Three’s a Crowd for Instrumentation Amplifiers

Three op amp instrumentation amplifiers (IAs) have long been the industry standard for precision applications that demand high gains or high common-mode rejection ratio (CMRR), or both. They have serious limitations, however, when powered from single supply voltages. This article explains the architectural limitations of conventional three op amp IAs, and introduces an indirect current-feedback circuit topology that provides specific advantages for single-supply operation.

IA Applications
IAs amplify small differential voltages in the presence of large common-mode voltages, while offering a high input impedance. This characteristic has made them attractive in a variety of applications, such as strain-gauge bridge interfaces for pressure and temperature sensing, thermocouple

Click for larger image Figure 1. Internal architecture of the MAX4194-MAX4197 family of three op amp IAs (Click image for larger version)
temperature sensing, and various other low- and high-side current sensing applications.

The classic three op amp IA (Figure 1) offers an accurate differential gain programmable by a single resistor, and excellent common-mode rejection. Its architecture is based on a two-stage configuration: the first stage provides most (or all) of the differential gain plus unity common-mode gain, while the second stage provides all of the common-mode rejection plus unity (or small) differential-mode gain (Figure 2).


Figure 2. Two-stage amplification of input signals. Input common-mode voltage is carried through to the intermediate stage

Although most modern low-voltage amplifiers have rail-to-rail outputs, they don't necessarily have rail-to-rail inputs. For the purpose of discussion, lets consider an extremely high-gain three op amp IA with rail-to-rail input and output, operating from a single-supply Vcc and using the circuit layout from Figure 1. Because Vout = Gain × Vdiff + Vref, Equation 1 follows:



Vout1 = Output voltage of amplifier A1
Vout2 = Output voltage of amplifier A2
Vcm = Input common-mode voltage, i.e., VIN++VIN–
Vdiff = Input differential voltage, i.e., VIN+–VIN–

To prevent Vout1 and Vout2 from hitting the supply rails, you need to ensure that 0 < (Vout1, Vout2) < Vcc, as in Equation 2:



Vcc = Single-supply power voltage

Note that 0 < Vout < Vcc.

Two common applications involve setting Vref either to zero (for unipolar input signals) or Vcc/2 (for bipolar input signals). For Vref = 0, the inequality reduces to Equation 3:


For Vref = Vcc/2, the inequality reduces to Equation 4:


These conditions are best understood graphically (Figure 3). The shaded areas in gray show the range of input common-mode voltage for which the amplifier's A1 and A2 outputs from Figure 1 do not saturate into the supply rails. This range depends on Vout and Vref, and because VoutVref is an amplified version of the input differential voltage, the input common-mode range varies with the input differential voltage. Outside the gray area, certain combinations of Vdiff and Vcm will cause the three op amp IA to malfunction, as we'll see later.



Figure 3. Usable Vcm range vs. the differential input voltage (Vout) for a single-supply three op amp IA with Vref = 0 (top), and Vref = Vcc/2 (bottom). The horizontal axis represents the amplified differential input voltage

As a practical matter, you'll want to make maximum use of the circuit gain, i.e., obtain the full output swing for Vout when the inputs see the maximum expected differential voltage. In Figure 4, the areas shaded in black indicate the range of input common-mode voltages for which the IA is able to amplify the maximum input differential voltage so that Vout ranges from 0 V to Vcc.



Figure 4. The black box shows the range of input common-mode voltages for which a conventional three op amp IA is able to use its gain to provide the maximum output voltage (at maximum input differential voltage)

As you can see, the input common-mode voltage is severely restricted in both cases. To fully amplify a unipolar-input differential signal (i.e., setting Vref = 0 and obtaining the full output swing from 0 V to Vcc), that signal's common-mode voltage can only be at Vcc/2. At common-mode voltages other than Vcc/2, the output voltage cannot swing all the way to Vcc. The maximum input differential voltage is thus reduced.

For the case of bipolar input differential signals (i.e., setting Vref = Vcc/2), the corresponding range of input common-mode voltages that allow a full output-voltage swing from 0 V to Vcc is from 1/4 Vcc to 3/4 Vcc. In both cases, if the input common-mode voltage is near ground (0 V), then the amplifier loses all ability to amplify input differential voltages.

In general, the desired input differential voltages are unrelated to the unwanted input common-mode voltages. For Vcm, the black areas in Figure 4 represent the allowable minima and maxima for which Vout can occupy its full range. In Figure 4A, note that if one requires a full-scale excursion for Vout, the tolerance on input common-mode voltage is zero and no common-mode variation of the input signal is allowed around Vcc/2.

Because of this behavior, three op amp IAs find limited application in single-supply systems. However, let's consider what happens if the internal amplifiers A1 and A2 saturate into the rails and also examine the effect of an architecture for which input excursions must be less than the rail-to-rail value.

Effect of Input-Amplifier Saturation
Let's assume the output of amplifier A1 has saturated into ground (Vin+ > Vin–), and the common-mode voltage is in the area marked X in Figure 4 (Vdiff is larger than that allowed by the gray area).

Because A1 in Figure 1 is saturated (Vout1 = 0), it transitions into a nonlinear (comparator) mode of operation in which the voltage at its inverting pin is no longer constrained to be the same as that on the noninverting pin (Vin–). Amplifier A2 then acts as a noninverting amplifier, with gain equal to (1+R1/(R1+Rg)) for voltages at its noninverting pin (Vin+). For high-gain amplifiers, Rg<< R1 and so amplifier A2 acts as a noninverting gain-of-2 amplifier.

The second-stage differential amplifier (A3) examines its inputs (Vout2 and Vout1), and presents the difference at its output (Equation 5):



Vref = External voltage used to bias the output of differential amplifier A3

Similarly, if A2 saturates to ground, then Vout = –(2Vcm + Vdiff) + Vref.

This is a potentially hazardous mode of operation for the three op amp IA: not only has it stopped amplifying the differential input voltage (gain changed from 1+ 2R1/Rg to unity) but, rather than degrading gracefully in some fashion, it transitions into a mode that amplifies the input common-mode voltage with a higher gain (gain = 2) relative to the input differential voltage. This effect is exacerbated by the fact that common-mode voltages are generally uncontrolled in real-life applications and cannot be guaranteed to occur only in the black regions of the graph. Common-mode voltages caused by fault conditions or unwanted noise can corrupt the output signal with input common-mode voltage—and the primary reason for using an IA in the first place is to get rid of the input common-mode voltage.

Effect of Non-rail-to-rail Inputs
The design of rail-to-rail input stages is especially tough for precision applications because the crossover between the circuit's operation with near-Vcc common-mode voltage and near-ground (GND) common-mode voltage is never perfect. Offset voltages (Vos) can arise during transitions between n- and p-type pairs in the input differential stage because they are not correlated. This change in Vos, caused by a changing input common-mode voltage at the cross-over region, severely degrades the amplifier's CMRR specification—a key differentiating factor when selecting IAs.

As a result, most precision IAs tend not to have rail-to-rail inputs, although they still include the negative rail (0 V) as part of the input common-mode voltage range. If we take into account the input common-mode voltage limitations of Figure 3 and redraw the graphs, we get those shown in Figure 5.



Figure 5. Usable input common-mode voltage vs. differential input voltages for single-supply three op amp IA, accounting for a non-rail-to-rail input stage

Indirect Current-Feedback Architecture

Click for larger image Figure 6. IA with indirect current-feedback architecture (Click image for larger version)
The indirect current-feedback architecture is a new approach in IA design that offers multiple benefits. Figure 6 shows this architecture as implemented in the MAX4462 and MAX4209 IAs. The devices each contain a high-gain amplifier (C), and two transconductance amplifiers (A, B). Each transconductance amplifier (Gm) converts its input differential voltage into an output current, and rejects all of its input common-mode voltage.

At the amplifier's stable operating point, the output current sourced from Gm stage A matches the input current sunk by Gm stage B. This matching is enabled by feedback action through the high-gain amplifier C, which forces the differential voltage at the input of feedback amplifier B to be the same as the differential voltage at the inputs of amplifier A. In turn, that action sets up a defined current in the output resistor chain (Vdiff/R1), which also flows through R2. Therefore, the voltage at OUT is an amplified version of the input voltage (Gain = 1+R2/R1). The output can be offset by applying an arbitrary bias voltage at REF.

After translating this operation to a high-level block diagram (Figure 7) and comparing it with Figure 2, we see that, whereas the intermediate signal in the three op amp IA contained the gained-up differential voltage as well as the input common-mode voltage, the indirect current-feedback architecture contains a representation (in current) of only the input differential voltage. That is, the first stage provides all the common-mode rejection and the second stage provides all the differential gain, while reinforcing the common-mode rejection and thereby allowing the output to be offset by a reference voltage if necessary. As a result, the input common-mode voltage limitations found in a three op amp IA are absent in the indirect current-feedback architecture.


Figure 7. Operation of indirect current-feedback IA. Note that the first-stage output has no common-mode voltage

Taking the input common-mode voltage limitations of a non-rail-to-rail input stage into account, the transfer characteristics are similar to that shown in Figure 8. The area in black depicts the limits of input common-mode voltage for which the full range of output voltage can be achieved. The area in gray depicts the range of input common-mode voltage for which the IA operates as expected. That is, its output voltage is proportional to a gained-up version of the input differential voltage, and it rejects all input common-mode voltages. The operational range of input common-mode voltage is thus greatly expanded.



Figure 8. The usable range of input common-mode voltages for an indirect current-feedback IA comprises the gray and black areas. The area in black represents only the input common-mode voltages for which the full range of output voltage is achievable

Experimental Results
Experimental results obtained with the MAX4197 and MAX4209H IAs are included to support the above discussion. Both devices have gains of 100. The MAX4197 uses a three op amp architecture, while the MAX4209H uses an indirect current-feedback architecture. Both parts are supplied with Vcc = 5 V, and with VREF = 2.5 V to offset the zero output. Two types of waveform are fed to the IA:

Case 1: A small 1 kHz differential voltage in the presence of a large 100 Hz common-mode voltage. The output of the IA is expected to contain only the 1 kHz signal, with no 100 Hz component. The waveforms are described as follows:

Vin+: sinewave amplitude = 2 Vpp, offset = 1 V, frequency = 100 Hz. (Vin+Vin–): sinewave amplitude = 30 mVpp, offset = 0, frequency = 1 kHz.

Case 2: A small 100 Hz differential voltage in the presence of a large 1 kHz common-mode voltage. The output of the IA is expected to contain only the 100 Hz signal, with no 1 kHz component. These input waveforms are described as follows:

Vin+: sinewave amplitude = 2 Vpp, offset = 1 V, frequency = 1 kHz (Vin+Vin–): sinewave amplitude = 30 mVpp, offset = 0, frequency = 100 Hz.

The results are shown in Figures 9 and 10. Channel 1 is Vin+, Channel 2 is Vin–, and Channel 3 is the IA output.

For Case 1, results for the MAX4209H IA are as expected (Figure 9A). MAX4197H (Figure 9B) amplifies input differential voltages only when the input common-mode voltage is well above ground.


Figure 9. Results for Case 1, featuring IAs with indirect current-feedback architecture (MAX4209H, A), and three op amp architecture (MAX4197, B). The 1 kHz Vdiff is too small to be visible on the INPUT1 and INPUT2 traces, which are dominated by the 100 Hz Vcm

For Case 2, again, results for the MAX4209H are as expected (Figure 10A). The MAX4197H, on the other hand, (Figure 10B) amplifies the input differential signal only when the common-mode voltage is well above ground. When the common-mode voltage is close to ground, the output voltage either inverts the common-mode voltage or simply buffers it, depending on whether A1 or A2 saturates.


Figure 10. Results for Case 2, featuring IAs with indirect current-feedback architecture (MAX4209H, A), and three op amp architecture (MAX4197, B). Note a breakthrough of the 1 kHz Vcm on the desired output for the three op amp IA

In Closing
The indirect current-feedback architecture holds multiple benefits for single-supply applications that require the use of high-gain IAs. To enable longer battery life and greater energy efficiency, today's consumers demand not only better performance, but also more intelligent power-management schemes. A transition from dual-supply analog designs to single-supply architectures is already underway, and is changing the design and use of electronics systems.

Suggested Articles

The world’s largest chipmaker saw a 47% decline in data center sales to enterprise and government, even as it forecast a full year 2020 record of $75B

Working with Jacoti of Belgium, Qualcomm wants to make earbuds recognize the hearing anomalies of users.

Tally upgrade from Simbe Robotics uses Nvidia Jetson GPU for edge processing and Intel RealSense LiDAR for higher resolution images