Sensor measurements typically translate physical phenomena of interest into electronic-circuit parameters such as resistance and capacitance, which can then be read with a bridge circuit. Bridge circuits produce an output voltage or current signal that is ratiometric with respect to temperature and powersupply voltages, thereby enabling the measurement system to be immune to these variables. Sensor examples can include:
- Thermistors for temperature sensing
- Resistive/capacitive strain gauges for pressure sensing
- Magnetoresistive sensors for direction/position sensing
Sensors that produce a signal voltage or current directly don’t require a bridge circuit to transform the physical variables. Examples include thermocouples, ECG-based medical instrumentation, and voltage across the currentsense resistor in a power-monitoring circuit.
Today’s sensor applications range from consumer electronics (thermometers, pressure scales, GPS systems, etc.), to automotive (fuel sensors, knock sensors, brake-line sensors, window pinch control), to industrial and medical instrumentation (valve-position sensing, temperature-based system calibration, and ECG). Their operating environment is rich in EMI noise, power-supply harmonics, ground-loop currents, and ESD spikes, while the signals of interest to be extracted are relatively small.
Thus, the analog-sensor interface becomes non-trivial and must maintain exacting specifications while rejecting these environmental phenomena. For commercial success, it must also deliver low cost, small size, and (for battery-operated meters) low supply current.TO AMP OR NOT TO AMP System designers like to keep analog chains short in the hope of improving the signal’s immunity to external noise phenomena. (Digital circuitry is generally immune to noise, but not always.) Lengthy analog chains in the past tackled a given signal-processing task in sequential stages.
One stage, for example, provided differential gain without common-mode rejection, and another provided common- mode rejection without differential gain, etc. Dual and high-voltage supply rails also helped relax the signalto- noise constraints on analog circuits. The requirements for shorter analog chains and single-supply, low-voltage, analog power-supply rails have forced the evolution of innovative architectures to meet these challenges.
One decision that arises early in a system design is whether or not the analog-to-digital converter (ADC) and sensor can interface directly. Such direct connections offer an advantage in some applications.
High-resistance ratiometric bridges, for instance, can use the rudimentary internal reference present in many ADCs, and some modern ADCs contain a high-impedance buffer or PGA that can be used to isolate the sensor signal from loading and from current spikes caused by the ADC’s sampling circuitry.
On the other hand, a substantial case can be made for using an instrumentation amplifier (IA) to interface the sensor to an ADC:
- Amplifying small analog signals at their source improves the overall signal-to-noise ratio in some applications, especially if the sensor is not close to the ADC.
- Many high-performance ADCs lack high-impedance inputs and must therefore be driven by an amplifier of low source impedance to get the full benefit of their specifications. Without an intermediate amplifier for such configurations, aberrations like input current spikes and mismatched source resistances can introduce gain errors.
- An external amplifier allows the user to optimize the signal conditioning (filtering) for an application.
- The best semiconductor process for fabricating an ADC isn’t necessarily the best one for fabricating amplifiers.
- The gain offered by an IA makes for an easier interface between sensor and ADC, both by easing the system design constraints and by lowering the overall system cost. For example, a much higher-resolution and expensive ADC would be required to read an un-gained sensor signal than that required for an amplified sensor signal.
Various sources of dc error are encountered when using IAs to read sensor signals. Perhaps the most critical of these is the effect of input offset voltage. In fact, every other source of dc error is modeled in terms of the input offset voltage: dc CMRR represents the change of dc input offset voltage with input common-mode voltage, and dc PSRR represents the change of dc input offset voltage with variation in powersupply voltage.
Continue on Page 2
Even if VOS can be calibrated out during manufacturing, the drift of input offset voltage (with temperature and time) can be of greater concern than the initial dc offset itself. Such drift errors are best tackled through the use of active circuitry within the chip.
Perhaps the single most important source of ac error is noise, which is inherent in the semiconductor chip design and process. Because most sensor signals are amplified by highgain blocks, the input referred noise is also amplified by that same gain. Noise comes in two forms: pink noise (also called 1/f or flicker noise) and white noise. Pink noise is more critical at lower frequencies (
In traditional low-noise analog-circuit design, bipolar transistors are often preferred for use in input-stage circuitry, especially if low levels of pink noise must be achieved. Pink noise originates as recombination effects at defect sites on the semiconductor surface. Therefore, the noise of CMOS devices tends to demonstrate a larger magnitude and a higher corner frequency than the noise developed by bipolar devices. (The frequency at which pink noise density equals white noise density is defined as the noise corner frequency).
Most sensors prefer high-impedance inputs, which forces the use of CMOS front ends on IAs. This, in turn, would seem to force one to live with the accompanying higher levels of lowfrequency noise. Fortunately, zero-drift circuit-design techniques that continuously cancel out input offset voltages also tend to cancel the low-frequency input pink noise.COOL NEW ARCHITECTURES ARE REALLY HOT A traditional IA uses three op amps to create an input buffer stage and an output stage (Fig. 1). The input buffer stage provides all differential gain, unity common- mode gain, and a high-impedance input. The differential amplifier output stage then provides a unity differential gain with zero common-mode gain. This IA works quite well in many applications, but its simplicity hides two significant drawbacks: the usable input commonmode voltage range is limited, and its ac CMRR is limited.
IAs based on three-op-amp architectures suffer a restrictive transfer characteristic (Fig. 2). Their architecture can allow the outputs of buffer amplifiers A1 and A2 to saturate into the power-supply rails during a certain combination of input common-mode and input differential voltage. In this condition, the IA no longer rejects input common-mode voltages.
As a result, the data sheet for most three-op-amp IAs shows a plot of the usable input common-mode voltage versus output voltage. Because output voltage is simply a scaled version of the input differential voltage, the two axes of this plot could also be labeled “input common-mode voltage versus input differential voltage.” The gray area within the hexagon depicts the “valid” zone of operation, where the outputs of amplifiers A1 and A2 aren’t saturated into the power-supply rails.
Note that the graph of Figure 2 has an important implication for single-supply applications. Common-mode voltages can easily approach the circuit ground, to which the gray zone doesn’t extend! Certain applications (such as low-side current sensing) can’t use a traditional three-op-amp IA, because the input common-mode voltage equals the ground potential.
Three-op-amp IAs achieve high common- mode rejection at dc by matching on-chip resistors around the differential amplifier, but the feedback architecture of such IAs can substantially degrade the ac CMRR. To overcome this and other drawbacks, alternate IA architectures have been developed. The 2-gm indirect current-feedback approach, for instance, has found considerable success (Fig. 3).
This architecture consists of two matched transconductance amplifiers and a high-gain amplifier. Because the matched amplifiers have the same gm, they develop equal differential voltages at their inputs, and the output voltage is therefore determined by the resistor divider ratio Rf/Rg. The output commonmode voltage is set by the voltage at the REF pin. Voltage-to-current conversion implemented by the input gm amplifier inherently rejects the input commonmode voltage, giving the amplifier a high dc and ac CMRR.
The indirect current-feedback IA architecture allows a full output-voltage swing even when the input commonmode voltage equals the negative supply rail. Thus, it offers an expanded range of operation not obtainable with the threeop- amp IA architectures. Examples of this IA type from Maxim Integrated Products include the MAX4460/1/2 and the MAX4208/9.OFFSET-CANCELLATION TECHNIQUES: CATCH THE DRIFT? As mentioned above, two important specifications for IAs are pink noise (also called 1/f or flicker noise) and input offset voltage and its drift (versus temperature and time). Because 1/f noise is a low-frequency phenomenon, many of the circuit techniques used to achieve “zero drift” and cancellation of input-offset voltage also remove 1/f noise. These techniques include sampling amplifiers, auto-zeroing amplifiers, chopper amplifiers, chopper-stabilized amplifiers, and chopper-chopper-stabilized amplifiers (e.g., the MAX4208).
Sampling techniques based on flying capacitors have also been applied to IAs for the purpose of auto-correcting input offset voltages. However, since a sampled input isn’t a true high-impedance structure, system-level accuracy can be compromised by a mismatch in the source resistances (such as those found in certain unbalanced bridges).
Continue on Page 3APPLICATIONS This section describes two IA applications: a ratiometric bridge circuit and a low-side current-sense amplifier.
1. BRIDGE OVER TROUBLED WATERS
A variation of the standard bridgemeasurement system is the ratiometric bridge, which delivers similar high accuracy but at a lower cost. Cost is lower because the ratiometric bridge doesn’t require a precision reference source for driving the bridge and ADC reference input. Instead, a “free” but relatively inaccurate and high-ppm/°C reference source, such as the power-supply rail, can be used to drive both the bridge and the ADC.
It’s well known that even an op amp with “rail-to-rail” output has trouble maintaining full accuracy while driving its output to within a few hundred millivolts of either rail. For an amplifier with high dynamic range and unipolar-signal inputs, it’s therefore necessary to bias the output above ground, by 250 mV or so. This bias voltage drives one end of a resistor chain and thus should be driven by a buffer of low output impedance to avoid introducing unintentional gain errors. To minimize output errors, this unity-gain op-amp buffer should also have low dc offset and low drift.
An IA from Maxim (the MAX4208) integrates a precision, zero-drift op-amp buffer with a 2-gm indirect current-feedback IA in a small µMAX package. This buffer allows a simple external resistor divider to be used to create a stable bias reference voltage that’s ratiometric with the ADC reference voltage. It’s also able to drive one of the inputs of a differential input ADC. The internal chopperchopper- stabilized architecture of the IA eliminates pink-noise effects in both the op-amp buffer and the gm amplifiers of the main (forward) and feedback paths. In addition, the part includes a shutdown mode that’s useful for powersensitive applications.
2. MAKE PERFECT CURRENT-SENSE
The increasing need for active power management in today’s portable electronic devices has led to a renewal of interest in current-sensing amplifiers. A ground-sensing IA can be used as a high-side current-sense amplifier in the core-voltage path of a memory module or microprocessor (Fig. 4) or as a low-side current-sense amplifier in the return path of an H-bridge power electronic converter.
The extremely high currents in these applications (sometimes approaching 90 A) imply that the sense voltage must be extremely small to prevent excessive power loss in the sense resistor. Quite often, this sense resistor is simply the ESR of the power-supply inductor itself. To read these small sense voltages accurately, the input offset voltage must be extremely small in comparison with smallest sense voltage (i.e., smallest load current) that’s required to be amplified with accuracy.
Core voltages in computer hardware can vary from 0.9 to 1.5 V, and so the small sense voltage must be measured in the presence of a low and varying common-mode voltage. An IA such as the MAX4208 with low VOS, high CMRR, and an architecture optimized for single- supply applications is thus ideal for this purpose.