"The term 'ground' too often seems to be associated with a sort of cure-all concept, like snake oil, money, or motherhood. Remember that, while you can always trust your mother, you should never trust your 'ground.' Examine and think about it."

Veteran analog designer Paul Brokaw of Analog Devices wrote this in 1982 (Ref 1), and his words still apply today. Unfortunately, the fact that ground is not a constant entity is easy to overlook in the design of systems with more complex signal paths to manipulate and control.

However, nothing is more critical to overall design quality than the integrity of that point that acts as the reference for all other signals. Although digital signals have some built-in tolerance for a varying ground potential, many analog systems do not. You have to work particularly hard to maintain a clean ground for any sensitive analog portion on boards and in systems with analog and digital components. As frequencies increase in all systems, the problem becomes much more critical. A/D converters can easily pick up noise from a neighboring DSP µP that's running at 20 MHz.

The culprit is current

When planning grounding schemes, you need to think about two quantities: current and impedance. Current has to flow in a circle. Despite whatever path current takes into a circuit, current has to flow through ground to get back to its source. As Brokaw states, "Think where the currents will flow."

Ground paths and connections always have some finite impedance. Current flowing through this impedance causes voltage drops. These voltage drops, which are the differences between the ultimate reference ground and a local circuit's operating ground, are the crux of the problem (see box, "The origins of noise"). The ideal role of ground is to act as the reference point for all other signals in the system; the practical role is that ground carries current.

Because it's impossible to remove the current or make the impedance zero, your job becomes tightly controlling the current flow and minimizing the impedance. If you can identify and control your circuit's largest currents, you've dealt with the largest potential polluters of ground. Ground planes are the best approach when dealing with high-speed signals, but not for precision circuits with 16-bit or greater precision.

Basic principles still apply

None of these principles is new to experts. If Brokaw were to rewrite his original application note, he wouldn't change much. Rather, he would add to the original. He first discussed grounding and decoupling of the then-most-common types of op amps. Today's op amps have ever-increasing bandwidths and may have to work in single-supply systems.

The grounding principles of single-supply components and systems differ little from those that use dual-supply voltages. However, single-supply operation can force some differences into the overall design. One key element in analog single-supply systems is your choice of a reference point. If your circuit has to deal with bipolar signals and single-supply components, you must create a new reference point.

The most obvious reference choice is halfway between V_CC and ground, typically 2.5V for a 5V system supply. Creating this point can seem straightforward. Figs 1a and b show how to create the reference using a simple resistive divider or reference IC and an op amp.

In essence, these circuits produce a substitute ground to serve as a new reference point for single-supply components. Although these schemes look fairly simple and innocuous, each has subtle differences. Fig 1a's resistive supply splitter operates directly off the main supply line. Thus, a noisy power supply affects V_CC, ground, and the 2.5V substitute ground. The substitute ground point is always exactly half of V_CC(or whatever percentage of V_CC you choose) by setting resistor values.

However, the substitute ground in Fig 1b is always 2.5V up from 0V because the circuit uses a voltage reference. The noise in this circuit exists only on the positive side of the supply-splitting circuit.

The circuit that works better with various types of op amps depends on the connection of the op amp's internal compensation capacitor. Unfortunately, information about this connection is not readily available on data sheets, which is why Brokaw included the connections of the then-popular op amps in . Ref 1

You can get some clues by looking at the PSRR (power-supply rejection ratio) for the positive and negative supplies. For op amps, PSRR is not the same for both supplies. The supply that has better PSRR depends on the compensation-capacitor connection. The supply with better PSRR at high frequencies usually means that the compensation refers to the other supply.

Some single-supply op amps refer the compensation capacitor to the negative-supply line, and positive-supply PSRR is typically better for most, but not all, op amps. The circuit in Fig 1b offers better noise rejection because the negative supply (ground) is less noisy than the positive supply.

If the op amp refers its compensation to the positive supply (making negative PSRR better than positive PSRR), Fig 1a is still noisy but less so than Fig 1b. This circuit essentially splits power-supply noise between V_CC and ground.

You have to be careful when using local decoupling for this new substitute ground. The optional ground-decoupling capacitor could cause problems for amplifiers that aren't designed to drive a capacitive load. Fortunately, numerous available op amps can easily drive capacitive loads. Also, this capacitor helps to maintain a fairly low output impedance for single-supply op amps operating at high frequencies.

Note: In Fig 1b, one of the output capacitors goes from 5V to ground instead of from 5 to 2.5V substitute ground. The goal is to avoid introducing any noise from the 5V supply into the new and clean substitute ground.

You can build the circuits in Fig 1 using the exact circuit elements the figures show, or you can buy an integrated version of the same concept. About three years ago, Texas Instruments ((800) 477-8924, ext 3437) introduced what it calls virtual-ground generators. The TLE2426 is a simple rail splitter, dividing the input voltage in half; the TLE2425 includes a 2.5V reference. These low-cost parts ($0.69 (1000)) can supply 20 mA, typically draw just 170 µA of current, and come in three-pin TO packages. Various manufacturers offer buffered references that you can use to implement the circuit in Fig 1b.

Splitting the splitter

Remember: No matter how you produce this substitute ground, it is only a reference point. If current finds a way into this "ground," this reference point easily moves around. In circuits that carry high currents, Brokaw suggests, duplicate the supply-splitter circuit. You can use one circuit for the return of high-current signals (such as the load) and the other for more sensitive analog signals (such as low-level inputs). Fig 2 illustrates this idea.

Many single-supply systems have signals that relate to the system's common ground and the new reference ground. Fig 2a takes an input signal--possibly from a transformer and referred to common ground--and shifts its reference to the new substitute ground so the single-supply components can deal with it. Amplifier A₁ acts as a common-mode stripper and refers its output signals to whatever point R₄ connects to. Typically, R₁ through R₄ are the same value, and R₅ equals R₆).

If a significant amount of current flows through the load, in this case, any circuitry using the new substitute ground, A₁ and the reference ground producer, A₂, may battle each other. Whatever current one amplifier puts into the load, the other has to carry off. Thus, it's fairly easy to disturb the reference ground. A₂ must be a stout amplifier, capable of sourcing and sinking large currents.

The circuit in Fig 2b provides some protection against disturbances of the substitute ground. In this case, the output of A₁ may be driving a heavy load, but the circuit returns the current to the A₃-produced substitute-load ground. If you don't use this load ground for any circuitry sensitive to noise, it doesn't effect circuit operation if large currents cause its potential to move around.

For signals susceptible to noise caused by large currents flowing through the load ground, Fig 1b's circuit uses A₂ to produce a clean ground. You can refer other signals that don't have much signal current to the clean ground. Now, buffer A₂ can hold it's output fairly steady because it doesn't have to deal with much signal current. If you can keep the load ground from connecting to any other signal inputs, this circuit can improve overall S/N ratio.

Decoupling isn't a cure-all

The issue of decoupling can seem deceptively straightforward. Most analog-component data sheets list the same recommendations: Place a 0.01- to 0.1-µF capacitor close to the supply pins to ground and also attach a larger 1-µF capacitor. Brokaw warns that, depending upon how you ground the feedback and signal sources, the decoupling capacitors may cause an effective disturbance larger than the disturbance that the capacitor intended to prevent (Ref 1).

High-speed amplifiers have stringent decoupling requirements because decoupling is usually necessary to minimize noise and prevent outright amplifier instability. Improper decoupling won't just impair high-speed amplifiers but can affect lower speed devices as well. Jerry Graeme, staff scientist, and Bonnie Baker, applications manager, of Burr-Brown Corp (Tucson, AZ) tell of a customer whose design included an op amp that was oscillating at a frequency beyond the amp's unity-gain crossover frequency.

Graeme and Baker found this higher oscillation puzzling: How could the amplifier produce a frequency higher than its unity-crossover bandwidth? The answer is that the oscillation didn't stem from the amplifier's main signal path. Instead, the feedback path sustaining the oscillation consists of the inductance of a too-long connection to a decoupling capacitor and amplifier (about 3 in.), through the positive supply pin, the amplifier's output stage, the load, and, finally, to ground (Fig 3). The easy solution is to move the decoupling capacitor much closer to the supply pin. When using through-hole parts, Burr-Brown also recommends, place the capacitors on the component side of the board.

Although it's easy to suggest ideal grounding and decoupling strategies for individual components, these strategies become more complex when taken to a larger scale. Decoupling each and every op amp, for example, can create a complex LC-resonant circuit on your board.

"In circuits handling fast signal wavefronts, decoupling networks paralleled by centimeters of wire generally mean trouble," writes Brokaw. You can avert this problem by inserting small-valued resistors to lower the Q of these resonant circuits.

Combining analog and digital

Also, considering grounding and decoupling of analog components is challenging, but the challenge increases when you add digital components to the mix. Fast switching speed and wide-signal swings, which both mean fast slew rates, and switching-regulator-based power supplies contribute to increased noise on the power-supply line. How to separate analog and digital signals and where to ultimately tie analog, digital, and the main reference ground together can be a brain teaser (Fig 4).

Morty Tarr, a staff consulting engineer at Avid Technology (Tewksbury, MA), has years of experience building products for desktop computers. When it comes to grounding, Tarr's job is often to make the most out of a less-than-ideal situation, where the designer has little control over system ground and power-supply connections. Generally, what you see at the motherboard is what you get.

Tarr contrasts this PC plug-in-card situation with that of benchtop instruments in which you can completely isolate the analog and digital grounds. Even if you can't reach this ideal, these instruments make it easier to control specific ground connections. For a desktop-computer peripheral card, explains Tarr, it's possible to isolate the analog power with a dc/dc converter, but it's difficult to isolate the grounds. You have the choice to combine the analog and digital ground returns at an ADC or DAC, the edge connector, or the backplane.

Often, you can dedicate several pins of the edge connector to analog return exclusively. Thus, the return currents join in the backplane only. Tarr used this approach successfully in systems requiring both high-resolution and -speed digital operation. Note: You need to place a resistor from analog to digital ground near the connector to provide some impedance when you unplug the card.

Whether you connect the grounds at an A/D converter or at the edge connector involves trade-offs. If you join the grounds at the edge connector, you may have a significant difference in ground potential between an ADC's or a DAC's analog and digital ground pins. But the switching currents of the digital section introduce a minimum amount of noise for the analog ground plane. Lowering the noise in analog ground should improve the performance of the ADC and associated analog circuitry.

If you join the grounds together at the ADC or a DAC, more noise may occur at that point than at the edge connector. But the direct connection between analog and digital ground does guarantee identical potentials at those points. In all cases, you should choose only one connection between analog and digital ground to prevent circulating ground currents or loops.

You may have difficulty in deciding where to connect the analog and digital grounds of data converters. Some converters internally connect the analog and digital ground and essentially force you to connect the two grounds at the components. A second connection elsewhere could set up a ground loop. Other converters' analog and digital grounds are separated by back-to-back diodes that give you more freedom in your ground-connection choice.

When choosing the dividing line between analog and digital circuits, you need to locate a point in the circuit where the ground can tolerate a slight voltage differential across it. To increase noise immunity, you can play with the location of this isolation barrier a bit and connect a small amount of logic to analog ground.

Carefully considering your logic-family choice can also help you minimize noise in a mixed analog-digital system. For low noise, use the slowest possible CMOS logic. Or, for the lowest possible noise, use ECL--if you can live with the higher power consumption.

1.Brokaw, Paul, "An I.C. Amplifier Users' Guide to Decoupling, Grounding, and Making Things Right for a change," Application Note, Analog Devices, 1982. (For a copy, fax Analog Devices' Literature Center at (617) 821-4273.)

2.Loaiza-Montiel, Rosie, "Layout techniques boost dynamic range for high-speed ICs," EDN, June 23, 1994, pg 99.