Op amps: to dual or not to dual? (Part 1 of 2) established the single versus dual context, and looked at layout considerations, the input stage, and package pinouts.
With all of these complex interactions between channels, when does it make sense to use the matching characteristics of a dual?Two common applications come to mind; building your own three op amp Instrumentation Amplifier, and phase compensation for critical applications. The classic three op amp instrumentation amplifier is shown in Figure 6 .
There is a tendency to use a quad for this application, but note that A1 and A2 may operate at a noise gain of five, ten, or higher. This means that input offset voltage and input voltage noise are important. A3 has different requirements, so it makes sense to use a different type of op amp (Reference 7 ).A3 is usually operated at a much lower gain, and its input noise, referred to the overall instrumentation amplifier input, is divided by the gain of the first stage, so it is much less important. Finally, the load on the third op amp is generally heavier than the first two.
The offset of the input section will depend on the Vos of A1 and A2. There are a few dual op amps on the market that have guaranteed matching between the two sections. Even if matching is not guaranteed, the two op amps will tend to match to some degree. For example, the maximum delta Vos on the AD8599 datasheet is 2.2 µV/°C, and although the matching is not specified, a random sample of 100 parts showed a maximum difference of less than 1 µV/°C.
A worst-case design should use the maximum Vos from the datasheet, with the monolithic matching giving extra margin for a solid design. One of the most important parameters of instrumentation amplifiers is Common Mode Rejection Ratio (CMRR). Pallás-Areny (Reference 8 ) showed that matching A1 and A2 for CMRR will improve the overall CMRR. This is the main reason to use a monolithic dual for the input stage.
The loading on A1 and A2 is low, but the loading on A3 could be quite heavy, so electrically and thermally, a monolithic dual and a single are better. In addition, routing considerations would also argue for using a dual and a single in this application. As a side note, the DC and AC CMRR of the output section is heavily dependent on the resistor matching and stray-capacitance matching, a fact often overlooked.
With the improvements in semiconductor manufacturing over the years, a monolithic difference amplifier with laser-trimmed thin film resistors, such as the AD8271, could actually cost less and give better performance than a discrete op amp and four 0.1% resistors (Reference 9 ).Depending on the desired CMRR vs. frequency, pc board space, overall accuracy, and total supply current, a complete monolithic instrumentation amplifier, such as the AD8226 (Reference 10 ), may be the best choice.
Power rail monitoring
For a single pole system, it is commonly known that the phase shift is 45° when the amplitude has decreased by 3 dB. Another useful rule of thumb is that the phase will be 5.71° away from zero or away from 90° for a decade below, or a decade above the corner frequency respectively, Table 1 .
Note that at even at a frequency 100 times lower than the corner frequency, the phase shift is still more than one half degree, and the amplitude is slightly less than assumed.
For systems that need extreme accuracy, in both amplitude and phase, such as power line monitoring, it is possible to use the AC characteristics of one op amp section to compensate for the phase response of the other op amp section.
The basic concept is shown in Figure 7 , and the phase response, for a normal single pole system (labeled uncompensated), and the Figure 7 system (labeled compensated) are shown in Figure 8 . The math is a little involved, so for more details, see References 11, 12, and 13.
Quads in signal chains
With millivolt signal sources, the signal chain must be low noise to maintain an acceptable signal-to-noise ratio (SNR). The gain distribution and selection of appropriate singles, duals or quads can improve performance and lower overall cost. For example, with a maximum input signal of 50 mV and a 10V output into 2 kO , a gain of 200 is required.
The four blocks in Figure 9 could be configured as a buffer, an inverting summing amplifier with a gain of -1 for adjusting the entire signal chain offset, a Sallen-Key filter with a gain of one, and a gain stage of 200.
A quad op amp could be chosen for the entire four op amp requirements. This would be a poor design for several reasons:
1)A low noise quad, such as an AD8674 would have to be chosen to get low noise int the first stage,
2) There will be electrical coupling on the pc board between the output stage and the input stage and thermal coupling on the silicon between sections,
3) A large gain bandwidth is required for the last stage.
A better arrangement, (although not the only one!) would be to take some gain earlier in the signal chain. Taking too much gain earlier can result in an intermediate stage overloading. If the gain of the first stage is ten, then the noise contributed by the second stage, referred to the input, is the noise of the second stage divided by ten. As each stage adds gain, the requirements of the following stage are reduced.
To buy an expensive, low noise quad, and use it for all four blocks is not as cost effective as using a low noise dual for the first two stages, and a low cost, general purpose dual for the last two stages.
Even if it were possible to build a perfect dual op amp in silicon, there are some packaging and pc board considerations. Duals and quads have one set of power pins, not two or four. The resistance of the bond wires can be 50-100 mO , so using one section of a dual op amp to supply 100-200 mA to a low-impedance headphone can present problems.
Multiple ground symbols on a typical schematic are all assumed to be at zero volts, but this is not true. One symbol is zero volts, and due to IR drops, all the other symbols are µV higher or lower. One inch of pc board trace can easily be over 50 mO creating an IR drop in unexpected places. Figure 10 shows the ideal world for a stereo headphone amplifier with theoretically infinite channel to channel separation.
What really existed in the real world was Figure 11 , with only 60 dB of separation. Simulation showed that the bond wires and on chip metalization did contribute some cross talk, but the major contributor was one quarter inch of pc board trace that was common to the ground returns for the left channel load and the right channel signal source. Two singles would have the advantage of better performance, lower junction temperature, better reliability, and easier pc board layout.
For the best system performance and lowest system cost, each socket must be evaluated as to the appropriate op amp required. With automatic assembly and reduced package size, using singles and duals instead of quads may have no effect on overall cost. Considering pc board layout, performance over temperature, channel separation, phase matching and cost, the best combination of singles or duals could then be chosen.
- Solomon, James “The Monolithic Op Amp:A Tutorial Study,” IEEE JSSC Vol. SC-9, No. 6 Dec. 1974
- Page 9, LT1013 datasheet at http://www.linear.com
- Page 2, http://cache.national.com/ds/LM/LMV774.pdf
- Page 2, http://datasheets.maxim-ic.com/en/ds/MAX4162-MAX4164.pdf
- Pease, Robert “What’s all this Common-Centroid stuff, anyhow?”http://www.national.com/rap/Story/0,1562,29,00.html
- Hastings, Alan “Art ofAnalog Layout” 2nd Ed., copyright 2005, Prentice Hall
- Pallás-Areny, Ramón and Webster, John G. “Common Mode Rejection Ratio in Differential Amplifiers,” IEEE Transactions On Instrumentation and Measurement, Vol. 40, No 4, August 1991, pp 669-676
- Wong, James “Active Feedback Improves Amplifier Phase Accuracy,” AN-107 at http://www.analog.com
- Soliman, Ahmed M, “Design of High-Frequency Amplifers,” IEEE Circuits and Systems, June 1983.
- Soliman, AM and Ismail, M, “Active Compensation of Op Amps,” IEEE Transactions on Circuits and Systems, February 1979