A Quick Guide to Signal Quality

Preserving signal quality in measurement applications using off-the-shelf data acquisition equipment–for signal bandwidths up to 1 MHz and sampling rates up to a few million per second–can be challenging. These systems are inclined to pick up all signals present, both the ones you want to measure and the ones you don't, making accurate measurements difficult.

To effectively address these challenges, start with an assessment of need. Suppose you can achieve 1 part/1,000 measurement error, about 10 bits precision. Is this success or failure? It depends on your requirements.

You won't get good information from bad signals. A 16-bit conversion with 2 low-order bits of random noise or a 24-bit conversion with 10 low-order bits of random noise still means only 14 reliable bits. Once you reach the noise level, the rest seldom makes much difference.

So, obtaining good measurements is primarily about obtaining and preserving high-quality signals. The economics are about selecting equipment that operates at the level you must achieve.

There are two ideal conditions for making measurements: low drive impedance, which is little affected by noise sources; and high measurement impedance having minimal interaction with a signal. It is difficult to get good performance if you must compromise on both conditions.

The factors limiting measurement performance include the following:

• Measurement systems deal with outside phenomena and cannot be completely isolated.
• Sensitive, high-impedance converters respond to the slightest disturbances.
• Signals from different sources can interact before measurement.
• The measurement process slightly alters what is measured.

All of these factors come into play all of the time. The general plan is to apply the most cost-effective measures to the most likely problems first.

Signal Source

Start with the signal source: the sensor and its connections. To deliver the sensor signal accurately, the source should drive hard and transfer across only a short distance. A strong signal is characterized by a low drive impedance.

Physical constraints often will determine where physical signals must be routed while the device characteristics will limit the amount of signal drive available. It is difficult to maintain signal quality when both the drive power and the transfer distance are compromised. Here are two common examples:

• A load cell has naturally low resistance but requires excitation power. The signal level is small, so low-level noise has a disproportionate effect. The casing of the load cell typically is a conductive material, and any circulating currents can couple noise into sensor leads.
• A piezoelectric sensor displaces some charge, but not very much. It has a high impedance and is relatively vulnerable to interference. It benefits greatly from having a gain amplifier close to the sensor. But then power is required to operate the amplifier.

Integrating power amplifiers at the sensor source can eliminate weak signals, but this is not a total solution. Watch bandwidth and dynamic performance carefully.

Bandwidths often are restricted to reduce sensitivity to high-frequency noise, but this is not helpful if your goal is to measure high frequencies. Any interfering signals that get into the sensor will be amplified along with the desired signal.

The 4- to 20-mA current loop converters are useful for transferring signals over a relatively long distance. However, they can add their own dynamic response to the signal, and their high impedance is somewhat vulnerable to noise pickup.

Integrated amplifiers are mostly a blessing, but beware of additional power supply noise and new paths for this noise to enter your system. The more channels, the more power demand, and the more difficult the problem of supplying quiet power.

Despite improved designs and manufacturing processes, providing power amplifiers for hundreds of channels adds significant costs. Sometimes this leads to configurations powered from the receiving end.

A sometimes-overlooked cost is incremental damage. Overvoltage conditions can produce short-term avalanche currents at sensitive amplifier inputs. Even if these do not cause catastrophic failure, they still can be enough to unbalance precision device inputs, elevate noise levels, and degrade signals.

Humans can carry a large dose of static charge, and just touching signal lines or connectors can cause problems. The best place to eliminate these problems is at the source with discharge and bypass paths.

Measurements captured at the high-impedance receiving end will be voltages, always with respect to a reference (ground) voltage. The reference should connect through a single low-impedance path to a common ground point.

A common ground point might not be as solidly grounded as you think, and ground path impedances might not be exactly zero. Currents through ground paths, often driven by external sources such as the power grid, can result in voltage shifts that corrupt your measurements. This is called the ground loop problem (Figure 1). In severe cases, disturbances larger than 1 volt are possible. Ground disturbances can be devastating to measurement accuracy.

A tempting solution would be to tie the ground points together more solidly. Stray currents then will travel through less resistance, producing smaller voltage offsets.

This solution does present a problem: The strong new connection couples your measurement system even more tightly to the power grid and the noise sources. It is even possible for measurement equipment to be severely damaged.

The single most effective thing you can do to reduce effects of ground disturbances is to use differential signals. Balanced differential signals are the best. But even if you measure from the hot side to the ground of a sensor device and treat this like a differential signal, there is a big benefit.

So why not just use differential signals all of the time? Here are some reasons:

• Cost: Each channel requires supporting two signal lines.
• Availability: Acquisition boards might not support it.
• Density: There are twice as many lines to terminate at connector panels.
• Noise: If signals are already extremely quiet, it doubles the number of noise paths.

Improved packaging increases the signal density and makes differential signals much more attractive for multichannel applications. But, there is a trade-off. With more signal paths to manage, you can expect less flexibility for isolating power and ground connections.

After providing a strong signal drive and protecting your signal sources from ground-coupled noise, you can add more levels of protection:

• Physically separate lines from disturbance sources.
• Shield physical paths of signal conductors.
• Shield individual signals (pairs).

For single-ended signals, an important way to protect signal quality is protective ground lines. Separator lines tied to a common, clean, low-impedance ground should be placed between adjacent signal lines. This is particularly true in ribbon cables where adjacent conductors are vulnerable to crosstalk.

This kind of signal protection is one of the first things that low-cost data acquisition boards will sacrifice to reduce costs. The differences might not show up in the device specifications, but your signal quality will suffer.

If you have a lot of panel space, coaxial cables are very good for single signal paths. Twin-conductor coaxial cables are superb for protecting balanced differential signals.

Separator ground lines are not as effective for differential connections. The interaction within a pair provides part of the noise immunity. It is better to use twisted-pair cables for differential signals; more effective are twisted-pair cables with the pairs individually shielded.

A common mistake is bundling low-voltage signals along with high-voltage signals, which radiate a great deal of energy. The radiated energy will go somewhere, some of it ending up in the low-level signal paths. For that reason, bundle and shield the low-voltage signals separately.

To further protect signals in harsh environments, keep signal paths entirely within protective enclosures and conduits along with your marshalling panels and instrument racks. If you can't justify the expense of shielded enclosures and conduits, protect cables as much as you can with screen mesh or foil wrap grounded on one end.

When there are many channels, there are many more possible grounding and interaction problems. Isolation can cut off stray circulating currents and isolate the transmitting and receiving ends of the paths from large voltage shifts.

Precision regulators help to keep noise levels low, but the additional power circuits increase the chances of ground interactions. An isolation device breaks all direct electrical connections while continuing to drive signals with low impedance. The isolation devices often are active, drawing power from one or both sides of the connection.

Widely available and very reliable 5B modules make isolation very easy. The drawbacks are in performance, packaging, and cost. The module carriers will consume rack space in a hurry. Also be wary of limited linearity, accuracy, and bandwidth. If the limitations are acceptable, 5B modules are an excellent way to isolate signals in a harsh environment. Otherwise, look for alternatives.

Connecting to the System

The next critical point is the connection to the data acquisition system. In many applications, this is the only accessible point of connection. It usually is necessary to drive the signals from this point through additional connections. Amplifying signals helps, but there is a definite increase in cost per channel.

Once again, load cells are a good example. If it is too difficult to supply the required excitation power at the sensor, the power must be routed through the cable along with the signal lines. At the termination end of the cable, there is the combined problem of supplying the well-regulated power while preserving the quality of the low-level strain signals.

The eight independent signal sections are clearly visible on the board shown in Figure 2. Boards like this one perform termination-end signal conditioning and filtering independently on each channel to preserve signal quality.

If your termination boards don't have integrated termination networks, look for breadboarding support. Balancing and charge dissipation networks for signals or loading resistors for 4- to 20-mA loops can be placed on the board.

With many signals merging into your acquisition system at the termination panel, it is necessary to select the appropriate connectors:

• DB Connector Family: popular, solid, inexpensive, widely available; supports moderate signal densities; good for single-ended signals.
• HD Connector Family: requires about half of the panel space of the DB connector family but harder to prepare.
• BNC Connector Family: popular and inexpensive but bulky; works well with shielded coaxial cables for maximum configurability.

With a 19″ rack-mount panel for a 21-slot, 3U package at maximum connector density, you can bring in about 640 differential signals with HD 78 connectors, 320 with DB 37 connectors, and 80 with single-conductor BNC connectors.

When you have regulation equipment, signal amplifiers, and loading resistors behind your termination panels, you might need to dissipate a significant amount of heat. Watch power supply capacity, regulation, and operating temperatures.

Even in the conversion of the measurements to digital form there are trade-offs. For the best possible conversion accuracy, each channel needs its own dedicated digitizer, but there is a cost for this. To reduce cost, digitizers can be shared. Some strategies include the following:

• Multiplexer circuits select the signals and route them to the digitizer. For more input signals, the sampling rate on each individual signal must be lower.
• Organize signals in groups, and use a board having one converter for each channel group.
• Use multiple boards, with each board capturing a group of signals.
• Use multiple hosts, each supporting multiple boards, circumventing bus and storage capacity limits when capturing data at high rates with high channel count.

For multichannel applications, the data acquisition converters must be designed to operate together. Otherwise, you can face some difficult timing and synchronization problems.

Multiplexer and instrumentation amplifier circuits typically perform well, but there are two significant limitations.

Settling Time
After switching between two signals having significantly different input levels, the input amplifier circuits undergo a transient period, with slewing and linear settling time required to stabilize at the correct level for the next sampling operation. You probably will not get full conversion accuracy at maximum operating rates.

Charge Injection
Switching from one signal line to another is much like charging a very small capacitance on one line and then discharging it into the next. Do this rapidly, through many measurement cycles, and accumulated charge can affect accuracy. Unless your signals are strong, you might need to operate at a reduced rate to achieve full accuracy.

A common mistake is to assume that once measurements are digitized, the data is good from this point onward. Many resources that should be invested in signal quality are used for pushing vast amounts of data with little value.

Measuring temperature with a thermocouple is a classic example. Each measurement is very noisy and not very useful, but averaging 1,000 measurements yields a very good temperature measurement.

Rather than clogging your data buses with millions of relatively useless raw measurements, analyze the data at a low level and transfer just the few thousand temperature measurements that you care about. Intelligent preprocessing to reduce and refine bulk data should be considered part of the signal capture process. Some cases where this is appropriate include the following:

• Compressing and encoding data.
• Compensating for known signal deficiencies.
• Removing interfering signals.
• Compensating shifts in time and phase.
• Averaging or other statistical processing.
• Calibrating adjustments.
• Converting into engineering units.
• Estimating and correcting for long-term trends.
• Taking relevant data selectively.

Picking the Right Equipment

High-end, high-quality products sometimes pay for themselves many times over. Less expensive equipment is better if you can stay within its capabilities and those capabilities are all you really need. Here are two examples.

Automotive Engine Applications
Engines are exposed to high-energy firing discharges that will affect sensors. Engine steel acts like a transformer core to induce noise in conductors. You likely will have less than 10 meaningful bits on any given measurement.

Intensive post-processing of captured data for frequency and timing analysis rarely depends on absolute accuracy of every sample. You should do very well with 12-bit bit converters that are well calibrated. Pick a board that supports multiple signals with good timing features.

Wind Tunnel Applications
Low-level strain or pressure sensors are subjected to numerous sources of noise from industrial equipment driving the air stream. The relevant phenomena are hard to detect under the best of conditions.

With maximum effort to protect signals, a careful statistical analysis can deduce important information at the level of the least significant bits. Typically, extra expense for every possible signal quality effort is justified.

When every aspect of signal quality is important, system development costs can greatly exceed those of the equipment. Rather than building everything from scratch, choose a signal conditioning system that provides dedicated differential amplifiers, independently-regulated excitation on every signal, isolation of signal and converter grounds, independent converters on every channel, self-calibration, and a mix of digital and analog filtering for optimized signal quality (Figure 3).

Summary

• Start with a realistic assessment of needs. What information is needed? What error tolerance is accepted?
• Consider the sources of corrupting noise. Apply countermeasures to address likely problems.
• Right-size your acquisition equipment. Neither over-specify nor waste money on underpowered equipment that can't deliver needed performance.
• Apply relatively simple and inexpensive signal quality measures such as differential signals and protecting your cables first.
• Use processing as part of signal capture to reduce bulk data to clean information.
• Follow the accuracy guidelines given in the sidebar.

About the Author
Larry Trammell is a senior software engineer at Microstar Laboratories. He joined the company in 1992. Mr. Trammell received a B.S.E.E. from Washington State University and an M.S.E.E. from Oregon State University. Microstar Laboratories, 2265 116th Ave. N.E., Bellevue, WA 98004, e-mail: [email protected]