Editor's Note : This lengthy and insightful article is presented in four parts:
In Part 2 we related the impact on signal quality to some possible real-world basic signal and shield grounding choices (both the good and the bad). But, even in that brief discussion, it is clear that simply wiring a sensor to the receiving instrumentation does not guarantee a quality signal path. In Part 3 we continue this theme by extending the discussion to how the signal source (sensor) dictates optimally configuring the receiver (op amp, in-amp) such that signal integrity is preserved. From this we will develop some rules for choosing the appropriate grounding scheme for the cable shield.
To produce accurate and interference-free measurements, the signal path as a whole must be evaluated. This means properly matching of the sensor configuration to that of the receiver. Often is impossible to control whether the sensor is grounded, as, for example, in the case of a pressure transducer screwed into the side of a metal water tank. If the water tank is connected to ground, the transducer's case unavoidably becomes grounded as well. This places constraints on the over-all system's grounding scheme, and may force a less than optimal grounding of shield and receiver amplifier.
The same is often true of the mounting limitations for thermal couples in industrial plant applications. These, and many more examples, require that the engineer is aware of the best way to mate the input amplifier to what ever sensor arrangement he is handed, so as to avoid ground loops and optimize the EMI and RFI immunity of the signal path.
Signal sources vs. measurement systems
Figure 4 showed the pitfall of using a ground-reference signal source, and single-ended measurement inputs with a poor grounding scheme. Note that in Figure 5, we have now used differential inputs (balanced inputs) for the measurement system. Grounded signal sources are best measured with a differential system because no ground current passes down the two measurement wires. However, beware if the common-mode voltage, VCM , exceeds the common-mode rating of the amplifier, signal feed-through will still introduce large errors.
Ensuring proper connections between the sensor and the instrumentation front-end and ground are essential to obtaining accurate measurements.
Rule 1 : If the sensor is floating, the measurement inputs will need a reference (ground)
Rule 2 : If signal source must float, use an instrumentation amplifier, and ground-reference the inputs with large value resistors (>1 MΩ). Keep in mind that for low input voltage offset (VOS ) and good CM rejection, the resistors should match to 1% or better and they should be oriented in the same direction.
Rule 3 : A grounded signal source is best measured with a differential measurement system.
Rule 4 : The in-amp inputs do not need to be grounded through impedances if the source is ground referenced, such as a grounded thermal couple.
These rules will be illustrated in Figures 7 through 11.
More on cable shielding
The success of shielding long cable harnesses, and signal wires against harsh environmental EMI/RFI, is centered on choosing the cable grounding technique that fits the specific application. However, when it comes to connecting sensors and shields to ground, you may need to experiment with different techniques. A method that works well in one situation might not work in another.
One of the hardest-held beliefs among engineers is that, in order to keep from setting up troublesome ground loops, the cable shield should be grounded at one end only, and we have shown a case where this is true. However, later we will show another case where it is desirable to ground the shield at both ends, and in so doing, the disadvantages will be outweighed by the advantage we gain from lower noise at the instrumentation front-end.
Knowing the location of all the grounding points in the system that the cable is connected to is key to choosing the best method (not easily achieved). Since you can't always know how the ground system is connected, you may have to try each grounding alternative.
Single-ended vs. differential signal measurements
If the sensor provides a single voltage output (one signal wire) with respect to ground, it is known as a single-ended output . On the other hand, sensors that supply a two-wire output signal, with the two halves of the signals being of opposite polarity, the sensor is said to have balanced , or differential outputs. The measurement front-ends will be of matching configuration. A balanced measurement system must meet three conditions:
- The signal source is balanced. The upper and lower terminals of the source have equal impedances to ground over the operating frequency range.
- The signal cabling is also balanced. Both wires will have equal impedance to ground.
- The instrumentation front-end, most often an instrumentation amplifier, will be balanced. The two input terminals will have equal impedances to ground, as well as to adjacent circuitry.
Coaxial cable, on the other hand, is not balanced, because the two conductors, the center signal wire, and shield, present different impedances to ground. Thus, although they add a measure of simplicity, single-ended signal paths, with their use of coaxial cables, are more susceptible to interference and signal degradation than differential systems, and thus are not appropriate when sensor and instrumentation are widely separated, and/or are in high EMI surroundings. The problem is that the shield is now part of the signal path, and any noise voltage developed across the shield will fully add to the sensor signal. Therefore, signal connections from sensors to data acquisition instrumentation, are almost always implemented with a balanced signal path.
Grounded single-ended source and single-ended measurement
As an illustration, a single-ended measurement system is shown in Figure 6 , the cable shield is the return path for the single-ended sensor signal. It is common practice to ground the instrument side of the cable shield to the instrumentation/measurement system ground. As can be seen, a ground loop will be set up if there is a difference in the two ground potentials due to ground noise. This ground noise will add directly to the signal current and completely corrupt the nominal sensor signal. Under these circumstances, for example, ground noise could be superimposed on a sensor signal and be read by the instrumentation as the sensor output.
Additionally, RFI can induce voltages on the shield if the two ends are at different potentials, that then couple to the signal wire These induced voltages, on the outside surface of the shield, will produce CM noise currents on the inside of the shield, which will then mix with the signal return currents.
Figure 6: Using a grounded signal source with a single-ended measurement system has set up an inevitable ground loop. Here the choice of a coaxial cable allows the return signal current to flow in the cable shield. Noise voltage, vn, appearing between the two grounds will be directly imposed on the signal sensor output, vs , as will low frequency RFI-induced shield noise.
(Click on image to enlarge)
Grounded source and differential measurement
By eliminating the coaxial cable, and using a balanced or differential input approach for the instrumentation (Figure 7 ), the loop is completed without signal currents in the shield. Any noise voltages on the shield set up currents that are directed to ground, not to the signal path.
Figure 7: Grounded sensor with fully differential instrumentation. Any ground voltage impressed across the shield will have much less effect on the sensor output owing to the balanced nature of the circuit.
(Click on image to enlarge)
The use of twisted pair cable is the key to insuring that the noise and cross-talk coupling is induced equally in both signal lines of a two-wire cable. The desired sensor signal generates equal but opposite currents on a balanced pair of wires while noise that is picked up is typically induced equally in both signal lines. Since the differential amplifier is designed to amplify only the differential mode signals between each line and attenuate the common mode signals on each line, it can be crucial to use an amplifier with a high common-mode rejection ratio (CMRR). The best instrumentation amplifiers today have a typical CMRR of at least 100 dB, to beyond 100 kHz. For example, if a common-mode signal and a differential-mode signal of equal amplitudes appeared at the inputs of such an amplifier, the noise signal would be attenuated below one LSB if the signal were digitized by a 16-bit A/D converter, and thus would not affect the accuracy of the conversion.
However, above about 100 kHz, the CMRR rolls off and the common-mode noise is converted to normal-mode noise. Whatever the cause, once interference noise becomes a normal-mode voltage, the only way to eliminate it is by filtering.
Floating (isolated) source and differential measurement
With an isolated sensor and differential inputs (Figure 8 ), the shield should be grounded at the instrumentation end. This gives the lowest noise level appearing at the amplifier inputs.
Figure 8: Floating sensor/non-ground referenced measurement amplifier (in-amp)
(Click on image to enlarge)
Since the shield and the instrumentation are grounded at one point, there is no ground loop. In this configuration, the amplifier inputs must be provided a DC reference. Otherwise, if there is no path for input-bias return current, the inputs will drift to the power supply rails. A ground reference connection is made to the midpoint of the sensor, which provides a reference point for the in-amp front-end.
Because some sensors, such as thermal couples, must be grounded (established externally by the sensor itself), the situation starts to get more complicated. Data acquisition instrumentation manufacturers have done detailed studies of the effects of different grounding schemes, and have come up with some surprising results.
As a rule, in cases of grounded sensors, the shield is grounded at only one point, usually to the sensor end, as shown in Figure 9 . An alternative is to ground the shield at the instrumentation end of the cable.
Figure 9: Grounded sensor with shield grounded at sensor
(Click on image to enlarge)
This is shown in Figure 10 . In both of these configurations, there is no ground-loop. If we connect the shield at both ends, as in Figure 11 , we seem to have created a ground-loop. Nevertheless, in nearly all cases, this configuration gives the lowest noise, and is the recommended technique in most applications. Tests by data acquisition card suppliers have shown that this method gives far superior results to either of the configurations with one end of the shield open. In fact the circuit in Figure 11 has a demonstrated noise immunity of at least 40 dB better than the last two methods.
Figure 10: Grounded sensor with shield grounded at instrumentation end.
(Click on image to enlarge)
How can a ground loop give better noise performance? The key is that no signal current flows through the shield . And, the shield being grounded at both ends reduces the standing wave that appears on the open shield end, and allows all of the shield current to drain to ground.
Figure 11: Grounded sensor with shield grounded at both ends.
(Click on image to enlarge)
There is one important caveat to using this technique, however: If the two end ground points are at considerably different voltage (i.e. outside the common-mode range of the amplifier) the double grounding cannot be used. In that case, some form of isolation block will need to be used. Extensive literature exists on the subject of isolation-barrier techniques.
A final point on ground loops and shielding
Good grounding practice states that multi-point grounding is necessary for high frequency interference. Unfortunately this creates a low frequency ground loop. On the other hand, single-point grounding breaks the low frequency loop but is useless at high frequencies.
This may present a dilemma, which can be explained as follows: signals from many sensors are very low frequency relative to the RFI on the signal lines.
For example, pressure transducers, strain gages, thermal couples, and the bio-electric signals originating from patient monitoring, tend to be low frequency (i.e., less than about 200 Hz) thus it is a relatively simple matter to filter the RFI on these signal lines without compromising the quality of the desired signal. One exception is the 50 to 60 Hz frequency of power lines If useful sensor measurements are likely to include signal frequencies above that of the power lines, and it is possible that ground loops could be become a problem, you may make a direct ground connection at the sensor, but at the instrumentation side tie the shield to ground through a high frequency capacitor.
This will break the low frequency ground loop while keeping the multipoint ground connection intact (the shield is open at low frequencies and shorted to ground at HF). Typical capacitor values are 0.01 to 0.1 μF. Keep in mind, however, that the capacitor should have a resonant frequency well above the highest frequency in the interference signal. Above resonance, capacitors become inductive and are useless. The larger the capacitor the larger its parasitic inductance and the lower its resonant frequency.
(to be continued. . . . )
About the author
Jerry Freeman is an amplifier applications engineer at National Semiconductor Corp., Santa Clara, CA. He received a BSEE from Heald College of Engineering in 1961.
Editor's note : If this article was of interest to you, also check out:
“Understanding noise optimization in sensor signal-conditioning circuits (Part 1a of 4 parts)”,
by Reza Moghimi, click here; note that Parts 1b, 2a, and 2b are linked.