Op-amp designers are scrambling to keep pace with changes in ADCs, changes that often make driving these converters more difficult. Ever-increasing speed and resolution, more ADCs with single-supply voltages, and more ADCs with switched-capacitor input structures are forcing system designers to carefully evaluate the drive amplifier's performance.

Depending on the input signal and source, many sampling A/D converters--particularly the switched-capacitor types--require an input-drive amplifier to amplify extremely low-level signals, to provide a low-impedance source for the ADC inputs, or to provide the necessary drive current for the ADC.

You might ask why manufacturers of monolithic, hybrid, and modular products don't routinely include op amps or buffers in front of their ADCs. In the case of an IC, ADC manufacturers use processes that may not be compatible with drive-amplifier requirements. For example, using a CMOS process makes it difficult to meet the low-drift and -noise requirements of high-performance analog functions. In the case of hybrid and modular ADCs that combine high resolution and speed, manufacturers don't include amplifiers because applications often require different amplifiers, such as one optimized for low noise or another optimized for low distortion.

ADC manufacturers, particularly those that also produce op amps, often make specific drive-amplifier suggestions. However, the drive amp you ultimately choose needs to closely match your system requirements. Choosing the optimum drive amplifier requires looking at the problem from two perspectives. The first is sheer overall performance. The second is the specific ADC input structure and how it can affect the performance of the op amp.

To provide the best overall performance, the op amp ideally should contribute no additional error to that of the ADC. One way to ensure this is to use an amplifier whose noise, as measured in S/N ratio, is much better than the ADC's theoretical, best-case dynamic range. The familiar equation for this range is:

where N equals the number of bits. For a 12-bit system, then, an amplifier should have an S/N ratio of 74 dB.

Willie Rempfer, design manager of ADCs at Linear Technology, recommends that you also keep in mind the results of adding the noise power of two sources--in this case, the op amp and the ADC. A simple calculation tells you what the combined S/N ratio is. First, transform the S/N ratios in dB to voltages, that is, divide by 20, then calculate 10^x. Next, calculate the square root of the sum of the squares and then convert the result back to dB by multiplying the log by 20.

The results from a series of these calculations demonstrates how an op amp can affect the overall performance. For two components with equal S/N ratios, say 93 dB, the joint S/N ratio is 3 dB less at 90 dB. If you choose an op amp with an S/N ratio 3 dB higher than that of the ADC, you knock the overall S/N ratio down by only 1.8 to 91.2 dB. With a 6-dB difference, an ADC of 93 dB and an op amp of 99 dB produce an overall result 1 dB down, or at 92 dB. A difference of 10 dB produces an almost-negligible difference--0.4 dB.

The point of these theoretical calculations is that the op amp has to be much have a higher S/N ratio than the ADC to have no detrimental effect on the ADC's performance.

Another way to evaluate op amps, suggest Texas Instruments engineers Al Miller and Paul Nossaman, is to compare an op amp's performance to the weight of an ADC's LSB in volts. For example, the LSB weight of a 10-bit converter with a 4V input range is 3.9 mV (4V/1024). Compare this number to amplifier specifications, such as input- offset voltage, drift, and noise, all multiplied by the closed-loop gain, to get an idea of the errors the amplifier introduces. For example, an amplifier with a gain of 10 multiplied by an offset of 0.5 mV produces 5 mV, or 1.28 LSBs, of error.

Bandwidth, settling time

To determine the speed requirements of a drive amplifier, you need to match the amplifier's settling time to the ADC's acquisition time. Also, bandwidth requirements can be much more than you expect. Burr-Brown Applications Manager Bonnie Baker says that many customers drastically underestimate the bandwidth necessary to sustain gain accuracy. Without substantial amounts of gain over the input signal bandwidth, you can easily introduce errors that exceed 1 LSB.

Fig 1, which applies only for single-pole systems and for accuracy of 0.25 LSB, provides a general idea of how high a single-pole frequency must be to produce a particular level of accuracy. For example, consider a unity-gain amplifier with a single closed-loop pole at 10 MHz (F_P=10 MHz). According to the plot, the highest signal frequency (F) you can amplify to 12-bit accuracy is 100 kHz because F_P/F is approximately equal to [approx. sign] 100 at 12 bits. Worse yet, for 16-bit accuracy, the plot indicates that you would need almost 400 times greater amplifier closed-loop bandwidth than signal bandwidth. (Remember, amplifiers can have more than one pole, the pole of the amplifier may not be the dominant pole, and the plot applies to 0.25 LSB accuracy. The graphs for 0.5 and 1 LSB accuracy would be flatter.)

Certain applications require amplifiers with higher performance levels than those that the ADC typically dictates. Undersampling, for example, requires that an amplifier's bandwidth be compatible with the high-frequency input signal, not the slower sampling rate of the ADC.

After defining your requirements, you hope to find an amplifier that meets those specifications. Unfortunately, this may not be easy. Engineers at Analogic and Datel know all too well that the choice of ADCs to drive the companies' high-resolution and high-speed ADCs (Ref 1) is limited. Although numerous suitable op amps are available for 12-bit ADCs, only a handful of parts is suitable for driving the 14- and 16-bit, above-500-kHz, hybrid and modular ADCs that these companies produce. According to Don Travers, Analogic's product marketing manager, engineers at Analogic can spend as much time working on the front end as on the ADC itself.

When recommending op amps to their customers, both of these companies choose from a select group. Depending on the applications, the choice requires a tradeoff between distortion, noise, and settling time. Settling time is particularly difficult because few manufacturers test to a settling time of 0.001% that's approximately equivalent to 16-bit performance.

For example, Analogic recommends Analog Devices' AD843 ($3.70) (all prices quoted are for 1000-piece quantities) for applications requiring the fastest settling, but this choice doesn't result in the lowest noise performance. For lowest distortion, they recommend the AD845 ($2.76). Datel engineers also recommend the AD845, which, according to the company's tests, settles to 14-bit accuracy in 400 to 500 nsec. The AD811 ($2.85) has even faster settling to 14 bits, or 200 nsec. Even faster is Comlinear's CLC402 ($5.25), which for 2V signals settles to 14-bit accuracy in 50 nsec. Burr-Brown's OPA627 ($7.35), which is somewhat slower but accurate to 14 bits, also makes Datel's recommended list.

Some of these recommendations may be about to change. Analog Devices has just introduced a new generation of extremely low-distortion amplifiers, the AD9631 and AD9632 (both $4.12). At 1 and 5 MHz, respectively, these amplifiers exhibit typical distortion of -113 and -95 dBc. Spectral noise density is 7 nV [square root sign] Hz. Settling time to 0.01% is typically 16 nsec.

Lower input ranges

The group of high-accuracy and high-speed amplifiers starts to dwindle as power-supply voltages decrease. The current crop of 14- and 16-bit ADCs with sampling rates of 1 MHz and above typically have wide input ranges of ±5 or±10V and work from high supply voltages of ±15V and sometimes an additional +5 or 5V supply. However, these companies are considering designing lower voltage parts that might present a problem in accuracy rather than speed. For example, Analogic designer Tony Diciaccio knows of no amplifier that settles to 16-bit level without a thermal tail in the ±5V amplifier supply range.

Supply voltage has a decided impact on amplifier performance. Numerous amplifiers can meet 12-bit accuracy and distortion specs at 1 MHz, but few can do this while operating from a single 5V supply. Although many companies have already announced numerous single-supply amplifiers--at the latest count, Analog Devices has 11 families and 25 products--most are still designing op amps that can drive the higher performance, single-supply ADCs.

Designing an amplifier with wide input and output ranges and other characteristics, such as low distortion, wide bandwidth, fast settling, and capacitive drive, is no easy task. According to Walt Kester, staff applications engineer with Analog Devices, an op amp doesn't have to be rail-to-rail on both the inputs and outputs. In general, it's more important for an op amp to have an input that can go to ground than one whose input can go to the positive rail. Another important asset of an amplifier is that its output swing within V_CE(SAT) of either supply rail, which implies common-emitter or totem-pole outputs instead of the traditional emitter-follower outputs.

To achieve higher ac performance for their single-supply amplifiers, companies are designing new input and output stages. For the new OP279 ($1.31), designers at Analog Devices altered the usual input biasing to reduce distortion. The amplifier can typically supply±80 mA and has a THD specification of 0.01%. A patent-pending output stage in Maxim Integrated Products' MAX492 ($2.25) and MAX493 (from $1.45) family helps these rail-to-rail output amps drive capacitive loads. Maxim recommends the 500-kHz MAX492 family for driving ADCs, such as the company's 12-bit, 75k-sample/sec MAX187.

Still, single-supply op amps have a long way to go before they can adequately drive 16-bit ADCs or those faster than 1 MHz. For example, the current crop of single-supply amplifiers cannot meet the noise or open-loop gain requirements for driving 16-bit delta-sigma ADCs.

On the high-speed side, National Semiconductor is working to design a better amplifier to drive its 12-bit, 1-MHz ADC12062 ADC. Finding an amplifier that swings 5V rail-to-rail at this speed is difficult. Until the last few months, the company recommended its LM6361 ($1.75) amplifier, which requires bipolar supplies. However, the company recently announced the rail-to-rail LM6142 ($2.10). Compared with driving the 12062 with a perfect source, driving the ADC with the 6142 reduces S/N ratio plus distortion by just 1 dB, from 70 to 69 dB. Still, the company isn't satisfied and is working on a higher speed amplifier, the LM7131 (from $1.55 to $1.85), for release next month.

The basic problem with rail-to-rail amplifiers, says National Application Engineer Bill McDonald, is that there isn't much gain in the amplifier's feedback when the op amp is operating near the rails. This lack of gain causes settling problems, particularly if the ADC input kicks back any unusual current or charge spikes. Thus, at peaks of a sine wave, settling becomes more difficult and can potentially lead to increased second-harmonic distortion. The LM7131's design provides adequate gain bandwidth even for signals close to the supply rail. The unity-gain crossover frequency is around 70 MHz.

Performance is not the only reason that lower supply voltages complicate your choice of a drive amplifier. The input-voltage range of new high-speed, single-supply ADCs is not ground-referenced but is centered around some common-mode voltage between ground and the positive supply. For instance, the input-voltage range can be 2V p-p centered around 3V. Thus, if you start with a ground-referenced input signal, you need to either ac-couple or level-shift the input with a single-supply op amp that also has the necessary distortion specifications and drive capability. Level-shifting is important not only because of changing supply voltages but also because of changing ADC input ranges. Gone are the days of wide input ranges for most ADCs. Many input ranges are now 1 to 2V, which you may need to shift before the ADC input. You can level-shift using an op amp and resistors or using difference amplifiers, such as Analog Devices' AD830 ($2.42) and Linear Technology's LT1187 ($2.85).

You don't necessarily have to drive a 5V ADC with a 5V op amp. As long as you pay attention to the ADC's input and common-mode range, you can use a ±5 or ±15V amplifier. You may need to take steps to protect the ADC. For example, each member of the Burr-Brown's ADS family includes a front-end resistor that provides inherent input-overvoltage protection. However, all ADCs do not offer such protection. For such cases, clamped amplifiers can prevent the amplifier from driving the ADC with an out-of-range signal. Most clamped amplifiers and those with output limits need to recover from saturation quickly to keep the ADC from going into saturation.

Comlinear, Harris, and Analog Devices produce amplifiers with output limits. Harris' 350-MHz HFA1135 (around $3) runs on ±5V supplies and has a maximum saturation recovery time of 1.5 nsec. Analog Devices has also just released two clamped amplifiers, the AD8036 and AD8037 (both $4.12), which feature high-speed and second- harmonic distortion around 72 dB at 20 MHz. Thus, these devices are suitable for 10-bit systems and implement the clamping at the amplifier's input, which the company says produces better clamp accuracy and linearity than do competing products.

Focus on ADC inputs

In addition to picking the right amplifier from strictly a performance point of view, a second major factor governing the amplifier choice is the ADC's input structure. Flash converters, with their notoriously nonlinear and high input capacitances, were formerly the most difficult ADCs to drive. Now switched-capacitor input structures are usurping that reputation.

In general, all ADCs fit into one of three groups, depending on whether they have benign, flash, or switched-capacitor inputs.

Benign ADCs have reasonably high and mostly constant input impedances and cause no unusual perturbations at the output of the op amp during sampling intervals. Many of Analog Devices' bipolar ADCs, such as the 1671, 871, 872, 9022, and 9023, fit into this category.

Although flash converters used to routinely require high current at high speed to drive their high input capacitances, manufacturers such as Signal Processing Technology have reduced the problem in new flash converters. Five years ago, a 150-MHz part from the company had a 45-pF input capacitance. Now, newer architectures and processes make possible a 150-MHz ADC with 10-pF capacitance.

Converters with switched-capacitor sampling inputs are the newest genre of ADCs. Many types of ADCs, including almost all CMOS types and high-resolution delta-sigma and audio ADCs, now feature these inputs. An ADC with this switching structure doesn't automatically have high input impedance, and the input impedance can change during the sampling cycle. These inputs also cause transient currents that shock and disturb the op amp's output. The op amp then must settle back to its original buffered or amplified version of the input signal before the next conversion.

Unfortunately, it's virtually impossible to tell from op-amp settling-time specifications what the settling performance will be under these circumstances. Settling time after a transient event at the output is not the same as the settling time a company specifies on op-amp data sheets. Normal settling time refers to how long an amplifier's output takes to settle based on a step change at the amplifier's input. In driving switched-capacitor ADCs, the output is at the desired level, but the ADC perturbs this output. No op amp data sheet directly addresses this type of output settling time.

However, it is possible to get some idea of how an amplifier will react to transients by looking at phase-vs-frequency curves. If the phase response rolls off in a smooth linear fashion, the op amp will likely settle fairly effectively after an output glitch. Without a smooth roll-off, especially near the crossover frequency, peaking in the response of the amplifier will occur. Too much peaking implies that the amplifier lacks a well-behaved transient response and will have difficulty driving the ADC's transient load.

You can also take steps to minimize the effects of the output transient. The most important step is to ensure that the amplifier maintains a low output impedance over all input frequencies of interest. Op amps with high output impedances can't quickly respond to changes in an ADC's input capacitance or handle the transient currents the ADC kicks back. If the op amp doesn't settle in time for the next conversion, nonlinearities can result.

By looking at single data-sheet numbers, you might assume that most op amps have a fairly low and constant output impedance. However, most data-sheet numbers apply only when the amplifier has sufficient loop gain. Also, output-impedance-vs-frequency curves often stop short of revealing what happens at high frequencies. If the dynamic load placed that the ADC places on the amplifier is beyond the amplifier's unity-gain crossover frequency, the output impedance can be quite high.

Remember that high loop gain is necessary for low output impedance, according to the following equation:

where R_o is the open-loop output impedance and A[beta] is the loop gain. As you get closer to the unity-gain crossover frequency of the op amp, A[beta] decreases, leading to increased output impedance (Fig 2).

This impedance requirement then translates directly to a bandwidth requirement. A higher bandwidth op amp has higher loop gain and thus lower output impedance at higher frequencies. According to Harris Semiconductor linear-product-marketing engineer Chris Henningsen, this is one reason to use an 800-MHz amplifier in front of a 20-MHz ADC, which is sampling a 5-MHz video-input signal. The high-bandwidth op amp more effectively swamps out the ADC's kickback signals than does a lower bandwidth amplifier.

According to Linear Technology's Rempfer, op amps with emitter-follower outputs running with lots of current usually have the necessary low output impedance. One thing to watch out for, however, is amplifiers with inherently higher output impedances, such as those with collectors driving the output--specifically, amplifiers with rail-to-rail outputs.

Manufacturers are also doing their part to minimize the effects of the ADC's transient glitch by modifying the switched-capacitor input. You now find ADCs with the following three types of inputs: (Fig 3):

• input directly connected to sampling switch,
• input series resistor before switch,
• coarse-charge buffer.

Burr-Brown's ADS family of pin-compatible 12- and 16-bit ADCs are an example of the input-series-resistor-before-switch structure. The designers put a series resistor on the input so that charge doesn't go straight into the op amp. The ADC designers also reduced the switching current by 10. The result, according to Burr-Brown, is that any amplifier can drive members of the ADS family.

Crystal Semiconductor uses the coarse-charge buffer approach in its CS5101A ADC and all its dc-accurate delta-sigma parts (CS5504 through CS5509). The input switch connects either to a CMOS unity-gain buffer amplifier or to the input signal. The coarse-charge buffer charges the hold capacitor to a signal near the input. When the switch connects to the input, the driving circuit has only to provide charge to compensate for the buffer's offset voltage. The result is that the ADC produces low transient current.

These improvements don't mean you can ignore the amplifier, however. Crystal Semiconductor Applications Engineer Jerome Johnston still fields calls from customers who complain about missing codes, presumably caused by a poor-performing ADC. Johnston says the problem isn't usually the ADC, but a poorly selected amplifier that has difficulty settling properly, particularly around bipolar zero. The company recommends placing an RC network between the op amp and ADC to buffer the dynamic transient current from the ADC. Choose the RC values so that the time constant of the network isn't too long, which would produce averaging and offset errors.

Linear Technology and Maxim Integrated Products also often recommend a 100-pF capacitor to ground between the op amp and ADC to absorb the transient glitch. Adding this capacitor means that the amplifier has to be able to drive this 100-pF load.

In fact, Linear Technology has just introduced a rail-to-rail amplifier, the LT1368 (no prices are available as this article goes to press) that requires and, thus, is happy driving, a 0.1-µF compensation capacitor on the output. When driving an ADC--typically a low-power and low-frequency device such as the LTC1288--this capacitor forms a filter that reduces the amplifier's output impedance and swamps the current spikes from the ADC. In ADC tests with input signals under 100 Hz, the LT1368 shows less than 1 dB of distortion.

Some final words of advice: Beware of what may happen if you run an op amp at other than the tested and specified power-supply voltages; don't scrimp on testing your own op-amp/ADC pair.

References

1. Swager, Anne Watson, "High-speed, high-resolution ADCs advance a spectrum of applications," EDN, Oct 28, 1993, pg 76.

2. Kester, Walt, "Designer's guide to sampling A/D converters," Parts 1 and 2. EDN, Sept 3, 1992, pg 135 and Oct 1, 1992, pg 97.

Acknowledgments

Many thanks for useful discussions. Paul Errico, Jay Cormier, and Steve Ruscak of Analog Devices; Jerome Johnston of Crystal Semiconductor; and Willie Rempfer and Bill Gross of Linear Technology.