1, i.e. 100 Ohm differential output to 50 Ohm
single-ended input. Other impedances such as 150 Ohm/
75 Ohm, would require the use of 75 Ohm coax components.
The Model 5320A BALUN was originally designed in the
mid 80s by PSPL for use in an ultra-wide bandwidth,
traveling-wave tube oscilloscope. It split a 50 Ohm singleended input signal into two 50 Ohm push-pull signals to drive
the scopes CRT 100 balanced deflection plates. The
scope was only used to measure extremely fast risetime,
single transient pulses in a time window of 5 ns or less. The
customers design requirements for the 5320A BALUN were
that it needed to provide the fastest possible risetime, flat
pulse response, and balanced drive only for a 5 ns time
window. It did this extremely well. Its differential risetime was
Figure 4: PSPL Model 5320A BALUN Pulse Response
Input Step has 10 ps Risetime. Time Scale is 20 ps/div.
PSPL still sells the 5320A as a standard catalog item. Today
however, customers are using the 5320A for dramatically
different purposes. The two major uses are for (1) differential
measurements of balanced transmission lines and other
balanced components with a single-ended, 50 Ohm network
analyzer (see Figure 8), and (2) splitting or combining Gigabit
data signals into differential or single-ended signals. For
these purposes, the 5320A performs in a less-than-optimum
manner. Users of the 5320A are cautioned to carefully study
the 5320As limitations. PSPLs future plans are to design
new BALUNs that will be optimized for Gigabit data
applications.
5320A LIMITATIONS
The fundamental limitation of the 5320A is an impedance
mismatch in the internal pulse-inverting transformer. This
mismatch is shown on the TDR plot found on the 5320As
specification sheet. A TDR measurement made looking into
the (-) Inverting Output port shows a -12% reflection
coefficient. Reflections from the transformer are also seen
on the TDR plots for the input port and (+) Non-Inverting
output port. The reflections seen on the input and (+) output
are smaller due to the attenuation introduced by the 6 dB
power divider. An input signal will be partially reflected from
the transformer and will eventually appear later in time on
the (+) output. A step response measurement, similar to
Figure 4, but at a slower sweep speed shows that starting at
t = 7 ns, the (+) output and also the differential output have
a small negative shift in voltage level. For t > 7 ns there is an
imbalance of about 0.5 dB (i.e. -6%) between the (+) and
(-) outputs.
PICOSECOND PULSE LABS P.O. BOX 44 BOULDER, CO 80306, USA TEL: 1.303.443.1249 FAX: 1.303.447.2236
WWW.PICOSECOND.COM
Page 2 of 5
AN-3048, Revision 1, 11/99
Application Note AN-8
Copyright November, 1999
When one single 5320A BALUN is used to split a signal into
separate (+) and (-) signals, the single 7 ns reflection is not
too severe. A 5320A BALUN can also be used to combine
differential signals into a single, unbalanced signal. If the
differential signals come from a well matched source, then
the results are satisfactory. If the source is not well matched
to 50 Ohm, then the results are not satisfactory due to
multiple reflections.
The real problem arises when one uses a pair of 5320A
BALUNs in a back-to-back configuration. Figure 5 shows
the resultant thru signal at a sweep speed of 10 ns/div. for
a pair of 5320As joined with a pair of 6, 50 Ohm coax cables.
This step response is objectionable for intermediate times
from t = 7 ns to about 50 ns. For times less than 7 ns, the
output is very well-behaved. Likewise for times > 50 ns, the
output is very well-behaved. For intermediate times, the
ragged step response (Figure 5) is due to multiple reflections
occurring due to the -12% mismatches of the 1:1 inverting
transformers used in the BALUNs. The intermediate time
zone of 7 ns to 50 ns will also cause severe problems in the
frequency domain between about 150 MHz and 20 MHz (i.e.
f = 1 / T). The problem shown in Figure 5 limits the usefulness
of a pair of 5320As for Gigabit data applications. However,
if additional losses can be tolerated, the multiple reflection
problem can be minimized by adding 6 to 10 dB SMA
attenuators between the two BALUNs.
5320A FREQUENCY RESPONSE
The frequency response of a single 5320A BALUN is shown
on the specification sheet. The insertion loss and return loss
plots appear to be noisy. The noise is, however, an actual
ripple in the frequency response that is due to the 7 ns internal
mismatch reflection. Figures 6 and 7 show the insertion loss
and return loss frequency responses from 10 kHz up to 10
GHz of a pair of 5320A BALUNs connected back-to-back.
The input and output ports were the unbalanced 50 Ohm
input ports of the BALUNs. The 7 ns mismatch problem
discussed earlier is very apparent in these plots. Figure 6b
is a smoothed response which suppresses the mismatch
ripple. The 50 Ohm input and output ports return losses are
-15 dB. The insertion loss of the back-to-back pair is -7 dB
with a -3 dB bandwidth of 6.5 GHz.
Figure 6: High Frequency Response of two PSPL
5320A BALUNs Connected Back-to-Back
Plots on right are smoothed data to show general trends.
Measured on Wiltron 37360A VNA. 1 GHz/div.
Figure 5: Pulse Step Response of two 5320A BALUNs
Connected Back-to-Back. 10 ns/div.
PICOSECOND PULSE LABS P.O. BOX 44 BOULDER, CO 80306, USA TEL: 1.303.443.1249 FAX: 1.303.447.2236
WWW.PICOSECOND.COM
AN-3048, Revision 1, 11/99
Page 3 of 5
Application Note AN-8
Copyright November, 1999
5320A BALUNs, unacceptable results are usually obtained
with a scaler network analyzer. New modern vector network
analyzers can provide acceptable performance with poorer
hardware due to their use of elaborate mathematical
calibration routines which calibrate out the hardware
imperfections. Caution should be exercised when using an
older VNA that does not include an elaborate calibration
scheme. PSPL has a new Wiltron Model 37369A, 40 MHz
to 40 GHz vector network analyzer which uses a 12-term
calibration scheme. This instrument was found to be capable
of calibrating out the mismatch errors of the PSPL 5320A
BALUNs and providing good differential measurements.
Figure 7: Low Frequency Response of two 5320A
BALUNs Connected Back-to-Back
Log sweep 10 kHz to 200 MHz. Ripple noted above 10
MHz is due to 7 ns mismatch. Measured on HP 3577A
network analyzer
DIFFERENTIAL NETWORK ANALYZER
PSPLs Model 5320A BALUNs can be used successfully
with modern vector network analyzers to perform differential
measurements. See Figure 8. We do not recommend that
they be used with scaler network analyzers. The accuracies
of scaler network analyzers are strongly dependent upon
the quality of the hardware used and in particular are strongly
influenced by the impedance mismatch of the source and
loads. Due to the poor mismatch of a back-to-back pair of
Figure 9 shows the results of a Wiltron 37369A differential
measurement system calibration up to 10 GHz. The S21 and
S11 scale factors are 0.2 dB/div and 10dB/div respectively.
The noise levels are < 0.05 dB and -50 dB respectively. To
perform the differential calibrations at the 100 differential
test ports, Figure 8, we used SMA calibration standards for
opens, shorts and broadband terminations. Two of each
standard were required to simultaneously terminate each of
the BALUNs (+) and (-) ports. To demonstrate the
usefulness, we measured two differential cables. The first
test was a differential cable made up of two identical PSPL
Model 5915-59, 50 Ohm, SMA semi-flex coaxial cables.
Figure 10 shows the S21 and S11 of this very high quality, 100
Ohm differential cable up to 10 GHz. The second differential
cable tested was a typical, low-cost cable commonly used
for differential interconnects. It was a 20-foot length of
shielded, twisted pair of #24 wires. Four SMA connectors
were attached to this cable to connect it to the 5320A
BALUNs. Figure 11 shows the measured differential insertion
loss, group delay, return loss and VSWR of this low-cost
cable up to 1 GHz.
Figure 8: PSPL/Wiltron, 100 Ohm Differential, 10 GHz
Vector Network Analyzer, Test Set
PICOSECOND PULSE LABS P.O. BOX 44 BOULDER, CO 80306, USA TEL: 1.303.443.1249 FAX: 1.303.447.2236
WWW.PICOSECOND.COM
Page 4 of 5
AN-3048, Revision 1, 11/99
Application Note AN-8
Copyright November, 1999
Figure 9: System Calibration Results for PSPL 100
Ohm Differential 10 GHz VNA
Linear sweep from 40 MHz to 10 GHz. Insertion loss, S21
scale is 0.2 dB/div. Return loss, S11 scale is 10 dB/div
Figure 10: Differential Measurement of S21 and S11
Differential 10 GHz VNA on high quality, 100 Ohm
Differential Coaxial Cable
Cable consisted of an identical pair of PSPL Model 591550, 50 Ohm semi-flex cables. Insertion loss, S21 scale is
0.5 dB/div. Return loss, S11 scale is 10 dB/div. Linear
sweep from 40 MHz to 10 GHz
Figure 11: Differential Measurements of a Shielded #24 Twisted Wire Pair Cable
Cable length was 20 feet. Plots are of S11 VSWR (0.5 /div), S12 insertion loss (5 dB/div), S21 group delay (1 ns/div) and
S22 return loss (3 dB/div). Linear frequency sweep from 40 MHz to 1 GHz.
PICOSECOND PULSE LABS P.O. BOX 44 BOULDER, CO 80306, USA TEL: 1.303.443.1249 FAX: 1.303.447.2236
WWW.PICOSECOND.COM
AN-3048, Revision 1, 11/99
Page 5 of 5
