Much has happened recently in the area of sensor technology. Since the launch of the first Apple iPhone on January 9, 2007, MEMS use in portable electronics has exploded. In the automotive arena, successful applications of MEMS gyros now include dynamic stability control systems, rollover airbag detection systems, and navigation and emergency response systems for passenger cars.
Additional solutions requiring ever-higher position accuracies, and often with integration to vehicle support systems, are emerging in commercial vehicles and machines used for agriculture, construction, and lifting purposes. However, for new sensors to successfully meet these application challenges, some important sensor accuracy questions must first be answered. For example: Have the advances across the range of sensors improved application accuracy and if so by how much? Where are the most compelling application performance improvements and what other accuracy improvements can be expected in the near term?
This article will address these and other sensor related accuracy questions, with particular emphasis on accuracy in MEMS-based sensors systems.
Although the accuracy of sensors is determined through stringent testing and the resulting specs are published on datasheets, several factors either enhance or leave accuracy results hamstrung. All too often the specification will read something like "accuracy within ±1.0 percent". My question then is: "Percent of what?" If it is meant to be percent of indicated value, we need to clarify a few things. Let’s take temperature as an example. There are four primary temperature scales in use today: Kelvin and Rankine, which are absolute temperature scales, and Celsius and Fahrenheit, which are not. Let's take the freezing point of water in Celsius, for example. What is ±1.0 percent of temperature accuracy at 0°C? A perfect reading? Possible, but not likely. If we were reading this in Fahrenheit, the tolerance would be ±0.32°F; in Kelvin, it would be ±2.73°K, which equals ±2.73°C. So which is right? None of the above. Simply stated, the specification was poorly written; it is acceptable, however, to use percent if you clearly state what the scale will be.
Some of the external influences impacting accuracy are characteristics of the environment, PCB layout, noise, sensor placement, range of conditions within which the sensor will measure, and whether or not standards are adhered to or even specified.
Common types of errors across the sensor spectrum include:
Increasingly, calibration is provided by the manufacturer rather than having it become an exercise left to the product development team. Yet, whether performed in-house or pre-use, there are still errors in calibration that occur frequently enough that calibration must be performed, or the impact of external factors on the calibration of the device. For example, with a glucose sensor, the rate of glucose change that takes place during calibration can throw the measurement off substantially.
Delay can be the result of the inherent characteristics of the underlying sensor, as well as a lag in data filtering.
Drift occurs in a variety of applications and there are an equal number of reasons why. Pressure, foreign bodies, oscillation, and coatings, can all have a potential impact on drift.
Lack of redundancy:
While redundancy probably resulted after the first catastrophic fire/earthquake left humans scratching their heads at data/assets foolishly lost, the redundancy provided by two sensors in an individual setting does seem to increase accuracy. Not just in case one does not work, but also in the likelihood that one will work better than the other, even if they appear to be exact. The question becomes one of how to use two sensors in a redundant setting. Is the best output an average of signals? Does the discrepancy between two sensors provide too wide of an output to assist with accuracy? Could a voting situation with three or more sensors be even more accurate?
Since drift, errors in calibration, and delay do exist in certain situations, two sensors have an advantage over one if the accuracy of one is in question. Two can also be used with a goal of selecting the more accurate of the two and eliminating the second, less-accurate sensor entirely. The use of an array in certain situations may also be beneficial.
Smartphones and tablets today all include some combination of an accelerometer, magnetometer and gyroscope, to enable a wide range of motion aware and motion controlled applications, and this trend is expected to grow dramatically.
Drift, calibration and external magnetic fields can play havoc with motion sensors. It is not a matter of throwing one or two more sensors at the problem. Instead, interoperability between providers and application processors across multiple OSes, as well as power, performance, and cost considerations, will go hand in hand to begin to solve sensor shortcomings. Then, the use of sensor fusion (see the TechZone article “Sensor Fusion: The Basics”) and multi-sensor solutions offer much needed accuracy improvements that are initiated squarely on solid ground.
One look at the feature list of accelerometers/inclinometers, and one can see that there is a focus on providing accuracy. The ADIS1620 iSensor, for example (Figure 1), is a digital inclinometer system for use in platform control, stabilization, alignment, tilt sensing, inclinometers and leveling, and motion/position measurement. Applications include security, medical, safety, and navigation. The iSensor provides precise measurements for both pitch and roll angles over a full orientation range of ±180°. Combining a MEMS triaxial acceleration sensor with signal processing, addressable user registers for data collection/programming, and an SPI-compatible serial interface, the production process includes unit-specific calibration for optimal accuracy. The part also offers digital temperature sensor and power supply measurements with configuration controls for in-system calibration, sample rate, filtering, alarms, I/O configuration, and power management.
Figure 1: ADI1610 iSensor is a precision triaxial inclinometer and accelerometer with SPI.
MEMS: Toward higher precision
MEMS gyroscopes have achieved improvements in cost, size, and ruggedness. However, gyroscopes cannot compete in high-accuracy critical apps such as guidance and navigation where bias drift is seen. Multiple sensors began to take care of the problem, as did interface circuits or mechanical sensing schemes, and accuracy has been enhanced as a result of better algorithm design.
MEMS accelerometers and gyroscopes currently are being used in commercial products including the Nintendo Wii video game, the iPhone, walking robots, and automotive airbags. Those applications work well because they do not need very high precision or accuracy. However, it is still difficult for conventional MEMS technology to accurately measure very small forces, such as van der Waals forces between molecules. These forces are measured in "piconewtons," an SI (International System of Units) unit of force equal to 10−12
A key aspect of MEMS devices is that they incorporate moving components. A MEMS device must either cause one of its components to move, or allow a component to move based on external influence, and then sense that motion. The moving component of a MEMS device typically employs flexure, and detection of the motion often involves sets of interleaved "fingers." For precise motion and motion detection, knowing the dimensions of the flexures and motion detection components is critical. Researchers at Purdue University now believe MEMS devices will soon be able to self-calibrate without external references and in doing so they will become very accurate sensors or actuators. Led by assistant professor Jason Vaughn Clark of the School of Electrical and Computer Engineering, the Purdue team has come up with a technology called electro micro metrology — or EMM — enabling engineers to account for process variations by determining the precise movement and force being applied to, or sensed by, a MEMS device.
This innovation might enable researchers to create a "nose-on-a-chip" for tracking criminal suspects (see the TechZone article “The Five Senses of Sensors—Smell”), sensors for identifying hazardous solid or gaseous substances, as well as a new class of laboratory tools for specialists working in nanotechnology and biotechnology.
EMM defines mechanical properties solely in terms of electrical measurements, for example, by measuring changes in capacitance, Geometric, dynamic, and material properties can be represented as functions of electrical measurands—f(ΔC, ΔV, etc.). Through EMM, the Purdue team has claimed it can obtain a microstructure's shape, stiffness, force, or displacement with high accuracy and precision.
MEMS-based sensor elements, such as pressure sensors for hydraulic control systems, humidity sensors (see the TechZone article “Humidity Sensors and Signal Conditioning Choices”) and liquid level gauges (see the TechZone article “Selecting a Fluid Level Sensor” What You Need to Know”) convert mechanical energy into an electric signal. Since most high accuracy signal processing is done in the digital domain, the sensor output signal needs to be converted to a digital signal by an analog-to-digital converter (ADC). Dresden-based Zentrum Mikroelektronik (ZMD AG) has developed a capacitive sensor signal conditioning device that has 14-bit resolution and 0.25 percent accuracy over a range of sensor capacitances and temperatures.
The device can be configured to interface with capacitive sensors from 0.5 to 260 pF, with sensitivity as low as 125 attofarads (aF) per digital bit. The chip employs a digital technique to correct both first-order and third-order nonlinearity errors that were difficult or impossible to correct with a purely analog signal path. Third-order correction in particular is said to be especially useful in humidity and pressure sensor applications. The device offers 14-bit resolution for compensation of sensor offset, sensitivity, and temperature.
While there are issues of accuracy in most sensors, either from an inherent characteristic point of view or an environmental one, great strides are being made to improve accuracy. Perhaps the most important of these is the promise that sensor fusion may deliver. Coming at the challenge of measurement in multiple ways truly does make for an improved result, and it will be interesting to see just how accurate sensors become in the very near term.