Data acquisition equipment and software are experiencing the same type of transformation that has occurred with personal computers and generic peripherals. Higher performance, simplified operation, and lower cost now provide inexperienced people access to tools previously reserved for technically competent users.
There is, however, a downside. The downside is that new users may not have the time or expertise to master every technical detail. As a result, they run a greater risk of assembling data acquisition systems that violate proven rules of measurement theory.
Understand Equipment Limits
Next, check ADC specifications to determine the measurement uncertainty you can expect for a selected range. Accuracy figures represent accumulated errors in gain, offset, drift, nonlinearity, and noise. Manufacturers express accuracy in a variety of formats, a common one being gain error plus or minus some offset. Typical format examples are 0.01% ±10 µV or 0.024% ±3 least significant bits (LSBs).
On more sensitive ranges, offset error often represents an increasing portion of the total error. For example, a 16-bit multifunction board offers an accuracy of 0.0125% ±500 µV on its ±10-V input range and a resolution of about 300 µV per bit. A 500-µV offset corresponds to one to two ADC counts.
On the board’s ±0.0125-V input range, however, accuracy is 0.020% ±10 µV, and resolution is about 0.38 µV. The offset figure now represents approximately 26 ADC counts. Someone new to data acquisition might consider 26 counts to be excessively noisy when, in fact, the board is operating properly.
You gain little from a high-resolution ADC board if the application can’t take advantage of the resolution. Some sensors provide readings accurate to only 1 or 2%. A 12-bit ADC resolves one part in 4,096, which corresponds to 0.024% of full scale. A 16-bit ADC has no practical advantage here and actually might impose limitations such as a lower maximum sample rate.
Consider the Complete Measurement Circuit
During a measurement, the instrument input absorbs a small bias current that the source must be able to generate. The interconnect cabling is an essential part of this circuit and can introduce resistance, capacitance, and inductive effects that depend on length, gauge, composition, routing, and environment. Figure 1 shows a dashed box around a portion of the signal path that represents the sum of these effects.
The impedance of the source in an ideal measurement circuit is zero while the input impedance of the instrument is infinite. In practice, the source impedance of most transducers and signal sources is less than a few hundred ohms; the input impedance of most instruments and data acquisition boards is tens or hundreds of megohms—close enough to ideal for most applications, but not all.
The pH sensor is an example of a very high-impedance, low-level sensor that requires an instrument with an input impedance well into the gigohm range to avoid signal loading. As a result, to make an accurate measurement, you need to know the nature of the signal source and then select a measurement device with an appropriate input impedance. When an instrument’s input impedance is too low with respect to the source, readings lower than the actual value can result.
For high-speed, rapidly changing signals, circuit inductance and capacitance also can be serious obstacles to accuracy, even if signal and instrument resistances are properly matched. Generally, signals from high-impedance sources take longer to stabilize at the instrument because the signal’s limited current requires more time to charge cable capacitance. Likewise, inductance in the circuit progressively impedes the measurement of higher-frequency signals.
For some types of signals, cable resistance alone can affect accuracy, regardless of measurement speed. Powered resistive sensors, such as resistance temperature devices (RTDs) and strain gauges, usually are located some distance from the data acquisition equipment. In a simple two-wire measurement setup, both the sensor power and the signal flow through the same conductors. The resistance of the intervening cable can introduce serious error because the instrument sees this resistance as a property of the sensor.
For example, the resistance of 100 ft of 28-gauge wire is about 7 W, which corresponds to an error of about 18°C for a 100-W RTD. Using 12-gauge wire reduces the cable resistance to less than 0.2 W and the error to less than 1°C.
As a general rule, use the shortest, heaviest-gauge cable practical to minimize cable resistance. For critical resistive sensor applications, use a four-wire measurement setup (Figure 2, see the December 2000 issue of Evaluation Engineering). The four-wire technique uses separate sense leads and source leads to power and read the transducer. When the sensor is powered by a current source, resistance in the source circuit has no effect on current through the sensor. A high-impedance meter in the sense circuit will draw virtually no current, so any resistance in the sensing leads has a negligible effect on measurements.
Manage External Noise
The effects of radiated energy decrease as the square of the distance, so relocating wiring or sensors a short distance away can reduce noise pickup significantly. Also, coupling electromagnetic noise to conductors increases with the area enclosed by the conductors.
A classic twisted pair of unshielded wires picks up less noise than the same two wires spread far apart. Coiling extra wiring into neat bundles also can inductively couple noise to the signal, even with shielded cable.
For that matter, not all shielding is equally effective. Cable shields vary from a single, relatively open, spiral-wrapped wire layer to multiple layers of tightly braided mesh. Tighter braiding and multiple layers usually are more effective, especially at higher frequencies. For critical applications, ask cable vendors for detailed specifications about noise attenuation vs. frequency for different types of cable.
Where external noise is random in nature and the data sample rate is relatively slow, a simple averaging method using 50 to 100 samples can statistically cancel noise from a reading. It’s also possible to use an RC input filter to eliminate higher-frequency noise components.
The cable-effects box in Figure 1 is the equivalent of a simple RLC low-pass filter. The inductor can be eliminated and the resistor and capacitor values chosen to provide the desired high-frequency rolloff. Some measurement instruments, notably bench-type, have ADCs that use line cycle integration. By nature, this method minimizes noise from AC power lines.
Short Circuit Ground Loops
The resistance of cable shields, power cords, and pseudoground points can cause voltage gradients and current flow through the ground side of a measurement system (Figure 3). One consequence is the buildup of common-mode voltages at the input of the ADC board, which can result in noisy readings, reduced input range, or inoperative analog inputs. The solution is to establish a single ground point for all signal sources and measurement instrumentation. Where possible, shielded cables should be grounded only at one end.
Eliminate Self-Generated Errors
Phenomena such as thermal, triboelectric, and piezoelectric effects can generate currents or voltages in the signal path that cannot be easily distinguished from the signal. Table 1 describes these error-generating mechanisms in their approximate order of magnitude along with some methods for minimizing or eliminating their effects.
Table 1. Types of Self-Generated Measurement Errors
| Effect | Description | Magnitude | Counter- measures |
| Thermal | Voltage generated by the junctions of dissimilar metals varies with temperature | 10-4 to 10-2 V | • Avoid using connectors and conductors made of different metals; use correct thermocouple cables • Keep temperatures of sensors, data acquisition equipment, and cabling uniform |
| Electrochemical | Current generated by ionic chemicals, humidity and other contaminants on conductors | 10-13 to 10-8 A | • Avoid touching circuit areas of data acquisition cards and sensors • Thoroughly clean flux and contaminants from user-built circuits • Protect equipment from moisture and atmospheric contaminants |
| Piezoelectric | Currents generated when certain insulating materials are subjected to mechanical stress | 10-15 to 10-8 A | • Minimize mechanical stress in cable connectors and insulating materials |
| Triboelectric | Currents generated by friction between cable conductors and insulators, particularly coaxial cable | 10-16 to 10-13 A | • Use low-noise cable • Tie down cables, wires, and components • Remove sources of vibration |
The magnitude of most of these errors is extremely low, on the order of about 10 nA. However, 10 nA is sufficient to generate millivolts of error on a system having a 100-MW input impedance, which is typical of many DMMs and data acquisition boards.
In the case of thermal EMFs, the junction of dissimilar metals is, in effect, a thermocouple. This junction can introduce errors into a temperature measurement that are similar in magnitude to the temperature of interest.
Summary
Assembling a data acquisition system presents several possible areas for error. In many cases, the solution consists of applying a few qualitative rules about equipment and measurement theory:
- Determine realistic instrument accuracy, resolution, and repeatability in your specific application.
- Understand the dynamics of the measurement situation, including how the signal source, conductor path, and instrument input interact.
- Position equipment and conductors to minimize noise pickup. Review cable shielding to assure it will be effective against noise.
- Design data acquisition setups with a single, centralized ground to reduce the risk of ground loops.
- Use the shortest, heaviest-gauge hook-up cabling practical. Where necessary, use four-wire measurement techniques to eliminate the effects of lead resistance.
- Be aware of thermal, piezo, and triboelectric effects. Design data acquisition systems to eliminate error-generating mechanisms.
When these suggestions are proactively built into your standard approach to data acquisition, they will reduce debugging time and provide a higher confidence level in your results.
About the Author
Mike Bayda is a data acquisition marketing engineer at Keithley Instruments. He has 10 years of experience in test and measurement and previously worked as a test engineer at NASA Lewis Research Center. Mr. Bayda holds an M.S.E.E. from Cleveland State University and a B.S. in physics from University of Orleans, France. Keithley Instruments, 28775 Aurora Rd., Cleveland, OH 44139, 440-248-0400.
Published by EE-Evaluation Engineering
All contents © 2000 Nelson Publishing Inc.
No reprint, distribution, or reuse in any medium is permitted
without the express written consent of the publisher.
December 2000