At times, data acquisition may seem much more like juggling than simply balancing because there are so many variables to consider. So, to get a better picture of some of the problems you may encounter, here is an interesting example: The number of channels, the sample rate, the signal bandwidth and dynamic range, resolution, and accuracy all are involved.
You have a challenge on your hands. The first full-size prototype of your company’s variable-height, self-extending passenger jetway is being built, and it’s your job to instrument and test it. There are interacting electrical, hydraulic, and heating, ventilation, and cooling (HVAC) systems to provide passengers a continuous environment from the plane to the terminal. But most of all, it’s a telescoping structure that deflects under load and must have a very long, safe service life.
The first step is to consider all the systems and subsystems and the types of tests that must be performed. Perhaps some of them can be avoided: the auxiliary generator system is the same as one already in use which has thousands of hours of satisfactory service data. There also is sufficient history of the traction system performance.
Which tests are unique to this project? Recent refinements of computer models have shown a tendency for panels to resonate as the structure extends. The whole jetway also may oscillate in response to a troop of soldiers marching through it. So, a large-scale test using many strain gages is in order.
What are the implications of this requirement? Obviously, many points must be monitored. Strain gage placement should be coordinated with the computer simulation engineers. With their guidance at an early stage, the results of the structural testing should correlate with the original computer modeling.
Because hundreds of sensors will be involved and because the required frequency response is no more than a few hundred hertz maximum, some type of distributed, multiplexed system is suggested. A number of such data acquisition systems are available commercially which provide local bridge completion and excitation for the strain gages.
Local buffering and digitizing are desirable because dealing with the low-level signals at the source reduces noise pickup. A digital connection such as Ethernet also can eliminate most of the remote-to-central cabling in a large, distributed test setup. And because it is so widely used, Ethernet seems to be the logical choice.
Thinking It Through,
Carefully, Again
What pitfalls have you overlooked? In the first place, you’ve oversimplified the problem and come to a conclusion too quickly. Surely, the overall goal, which hasn’t been stated, is to centrally acquire time-related data from the entire structure.
Analysis involving all the data only can be undertaken if the time reference is consistent throughout. How much timing margin is there? Using the usual engineering 10:1 rule of thumb, if a 10-Hz response is desired, the timing of all the data must align to within 10 ms. For a 100-Hz response, the minimum needed to record any impulses and harmonics, alignment to within 1 ms is required.
Although the absolute timing of an Ethernet link is indeterminate, a precise time reference can be added to the data within each Ethernet packet. The central computer will recover data from the packets and transfer it to a database according to the embedded time code. As a result, collisions and retransmissions become totally transparent. Of course, this assumes that sufficient throughput can be achieved.
Now we seem to be getting somewhere. The absolute timing accuracy of individual data points can be as close to a centrally distributed timing reference as required. What have we given up, and are there other problems lurking to trip us up later?
We have traded some system latency for timing accuracy because of the type of digital communications link we chose to minimize noise pickup. We’ve gained flexibility and robustness. Was it a good trade-off? Could we have chosen a different type of digital communications bus better suited to real-time data acquisition? Is true real-time important to us anyway?
With regard to the other lurking problems, details of the actual signal conditioning and conversion have yet to be considered. A bridge completion circuit was mentioned, so decisions must be made about whether a full, half, or a quarter bridge circuit is best for this application.
A quarter bridge uses just one strain gage and three balancing resistors. It has the lowest output of the three configurations. It is the most sensitive to noise and, together with the half bridge, suffers from greater bending-induced nonlinearities than a full bridge.
A half bridge uses two strain gages and is less susceptible to noise. But both quarter and half bridges are temperature sensitive because the balancing resistors are made from different materials than the gages themselves.
A full bridge has the highest output, making it the least susceptible to variations in the bridge excitation voltage and noise. But four strain gages are required.
Is there a need to use a nonstandard strain gage resistance value? And how stable and noise-free does the bridge excitation supply need to be? Will mx + b type linearizing be required? Is it desirable to store raw data in the data- base, or would it be advantageous to have a digital signal processor (DSP) in the system that can perform conversion directly to engineering units including calibration factors?
These are questions that need to be answered. The answers, in turn, will raise more questions. For example, if you need to acquire data from many hundreds of points, you may wish to consider a VXI system rather than a PC-based one. Making this high-level decision has major consequences.
Acquiring the Data
Can a multiplexed analog-to-digital converter (ADC) be used, or is an ADC per channel needed? Because we think that a 100-Hz bandwidth is required, a very low-cost, delta-sigma, high-resolution converter may be a good choice on a per-channel basis. Another type of ADC will be better suited in a multiplexed, multichannel approach-perhaps a voltage-to-frequency converter or a single- or dual-slope integrating converter.
According to Paul Worrell, product marketing manager at Hewlett-Packard, “For best results, the delta-sigma ADC should be dedicated to one channel. Or to compensate for pipeline delay, the system must allow time to clear information out of the ADC before another channel can be correctly measured. Scanning data acquisition systems that have a measurement rate below 100 S/s typically will use an instrument-grade delta-sigma converter. Systems that run faster than 100 Hz typically will have a successive approximation register (SAR) ADC.”
Voltage-to-frequency converters and integrating ADCs are good choices for our jetway application. Like the delta-sigma converter, they reject noise much better than a successive approximation or flash converter. Integrating and voltage-to-frequency converters average the effects of noise over the conversion cycle and can be especially effective at reducing AC power supply pickup.
In contrast, flash converters and successive approximation converters, preceded by a sample-and-hold (S/H) circuit, sample the input voltage at a single point in time. If there is a noise spike on the signal right then, it adds to the signal and appears full size at the converter output.
“For low-level and cost-effective solutions up to 10 kHz, it’s hard to beat the voltage-to-frequency converter,” said Walt Maclay, president of Strawberry Tree. “Although the successive approximation converter technology is available at even lower prices, the voltage-to-frequency converter provides many advantages. It can have a resolution of more than 20 bits at 10% of full scale and simultaneously offer very low noise.”
Also commenting on the suitability of different kinds of ADCs defined by speed requirements was Jim Borton, data acquisition product manager at Keithley Instruments. “At lower speeds, integrating ADCs are adequate up to about 10 kHz and even work well for low-level signals if other circuitry is designed properly. This also tends to be the most cost-effective design. Above 10 kHz, successive-approximation ADCs provide better performance at moderately higher cost.”
Data-sampling considerations involve antialiasing and noise band-limiting filtering. Digitizing the desired signal, no matter how it is accomplished, will produce aliasing if the signal contains frequencies above the Nyquist frequency-half the sampling rate. For example, Cauer elliptical filters have very fast transitions from passband to stopband so you don’t have to allow an excessive margin between the 100 Hz you require and the Nyquist frequency that you want to be as low as practical.
What sampling rate is needed? Again, the 10:1 rule of thumb says that a 1-kS/s rate is needed to support a 100-Hz bandwidth. This is conservative but will work well and can easily accommodate the required antialiasing filter. Should you be using such a sharp cutoff filter? An eight-pole Bessel filter with a -3-dB, 100-Hz corner frequency will have a slower roll-off than the Cauer elliptical filter, but it will not introduce ringing on the edges of a pulse. The Cauer elliptical filter will. The Bessel filter stopband response still will be down by 65 dB at 500 Hz, which is probably adequate.
Transferring data from hundreds of channels sampled at 1 kS/s may not be as easy as first thought using Ethernet. Robert Winkler, product marketing engineer at Intelligent Instrumentation, said that forming acquired data into packets and adhering to the necessary protocols limited data transmission rates in his company’s EDAS systems to about 8,000 16-bit words/s.
Unless you can improve upon the speed of the packetization process, you will need to perform packetization separately for each group of eight strain gages. The assumed 10Base Ethernet link itself is not the bottleneck, so the packetized outputs of many groups of gages can feed onto one common network.
If each channel in a multiplexed system needs to be sampled at a 1-kHz rate, the minimum sampling rate of the converter must be at least n x 1 kHz, where n is the number of channels being multiplexed. Multiplexers don’t switch instantaneously, so the actual maximum speed of the ADC will have to be higher to allow multiplexer settling time and still achieve the n x 1,000 S/s. Also, if nothing else is done, the overall sampling-time skew among all n channels will be 1 ms.
If the ADC can sample at a much higher rate and the multiplexer also switches more quickly, the required n samples can be taken in a burst. The skew among the n channels could be as small as several tens or hundreds of microseconds if the ADC converted quickly enough. The n-sample sequence would be initiated again every millisecond, but the alignment of the data samples would be much better.
The ideal is simultaneous sampling, a distinct advantage of an ADC per-channel system. Some multiplexed systems use separate S/H circuits per channel to eliminate sampling skew, so this is yet another choice that must be evaluated.
Noise filtering is a separate subject. If a multiplexed successive approximation or flash converter is being used, the signal presented to the ADC must be free of noise. These types of ADCs will not reduce whatever noise may be present at the input. Amplifying the signals before they are filtered will reduce noise better than the other way around.1
Make sure you have arranged the system gain to be optimum for both the multiplexer and the following ADC. For example, there probably will be an instrumentation or differential amplifier ahead of the multiplexer on a per-channel basis to convert the strain-gage bridge output from differential to single ended or simply to buffer the few hundred ohm bridge output resistance.
You also may require this amplifier to provide gain ahead of the multiplexer. And gain may be required after the multiplexer. To achieve the highest resolution, drive the ADC with the correct full-scale amplitude. This level may be high enough to overdrive the multiplexer and cause distortion.
You can resolve this problem by selecting a multiplexer with a voltage range that exceeds the ADC full-scale input or by inserting a gain stage after the multiplexer. Second-order effects such as multiplexer settling-time dependence upon the signal level still could benefit from careful apportionment of system gain. Yes, that’s another factor to check.
The interactions among ADC, amplifier, filter, and S/H are encompassed in the single effective number of bits measure of a digitizing system’s performance. How closely this number agrees with the nominal number of bits used to describe the system is a good performance indication.
Conclusion
There are many trade-offs to consider in developing the solution to a typical data acquisition problem. And in the case of the new jetway, we only have considered some of the vibration testing requirements. There still are electrical and HVAC tests to design, plan, and review.
For vibration test data acquisition, we have developed a possible solution. Using Ethernet, data from groups of eight strain gages will be sent to a central computer to be stored in a database. Bridge completion circuitry, buffering, and filtering will be provided for each strain gage. A multiplexed voltage-to-frequency ADC will be used to achieve noise reduction at low cost.
Is this the only solution? Absolutely not. Is this the best solution? Maybe.
The role of the test manager is to run through virtually all of the possible test scenarios. We have only considered one. A spreadsheet can help keep track of requirements, decisions, trade-offs, and conclusions. For large projects, a planning program can account for personnel, activities, and specification changes and determine the critical path.
The manager also must be clear in his own mind that he is trying to solve the instrumentation and measurement problems at hand and not trying to design and build an instrumentation system. Although he needs to understand all the trade-offs in detail, in most cases, he will want to buy commercially available equipment rather than make his own. Realizing this at the beginning of the job will help him constrain the technical solution to proven, reliable hardware and software.
No doubt, he will begin the job influenced by the data acquisition systems already at his disposal within the company. He also will have to justify to his managers how the present investment in equipment will serve the company in the future.
Many new opportunities exist today for improved performance and lower costs. According to Ed McConnell of National Instruments, “Consumer electronics and the communications industry are driving the performance, power, and price of ADCs. The test, measurement, control, and data acquisition markets benefit from the new, high- volume ADCs that were designed initially for use in the higher-volume consumer and electronics markets.”
Reference
1. Jacob, G., “A Primer on Signal-Conditioning Issues,” EE-Evaluation Engineering, November 1996, pp. 20-24. NOTE: This article can be accessed online. Select EE Archives and use the key word search.
About the Author
Tom Lecklider is a Technical Editor for EE.
Click here to learn about new Data Acquisition Products.
Return to PC-Based Test Online – Return to EE Home Page
Published by EE-Evaluation Engineering
All contents © 1999 Nelson Publishing Inc.
No reprint, distribution, or reuse in any medium is permitted
without the express written consent of the publisher.
May 1999