Grounding and Noise Coupling Issues in Automated Measurement Systems
by Joe Czapski, rev. Apr. 2000

Noise Problems and Document Scope

In an automated measurement system, noise problems take the form of excessive variability of samples or data points. One can see the noise magnitude by taking the standard deviation of a set of samples, by plotting a histogram, by plotting measured values vs. sample number (time), or by plotting the power spectrum (FFT -- showing the frequency content of the noise). One tries to make the noise insignificant by averaging enough samples (noise drops by a factor of root N). But, commonly, one is taking measurements vs. quickly swept parameters, and there are only n milliseconds to acquire each data point, not enough time to do significant averaging.

The best solution: fix your measurement system -- or preferably design it right in the first place -- to have as little noise as possible. There are two noise areas to consider. One area is component noise -- noise of A/D converter chips, op-amps, transistors, resistors -- which requires the circuit designer to choose his parts and values carefully. The other area, comprising the vast majority of noise problems, is noise coupling, which is what this web page is about. Noise coupling is an unwanted signal from some source adding into a wanted signal with enough intensity to cause a problem. The source of the unwanted signal may be an alien electrical beast or may be, more commonly, another part of the measurement system. A major contributor to noise coupling problems is the design and layout of ground and return paths.

In the guidelines for grounding and coupling listed further on, little or no supporting explanation is given for each item. It is assumed that the reader has a firm grasp of circuit and electromagnetic concepts. Please contact the author via e-mail or phone with questions about any of these guidelines.

The Automated Measurement System

An example of an automated measurement system appropriate to this discussion is: a controlling PC (personal computer), analog and digital I/O boards in the PC, interface circuit boards external to the PC, sensors & transducers, the DUT (device under test), cables and connectors linking everything, and the chassis or enclosures.

Designing for low noise

The best time to combat noise problems is in the design phase, before the system is built. First, one should refresh oneself on grounding and coupling principles so everything one needs to look for is foremost in ones mind. Then go over every signal and trace its signal/return current loop path. Examine its sensitivity, bandwidth, and transients. Examine its source and load impedances. Examine its length and proximity to transient or sensitive signals. Pay extra attention to each signal's return connection, any ground connections, and cabling structure.

Grounding Principles

"Ground" in this discussion is reserved for earth ground, or the single connection to earth ground, only. Other nodes commonly called "ground," such as 0V planes, buses, and cables, are referred to here as "return." The frequent use of the term "ground" for return lines may contribute to inexperienced engineers' belief that these lines sit solidly at earth ground potential. Here are some grounding principles:

• One and only one earth ground connection point for the system. Main chassis should be connected at same point. With most benchtop automated measurement systems, the PC's earth ground (the ground line in the power cord) must be the single connection for the system. All other power supplies external to the computer must be floating.
• If single-ended return pins appear in external connectors going to other systems (e.g., for RS-232 communication), tie the connector "ground" pin to earth ground at the connector. This arrangement forces the single earth ground connection point to be near the external connector.
• Electrically isolate electronics from chassis, brackets, plumbing, etc. except at the single earth ground point. All brackets and plumbing should be electrically connected to chassis.
• No floating metal allowed in the system, unless specifically required as magnetic shielding between two circuits.

Coupling Mechanisms -- Causes and Cures

Conductive coupling

Conductive coupling is caused mostly by shared return paths for signals, especially high current, high current switching, or fast current transient signals. To minimize conductive coupling:

• Minimize sharing of return paths. That is, make separate return paths for separate circuits or components.
• Use lowest resistance (for high currents) and lowest inductance (for high dI/dt's) as practical for return paths.
• Use very low impedance nodes (planes) for DC supplies and their returns, so that varying supply current draw for one circuit will not couple into another circuit sharing the supply. A local bypass capacitor between the supplies and returns will keep high-frequency current loops local to the circuit.
• Lengthen rise and fall times.

Inductive coupling – via magnetic field

Inductive coupling is caused by fast current transients (high dI/dt), and is commonest in low source impedance, high load impedance circuits. To minimize inductive coupling:

• Minimize loop areas of signal/return paths. Use net-zero-current cable bundles (twisted or coax). Use return planes or paired signal/return traces. Return current in a plane will stay right under the signal trace for frequencies >10kHz, but will fan throughout the plane for frequencies <1kHz.
• Move different loops farther apart from each other.
• Lengthen rise and fall times.

Capacitive coupling – via electric field

Capacitive coupling is caused by fast voltage transients (high dV/dt), and is commonest in high source impedance, high load impedance circuits. To minimize capacitive coupling:

• Minimize line length and move lines farther apart.
• Use electric-field shield (one end, or one point, grounded).
• Instead of twisted pair, coax cable is good, shielded twisted pair is better, and triax is best.
• Lengthen rise and fall times.

Electromagnetic field coupling

Electromagnetic fields are almost never the noise problem in automated measurement systems. The main reason is that the typical system's circuits have a maximum bandwidth of about 3 MHz, and no source within less than 60 feet of your system will be able to electromagnetically couple, because the distance is "electrically short" compared to the 100 meter wavelength of a 3MHz wave.

For those systems with bandwidths in the 100's of MHz, or for those next to TV and radio stations, you can't fool around. Every part of the system must be enclosed. Your chassis needs to be sealed, with no gaps but the tiniest of pinholes, and all cables need to be in conductive conduit, with EMF gaskets sealing them to the chassis.

Inter-Board Signal Connections

When you have to run signals between circuit boards in your system, it's best to make them differential, with a differential transmitter circuit on one board and a differential receiver circuit on the other. The signal and return (+ and -) need to run together in the cable & connectors and should have equal and opposite currents.

Many designers feel obliged to run inter-board signals single-ended, mainly for simplicity. In these unfortunate cases, you can follow these guidelines:

• Connecting the return on both ends causes the low frequency return currents of all the circuits on a board to partially route through the other board, making a large loop area path, possibly causing conductive or inductive coupling.
• For low-bandwidth signals, do not connect the return (return current will flow back through the power supply). Coax shield can be connected on one end as E-field shield. Filter low-bandwidth signals, as the large signal/return loop area may pick up magnetic fields.
• For high bandwidth or transient (switching) signals, connect the return through a series capacitor (2uF ceramic usually fits) to route high-frequency return current back along the cable. Reduces high-frequency magnetic field emissions from a large signal/return loop, and reduces overshoot in transient signal edge.
• Neat fix for low-bandwidth signals: include signal cable in power bundle, all the way back to power supply then off to the other board. Minimizes signal/return loop area.

Intra-Board Signal Connections

In reviewing your circuit board layout, look for:

• length and proximity of coupling-sensitive pairs of traces.
• large signal/return loop areas, such as in +/- supply buses routing.
• single-ended traces crossing ground-plane gaps. High frequency return current can’t come back under the trace along its whole length, causing a lateral signal/return loop.

©2000 AutoMeasure, an Arlington, Massachusetts company. All rights reserved.