How to connect cable shields
a.k.a: A short summary of Henry Ott's approach in Electromagnetic Compatibility Engineering (2009).
Foreword
Does it matter if the shield gets shorted to ground on the host
side versus the device side? I'm pretty sure you're not supposed
to do it on both, but I don't know if it practically matters which
side.
This is a dangerous question to ask. On one hand, EMI/EMC textbooks
suggest that connecting the shields to chassis at both ends is always
the best option from the perspective of RF shielding... On the other
hand, for low-frequency applications, the conflicting requirement of
avoiding ground loops mandates that the shield should only be connected
at one side.
In my original answer, I believed I already carefully navigated around
the nuisances of the pros and cons of connecting the shield at one side,
or at both sides. Unfortunately, I completely ignored the problem of
another related question: the connection between the shields and the
circuit grounds. My failure to take that into account caused a lengthy
3-way debate with over 20 exchanged comments.
This new answer has been fully rewritten, and would hopefully address
the deficiencies in my original answer. This answer is meant to be a
faithful summary of the apporach advocated by Henry Ott in the textbook
Electromagnetic Compatibility Engineering (2009), and would serve as
a general reference on shield connection to all future questions.
Nearly every single claim in each every paragraph is backed up
with quotations from Ott (2009), and please notice me immediately
if I made any misrepresentation of Ott. However, disagreements
exist even among experts. If you disagree with Ott, debating
with me would be unproductive.
Overview
The connection of the shield is subjected to multiple and often
contradictory requirements: (1) effective RF shielding, (2)
avoiding ground loops that cause low-frequency noise and hum,
(3) ESD immunity, (4) radiation due to common-mode current
flowing across the shield or chassis.
Shield-to-Chassis Termination
The first problem is whether the shield should be terminated to
the chassis, at which location, and whether one-side or two-side
termination should be used.
RF Shielding
For high-speed digital and RF systems, including most USB devices,
RF shielding (Requirement 1) is usually the most important factor.
[1] In this case, the shield of a cable should be seen as an
extension of the metal enclosure of the equipment [2] - if a cable
is used to connect two devices, this situation is equivalent to a
single device inside a shared outer chassis. This necessarily
mandates the termination of shield and chassis at both sides. [1][2][3]
Thus, a shield should satisfy the following requirements:
Terminate the shield at chassis on both ends, never the circuit
ground. [1] The shield-to-chassis connection should be the preferred
path of the RF noise current.
Make complete 360-degree contact with the chassis. [3]
Connect other non-shield conductors (such as power, signal,
power ground, signal ground) to the circuit, as usual.
Ground Loop
Connecting the shield at both ends create a
ground loop, a small difference of ground potential causes
a noise current to flow from one end to another. For
high-speed RF and digital systems, this is usually considered
to be an acceptable cost - functional RF shielding is much more
important than a few millivolts of negligible noise, which is
below the logic threshold or can be filtered out. [1]
Unfortunately, for low-frequency or analog systems, the
low-level noise can cause serious interference. The classic
problem is the 50/60 Hz mains hum in audio. [4] Analog data
acquisition systems may experience similar problems.
Coincidentally, this problem also occurs at a much greater
scale in industrial installations across buildings where a
significant difference of "Earth" potential between locations exist.
Thus, the shield is often connected at one end, and disconnected
at the other side. [4] This sacrifices effective RF shielding,
reducing the Faraday cage that is capable of blocking
high-frequency radiation to a simple electrostatic screen,
only capable of blocking low-frequency electric fields, like
mains hum.
Terminating the shield to the chassis, instead of the circuit
ground, somewhat mitigates but does not solve the problem of
the lack of RF shielding. [5]
Best of both worlds?
If terminating the shield at both sides is required for high-frequency
shielding, while terminating the shield at one side is required for
low-frequency analog systems to avoid mains hum. What is the solution
for mixed-signal systems? Is there a compromise between Requirement 1
and Requirement 2?
Three options exist:
- Use a shielded twisted pair. Twisted pair, even unshielded, provides
inherent suppression of electromagnetic interference, especially when
combined with differential signaling. This way, the necessity of an
RF shield is reduced. Thus, the shield for the twisted pair can be
dedicated for low-frequency shielding only, and still providing acceptable
EMI/EMC performance. [4] Nevertheless, when there's a chance to connect
the shield at both sides properly, RF suppression would be even higher.
This possibly explains the reason that disconnecting the USB shield at one
side is not a deal-breaker, despite that it's not elegant in theory.
Use a large number of SMD capacitors to connect the chassis and shield.
This way, the shield is disconnected at DC and low frequencies, but
it's reconnected at RF. However, to be effective at RF, the parasitic
inductance of the capacitors must be as low as possible. [7] Thus, it
needs to be considered on a case-by-case basis, and it's a non-standard
solution.
Use a triaxial cable with two layers of shields, one is connected
at one end for low-frequency shielding, another is connected at both
ends for RF shielding. This requires the use of awkward and non-standard
cables and is unpopular today. [8]
Chassis-to-Circuit Ground Connection
The next problem is whether the chassis should be connected to the circuit
ground (usually the ground plane of a circuit board), and if so, at which
location. Because the shield is already connected to the chassis, the problem
of whether the shield should be connected to the circuit ground is also implied
here.
Connection vs Termination
Didn't I just write that "terminate the shield at chassis on both
ends, never the circuit ground"? Why am I now discussing the connection from
shield to the circuit ground?
There's a difference between termination and connection. Any electrical path
would be a connection, but termination emphasizes the first location a contact
is made. A shield should first and foremost be connected to
the chassis via a solid, low-impedance, 360-degree bond to the chassis. This is
the preferred path of current flow in the shield.
But an eventual electrical connection between shield and circuit ground (as a
result of bonding the circuit ground to the chassis) is still permitted. For
example, a coaxial connector should ideally be screwed onto the chassis directly,
before the same "shield/ground" and center conductor wires reach the circuit board.
ESD Immunity
Theoretically, an RF shield works by itself and does not need an electrical
connection to anything else. From the perspective of RF interference only,
the shield can be left floating. However, bonding circuit ground and chassis
is often desirable due to other practical problems, mainly ESD.
Imagine a circuit board fully enclosed by a Faraday cage. When the cage is
zapped by ESD, although the absolute potential of the circuit relative to
the Earth ground increases, the relative potentials remain the same, and
the circuit board is perfectly protected. Unfortunately, real circuit boards
have external cables attached, and one of the cable may attach the circuit
ground to an external ground, possibly an Earth ground.
After the metal enclosure is zapped by ESD, the circuit ground potential
is held by the cable, enabling a secondary ESD strike may develop from the
chassis to the circuit ground, finally leaving the system via an attached
cable. It may crash the system in this process.
Bonding the chassis and circuit ground with a solid connection is often
used as the solution of this problem to satisfy Requirement 4.
Location of Bonding
The location where the bonding is made requires attention.
Due to the flow of current, there exists a voltage gradient across the
circuit ground plane of the circuit board. If the ground plane is bonded
to the chassis at the right side of the board, while the cable enters at
the left side of the circuit board, this potential difference would cause
a common-mode noise current to flow, degrading the EMI/EMC performance
of the system. [9]
To mitigate this problem, Ott recommends creating a separate area on
the circuit board, dedicated to I/O connectors. At this I/O area, a
solid connection is made between the chassis and the circuit ground,
simultaneously, the cable shield is terminated to the chassis at the
same location. [10] By connecting the shield, chassis, and circuit
ground at nearly the same location, a voltage gradient is largely avoided.
This area of the PCB also uses its own ground plane, largely but not fully
isolated from the main circuit ground plane. Only a small bridge is used
to connect both planes, allowing high-frequency signals to flow on top of
the bridge without crossing a slot in the plane, while providing a degree
of isolation between the circuit ground of chassis gruond.
This is an attempt to solve the problem of making a shield-to-chassis
connection to be the preferred path of the RF noise current. In an
old-school design the connectors are screwed onto the chassis, so a
shield-to-chassis connection is almost always the prefered path for noise
current. But in modern designs, connectors are mounted onto the circuit
board, not the chassis. Thus, avoiding injecting noise from the shield
to the circuit ground becomes a problem. The isolated I/O ground plane
is the author's best attempt to solve this problem.
Also, note that other connections between the chassis and the circuit
boards are permitted. But they cannot replace the connections at the
I/O area of the PCB, which is of crucial importance.
In my observation, the design of desktop computer motherboards largely
reflects the principles behind this method. the motherboards come with
a separate I/O cover. When the board is installed, the I/O cover is
pushed forcefully onto the chassis.
Perfect Shield Connection
To summarize, if you're connecting two high-speed digital devices
together (with no analog or mixed-signal circuit, and both with
metal chassis), according to Ott, theoretically, the purest way to
do that would be:
Mount the connector onto the chassis, creating a solid shield-to-chassis
termination.
Ideally, the connector should be mounted directly onto the chassis
first. For many coaxial and circular connectors, they can be screwed
onto the chassis directly.
If the connectors are mounted onto the circuit board, use metal
I/O cover, EMI gaskets, grounding fingers, or other means to
create a solid connection between the metal shell of the connector
and the chassis.
If the above is not possible, the I/O ground plane to chassis connection
serves as the final fallback. At the very least, multiple metal screw
standoff is used.
Create a seperate I/O area and I/O ground plane on the circuit board,
allowing the chassis-to-circuit ground connection to be made with minimum
common-mode current flow.
The I/O area of the PCB also uses its own ground plane, largely but not fully
isolated from the main circuit ground plane. Most copper between the two regions
are removed, only a small bridge is used to connect both planes, allowing
high-frequency signals to flow on top of the bridge without crossing a slot in
the plane, while providing a degree of isolation between the circuit ground of
chassis ground.
Connect other non-shield conductors (such as power, signal, power ground, signal
ground) to the circuit, as wires or traces.
This complete my summary of Henry Ott's Electromagnetic Compatibility Engineering,
the following sections are my own opinions.
Practical Issues
In the context of USB, there are many practical issues that force designers to devitate
from the ideal solution. My observation is that, they include:
It's difficult to terminate shield to chassis correctly. Sometimes designers simply
have no control over the I/O area. Furthermore, USB connectors are tiny, especially
the new Type C connectors, using gaskets or ground fingers is likely not practical.
The Ott's I/O-area solution is also a mitigation more than a solution.
For the ideal method to work, both sides of the shield must be designed correctly,
with the correct bonding of circuit ground, chassis, and shield. If one side is not
correctly designed, sometimes beyond our control, a common-mode noise current flows
and creates increased interference, as previously described.
The use of analog and mixed-signal circuits in USB device, such as audio or
data acquisition, may rule out connecting the shield at both sides as an option,
compromising RF shielding. Thanksfully, due to the use of shielded twisted pair
in USB, even with a compromised shield, the EMI/EMC performance may still be
acceptable, it depends. Galvanic isolation may be an alternative solution.
Many devices have no metal enclosure at all, invalidating the entire method.
No Metal Enclosure
The worst-case (yet common) scenario is when there's no metal enclosure.
In this case, it's impossible to divert the noise current on the shield away
from the circuit board. If a connection is still made from the shield to the
circuit ground, noise is injected directly into the circuit board's ground
plane. Combined with some mixed-signal or analog circuits on the board that
are vulnerable to ground loop, the situation becomes a total mess.
As a result, people invented various workarounds to overcome this problem,
they include:
Disconnecting the shield entirely.
Idea: Stop noise current from the shield from entering circuit ground.
Flawed.
Use RC circuits to connect the shield to the circuit ground.
Idea: Create a high-pass filter to stop low-frequency noise current, such as
mains hum, from flowing on the shield or entering the circuit ground. Flawed.
Use ferrite beads to connect the shield to the circuit ground.
Idea: Create a low-pass filter to stop high-frequency noise current, such as
EMI, from entering the circuit ground. Flawed.
Connect shield directly to the circuit ground.
Idea: Maintain shield and circuit ground at the same potential. Flawed. It
violate all of Ott's rules. But better than you think, and is a suitable
compromise for many applications.
Unfortunately, these methods may work in some particular situations, but each
has its flaws.
The source of my claim was Tim Williams, one participant of the previous
debate in the comment section, he was already known to me as a prolific
poster on the EEVblog forum, and I remember seeing his extensive writing on
the subject of the USB shield connection.
The most important flaw is that if the shield and circuit ground are
isolated from each other via capacitors or ferrite, during a ESD strike,
a large potential difference is created between the shield and circuit
ground, enabling a ESD strike across them, and causing the device to
fail ESD compliance tests. In Williams' anecdotal observations, floating
shield, RC and ferrite bead solutions performs poorly under ESD strikes,
and is a common cause of failure of ESD compliance tests. After bonding the
shield directly onto the circuit ground, these devices would pass ESD tests
immediately. Whenever the topic of splitting USB shield and circuit ground
is brought up, he would always say, "don't do it, it simply cannot pass ESD
tests."
Another flaw mentioned by Williams, if I remember correctly, was the issue of common-mode radiation when the cable shield and power/signal ground
is not at the same potential. The potential difference across the two conductors
create a noise current flow throughout the entire cable's length, creating common-mode
radiation. On the other hand, making a solid connection between the shield and
the circuit ground suppresses this potential difference, reduce radiation (of
course, this is not the only possible failure mode, and I can imagine that there
are other situations that it may create the opposite situation).
Another flaw is that isolating the shield from the circuit ground is in a violation
of the USB standard. The USB Type-C specification includes a blanket prohibition
against these practices (On the other hand, the USB Type-C specification also
includes extensive description on bonding the shield and the chassis, thus, I
imagine that the technical committee behind USB follows a method similar to the
one proposed by Henry Ott).
- The receptacle shell shall be connected to the PCB ground plane.
As a result, for a pure digital USB device, without analog or mixed-signal circuits,
without a metal chassis, connecting the shield directly to the circuit ground, while
violating all of the rules, in actually a suitable compromise for many applications with
justification. And indeed, many USB dongles are designed like that. When you don't
have a choice, a straight connection may be the only compromise here.
If one decides to use RC and ferrites solutions, solving the ESD problem is left as
an exercise for the reader.
Thus, USB shield connection appears to be a rather "polarized" problem: if you're
doing it wrong, you'd better to do it wrong all the way. If you're doing it right,
you'd better to do it right all the way. It's difficult to take the middle ground.
On Holy Wars and a Plea for Peace
I have a personal explanation of why EMI/EMC problems trends to generate flame
wars - the gap between idealized best practices and actual systems. The
problem of all "ideal" approaches is that, they must be followed precisely and
exactly, and all the necessary preconditions must also be satisfied for their
assumptions to be valid. Any deviation would change their applicability and result.
Using another problem in PCB design as an example: In PCB design, there's a rough
consensus that, if the PCB is already partitioned into digital and analog
sections, the return current in each section is mostly contained in their own area.
Thus, unless very high isolation is required, splitting the analog and digital ground
planes is often counterproductive. Hence, people start to describe it as an
anti-pattern and is something to be avoided - it's better to avoid the gaps in the
ground plane to reduce EMI, and "don't split planes" becomes a slogan. But if a circuit
board is not following the assumption behind this methodology to begin with (not
having partitioned sections), splitting the ground plane may actually improve
performance - an apparent contradiction. Soon, another fraction of designers would
soon to start attacking the supposedly "the best practices."
But it's not the theory that is in fault, it's just that it has to be followed
systematically.
I believe this scenario is extremely common in the field of EMI/EMC. Since every
system has quirks and limitations, it creates a lot of contradictory observations.
Thus, it generates never-ending debates and controversies.
References
[1] Page 91.
At frequencies above about
100 kHz, or where cable length exceeds one twentieth of a wavelength, it
becomes necessary to ground the shield at both ends. This is true for either
multiconductor or coaxial cables. [...] It is therefore common practice at high frequency, and with digital circuits,
to ground the cable shield at both ends. Any small noise voltage caused by a
difference in ground potential that may couple into the circuit (primarily at
power line frequencies and its harmonics) will not affect digital circuits and can
usually be filtered out of rf circuits, because of the large frequency difference.
[2] Page 90.
If you think of a cable
shield as being an extension of the enclosure’s shield, then it becomes clear
that the shield should be terminated to the enclosure not to the circuit ground.
[3] Page 555
Think of a cable shield as an
extension of the shielded enclosure. Therefore, how effective the shield is has
a lot to do with how well the cable shield is terminated to the enclosure. Use a
360-degree termination to the enclosure, not to circuit ground.
[4] Page 88
The main reason to shield
cables at low frequency is to protect them against electric field coupling
primarily from 50/60-Hz power conductors. As was discussed in Section
2.5.2, a shield provides no magnetic field protection at low frequency. This
points out the advantage of using shielded twisted pair cables at low frequency:
The shield protects against the electric field coupling and the twisted pair
protects against the magnetic field coupling. Many low-frequency circuits
contain high-impedance devices that are susceptible to electric field coupling,
hence, the importance of low-frequency cable shielding.
At low frequency, shields on multiconductor cables where the shield is not
the signal return conductor are often grounded at only one end. If the shield is
grounded at more than one end, then noise current may flow in the shield
because of a difference in ground potential at the two ends of the cable. This
potential difference, and therefore the shield current, is usually the result of
50/60-Hz currents in the ground. In the case of a coaxial cable, the shield
current will produce a noise voltage whose magnitude is equal to the shield
current times the shield resistance, as was shown in Eq. 2-33. In the case of a
shielded twisted pair, the shield current may inductively couple unequal
voltages into the twisted pair signal conductors and be a source of noise (see
Section 4.1 on balancing
[5] Page 89.
Grounding the cable shield at only one end to eliminate power line frequency
noise coupling, however, allows the cable to act as a high-frequency antenna
and be vulnerable to rf pickup. AM and FM radio transmitters can induce
high-frequency rf currents into the cable shield. If the cable shield is connected
to the circuit ground, then these rf currents will enter the equipment and may
cause interference. Therefore, the proper way to terminate the cable shield is to
the equipment’s shielded enclosure, not to the circuit ground. This connection
should have the lowest impedance possible, and the connection should be made
to the outside of the shielded enclosure.
[6] Page 78.
A shielded twisted pair has characteristics similar to a triaxial cable and is
not as expensive or awkward. The signal current flows in the two inner
conductors, and any induced noise current flows in the shield. Common-
impedance coupling is therefore eliminated. In addition, any shield current
present is equally coupled (ideally), by mutual inductance, to both inner
conductors and the two equal noise voltages cancel.
An unshielded twisted pair, unless its terminations are balanced (see Section
4.1), provides very little protection against capacitive (electric field) pickup, but it is very good for protection against magnetic field pickup. The effectiveness of
twisting increases as the number of twists per unit length increases. When
terminating a twisted pair, the more the two wires are separated, the less the
noise suppression. Therefore, when terminating a twisted pair, shielded or
unshielded, do not untwist the conductors any more than necessary to make the
termination.
Twisted pair cables, even when unshielded, are very effective in reducing
magnetic field coupling. Only two conditions are necessary for this to be true. [...]
[7] Page 92.
At low frequency, a single-point ground exists because the impedance of
the capacitor is large. However, at high frequency, the capacitor becomes
a low impedance, which converts the circuit to one that is grounded at both
ends. The actual implementation of an effective hybrid cable shield ground may,
however, be difficult, because any inductance in series with the capacitor will
decrease its effectiveness. Ideally, the capacitor should be built into the
connector.
[8] Page 93
2.15.2.4 Double Shielded Cable Grounding. Two reasons to use a double-
shielded cable are as follows: One is to increase the high-frequency shielding
effectiveness; the other is when you have both high-frequency and low-
frequency signals in the same cable. In the first case, the two shields can be
in contact with each other; in the second case, the two shields must be isolated
from each other (often referred to as a triaxial cable).
Having two shields that are isolated from each other allows the designer the
option of terminating the two shields differently. The outer shield can be
terminated at both ends to provide effective high-frequency as well as magnetic
field shielding. The outer shield is often also used to prevent radiation from the
cable, which results from high-frequency common-mode currents on the cable.
The inner shield can then be terminated at only one end, thus avoiding the
ground-loop coupling that would occur if grounded at both ends.
[9] Page 131
Chassis ground is any conductor that is connected to the equipment’s metal
enclosure. Chassis ground and signal ground are usually connected together at
one or more points. The key to minimizing noise and interference is to
determine where and how to connect the signal ground to the chassis. Proper
circuit grounding will reduce the radiated emissions from the product as well as
increase the product’s immunity to external electromagnetic fields.
Consider the case of a PCB, with an input/output (I/O) cable, mounted
inside a metallic enclosure as shown in Fig. 3-24. Because the circuit ground
carries current and has a finite impedance, there will be a voltage drop VG
across it. This voltage will drive a common-mode current out on the cable, and
will cause the cable to radiate. If the circuit ground is connected to the chassis at the end of the PCB opposite the cable, then the full voltage VG will drive the
current onto the cable. If, however, the circuit ground is connected to the
enclosure at the I/O connector, the voltage driving common-mode current out
onto the cable will ideally be zero. The full ground voltage will now appear at
the end of the PCB without the cable connection. It is, therefore, important to
establish a low-impedance connection between the chassis and the circuit
ground in the I/O area of the board.
Another way to visualize this example is to assume that the ground voltage
produces a common-mode noise current that flows toward the I/O connector.
At the connector, there will be a current division between the cable and the PCB
ground-to-chassis connection. The lower the value of the board ground to
chassis impedance, the smaller the common-mode current on the cable will be.
The key to the effectiveness of this approach is achieving a low impedance (at
the frequencies of interest) in the PCB-to-chassis connection. This method is
often easier said than done, especially when the frequencies involved can be in
the range of hundreds of megahertz or more. At high frequency, this implies
low inductance and usually requires multiple connections.
Establishing a low-impedance connection between the circuit ground and the
chassis in the I/O area is also advantageous with respect to radio frequency (rf) immunity. Any high-frequency noise currents induced into the cable will be
conducted to the enclosure, instead of flowing through the PCB ground.
[10] Page 625
A major source of radiation from electronic products is due to common-mode
currents on the external cables. From an antenna theory perspective, a cable can
be considered as a monopole antenna, with the enclosure being the associated
reference plane. The voltage driving the antenna is the
common-mode voltage between the cable and the chassis. The reference for
the cable radiation is therefore the chassis and not some external ground such
as the earth.
Because the potential difference between the cable and chassis should be
minimized, the connection between the PCB ground and the chassis becomes
important. The internal circuit ground should be connected to the chassis at a
point as close to the location that the cables terminate on the PCB as possible.
This is necessary to minimize the voltage difference between the two. This
connection must be a low-impedance connection at radio frequencies. Any
impedance between the circuit ground and the chassis will produce a voltage
drop, and will excite the cables with a common-mode voltage, which causes
them to radiate.
[11] Page 554
