Sunteți pe pagina 1din 6

Impact of Common-to-Differential Mode Conversion

on Crosstalk in Balanced Twisted Pairs


Giordano Spadacini and Sergio A. Pignari
Dipartimento di Elettrotecnica
Politecnico di Milano
Milan, Italy
giordano.spadacini@polimi.it; sergio.pignari@polimi.it

Abstract— In this paper, crosstalk in wire-interconnections terminations, and a statistical description of twist-pitch non-
involving twisted wire pairs (TWPs) with non-perfectly balanced uniformity in length. Predictions based on the proposed model
terminations is investigated. In the first part, properties of show that, when mode conversion is low, non-uniformity of
balancing devices are described by resorting to modal twist-length is the dominant effect in determining random
impedances, modal gains, and common-mode rejection ratio crosstalk levels. Conversely, when CM-to-DM conversion
(CMRR). In the second part, a prediction model based on a takes place at the TWP loads, crosstalk is severely affected and
statistical description of twist-pitch non-uniformity is exploited does not depend on line twisting. Experimental measurements
for crosstalk simulations, and validated via experimental and numerical simulations of a real TWP structure are used to
measurements. The combined effect of: a) common-to-
substantiate the proposed analysis.
differential mode conversion at the TWP terminations; b)
random non-uniformity of the twist-length along the TWP, is
discussed. II. MODAL ANALYSIS OF TWP TERMINATIONS
Balanced terminations for TWPs are often realized by
Keywords-balance; common-to-differential mode conversion; adding suitable interfaces between the wire pairs and the
twisted-wire pairs; crosstalk; statistical EMC models.
drivers/receivers. Typical examples are modern Ethernet
magnetic modules, usually composed of broadband isolation
I. INTRODUCTION transformers and CM chokes in various configurations, which
Among different interconnection technologies, the use of provide high degree of balance and CM rejection. This section
balanced twisted-wire pairs (TWPs) is a cost-effective and describes the basic properties of such terminations from the
largely employed cabling technique in modern electronic and viewpoint of modal circuit analysis. Without loss of generality,
communication systems. As a matter of fact, balanced concepts will be illustrated by resorting to an elemental
drivers/receivers make the equipment more immune to noise balancing device (balun) composed of a center-tapped
and –if used in conjunction with shielded/unshielded TWP transformer, as shown in Fig. 1(a) [1], [7]. In this figure, V1 ,
cables– offer intrinsic robustness to crosstalk interference [1]. V2 and I 1 , I 2 are voltages and currents at the input balanced
In past years, several studies focused on the derivation of ports connected to a TWP, and Vo is the output voltage across
crosstalk prediction models for these wiring structures. In the termination element, modeled by resistance Ro .
particular, in [2-3], a transmission line (TL) model was firstly
proposed by describing the TWP as a cascade of uniform TL Let’s introduce DM and CM voltages and currents at the
sections with abrupt wires’ interchanges. In [4-5], the impact of TWP termination, via the following transformations [1]:
non-uniformity in twist lengths on crosstalk was investigated.
Crosstalk reduction is highly influenced by balancing of the Vd = (V1 − V2 ) / 2 Vc = (V1 + V2 ) / 2
TWP terminations. This was evidenced in [1] by analyzing the DM:  ; CM:  . (1)
contribution of capacitive coupling at low frequencies. In  I d = ( I1 − I 2 ) / 2 I c = ( I1 + I 2 ) / 2
general, immunity to crosstalk interference can be seriously
compromised by poor interface circuits, that make the choice Hence, a general termination element can be represented in
of higher-quality cables fruitless [6]. As a matter of fact, the terms of modal impedances by the following constraints:
role played by the TWP loads in the crosstalk phenomenon is
tightly related to the concept of common-to-differential (CM- Vd = Z d I d + Z dc I c
to-DM) mode conversion, in which balancing appears as a  , (2)
specific aspect. Vc = Z cd I d + Z c I c

In this paper, the combined effect of twisting the line


conductors and CM-to-DM conversion on crosstalk is which lead to the modal circuit model in Fig. 1(b). In this
investigated by resorting to a modal representation of the representation, Z d and Z c are the DM and CM impedances,

This work was supported by the Italian Ministry of University (MIUR)


under a Program for the Development of Research of National Interest (PRIN
grant #2004093025).

1-4244-0293-X/06/$20.00 (c)2006 IEEE 145


2k:1+1 I1 TWP |Z |
100 |CMRR| 150
dc

V1
Vo Ro
V2 50 100

|CMRR|, dB

|Zdc|, Ω
I2
(a)
0 50
Id

Zd
Vd
Z dc I c +
_ -50
6 7 8
0
10 10 10
Ic frequency, Hz
Figure 2. Magnitude of CMRR and of modal mutual impedance Z dc , as
Zc derived from the experimental characterization of a broadband center-tapped
Vc transformer.
Z cd I d +
_

 Vc 
Vo = GdVd + GcVc = Gd Vd + . (4)
(b)  CMRR 
Figure 1. (a) Balanced termination of a TWP realized via a center-tapped
transformer; (b) Modal circuit of a termination. In (4), CMRR = Gd Gc is the so-called common-mode
rejection ratio [7]. For an ideal center-tapped transformer with
respectively, whereas Z dc and Z cd are the mutual modal turn-ratio 2k:1+1 one obtains:
impedances accounting for the contribution to the DM voltage
due to a CM current, and the contribution to the CM voltage Gd = 2k , CMRR → ∞ . (5)
due to a DM current, respectively. For reciprocal networks –as
those composed of transformers– the mutual impedances are It follows that such an ideal component provides full CM
equal, i.e., Z dc = Z cd . rejection, because the output signal depends only on the input
DM signal. It is worth to be noted that an infinite CMRR in (4)
As in common practice, a termination is said to be balanced
when the impedance between each input pin and ground is the is not a strict consequence of the aforementioned definition of
balancing. As an example, we may have perfect balance, i.e.,
same [1], [7]. One can easily show that in terms of the modal
representation in (2), this definition of balancing implies Z dc = Z cd = 0 , but not perfect CM rejection, i.e., a finite
Z dc = Z cd = 0 , that is, no mode conversion occurs. As a CMRR .
consequence, in an ideally balanced termination DM is not In general, real-world components deviate from the ideal
influenced by CM and vice versa. As an example, for an ideal behavior in proximity of the edges of the nominal frequency
center-tapped transformer with turn-ratio 2k:1+1, the modal range. As a consequence, CM-to-DM conversion may occur,
impedances in (2) result to be: and this lowers the interface performances. The first mode
conversion occurs at the input ports due to unbalance, and is
Z d = Ro /(2k 2 ) , Z c = 0 , Z dc = Z cd = 0 . (3) associated with Z dc = Z cd ≠ 0 . Additionally, if CMRR<∞ in (4),
a second contribution to CM-to-DM conversion occurs. It
Hence, the ideal center-tapped transformer provides a follows that CM interference due to crosstalk at the input pins
perfectly-balanced input, which is also characterized by the of balancing devices contributes to DM interference and affects
following properties (not related to balancing): (a) The output the output signal.
resistance Ro is seen as a DM resistance Ro /(2k 2 ) ; (b) The As a specific example, the magnitude of Z dc and CMRR of
CM is short-circuited. a broadband center-tapped transformer (Mini-Circuits T1-6T-
In general, every balancing device implements a transition KK81), for Ro = 50 Ω are reported in Fig. 2. In order to
between the input balanced interface and the output port. measure these quantities, the transformer was mounted on an
Referring to the example in Fig. 1(a), the single-ended output ad hoc PCB and characterized at the terminal ports with a
Vo can be related to the input DM and CM voltages via modal Vector Network Analyzer. Note that the performance is
gains Gd and Gc as: excellent from 300 kHz to 60 MHz, with CMRR > 35 dB and
Z dc ≈ 0 Ω .

1-4244-0293-X/06/$20.00 (c)2006 IEEE 146


β G

G TWP RS TWP
R1
d NE V1NE V1FE RL
FE
2∆h VS
h R2 V2NE V2FE

ground plane
Figure 4. Connections at the TL terminations. The shaded blocks represent
balanced loads at NE and FE, as those realized via center-tapped transformers.
Figure 3. Cross-section of the 4-conductor TL under analysis

III. PREDICTION MODEL FOR CROSSTALK IN TWPS HT of the second kind. It follows that the chain parameter
This Section describes a numerical prediction model, based matrix Φ of the entire wiring structure takes the form:
on the segmentation of the non-uniform TL into the chain-
N
connection of uniform sections. This model − known in the
literature as ‘click model’ [8] − is extended here in order to Φ= ∏Φ
i =1
(2) (1)
Φ HT
HT , N −i +1 , N −i +1 (7)
include effects of non-uniformity of the length of the twist-
pitch.
where Φ (HTp ),k , p = 1,2 , k = 1, 2,...N , is the chain-parameter
A. Structure under analysis matrix of the 4-conductor TL (wire G and TWP above ground,
The structure under analysis consists of a 4-conductor TL as in Fig. 3) for the k-th HT of the p-th kind of length ξ k( p ) . In
composed of a generator circuit G and a TWP receptor (R1, order to include the non-uniform TL structure into the model,
R2), running parallel and placed over a metallic ground plane. these chain-parameter matrices are approximated by chain-
The TL cross-section is sketched in Fig. 3. The TWP is connections of uniform-TL sections [8].
assumed to be wound into a bifilar cylindrical helix described
by the rotation angle β , and composed of N full-twists, or Namely, the k-th HT of the p-th kind of length ξ k( p ) is
2 N half-twists (HT). The total TL length is L , and the mean subdivided into n elemental sections with constant rotation
HT length is ξ m = L / 2 N . The TL terminations are shown in angle β . For each of those uniform-TL sections, the p.u.l.
Fig. 4, where the shaded blocks represent the balanced near- parameters are evaluated and the chain-parameter matrix is
end (NE) and far end (FE) loads, as those realized via center- computed. Then, matrix Φ (HTp ),k is obtained by multiplying the
tapped transformers. The generator circuit G is fed at NE n chain-parameter matrices of the uniform elemental sections
(voltage VS , resistance RS ) and loaded at FE by resistance RL . [8].
As in [4-5], twist-pitch non-uniformity is modeled by If the chain-parameter matrix Φ of the entire wiring
assuming that the HTs’ lengths randomly deviate from the structure [computed via (7)] is combined with the impedance-
ideal mean length ξ m . To this aim, the HTs are subdivided into matrix representation of the balanced terminal loads, the
two classes: (i) The HTs of the first kind with length crosstalk voltages V1 NE , V2NE , V1 FE , V2NE (see Fig. 4) are readily
ξ k(1) = ξ m + ∆ξ k(1) , corresponding to β ∈ [2(k − 1)π , (2k − 1)π ] ; derived. Then, CM and DM crosstalk voltages VdNE , VcNE ,
(ii) the HTs of the second kind with length ξ k( 2 ) = ξ m + ∆ξ k( 2 ) , VdFE , VcFE are obtained via (1), and the output crosstalk
corresponding to β ∈ [( 2k − 1)π ,2 kπ ] , k = 1, 2,...N . It is a voltages VoNE , VoFE via (4).
simple matter to show that the pitch deviations ∆ξ k(1) and
C. Probabilistic description of twist-pitch non-uniformity
∆ξ k( 2 ) are not independent quantities, since the following
relationship holds [4]: It has to be recognized that the HT lengths ξ k(1) and ξ k( 2 ) ,
k = 1, 2,...N , are unknown and uncontrolled parameters. They
N N
are generated during the TWP manufacturing process and are
∑ ∆ξ
k =1
(1 )
k =− ∑ ∆ξ
k =1
( 2)
k (6)
subject to possible modifications during installation (e.g., in
proximity of the TL ends). Direct measurement of the HT
lengths is clearly an unpractical process, and even if done, the
B. Non-uniform TL model
collected data would refer only to the specific TWP sample
According to the modeling scheme illustrated in the under analysis. In order to overcome such an unpractical
previous subsection, the TWP is described as a chain of N approach, the proposed model exploits a probabilistic
twists, each of them composed of an HT of the first kind and an description of the HT length.

1-4244-0293-X/06/$20.00 (c)2006 IEEE 147


-20

-40

-60
NEXT, dB

-80

-100

-120

-140
5 6 7 8 9
10 10 10 10 10
frequency, Hz
Figure 5. Experimental measurements of NEXT (30 TWP samples) Figure 6. Prediction of NEXT (10 repeated runs)
3

Namely, the twist-pitch deviations ∆ξ k(1) of the HTs of the -20


prediction interval
first kind can be considered as N independent and equally- µNEXT
distributed random variables (RVs). It is assumed that these -40 µNEXT-σNEXT
RVs behave as a RV ∆ξ ∈ [−∆ξ max ,+ ∆ξ max ] , where ∆ξ max ≤ ξ m ,
µNEXT+σ NEXT
with uniform probability density function δ ∆ξ (∆ξ ) [9], i.e., sample mean (measurements)
-60
NEXT, dB

1 /(2∆ξ max ) ∆ξ < ∆ξ max


δ ∆ξ (∆ξ ) =  . (8) -80
0 otherwise
-100
Hence, random samples ∆ξ k(1) , k = 1, 2,...N , are generated
according to (8). The deviations of the HTs of the second kind
are randomly generated in a similar fashion, consistently with -120
5 6 7 8 9
10 10 10 10 10
condition (6). The statistical analysis is based on repeated frequency, Hz
generation of HT sets, as specified above, and subsequent Figure 7. Statistical estimates of NEXT
crosstalk computation via the proposed TL model.
was measured. Fifteen samples of TWPs were tested, and each
IV. EXPERIMENTAL VALIDATION of them was used for the two experiments. In fact, each TWP
can be inserted in two different ways in the setup, by
Experimental measurements were carried out by exploiting interchanging the TWP ends. This leads to two different
an ad-hoc setup aimed at reproducing the structure sketched in geometrical structures, since non-uniformities in twist length
Fig. 3-4. This test-bench is composed of AWG 24 wires are oppositely and differently distributed along the line. Hence,
(obtained from Cat. 5 UTP cables). The geometrical parameters thirty curves of the NEXT versus frequency were measured
are: h = d ≈ 2 cm , ∆h ≈ 0.5 mm , L ≈ 1.9 m . The TWP is and reported in Fig. 5. Note the low-frequency range, where
connected to balanced NE and FE terminations realized via the TL is electrically-short and crosstalk increases with a 20
center-tapped transformers with ratio 2:1+1 and Ro = 50 Ω . dB/decade slope, and the high frequency range, where
These devices are mounted on PCB enclosed in metallic boxes, resonances of the electrically-long TL appear.
and connect to the TWP via RJ45 Stewart connectors. Both the Moreover, for frequencies up to 60 MHz a wide spread
NE and FE transformers were previously characterized in order between curves is observed, and only some of them show
to derive the modal impedances, gains, and CMRR defined in resonances. It was also observed that frequency behavior of
Section II (see Fig. 2 for specific results concerning the NE crosstalk can be modified by intentionally perturbing the twists
transformer). The generator wire G was loaded by [10]. In this frequency range, crosstalk shows a high sensitivity
RS = RL = 50 Ω . A Network Analyzer Agilent E5070B was to twist-pitch non-uniformity and behaves as a RV [4-5], [10].
connected to the NE (port no. 1 connected to the generator Conversely, the spread between crosstalk curves reduces as
wire, port no. 2 connected to the balancing transformer as Ro ), frequency increases, since all the curves tend to collapse into a
unique one, which results to be insensitive even to strong
and the NE crosstalk ratio (NEXT): perturbation of twists. In the following section, these
phenomena will be discussed and explained as an effect of
NEXT =| VoNE / Vs |=| s21 | / 2 (9) CM-to-DM conversion at the line terminations.

1-4244-0293-X/06/$20.00 (c)2006 IEEE 148


Figure 8. CM and DM crosstalk ratios at the input pins of the NE center- Figure 9.10.CM
Prediction
and DM crosstalk
of NEXTratios(10 (10
repeated-runs)
repeated-runs)
in as
case
in Fig.
of perfecty-
8, but in
3 3

tapped transformer (10 repeated-runs).


3
balanced
case of perfectly
loads and
balanced
high CMterminations
rejection over
(no the
modeentire
conversion).
frequency range.

Fig. 6 reports 103 crosstalk predictions obtained via the Additionally, it is observed that all the 103 curves of the CM
proposed repeated-run statistical model. These predictions crosstalk coincide (in the entire frequency range), whereas the
were obtained by setting the simulation parameters to: 103 curves of the DM crosstalk tend to overlap only at the
N = 130 (estimated value), ∆ξ max = ξ m = 7.3 mm (worst-case higher frequencies. Conversely, a high sensitivity of DM
condition for twist-pitch deviations). From these simulations crosstalk is observed at lower frequencies.
one obtains the statistical estimates of crosstalk reported in Fig. In order to investigate how CM-to-DM conversion at load
7, i.e., expected value µ NEXT (dashed line); expected value terminations affects crosstalk levels, in the following example
minus a standard deviation µ NEXT − σ NEXT (dash-dotted line); CM and DM crosstalk are computed in case of perfectly-
expected value plus a standard deviation µ NEXT + σ NEXT (dotted balanced terminations, i.e., by virtually making transformers
ideal balancing-devices. In line with this objective, repeated-
line). In Fig. 7, the solid line is the sample mean of the 30 runs reported in Fig. 9 refer to the same structure of previous
experimental curves reported in Fig. 5.
examples in which we have imposed Z dcNE = Z dcFE = 0 [see (2)],
The agreement between experimental data and predicted whereas the other load parameters ( Z c , Z d , Gd ) are set to the
estimates represents a validation of the proposed model. In
particular, note that the statistical model correctly predicts the values determined from the experimental characterization of
spread between crosstalk values in the low-frequency region. the center-tapped transformers. Note that: (a) CM crosstalk
Repeated-run analysis shows that the symmetric interval remains unaltered with respect to Fig. 8, i.e., as in the case of
non-ideally balanced loads; (b) DM crosstalk is generally lower
µ NEXT ± σ NEXT reported in Fig. 7 is characterized by a
than in Fig. 8; (c) DM crosstalk shows high sensitivity to non-
confidence coefficient variable from 70% to 85% (depending uniformity in the entire frequency range. By comparing Fig. 8
on frequency) [9].
and Fig. 9 and taking into account the behavior of Z dcNE in Fig.
2, one concludes that CM crosstalk does not depend on twist-
V. IMPACT OF CM-TO-DM CONVERSION pitch non-uniformity and on termination balance. On the other
The proposed crosstalk model is exploited here to hand, DM crosstalk experiments high sensitivity to twist-pitch
investigate the impact of CM-to-DM conversion, and to relate non-uniformity. Such sensitivity reduces as frequency increases
this phenomenon with non-uniformity effects which determine due to CM-to-DM conversion due to load unbalance.
the random behavior of crosstalk. Additionally, DM crosstalk is increased by load unbalance.
Firstly, let’s reconsider the experimental setup and According to Section II, a second contribution to CM-to-
numerical prediction described in the previous section. Instead DM conversion occurs as a consequence of poor CM rejection
of focusing on crosstalk at the output port of the center tapped from the input to the output port of the balancing devices. As
transformer, (i.e., NEXT = VoNE / Vs reported in Fig. 5-7) it is evidenced by (4), if the CMRR is small, a greater amount of the
input CM crosstalk | VcNE / Vs | is converted into output DM
worth investigating the behavior of the CM and the DM
crosstalk ratios at the input pins of the balancing device, i.e., crosstalk | VoNE / Vs | . Consequently, poor CM rejection makes
| VcNE / Vs | , and | VdNE / Vs | , respectively. These quantities output crosstalk large and insensitive to twist-pitch non-
combine to determine to output crosstalk level according to (4). uniformity. This can be clearly seen by analyzing Fig. 6 by the
Predictions of CM and DM crosstalk (corresponding to the 103 light of the frequency-behavior of CMRR in Fig. 2.
repeated-runs of the output crosstalk previously reported in Fig. As a final example, let’s predict the NE output crosstalk in
6) are reported in Fig. 8. Note that CM crosstalk is much case of both perfectly-balanced loads and optimal CMRR.
greater than DM crosstalk at the lower frequencies. Namely, repeated-runs of NEXT in Fig. 10 refer to the same

1-4244-0293-X/06/$20.00 (c)2006 IEEE 149


-20
µNEXT σNEXT (Zdc=0; CMRR=50 dB)

-40 µNEXT (real balancing devices)

-60

NEXT, dB
-80

-100

-120
5 6 7 8 9
10 10 10 10 10
frequency, Hz
Figure 10. Prediction of NEXT (10 repeated-runs) in case of perfecty- Figure 11. Statistical estimates of NEXT in case of perfectly-balanced loads
3

balanced loads and high CM rejection over the entire frequency range. and high CMRR (shaded interval); Expected value of NEXT for the real
balancing devices (solid line).

structure considered in previous examples, but where the crosstalk is effectively reduced by twisting the line conductors
following load parameters are set to constant values over the and demonstrates high sensitivity to twist non-uniformity.
entire frequency range: Z dcNE = Z dcFE = 0 (i.e., perfect balance),
and CMRR NE = 50 dB (i.e., high CM rejection). NEXT in Fig. REFERENCES
10 has to be compared versus NEXT in Fig. 6 resulting from [1] C. R. Paul, Introduction to Electromagnetic Compatibility, New York:
the real non-ideal balancing-devices, whose balance and CMRR Wiley & Sons, Inc., 1992.
depend on frequency (see Fig. 2). Note that NEXT in Fig. 10 [2] C. R. Paul and J. A. McKnight, “Prediction of crosstalk involving
twisted pairs of wires, Part I: A transmission line model for twisted wire
shows high sensitivity to twist-pitch non-uniformity in the pairs,” IEEE Trans. Electromagn. Compat., vol. EMC-21, pp. 92-105,
entire frequency range. Statistical estimates obtained from the 1979.
103 simulations in Fig. 10 are reported in Fig. 11, i.e., expected [3] C. R. Paul and J. A. McKnight, “Prediction of crosstalk involving
value µ NEXT (dashed line); expected value minus a standard twisted pairs of wires, Part II: A simplified, low-frequency prediction
model,” IEEE Trans. Electromagn. Compat., vol. EMC-21, pp. 105-114,
deviation µ NEXT − σ NEXT (dash-dotted line); expected value plus 1979.
a standard deviation µ NEXT + σ NEXT (dotted line). Finally, the [4] D. Bellan, G. Spadacini and S. A. Pignari, “Prediction of twist non-
uniformity and twist-residual effects on crosstalk in twisted-wire pairs,”
solid line refers to the expected value µ NEXT already reported in in Proc. EMC Zurich 2003 Int. Symp. Electromagn. Compat., Zurich,
Fig. 7 for the real non-ideal terminations. From Fig. 11 one can Switzerland, Feb. 18-20 2003, pp. 181-186.
see that the degradation of the performance of the balancing [5] G. Spadacini, D. Bellan and S. A. Pignari, “Impact of twist non-
devices, as the frequency increases (beyond 100 MHz), uniformity on crosstalk in twisted-wire pairs,” in Proc. 2003 IEEE
severely affects crosstalk levels (more than 40 dBs). Symp. on Electromagn. Compat., Boston, MA, USA, Aug. 18-22, 2003,
pp. 483-488.
[6] A. Axelrod, “On crosstalk mechanism in cables composed of several
VI. CONCLUSION unshielded twisted pairs,” in Proc. 2003 IEEE Int. Symp. on
In this work, crosstalk in balanced TWPs was investigated Electromagn. Compat., Istanbul, Turkey, May 11-16, 2003, pp. 1016-
1019.
by resorting to modal analysis of terminations. Statistical
[7] K. L. Kaiser, Electromagnetic Compatibility Handbook, Boca Raton:
properties discussed by exploiting a non-uniform TL circuit CRC Press, 2005.
model have been confirmed by experimental measurements [8] M. B. Jolly and C. R. Paul, Crosstalk in twisted-wire circuits, Tech. Rep.
and numerical simulations. The proposed model includes Rome Air Development Center, Griffiss AFB, NY, RADC-TR-82-286,
random twist-pitch non-uniformity effects and provides Vol. IV C (November 1982).
crosstalk prediction in terms of statistical estimates. [9] A. Papoulis, Probability, Random Variables, and Stochastic Processes,
Third Edition, New York: McGraw-Hill, 1991.
The combined impact of (a) CM-to-DM conversion due to [10] C. R. Paul and M. B. Jolly, “Sensitivity of crosstalk in twisted-pair
unbalance and low CM rejection at the line terminations; (b) circuits to line twist,” IEEE Trans. Electromagn. Compat., vol. EMC-24,
random twist-pitches, is discussed. It is shown that when pp. 335-364, 1982.
output interfaces are characterized by high balance and CMRR,

1-4244-0293-X/06/$20.00 (c)2006 IEEE 150

S-ar putea să vă placă și