Design Con 2015

Grounding and shielding: No size fits all

-August 01, 2001

Ask six electrical engineers how to correctly ground and shield a measurement system, and you’ll likely get 12 different answers. Sound farfetched? I used to think so, but after talking to engineers who work for manufacturers of data-acquisition products and after reading several papers on the topic, I’ve found that there are a few general techniques that apply to all applications. As with other areas of test and measurement, the procedures you follow will depend upon your application.

Fortunately, you can rely on a couple of truths when connecting sensors to PC-based measurement equipment. For example, shields can improve performance, and ground loops degrade performance. When it comes to connecting sensors and shields to ground, though, you may have to experiment with different techniques. A technique that works in one application may not work in another.

08F4fig1.gif (57799 bytes)
Figure 1. A signal on a shielded cable (yellow traces) contains considerably less noise than a signal on an unshielded pair of wires (red traces). Courtesy of National Instruments.

The plots in Figure 1 show how proper grounding and shielding can improve measurement performance. The plots compare a clean signal (yellow traces) with a noisy one (red traces) in both the time domain (top) and the frequency domain (bottom). For both plots, a sine wave was measured in the presence of a noise signal. In the yellow traces, a grounded shield protects the sine wave from noise.

Noise degrades millivolt-level signals from sensors such as thermocouples and bridges more than it degrades volt-level signals from other sensors. If you must measure a low-level signal, amplify the signal as close to the sensor as possible. You’ll boost the signal well above the noise level.

Unfortunately, the cost of signal conditioning (typically $150 to $200 per channel) often exceeds equipment budgets. So, you must run the low-level signals directly into your instrumentation.

Look at the circuit in Figure 2. It lacks a reference for the sensor. Without a reference, bias currents in the amplifier can cause the sensor’s voltage to rise or fall relative to the amplifier’s analog ground. That voltage change creates a common-mode voltage—one that appears on both sensor lines. Common-mode voltages can exceed the level that the amplifier can reject. When that occurs, the amplifier’s output will move toward either of its power-supply voltage rails. The ADC that follows the amplifier will never receive the signal of interest.


Figure 2. The sensor needs a reference level to negate the effects of the amplifier’s bias current. This configuration also lets the sensor’s wires pick up current from magnetic fields.

Get a reference

To eliminate common-mode voltages, you need a reference level. In some applications, you can connect a high-value resistor (such as 1 MV) from each sensor line to analog ground at the instrument end of the system. These resistors provide a path to ground for bias currents and keep common-mode voltages within the common-mode rejection (CMR) limits of the amplifier. Some measurement cards have built-in 1-MV resistors that connect to analog ground. Check your card’s data sheet for this information. If it’s not available, ask an applications engineer at the manufacturer of the instrument.

The resistors, however, reduce the input impedance of the measurement system. While they may provide a ground reference, they may load the incoming signal, pulling it toward zero volts. Check your signals with and without the ground resistors. If the voltage difference produces unacceptable errors, you need another grounding scheme. Where and how you reference the signals, though, remains open to debate.

Technical literature on grounding often suggests that you connect the low side of your passive sensor (thermocouple) to ground near the sensor. Some engineers, though, suggest that you run a separate line from each sensor’s low side to your instrument’s analog ground. They argue that if you connect the sensor to a local ground, you could easily have a difference in potential between the sensor ground and the instrument ground, which can result in excessive common-mode voltages.

I could find no consensus on which end of the system to connect to ground. Try each end to see which produces the more reliable measurements. You may find little or no difference if the ground potentials at the instrument and the sensor are roughly equal.

Passive sensors generate their own signal voltages, but active sensors such as bridges need an excitation voltage. Assume that the sensor’s power supply has no ground connection such as with a battery or an isolated AC-DC supply. Where should you connect the supply to ground the circuit so that it provides a voltage reference? I asked several engineers this question and again, I received no clear consensus.

All agreed that you can’t let the sensor float—you need a ground reference. But they disagreed on where and how you should connect the supply to ground. Some suggested connecting the low (return) side of the supply to power ground at the sensor end of the wires. They preferred a local connection, particularly if the power supply must drive two or more sensors—a local connection will keep the sensors at the same potential.

Others, though, suggested connecting the power supply’s return to analog ground at the instrument through a dedicated wire. Much of your success will depend on how the building’s ground wires connect between your sensors and your instruments. Unfortunately, you can’t always know how the ground wires connect, so you just have to try each grounding option.

Too many grounds

In some applications, you must connect your sensors to a local power ground for safety reasons. From a safety perspective, you can’t have enough connections to ground. From a measurement perspective, though, two grounds constitute one too many.

If safety precautions require you to connect your sensors and power supplies to a local ground, then do not connect them to ground at the measurement equipment. Never ground your system at more than one point. Doing so will create a ground loop. That loop can conduct current because of magnetic fields that pass through the loop. The loop current, which can reach hundreds of milliamps, is proportional to the loop’s area.

If the ground wires in the ground loop conduct current, it will combine with wire impedance to create a voltage difference, Vcm in Figure 3, between the two grounds in the loop. That common-mode voltage on your signal wires can exceed your amplifier’s CMR limit.


Figure 3. Never connect a measurement system to ground at more than one point. Two ground points create a ground loop that can create an error voltage (Vcm).

Another look at Figure 2 and Figure 3 shows that a sensor and its signal wires form a loop. Just as with ground loops, sensor-wire loops can intercept changing magnetic fields such as those from lights and computer monitors, which will produce current in the signal wires. The current will produce an error voltage at the amplifier. Magnetic fields generally add interference signals at frequencies below 1 kHz because inductance in the wires blocks high frequencies but passes low frequencies.

Because the current in a loop is directly proportional to the loop’s area, you can minimize the current by twisting your signal wires or by using twisted-pair wires—which are readily available from both wire manufacturers and electronics distributors. As a general rule of thumb, one full twist per inch of wire should significantly reduce noise from magnetic fields.

Twisted-pair wires, however, won’t reduce interference from high-frequency electric fields. Interference above 1 kHz often comes from electric fields. Electric fields couple into wires through stray capacitance. Depending on your measuring system’s sample rate, these signals can alias into your measurement’s pass band where you can’t remove them (Ref. 1). To reduce electric-field interference from sources such as microprocessors and broadcasts, you need to shield your signal wires.

Shields improve measurement performance because they provide high-frequency noise signals with a low-impedance path to ground. They also, through mutual inductance to signal wires, act like filters that reduce common-mode currents in signal wires at frequencies that exceed the cable’s shield-cutoff frequency. A cable’s resistance, inductance, and capacitance form an RLC filter, which defines the cutoff frequency, beyond which the shield shunts common-mode currents to ground (Ref. 2).

The lower a shield’s impedance at a given frequency, the better it will shunt signals to ground. Everyone agrees on that. But where and how do you connect a shield to ground to get the best noise filtering? That, I discovered, lends itself to debate.

Just as they do with unshielded wires, some engineers suggest you connect a shield to ground at the sensor end of a grounded sensor. That way, no potential difference can occur between the shield and its signals (Ref. 3). Other engineers, though, recommend that you connect the shield to ground at the instrument end of the wires to reduce any common-mode voltage at the equipment’s inputs. But all agree that you should never directly connect a shield to ground at both ends. You’ll certainly create a ground loop (Figure 4).


Figure 4. Never connect a shield to ground at both ends. Doing so would create a ground loop.

Some measurement instruments contain more than one “ground.” Plug-in data-acquisition cards, for example, provide connections to analog (measurement) ground and to digital ground. You also have a power ground available through the PC’s power supply. Ultimately, the analog and digital grounds must connect to power ground, but the card should connect them at one point only, as close to the computer’s bus connector as possible. A good design will use an analog ground plane under the analog circuits with a small, short connection to power ground. Digital grounds usually connect more directly to power ground through wires separate from the analog ground.

So, should you connect your analog-signal shields to power ground at the sensor, to analog ground at the instrument, or to power ground at the instrument? If you connect the low side of your sensor power supplies or cable shields to ground through the instrument card, you should ground them through the card’s analog ground. All of the engineers I spoke with who work at data-acquisition card manufacturers agree on this point. They say that if you connect your shields to power ground, then noise on the power ground could couple from the shield into your signal wires through the cable’s capacitance.

But if you’re using a stand-alone data-acquisition system, you might want to consider another option. In a paper referring to such systems, Agilent Technologies suggests that you connect cable shields to power ground to keep high-frequency currents in the shield from polluting your analog ground (Ref. 4). Try both techniques to see which works better in your application.

If you must connect your sensors to power ground for safety compliance, then connect your cable shield to that ground and check the performance. That single connection to ground will reduce some noise and may sufficiently reduce noise in low-resolution (12-bit or 16-bit) measurement systems.

For high-resolution (20 bits or more) measurement systems such as DMMs and some stand-alone data-acquisition systems, the Agilent paper says that you should never leave the instrument end of a shield open. If you must connect your shield to ground at the sensor end because of a grounded sensor, then connect the instrument end of the shield to ground through a capacitor. Typically, you’ll find that a ceramic capacitor from 0.01 mF to 1 mF works well. You may have to find the best value by trial and error.

The capacitor shorts high-frequency signals to ground at the measurement equipment but doesn’t create a low-frequency ground loop. See “Caps and Shields,” for experimental results that compare a shield left open at the instrument end with it connected to analog ground through a capacitor. You’ll see a significant difference, even with a 16-bit measurement system.

You can also use shielded, twisted-pair wires to deliver power from a supply to a sensor. Here, you should connect the power-supply cable’s shield to power ground at the supply end of the cable. Leave the sensor end of the power-supply cable open.

Connecting grounds and shields often requires some trial and error until you find what works best. Regardless of the connections you make, you can still improve measurement performance by following some basic rules:

• Use round cables with individual shields for each twisted pair.

• Use shields made from tinned copper rather than from aluminum. Copper is a better conductor, providing a lower impedance path to ground.

• Some sensors have shields, but those shields may not connect to the sensor’s power ground. If they don’t, you should extend your cable shield so it connects to the sensor’s shield. If the sensor has a shield that does connect to the sensor’s power ground, then connecting it to the cable shield will also create a ground loop if you ground the shield at the instrument end of the cable.

• Keep similar signals together. Don’t run signal wires in the same cable with power-supply wires. Don’t run signal wires adjacent to AC power cords.

• Physically separate analog measurement wires from those that carry digital signals. Run them in different cables and keep those cables away from each other. Capacitance in the wires will couple high frequencies from rising edges into your signal wires.

• Connect unused analog inputs to analog ground at the measurement instrument.

• Keep all connections from cable shields to grounds as short as possible. You need the lowest impedance to ground that you can get. T&MW

References

1. Moffitt, Paul, “Proper sampling ensures good data,”Test & Measurement World, July 2001.
p. 115. 

2. “Characteristic Impedance of Cable at High and Low Frequencies,” Belden Electronics Division, Richmond, IN. bwcecom.belden.com/college/college.htm (look in the Technical Papers section).

3. Rich, Alan, “Shielding and Guarding, How to Exclude Interference-Type Noise, What to Do and Why to Do It—A Rational Approach,” Analog Dialogue, issue 17-1, Analog Devices, Norwood, MA, 1983.

For more information

Fowler, Kim “Grounding and Shielding, Part 1—Noise,” IEEE Instrumentation & Measurement. IEEE, Piscataway, NJ, June 2000. p. 41.

Low Level Measurements, 5th Ed., Keithley Instruments, Cleveland, OH, 1998.

Morrison, Ralph, Grounding and Shielding Techniques, John Wiley & Sons, New York, 1998.

“Noise Reduction & Isolation,” Chapter 7 in Signal Conditioning & PC-Based Data Acquisition Handbook, IOtech, Cleveland, OH. www.iotech.com/prsigcon.html.

Ott, Henry W., Noise Reduction Techniques in Electronic Systems, 2nd ed., John Wiley & Sons, New York, 1998.

“Signal Conditioning Fundamentals for PC-Based Data Acquisition Systems, National Instruments, Austin, TX. zone.ni.com/libraries. Click on “Signal Conditioning,” then click on “General.”

Martin Rowehas a BSEE from Worcester Polytechnic Institute and an MBA from Bentley College. Before joining T&MW in 1992, he worked for 12 years as a design engineer for manufacturers of semiconductor process equipment and as an applications engineer for manufacturers of measurement and control equipment. E-mail: m.rowe@tmworld.com.

Caps and shields
In some applications, you must use a grounded sensor. If you use shielded wires with a grounded sensor, connect the shield to ground at the sensor. Grounding the shield at the instrument end would create a ground loop. A capacitor at the instrument end of a shield that connects to measurement ground improves performance because high-frequency signals look for the shortest path to ground. The capacitor blocks DC, thus preventing ground loops.

To prove that adding the capacitor helps, I asked Jim Stevens, VP of hardware engineering at Measurement Computing (Middleboro, MA), to run an experiment. Jim used a 3.3-battery and a voltage divider to produce a 3.3-mV signal. As the Figure shows, he connected the low side of the battery to the cable shield, which he then connected to ground through a power strip. He then took 10,000 samples at 200 ksamples/s with a 1-mF ceramic capacitor in the circuit. Then, he removed the capacitor and took another 10,000 samples. The plots show histograms of the ADC’s bins (raw counts).

The title bars of the plots indicate the number of ADC bins the 10,000 samples fill. In the plot on the left, just 11 bins activated. The plot shows nine lines because two of the 11 bins received too few hits to appear in the plot. (The software adjusts the spacing between bins to fill the plot area, hence the wide spacing between the bins.) Without the capacitor, though, the ADC activated 97 bins in the plot on the right. Because the plot had to fit 97 bins into the same space, it reduced the space between them.

A span of 11 bins represents far less noise than a span of 97 bins. Therefore, you can conclude that the capacitor reduced noise in the system. You may need to experiment with capacitor values between 0.01 mF and 1 mF to get the best results.—Martin Rowe
A capacitor in the circuit (top, switched closed) produces 11 ADC bins of noise (bottom left). With the switch open, the circuit produces 101 bins of noise (bottom right).

Loading comments...

Write a Comment

To comment please Log In

FEATURED RESOURCES