When one works as an analog applications engineer for many years supporting customer inquiries, some patterns begin to emerge. Inquires run the gamut from newcomers who have no business attempting analog design to those from analog design experts who encountered something new and unusual that challenges even the best support engineer. Unfortunately, there is also a class of inquiries that are bound to elicit a groan, mistakes the author has seen many times before and unfortunately will see many times again. It is hoped that this chapter will educate people to some of the most common mistakes and save designers from continuing to make them.

Does the phrase unity gain stable ring a bell? In early chapters of this book, the statement was made that an op amp is least stable at its lowest specified gain. Ideally, the engineer has read the material and understands the true nature of stability, and yet, during the book tour for the first edition of this book, a customer approached me with a problem: This person had a programmable gain op amp circuit where resistors were switched to program a gain of 1, 1/10, and 1/100 (see Figure 25.1). The unity gain case worked as expected, but ringing occurred with a gain of 1/10 and sustained oscillation at a gain of 1/100. In no way do I demean the individual who made this mistake; I have been known to grab an amplifier out of a bin to construct a quick circuit, only to have it oscillate uncontrollably. Invariably, a quick glance at the data sheet reveals it is a “gain of ten” stable op amp not unity gain stable. There is no other option at that point but to use a different op amp.

Fortunately, the solution to the problem is exceptionally easy. A voltage divider can be applied to the input of a noninverting op amp buffer, as shown in Figure 25.2.

As far as the op amp is concerned, it is operating at unity gain and is stable. The voltage divider rule is employed to calculate the correct degree of attenuation. The high input impedance of the noninverting op amp input does not affect the voltage divider to any degree unless extremely large value resistors are used.

If the signal must be inverted, then the inverting attenuator of Figure 25.3 can be used. It is a variation on the approach used in Figure 25.2 but takes into account the resistors used for the op amp feedback and input resistance.

Texas Instruments has a calculator available on its Web site to assist in the calculation.

A common objection to the solution of Figure 25.2 is that it introduces resistor noise to the attenuator. This is a fallacy for two reasons—first of all, the incorrect inverting attenuator also contains resistors that are going to be almost the same value. The second reason is that the carbon composition resistors that caused the problem are a fast fading memory. Most resistors now are metal film or thick film types that have much better noise specifications.

This misapplication usually occurs in cost sensitive pieces of equipment when a comparator is needed and a quad op amp has an unused section. I first encountered it when I discovered that the expensive telephone answering machine I had purchased quit working. “Why?” I asked myself, “is there an open loop op amp circuit on one quarter of an LM324, and why is it interfaced to a digital logic gate?” The answer was that somebody looked at the schematic symbol of an op amp and the schematic symbol of a comparator (Figure 25.4), saw that they look alike, and decided that they both work the same way!

Unfortunately, not even the internal schematic of the parts give much indication of what is going on (see Figure 25.5 and Figure 25.6).

The input stages look almost identical, except the inputs have opposite labels (a fine point discussed later). The output stage of the op amp is a bit more complex, which should be a clue that something is different. The output stage of the comparator is obviously different, in that it is a single open collector. But be careful—many newer comparators have bipolar stages that are very similar in appearance to op amp output stages.

So, if very little appears to be different in the schematic symbol or the internal workings, what is the difference? The difference is in the output stage. An op amp has an output stage optimized for linear operation, while the output stage of a comparator is optimized for saturated operation.

One of the easiest ways to unintentionally misapply an op amp is to misapply unused sections of a multiple section IC. Figure 25.7 shows the most common ways designers connect unused sections.

Many designers know how to properly terminate unused digital inputs, hooking them to the supply or ground. These designers may not have a clue how to terminate unused op amps. Figure 25.7 demonstrates techniques of Texas Instruments applications actually seen. I gave them titles:

Another way designers create problems is when they forget about DC components on AC signals. Figure 25.8 illustrates this problem. When an AC signal source has a DC offset, a coupling capacitor isolates the potential in the top circuit. The DC component is rejected, and output voltage is 1 VAC. If the coupling capacitor is omitted, the circuit attempts a gain of –10 on both the AC and DC components, which would be 1 VAC, –50 VDC. Because the power supply of the circuit limits the DC output to ±15 VDC, the output will be saturated at –15 VDC (minus the voltage rail limitation of the op amp).

The op amp current source circuit shown in Figure 25.9must always contain the load. Many applications put the load at the end of a cable, and the cable is on a connector. When the cable is unplugged, the op amp has positive feedback! It will hit the negative voltage rail.

The output of the current source is

It should be understood that R₁ through R₄ are much greater than R₅, and R₅ ≫ R_LOAD.

By far the most common mistake with current feedback amplifiers occurs when a designer shorts the output directly to the inverting input, as in Figure 25.10.

The designer is invariably trying to take advantage of the speed and bandwidth of the CFB to create a buffer. Shorting the output pin to the inverting input is always a bad idea, because it makes the CFB unstable. Stability criteria for the current feedback amplifier are different from that of the voltage feedback amplifier. The voltage feedback amplifier stability criterion is

The current feedback amplifier stability criterion is

As you can see, VFB stability depends equally on both R_F and R_G. But CFB stability is much more dependent on R_F. In fact, if R_F is zero, the denominator goes to zero and the stability criterion fails! This is seen graphically in actual data sheet plots (Figure 25.11).

The effect of changing R_F only slightly has an enormous effect on the CFB response on the left, with an alarming trend as the resistor is lowered. The effect is much smaller and opposite with the VFB plot on the right. The bottom line is this: Stick with the recommended value of feedback resistor for CFB op amps. That also makes a very easy solution when a noninverting buffer is desired. Just put the recommended value of feedback resistor between the output and inverting input, and the stage will work perfectly!

A capacitor often ends up in the feedback loop when the designer is attempting to do active filter design with CFAs (Figure 25.12).

Very few filter topologies work with CFAs. The Sallen-Key is one—if the proper value of feedback resistor is employed. The bottom line is that CFAs are not the best choice for active filter designs. Choose something else wherever possible.

One of the most common applications for a fully differential op amp is single ended to fully differential conversion. However, when the input signal must be terminated, the situation gets very complicated!

Looking at the circuit in Figure 25.13 and the equations that govern it, R₁ and R_t are cross defined. Solving this equation for the correct values requires a goal seeking algorithm. If the values are calculated incorrectly, the results can be

Fortunately for the designer, Texas Instruments provides a calculator on its Web site to do this task automatically.

The designer might want to consider the simpler design alternative in Figure 25.14.

Single supply operation of a fully differential amplifier is very easy to mess up (just see Figure 25.15).

What happened? The two outputs have a 3.3 VDC difference between their operating points! Remember that differential input circuits have two potential sources of DC. In this case, the designer correctly put an AC coupling capacitor between V₁ and R₁ but forgot to put one between R₃ and ground. Figure 25.16 shows what happens when the second AC coupling capacitor is installed.

When the second AC coupling capacitor is inserted, the correct DC operating point is established.

A very subtle but equally destructive problem often arises from incorrect application of the V_OCM input of the fully differential amplifier when the amplifier frequency response has to include DC, making AC coupling capacitors impossible. Consider the circuit in Figure 25.17.

The DC operating point appears to be correctly established, the outputs will swing around the V_OCM common mode point, which is established at 5 V by V₃. But, when an AC simulation is done, the results are terrible. What happened?

The problem comes when the input voltage range does not include the negative rail, in this case, ground. There are two solutions for the problem. One is to offset the inputs to the same DC level as V_OCM. The other is to choose a fully differential amplifier that includes the negative rail in its common mode range. Figure 25.18 illustrates the effect of V_OCM on the output signals.

V_OCM causes problems when it forces the outputs of the amplifier too close to the power supply rails. It is best to operate both inputs and V_OCM as close as possible to the same DC potential.

I saved the very best and most common mistake for last. And it doesn't even involve an op amp. It involves support components: the decoupling capacitors!

In Chapter 1, I mentioned some part numbers that are etched in the memory of every design engineer, at least those involved in analog design. There is one other: 0.1 μF.

Need to decouple? OK, everybody knows you put a 0.1 μF capacitor on every power supply input and the job is done, right? I can disprove that truism very easily with two words: cell phone.

Put your cell phone near your prototype circuit, which is bypassed with 0.1 μF, and make a call while monitoring the output on a high bandwidth oscilloscope. You will see horrendous 2.4 GHz leakage!

An alternative version of this problem came from some cellular telephone base station installers who called in a panic, “We have 90 MHz noise running all over our system—and can't figure out where it is coming from.” A suspicion on my part asked them to tell me the exact coordinates where they were installing the system, and they provided the exact latitude and longitude. A quick check of the FCC database revealed the problem. I asked them, “Are you anywhere near the tower for W____ 90.5 FM, a 100,000 W NPR station listed at those coordinates?” They told me on the phone that they could see the transmitter 5 ft away—they were colocating with the station!

The point of this is that their board was bypassed with 0.1 μF capacitors. While that worked fine for the digital portions of the board, the analog portions were being clobbered by radiation of the powerful 90.5 MHz FM station. Conventional thinking is that the lower the value of capacitance, the lower the frequencies it will filter. So, 0.1uF should get rid of just about everything because it is a very large value (relatively speaking). This conventional wisdom is wrong! The actual case is the exact opposite.

Where did the value 0.1 μF come from, anyway? A high technology store near me used to have antiquated computer boards as a wall decoration. Backlit with white light, the translucent green boards made a pretty sight. But, they were also populated with 0.1 μF decoupling capacitors. A quick survey of the circuitry revealed that the clock rate of the old computer had been 1 MHz.

So, the 0.1 μF capacitor value seems to have come from bypassing TTL logic in the 1960s! Isn't it time to rethink the issue a bit, in light of op amps and other analog components that can operate to frequencies of 3 GHz, especially when almost every engineer carries a 2 W, 2.4 GHz transmitter into the lab (cell phone)?

The reality of the situation is that a good 0.1 μF capacitor with an X7R dielectric exhibits a resonance in the 10 MHz region. This is due to parasitic inductance creating an LC circuit. Below 10 MHz, its impedance is capacitive, decreasing almost linearly on a logarithmic plot until it reaches the resonant frequency. Above the resonant frequency, the impedance is inductive. Since inductor resists the flow of high frequencies and passes only low frequencies, the decoupling capacitor is useless above its resonant frequency.

Looking at representative plots from capacitor manufacturers, at 100 MHz, the venerable 0.1 μF bypass capacitor has become an inductor with an XL of at least 1 Ω. By 2.4 GHz, XL has risen to above 10 Ω.

A good rule of thumb for effective bypassing is to put several capacitors in parallel. The standard 0.1 μF capacitor does quite nicely for frequencies up to 10 MHz, a 1000 pF NPO dielectric does nicely up to 100 MHz, and 33 pF NPO eliminates frequencies in the 2.4 GHz region. Bulk decoupling of the power supply as it enters the board eliminates low frequency ripple.

Here is a truism to replace the older one: When poor decoupling is suspected, decrease (do not increase) the value of the capacitance.