使用宽带电压和电流反馈运算放大器时的应用基础

使用宽带电压和电流反馈运算放大器时的应用基础

With both current and voltage feedback operational amplifiers (op amps) available to the system designer, how do you select which device to use? This two-part article discusses applications most suited to each type of op amp, and why certain applications are unsuitable for one or the other type of amplifier. Widely-used circuits are shown with example designs. The emerging fully differential amplifier type of amplifier (a special case of a voltage feedback device) is also shown along with several suitable applications. Part 1 reviews the internal differences between the two types of op amps and presents some key applications most suitable to voltage feedback type devices. 

Initial amplifier selection criteria

There is a bewildering array of possible op amps for a designer to select from in any given application. However, there are distinct internal differences between the two major types of amplifiers: voltage feedback (VFB) and current feedback (CFB). These differences may lead a designer to choose one over the other in certain applications. A newer type of op amp, the fully differential amplifier (FDA), is a type of voltage-feedback op amp that includes an output common-mode control loop. It has the same application emphasis as a standard voltage feedback amplifier, as well as several specific applications ideally suited to this type of amplifier. 

Before giving any attention to the type of amplifier to use, most designers first should consider the output signal requirements (for example, maximum desired output V_pp and output current, as well as load type) and necessary dynamic characteristics such as settling time, full power bandwidth, or a distortion requirement. Using these specifications, the universe of possible amplifiers can be narrowed to devices that can handle the correct range of supply voltages needed to deliver the output V_pp, and then limited further to a minimum supply current versus desired output dynamic characteristic. Normally, a CFB gives the best power efficiency for higher-frequency dynamic range needs, while a VFB generally has better noise and/or DC precision, and is the part of choice for certain types of circuits. 

Internal comparison of voltage and current feedback op amps

To understand why some circuits work better with one or the other type of amplifier, you need to first understand the internal topology of each amplifier and their resulting transfer functions. Compare a simplified expression for the Laplace transfer function written in loop gain format. Figure 1 shows the internal block diagram of a VFB along with the internal model and closed loop transfer function. A(s) is the frequency-dependent open-loop gain of the op amp. It is modeled here as a single dominant pole response. Figure 1 uses an inverting gain configuration for comparison purposes because the FDA device is most clearly understood as a differential inverting configuration. 

(Click to Enlarge Image)
Figure 1. Open loop to closed loop conversion for voltage feedback.

The transfer function has the desired gain in the numerator (-R_F/R_G) and a loop gain (LG) term in the denominator that determines the frequency response. One way to understand this LG is to plot [20 · Log (A(s))] and [20 · Log(1+R_F/R_G)] on the same grid. The key issues for this plot are: 1) the separation between these two curves at lower frequencies (this separation shows the magnitude of the loop gain); and 2) at what frequency they intersect. 

At the point of intersection, sometimes called loop gain crossover, the term in the denominator of the transfer function drops to 1 + 1e^-jθ (where jθ is the angle of that expression). The important check is that this angle is well away from -180 degrees to avoid closed loop oscillations or peaking. Figure 2 shows an example loop-gain plot of these two terms, where a simple single-pole response for A(s) is assumed and no phase added by the feedback network. 

(Click to Enlarge Image)
Figure 2. Loop gain and phase for a VFB op amp. X-axis is log frequency.

In Figure 2, the (1 + R_F/R_G) term is assumed to add no phase impact to the loop gain; only the open-loop phase of A(s) introduces loop phase shift in this simple example. This graph is the lower plot, where the loop gain crossover frequency is mapped down to find the remaining phase margin at that frequency. 

Note that it is impossible for the VFB to change the signal gain without changing the loop-gain characteristic. This effect is where the gain bandwidth product (GBP) concept arises. If the gain increases, the bandwidth must decrease. If the gain decreases, the bandwidth increases, and the phase margin will normally decrease. (See Reference 1 for a more complete discussion.) 

In contrast, the internal workings of a CFB op amp are quite different. Those blocks and the resulting closed loop transfer function for the inverting configuration are shown in Figure 3where, again, an inverting configuration is used and a single pole internal transimpedance gain is assumed for Z(s). 

(Click to Enlarge Image)
Figure 3. Current feedback internal structure and closed loop response.

The CFB uses a unity-gain buffer across the two inputs. This forces the inverting node voltage to follow the non-inverting input voltage. That buffer is intended to present a low impedance to the inverting port, where a low level error current may be sensed and passed on to the output through a transimpedance gain. It is this internal transimpedance gain, Z(s), which acts in the same fashion as the VFB A(s) to provide a high DC gain with a dominant pole. 

When the loop is closed, the same desired gain is achieved; but the loop-gain terms are very different. The CFB amplifier has a loop gain set by the forward transimpedance gain compared to the feedback impedance. Figure 4 plots the loop gain and phase for a typical CFB amplifier, where the feedback element is assumed to introduce no phase shift in this simplified analysis. 

(Click to Enlarge Image)
Figure 4. Current feedback (CFB) loop gain and phase plots. X-axis is log frequency.

This plot looks very similar to the VFB plot except that the external element setting the loop gain is the feedback impedance alone. The greatest difference between the VFB and CFB amplifier is that the loop gain can be set separately from the signal gain using the feedback impedance. The feedback impedance becomes an independent compensation element, where the gain can then be set using the normal gain equations from whatever impedance value is selected for R_F. This approach gives what is sometimes called gain bandwidth independence for the CFB amplifier. (See Reference 1 for a more detailed discussion.) 

The final type of amplifier to be considered here is the new fully differential amplifier (FDA). Figure 5 shows the configuration and closed loop transfer function for this type of amplifier. 

(Click to Enlarge Image)
Figure 5. Fully differential amplifier (FDA) structure and transfer function.

If the two feedback networks are allowed to be unmatched, the transfer function is fairly complicated. If they have a matched divider ratio (see Figure 5), the equations simplify to be the same as the inverting VFB transfer function. The effect of the separate common-mode loop is not shown. This loop acts to servo the average output voltage to a value set by a V_OCMinput pin voltage (see Reference 2 for a more complete discussion of the FDA topology). For applications considered in this article, the FDA is treated as differential VFB device. 

Application imperatives

In the range of possible applications for wideband op amps, several types must use VFB devices. These circuits can sometimes be forced to work using CFB devices, but usually at the cost of complexity and poorer performance. Any circuit that requires flexibility in the feedback element and/or capacitors in the feedback will have stability problems with a CFB device. This instability is the result of the loop gain depending on the feedback impedance. Therefore, any circuit that needs a lot of flexibility in that impedance is going to interact with the achievable frequency response if a CFB amplifier is used. 

The following example circuits should use a VFB device for implementation. 

A. Transimpedance amplifiers. These circuits take a current source input, typically from a capacitive source, and turn it into a voltage at the output. The feedback resistor is the gain element and normally needs a compensation capacitor in parallel for correct operation. Figure 6 shows an example using the OPA657, a very wideband JFET input device uniquely suited to the transimpedance application. This device is a non-unity gain stable VFB with relatively low input noise voltage and very high gain bandwidth product. For a given diode source capacitance, the amplifier gain bandwidth product (GBP) determines the achievable bandwidth and/or transimpedance gain (Reference 3). 

(Click to Enlarge Image)
Figure 6. Example transimpedance design using the OPA657.

In this example, the 500 kΩ along with the 200 pF diode capacitance gives a noise gain zero at approximately 1.6 kHz. With the feedback capacitor set to achieve a maximally flat Butterworth response, the resulting F_-3dB will be at the geometric mean of this zero and the 1.6 GHz gain bandwidth product of the OPA657. (Reference 3 gives a detailed analysis for compensation and noise in a transimpedance design.) Figure 7 shows the 1.6 MHz transimpedance bandwidth in a simulated frequency response of Figure 6. 

(Click to Enlarge Image)
Figure 7. Simulated transimpedance frequency response of Figure 6.

B. Integrator based circuits. These applications are looking for a capacitive feedback to implement an integrator function. A good example would be the Multiple feedback (MFB) active filter. Figure 8 shows an example using the OPA820, a low-noise, wideband voltage feedback op amp uniquely suited to this application. 

Embedded within this filter circuit is an integrator configuration that arises from the 200 Ω and the 12.5 pF feedback capacitor. 

At very high frequencies, that capacitor shorts out, giving a local unity-gain feedback for the amplifier. This result suggests that a unity gain stable VFB should be used in this type of circuit to avoid high frequency oscillations. This particular example is targeting a 10 MHz low-pass Butterworth response with an in-band gain of -4 V/V. (Reference 4 describes how to choose the R's and C's in the MFB filter for improved noise and distortion.) 

(Click to Enlarge Image)
Figure 8. Example MFB filter using the OPA820.

Since the local feedback capacitor shorts out at higher frequencies, this circuit normally needs to use a unity gain stable VFB. (Reference 4 suggests a path to improved phase margin in this circuit using non-unity gain stable amplifiers.) One of the advantages of the MFB design is improved stopband rejection (over the Sallen-Key filter), and very good filter accuracy versus finite amplifier gain bandwidth product. This filter tends to achieve a slightly lower Q than that targeted as the Ω₀ moves closer to the amplifier GBP. 

(Click to Enlarge Image)
Figure 9. Simulated 10 MHz Butterworth filter of Figure 8.

C. Circuits with simple feedback poles implemented as an RC in the feedback network. Designers often implement a low-pass pole in this fashion. Since the feedback impedance is now a parallel RC network, this circuit must be implemented with a VFB, and preferably a unity-gain stable VFB. This type of simple filtering works best for inverting signal paths. In that case, the input signal sees the gain resistor as an input impedance that converts the signal to a current into the inverting node. From there, it continues into the feedback impedance to set the gain to the output. 

This type of pole implemented for a noninverting configuration has a pole/zero pair because the gain transitions from a DC value set by the resistors to unity gain as the feedback capacitor shorts out at higher frequencies. Therefore, this circuit drops to unity gain if implemented as a non-inverting stage. In the inverting configuration, the gain continues down with a one pole response. It is important to notice that a unity-gain stable amplifier would be preferred because the noise gain drops to unity for either the inverting or non-inverting application. 

Figure 10 shows an example using the THS4281 (a low-power, rail-to-rail output VFB) in the inverting configuration. In this example, the signal path was AC-coupled through a blocking capacitor. This allowed the non-inverting bias voltage to be set to mid-supply and then have a DC gain of one to the output. Because the THS4281 includes the negative rail on the input, an alternative DC-coupled design is possible by changing the lower 8 kΩ resistor on the non-inverting input to 889 Ω and removing the input blocking capacitor. 

This AC-coupled design allows a direct comparison of this circuit to a low-power CFB that does not have an input range extending to ground. That device, the OPA684, also uses less than 2 mA supply current and gives greater than 100 MHz bandwidth in most applications. Figure 11compares the simulated response of this 1 MHz one pole rolloff circuit of Figure 10, using both the THS4281 and the OPA684 (where the OPA684 would use a single +5 V supply). Both parts give the expected one pole rolloff, but the OPA684 shows a very anomalous response at higher frequencies indicative of probable oscillations. 

(Click to Enlarge Image)
Figure 10. Inverting band-limited design using the THS4281.

(Click to Enlarge Image)
Figure 11. Simulated inverting low pass filter design.

D. Wideband DC-coupled amplifiers with good DC precision. VFB op amps have better input offset voltage and (for bipolar amplifiers) matched input-bias currents. As a result, considerably lower output DC-offset voltage and drift can be delivered as compared to equivalent CFB implementations. This consideration becomes more important as the required gain increases. Figure 12 shows an example high gain, DC-coupled inverting amplifier stage using a non-unity gain stable VFB. The OPA846 offers very good input offset voltage and offset current along with a low 1.2 nV/√Hz input voltage noise. 

To take advantage of this low noise, the gain of -20 circuit in Figure 12 implements a matched 50 Ω input impedance using only the input resistor and then a 1 kΩ feedback resistor. If a 50 Ω DC-coupled source impedance is assumed, the required bias current cancellation resistor on the non-inverting input would be 100 Ω| in parallel with 1 kΩ = 90.9 Ω. Decouple this resistor with a parallel capacitor to reduce its high-frequency noise contribution. 

Another interesting aspect of this circuit is the noise gain (NG), which is also the gain for the input offset voltage, is reduced below the inverting signal gain because of the 50 Ω source resistor. Including that value in the NG equation gives an NG = 11 while the signal gain will be -20. 

(Click to Enlarge Image)
Figure 12. Inverting gain of -20, low-output V_os design, using the OPA846.

This high-gain inverting implementation using the OPA846 can also be done using a low-power current feedback amplifier. For comparison, an OPA684 low-power CFB is compared in Figure 13for the simulated frequency response. Since high equivalent gain bandwidth is a natural advantage for CFB amplifiers, the OPA684 gives very similar small signal bandwidth to the OPA846 of approximately 150 MHz. The OPA684 uses 1.8 mA quiescent current while the OPA846 uses 12.9 mA. 

(Click to Enlarge Image)
Figure 13. Gain of -20V/V simulated performance using VFB and CFB devices.

A calculation of the maximum output DC offset error band at 25 °C for the OPA684 shows the relatively poor DC precision offered by the CFB implementation. That calculation must treat the two input bias currents separately, because they are not physically related in the CFB input stage as they are in most VFB input stages. Leaving the 90.9 Ω resistor on the non-inverting input, the total output DC error band, using the OPA684 specified maximum DC errors at 25 °C, is: 

Maximum V_os = 
±3.5 mV·11 V/V±10μA·91 Ω·11 V/V± 16μA·1 kΩ =
±64.5 mV

So, while the OPA684 can match the speed of the OPA846 in this higher gain application, if output DC precision (or lower noise) is desired, the VFB design offers considerable advantages. 

Conclusions to Part 1

In Part 1, we reviewed the key internal differences between VFB and CFB op amps as a means of identifying applications that must use the VFB topology over the CFB devices. These examples generally fall into two categories: applications that need a capacitive feedback for some reason; and circuits that need to emphasize DC precision and/or the lowest output noise. Part 2 will continue with applications uniquely suited to the CFB topology, and then introduce the newest member of the wideband op amp universe—the fully differential amplifier. Part 2 will conclude with illustrations of a few applications uniquely suited to this amplifier. 

(To go to Part 2 of this article, click here)

About the author
Michael Steffes is the Market Development Manager for High-Speed Signal Conditioning, and a Distinguished Member of the Technical Staff, at Texas Instruments Inc. With more than 25 years of experience in high-speed amplifier design, applications, and marketing, Michael currently provides product definition and customer design-in support. 

Michael earned a BSEE from the University of Kansas and an MBA from Colorado State University. He shares several basic patents in high-speed op amp designs and has written more than 85 product data sheets, scores of contributed articles, applications notes and conference papers. You can reach Michael at ti8728499steffes@list.ti.com. 

References
1. "Voltage Feedback vs. Current Feedback Op Amps," Jim Karki, Texas Instruments Application Note, SLVA051 (Voltage Feedback vs. Current Feedback Op Amps or click here). 

2. "Fully Differential Amplifiers," Jim Karki , Texas Instruments Application Note SLOA054D (Fully-Differential Amplifiers (Rev. D) or click here). 

3. "Control Frequency Response and Noise in Broadband, Photodetector Transimpedance Amplifiers," Michael Steffes, EDN, Design Feature, July 4th, 1996 pp. 113-125. 

4. "Design Methodology for MFB Filters in ADC Interface Applications," Michael Steffes. TI Application Note, SBOA114 (Design Methodology for MFB Filters in ADC Interface Applications or click here).

With both current and voltage feedback operational amplifiers (op amp) available to the system designer, how do you decide which one to use? This article is the second of two parts. It reviews several applications that are most suitable to the current feedback type, then it introduces the fully differential amplifier (FDA). All applications covered in Part 1 for voltage feedback op amps are also suitable for an FDA, but some particularly useful applications for this type of amplifier are shown here. 

Keep in mind that applications requiring flexibility in the gain setting network will benefit from the gain bandwidth independence of the current feedback (CFB) design. As described inPart 1, the loop gain equation for a CFB includes only the feedback impedance while the gain element can be varied freely with minimal bandwidth interaction. The following examples exploit that advantage in several situations where a CFB would be the preferred solution. 

Example A
If an inverting summing amplifier is required with a signal bandwidth that is independent of the number of channels summed or the required gains in those channels, a CFB amplifier should be used. Figure 1 shows an example of this approach using the high output voltage and current THS3091.

(Click to Enlarge Image)
Figure 1. Inverting summing amplifier using the high voltage/current THS3091.

Here, we assume a 0 Ω source for each of the signal sources where each channel would see a gain of "2 V/V. For these CFB designs, the feedback resistor is first picked to be close to the recommended value for that particular CFB amplifier. Then, each input resistor should be selected according to the gain required by that channel. 

Recall that this circuit implemented with a voltage feedback (VFB) has a bandwidth set by the noise gain (NG). For instance, five channels summed with a gain of -2 V/V will have a noise gain of 11. This condition would set a bandwidth for all five channels reduced to the [gain bandwidth product (GBP)]/11 even though each signal only sees a gain of -2 V/V. Using a CFB for this application retains the bandwidth much better because the parallel combination of all gain resistors does not enter strongly into the loop gain equation. 

By way of comparison, Figure 2 simulates a single channel of Figure 1 using first the CFB TH3091, then a similar VFB THS4031. The THS4031 is a low-noise, high-voltage VFB that offers approximately 200 MHz GBP.

(Click to Enlarge Image)
Figure 2. Comparative frequency response for one channel of Figure 1 using THS3091 and THS4031.

The circuit of Figure 1 produces a noise gain of 11, which shows up as a signal bandwidth close to 18 MHz for the THS4031, while the THS3091 gives about 200 MHz for each channel. 

Example B
Where frequency response peaking is required, a CFB amplifier permits this characteristic to be achieved with reduced interaction between the gain shaping and the amplifier bandwidth.Figure 3 shows an example single stage of a zero/pole pair using a high output current OPA691 CFB amplifier.

(Click to Enlarge Image)
Figure 3. Frequency response peaking circuit using the OPA691.

This example transitions from a gain of 2 V/V (6 dB) to a gain of +20 V/V (26 dB) over a 2 MHz to 20 MHz span. Implementing this design with a VFB requires a minimum gain bandwidth product (GBP) in excess of (20 × 20 MHz) = 400 MHz in order to not immediately roll off at the maximum gain setting. 

Figure 4 shows this simulation where the improved performance of the CFB OPA691 is apparent.

(Click to Enlarge Image)
Figure 4. Peaking circuit frequency response with CFB (OPA691) and VFB (OPA690).

As a comparison, Figure 4 also shows the high output current OPA690 VFB; note that the 300 MHz GBP of the OPA690 is not quite enough for this application while the CFB OPA691 achieves the maximum 26 dB gain at 20 MHz and remains there up to an approximate -3 dB bandwidth of 100 MHz. 

Example C
Sallen-Key, or voltage-controlled voltage source, active filters need a non-inverting gain amplifier that has a bandwidth far in excess of the desired filter bandwidth. While this type of filter can be implemented with VFB devices quite well, a CFB device would be preferred where higher frequency cutoff filters are needed, or where the amplifier gain needs to be flexible. The amplifier gain enters into the ideal filter transfer function as part of the Q setting equation. This gain also sets the low frequency gain in a low-pass filter design. 

The local bandwidth of the op amp used in this design moves the actual filter poles away from the design targets. In the Sallen-Key low-pass filter, the actual poles move down and to the right in the complex s-plane. This shift gives an actual filter that has lower 0 and higher Q than targeted. To the extent that the amplifier bandwidth changes with gain setting, as it would with a VFB amplifier, the actual filter poles are impacted more strongly using a VFB over a range of gains in the design. 

Using a CFB in this filter normally allows a more solid pole placement where the amplifier gain can be varied more freely in the design process because there will be minimal interaction between the required amplifier gain and the impact on the final filter pole locations. 

Figure 5 shows an example 20 MHz low-pass Butterworth filter implemented using the CFB OPA695 that gives a passband gain of +4 V/V.

(Click to Enlarge Image)
Figure 5. Second-order, low-pass, 20 MHz Butterworth filter using the CFB OPA695.

The resistor values have been adjusted slightly from an ideal analysis to account for the amplifier bandwidth effects and hit the desired frequency response exactly. 

In this design, the OPA695 gives an amplifier bandwidth >400 MHz at a gain of +4 V/V (12 dB). This bandwidth gives a 20x margin to the desired filter poles. It is this bandwidth margin that allows the filter poles to be implemented with minimal production variation and holds the stop band rejection down to higher frequencies. All Sallen-Key filters show an increasing gain (or reduced stop band rejection) at very high frequencies as the closed loop output impedance of the amplifier increases sufficiently to support a feed through path through the filter feedback capacitor. 

As a comparison, Figure 6 shows the filter of Figure 5 simulated first with the OPA695 and then with the VFB OPA820. Operating at a gain of +4 V/V, the OPA820 will have a bandwidth of about 80 MHz.

(Click to Enlarge Image)
Figure 6. Second order, low-pass 20 MHz Butterworth Sallen-Key filter.

The OPA695 implementation hits the desired maximally flat Butterworth design with a 20 MHz cutoff almost exactly. The OPA820 placed into the same circuit shows a slight peaking and reduced bandwidth as predicted. This performance contrasts with the MFB filter discussed in Part 1 of this article, where a VFB is preferred and the effect of finite amplifier gain bandwidth product (GBP) moves the actual poles on a constant 0 circle to lower Q. 

Example D
Where the gain needs to be adjusted, a CFB amplifier is preferred because it holds more constant bandwidth as that adjustment is made. Figure 7 shows an example non-inverting design where the adjustment is configured to provide a fine-scale tune from a gain of +2 to +4 V/V using a high output current OPA691.

(Click to Enlarge Image)
Figure 7. Adjustable gain using the high output current OPA691.

In all of these CFB circuits, the feedback resistor is selected and fixed near the recommended value for that particular CFB device. Any adjustments or frequency response shaping is then done using only the gain element. 

Since the loop gain does not depend strongly on the signal gain, this type of adjustment holds a more constant bandwidth using a CFB as opposed to a VFB implementation. Conversely, this circuit with the gain adjustment in the feedback resistor would have a significant frequency response variation when using a CFB. 

To compare, Figure 8 shows the circuit of Figure 7 simulated at the gain extremes using both the CFB OPA691 and a very similar VFB device, the OPA690. 

(Click to Enlarge Image)
Figure 8. Adjustable gain small-signal bandwidth comparison.

At a gain of +2 V/V (6 dB), both parts show about the same bandwidth (280 MHz) where the OPA691 is closer to a Butterworth response while the OPA690 is bit more peaked. Adjusted to a gain of +4 V/V (12 dB), the OPA691 still holds >220 MHz bandwidth while the OPA690 drops to around 100 MHz. 

Most simple circuits not mentioned thus far can generally use either a VFB or CFB device. It is sometimes suggested that some of these circuits cannot be implemented using a CFB device; this claim is often incorrect. One example that demonstrates this is the differencing amplifier. 

Example E
A single amplifier differential stage (sometimes called a 'differencing' amplifier) can use either VFB or CFB devices. The common-mode rejection ratio (CMRR) of a CFB implementation appears to be lower when compared to an equivalent VFB design. However, that CMRR is the effect of the buffer gain loss across the input stage and is quite repeatable for a particular CFB part number. It is possible to tune up the CMRR to a much higher dc level for a CFB differencing amplifier by slightly increasing the resistor to ground on the non-inverting input. 

Figure 9 shows a representative differencing amplifier using the very high slew rate, unity gain stable, VFB OPA690. Notice also that the resistors on the non-inverting side were adjusted down to achieve a matched input impedance for each source (if V₁ and V₂ are independent sources. See Reference 1 for an example of where they are not).

(Click to Enlarge Image)
Figure 9. Wideband differencing amplifier using an OPA690.

A series resistor into the non-inverting input is then added to achieve bias current cancellation, which would only work to improve output offset voltage in a VFB implementation. This is assuming 0 Ω sources for each source and two independent sources. 

This same circuit can be built using the CFB OPA691. Figure 10 shows a CMRR simulation where both inputs are tied together and driven.

(Click to Enlarge Image)
Figure 10. CMRR simulation for OPA690 and OPA691.

The resulting small output gain (large negative dB gain) is then input-referred and the negative taken to get a typical CMRR plot. The OPA690 shows slightly higher CMRR than the OPA691. In this case, the resistors of Figure 9 have been used in both simulations, and no adjustment for improved CMRR made in the OPA691 simulation. 

Differential input/output circuits
An emerging class of amplifiers called fully differential amplifiers (FDA) easily can take a single or differential input signal. and produce a differential output centered on a user-selected common-mode operation point. An alternative approach in going differential-in to differential-out has been to use standard dual op amps. A brief review of that approach helps to set the background for the FDA. These approaches are useful also because once they are understood, they open up a large range of dual op amps to the designer for possible application. 

Application A
Differential I/O circuits can be easily implemented using either a VFB or CFB. There is, however, some difference between a non-inverting or inverting input implementation with regard to how the common-mode voltage is treated. In the non-inverting input case, the two inputs show a high input impedance to the differential source (allowing filters or other passive circuits to be easily inserted up to these inputs). The common-mode gain from the non-inverting inputs to the differential outputs will be one. 

Figure 11 shows an example of this design where the wideband, high output current, dual OPA2614 is used to implement a DSL driver.

(Click to Enlarge Image)
Figure 11. Non-inverting differential I/O circuit using the OPA2614.

Here, the amplifier operates on a single +12 V supply and the filtered differential source is driven through blocking capacitors to the mid-supply referenced termination impedance. The differential gain is set to 6 V/V with a dc blocking capacitor in series with the gain resistor to further attenuate low-frequency noise and reduce output differential offset voltage. 

Whether that capacitor is present in the circuit or not, the mid-supply reference on each non-inverting input shows up as the output common mode voltage as well. This circuit demonstrates a typical 1:2 step-up transformer at the output to a 100 Ω load, giving a nominal differential load of 50 ohms.

This circuit is often implemented using dual current feedback op amps (such as the OPA2677) but can give lower output noise using a decompensated dual voltage feedback op amp (for example, the OPA2614). Figure 12 shows the simulated frequency response for this circuit showing about 50 MHz bandwidth—more than adequate for most DSL line driver applications. 

(Click to Enlarge Image)
Figure 12. Non-inverting differential I/O small-signal frequency response.

In the inverting input, differential I/O implementation, the differential input impedance is the sum of the two gain resistors; the output common-mode voltage depends on the dc voltage applied to the non-inverting inputs and the dc gain for that signal path, along with the dc common-mode voltage of the source. If the sources are capacitively or transformer-coupled, the common-mode voltage applied to the non-inverting inputs will have a gain of one to the output.

Figure 13 shows an inverting differential I/O using the very low power OPA2684 dual CFB in a single +5 V supply with a mid-supply common-mode reference and an ac-coupled input interface.

(Click to Enlarge Image)
Figure 13. Low power, inverting differential I/O using the OPA2684.

This circuit provides a common-mode output of 2.5 V with a differential gain of five and >100 MHz bandwidth while using only 3.6 mA total quiescent supply current. 

Figure 13 also shows a differential ac input impedance of 400 ohms. One advantage of the OPA2684 is that the source impedance (possibly a filter) will not interact with the amplifier bandwidth. A dual VFB amplifier used here will work, but the source impedance would then be part of the loop gain equation and might adversely impact the frequency response. 

Figure 14 illustrates a simulation of this low-power inverting input differential I/O example where a dual VFB device, the OPA2822, is also shown.

(Click to Enlarge Image)
Figure 14. Inverting input differential I/O frequency response.

That part is a unity gain stable, 200 MHz GBP device so this noise gain of six configuration shows about 35 MHz bandwidth versus >150 MHz bandwidth for the OPA2684. 

Fully differential amplifiers are essentially VFB devices with an added output common-mode control loop. Instead of an internal differential to single-ended conversion, such as standard op amps require, these devices continue the signal path to the output differentially. All of the applications discussed in Part 1 for VFB devices would also be suitable for an FDA device adapted to a differential signal path. However, there are at least two types of application circuits where the FDA provides a compelling solution as compared to standard VFB implementations. 

Application B
A dc-coupled, single-ended input to differential output with output common-mode control can best be implemented with an FDA device. One of the key considerations in this design is to match the feedback divider ratios on each side of the FDA circuit, including the signal source impedance. It is also important to understand that the common-mode control loop is level-shifted from the input to the output by setting up a common-mode current in the feedback paths. Therefore, the source must be able to sink or source a portion of this dc common-mode current. (See Reference 2 for a discussion of basic FDA operation and applications.) 

Figure 15 shows an example circuit using the THS4511, a very wideband, single, 5V-supply FDA that includes ground in its input common-mode range.

(Click to Enlarge Image)
Figure 15. Single-ended input to differential output using the THS4511.

This feature makes the THS4511 particularly useful for converting a single-ended, ground-referenced signal that swings only above ground into a differential output around a common-mode voltage. 

The THS4511 shows a very high full-power bandwidth with its 2 GHz GBP and 4900 V/μsec differential slew rate. These two characteristics together give an exceptional pulse response in a dc-coupled single to differential conversion. Figure 16 shows the simulated frequency response. 

(Click to Enlarge Image)
Figure 16. DC-coupled single to differential frequency response.

Application C
Differential-input-to-differential-output circuits with very low distortion can clearly benefit from the FDA topology. Where a low IF requirement needs the best third-order intermodulation spurious suppression using modest quiescent power levels, a transformer-coupled FDA implementation provides a surprisingly low-noise figure with exceptionally low harmonics. 

Figure 17 shows an example of this type of circuit using the very wideband THS4509.

(Click to Enlarge Image)
Figure 17. Very low noise and distortion IF amplifier using the THS4509.

This example gives a noise figure of 8.2 dB (from a 50 Ω source) while also giving >100 dB two-tone spurious free dynamic range (SFDR) through 70 MHz for 2 V_pp outputs. This performance is equivalent to a 54 dBm third-order intercept in an FDA using approximately 200 mW quiescent power. 

The 1:1.4 input turns-ratio transformer reflects the 100 Ω differential input impedance of the FDA circuit to a 50 Ω termination (Reference 3 discusses this circuit and measured performance in detail). In this case, the transformer becomes the bandwidth limiting element.Figure 18 shows the simulated performance, including the transformer model. 

(Click to Enlarge Image)
Figure 18. Small-signal bandwidth for the transformer-coupled FDA.

This single to differential gain of 7 V/V shows up as a low frequency gain of 16.9 dB. The first rolloff above is the transformer while the second break in the rolloff curve is the appearance of the THS4509 bandwidth limitation. As can be seen, this circuit gives very good flatness through 200 MHz IF frequencies. 

Conclusions
Today's circuit designers enjoy a tremendous selection of high performance wideband op amps. Newer and increasingly better devices are emerging steadily, showing constant improvement on the speed versus power tradeoff. Where the feedback element needs to be adjustable or is capacitive, a voltage feedback or fully differential device is the preferred solution. Where gain flexibility or frequency response shaping is desired in a low-power implementation, and dc precision is a secondary concern, a current feedback device would be the first choice. Many applications can use either a VFB or CFB device where issues such as the speed/power tradeoff, noise or dc precision become the deciding factor. 

About the author
Michael Steffes is the Market Development Manager for High-Speed Signal Conditioning, and a Distinguished Member of the Technical Staff, at Texas Instruments Inc. With more than 25 years of experience in high-speed amplifier design, applications, and marketing, Michael currently provides product definition and customer design-in support. 

Michael earned a BSEE from the University of Kansas and an MBA from Colorado State University. He shares several basic patents in high-speed op amp designs and has written more than 85 product data sheets, scores of contributed articles, applications notes and conference papers. You can reach Michael at ti8728499steffes@list.ti.com. 

References
1. "Wideband Complementary Current Output DAC to Single-Ended Interface: Improved Matching for the Gain and Compliance Voltage Swing," Michael Steffes, TI application note SBAA135, click here.

2. "Fully Differential Amplifiers," Jim Karki, TI application note SLOA054D, click here. 

3. "Low Power, High Intercept Interface to the ADS5424 14-bit, 105MSPS converter for undersampling applications," Michael Steffes, Xavier Ramus, TI Analog Applications Journal, 4Q2005, click here.