How to keep instruments accurate inside hot, noisy PCs

PCs

Paul Packebush, National Instruments -- EDN, 4/13/2000

The PC has earned its reputation as an awful place to put sensitive measurement circuits. However, as PCs become less expensive and more powerful, engineers use them in more and more measurement and control systems. In many of these systems, the instruments go inside the PC, because putting them there makes sense. But designing and packaging circuits that accurately measure low voltages and high frequencies from within a PC's hot, noisy environment pose no small challenge. And the challenge isn't restricted to designers of data-acquisition cards. Many of today's stand-alone instruments use the same digital components as desktop PCs and therefore present the same design challenges. Nevertheless, good measurement-component design can ensure accurate and precise measurements even when the environment seems bent on thwarting those objectives at every turn.

By their nature, PCs represent an ever-changing environment over which instrument designers can exert little control. The variability of components, chassis, power supplies, and other plug-in peripherals means that PC-based instruments must operate in unpredictable environments. Although designing for an unknown environment can at first sound daunting, a logical way to start is by identifying the major sources of trouble. Until you know what problems you're dealing with, you have little chance of solving them. The key areas of concern are temperature variations, electromagnetic fields, and the management of power buses—especially ground.

The possibility that power-supply dissipation, restricted airflow, and hot CPUs will cause overheating has been an issue since the earliest days of personal computing. Also, adding peripherals to the computer chassis reduces the overall airflow and leads to temperature gradients. A typical computer with adequate airflow and ventilation has an internal temperature of 5 to 15°C greater than the external ambient temperature. Nevertheless, elevated temperatures and temperature variations must not affect the measurements.

Electromagnetic interference (EMI) is another issue instrument designers must consider. Emissions from other devices can affect signal integrity and cause instruments to perform poorly or fail meet their specifications.

In designing a computer-based instrument, you must consider both radiated EMI and conducted EMI. Radiated EMI refers to interfering signals that arrive through the air, much as radio and TV signals do. Conducted EMI uses the system's ground or power bus to transfer interference to other devices. In a PC, the video system and the power supply contribute greatly to EMI problems.

Watch out for the monitor

Although video cards are major components of video systems and are most like PC-based instruments, they are still only part of the total system. Computer monitors and cabling make up the other components. Radiated emissions from computer monitors' flyback dc/dc converters represent a primary source of EMI. Because most monitors operate at horizontal-scan frequencies of 15.74 to 80 kHz and at vertical-scan frequencies of 40 to 80 Hz, designers must work with both high- and low-frequency interference.

Whereas the computer power supplies also act as sources of radiated emissions, their overall contribution is small compared with that of video systems. Power supplies contribute to emission problems by adding switching noise to the power buses. Therefore, you must condition the system power before supplying it to a computer-based instrument.

So far, this discussion has centered on the technical hurdles associated with designing products to operate within a computer. You can't create a good computer-based-instrument design without understanding the challenges the environment imposes. As benchtop instruments increasingly incorporate PCs, manufacturers of traditional instruments must understand these issues as well. Knowing the possible problems allows designers to systematically explore solutions.

Power supplies

The first aspect to tackle is managing the power bus. The voltage that the computer supplies is rarely stable enough for sensitive measurement circuitry. Instruments that use power available in PCs must filter out high-frequency noise that emanates from other peripheral boards and from power supplies. Unless an instrument's power-supply design is robust enough to manage the main supply buses' voltage fluctuations, the instrument's accuracy will suffer.

Before you design a power architecture for a PC-based instrument, you must be aware of what power is available to peripheral boards in various types of PCs. Table 1 lists the power that is available on each PCI-bus connector. You can draw quite a bit of power from the 5 and 3.3V supplies. The ±12V supplies provide limited current, however.

Appropriate power for an instrument requires stable supplies for analog components and for digital and control circuits. Usually, this requirement translates into ±15V supplies for analog functions and a 5V supply for digital subsystems. Depending on the available power, the best approach often is to design or purchase a dc/dc boost converter to increase the 5V supply to ±20V. You can then follow the converter with linear voltage regulators to achieve acceptable ±15V and 5V supplies.

Because most boost converters operate in the 100-kHz-or-greater range, you should add a simple lowpass filter between the converter and linear regulators. The filter removes high-frequency noise from the power supply. Because of their design, linear regulators remove low-frequency noise. The combination of a linear regulator and a lowpass filter provides a well-filtered instrument power supply.

Routing power throughout the circuit also requires an understanding of basic layout concepts including power-supply decoupling and creating clean digital-control signals.

Bypassing the bypass caps

Power-supply decoupling usually requires bulk and local bypass capacitors. Bulk capacitors, which provide distributed sources of stored charge, should be large-value tantalum devices. Adding a ceramic capacitor in parallel with a larger tantalum capacitor provides even better decoupling. This combination works better than a single tantalum capacitor because tantalum capacitors have an equivalent series resistance (ESR) of 0.5 to 1W at high frequencies, whereas ceramic capacitors' high-frequency resistance is negligible. For example, a 33-µF tantalum capacitor in parallel with a 1-µF ceramic device provides a low ESR but maintains a large enough capacitance to effectively remove noise on a 15V supply line.

You must add local bypass capacitors at the supply-voltage terminals of sensitive components, such as instrumentation amplifiers (Figure 1). The bypass capacitor provides a low-impedance path to ground for ac noise that may exist on the power line. Failure to remove this remaining noise from the power supply can result in noise appearing on an analog signal. You should place the bypass capacitor as close as possible to the IC's power-input pin. This layout limits the area in which external noise can couple to the circuit between the capacitor and the IC pin. A small 0.1-µF ceramic capacitor works well as a local bypass capacitor.

Another issue to consider is passing of digital-control signals to analog components. Such signals can spread to the analog circuits switching noise that digital circuits have injected onto the digital supply's output. Whereas digital circuits can ignore some variation in the power supply, supply variations often affect analog components' operation. This problem can be especially troubling if you use the digital signals to control triggering or instrumentation-amlifier gain.

An easy way to avoid the problem is to buffer the control signals prior to passing them to the analog circuitry. For such buffering to work, the design must contain a "clean" 5V supply rail that is separate from the main 5V digital supply. You can create the clean rail by using a lowpass filter to heavily attenuate noise on a branch of the main 5V supply. Only the buffers between the analog and digital sections draw power from this rail.

Noise-free grounds

A poorly designed grounding scheme can spoil an otherwise acceptable design. For example, currents from noisy portions of an instrument can loop through the ground of sensitive circuits and inject noise. Series-connected grounds can propagate noise from one circuit block to another. Inadequate grounding results in a computer-based instrument that does not measure accurately. Fortunately, designing a good ground structure is not difficult; it requires separate ground planes for the voltage sources and single-point connections among the planes.

When designing the ground scheme, you should consider current flow in the circuit. If the analog and digital circuits share the same ground plane, currents returning from the digital signals can pass through the analog circuitry. To properly design the ground, you need to divide the circuit board into ground-plane regions (Figure 2). Simply put, you must create a digital ground plane, an analog ground plane, and a plane for the boost dc/dc converter. This type of design forces circulating currents to loop only within their own areas.

At some point, you must connect the ground planes. Ideally, you should make this connection a small, single point. For a measurement product, it is best to join the analog and digital ground planes at the A/D converter. This location is ideal because it is where the measured analog signals are converted to the digital domain.

Another technique for minimizing ground noise is to locate the power supplies within the ground planes of the circuits they power. That is, you place the linear voltage regulator for the ±15V analog supplies within the analog ground plane and you place the regulator for the 5V digital supply within the digital plane. This approach confines returning currents to the ground planes in which they originate.

Interference

Because of circulating currents, differences may exist among areas of the ground planes. These potential differences induce currents in power traces that cross boundaries between planes. These currents add noise to any power-supply voltage present on the trace. To avoid this problem, place a small inductor in series with the power trace at the point at which it crosses the boundary.

Combining the inductor with the filtering capacitors creates a lowpass filter. Taking into account the resistance inherent in the capacitor, you can use the following formula to size the inductor:

To ensure a flat response at the lowpass filter's rolloff (–3-dB amplitude-response) point, Q must not exceed 0.707. The capacitor in the equation is the "bulk" capacitor connected between the power trace and the ground plane. For example, a power trace to which you connect a 100-µF tantalum capacitor that exhibits a 0.5W ESR yields a 12.5-µH inductor value. Once you have determined the inductor value, the last step is to choose the inductor's wire size to meet the power trace's current requirements.

You should also consider high-speed digital signals. The return current of such signals tends to follow the path of least inductance, which is generally the path directly under the conductor. If such a conductor crosses a ground boundary, the return current may travel in an unexpected region to get back to its source. To avoid this problem, make sure that the high-speed digital lines do not cross directly over a boundary region (Figure 3a). When high-speed digital lines must cross a boundary, you should route them to cross at the connection between the ground planes (Figure 3b).

Designing for EMI

EMI generated in and around the computer affects the measurement performance of sensitive circuitry. Radiated interference in the form of magnetic fields is, for the most part, beyond control in a computer-based measurement system. Monitors, video cards, and other plug-in peripherals generate magnetic fields. These fields couple to circuits, adding noise and ruining circuit performance. You can deal with radiated interference by shielding the circuit or by laying out the design in a way that hardens it against electromagnetic emissions.

A shield is simply a metallic partition between the susceptible circuits and the rest of the computer. The shield keeps electromagnetic radiation from entering sensitive portions of the circuitry. Ideally, you shield the entire instrument. However, in most cases, such shielding is neither practical nor cost effective.

After determining which areas require shielding, you must design an effective shield. Unfortunately, not all shielding materials are created equal. Some provide better results at high frequencies; others work only at low frequencies. Because it is impossible to control the emissions inside a computer, the shielding design should attenuate both high and low-frequency noise.

The combination of a shield material's reflective capabilities (conductivity) and absorption capability (permeability) define its effectiveness (Table 2). However, when the distance to the emission source is unknown, you can assume that the near-field magnetic-reflection loss is zero. This assumption allows you to limit the search to high-permeability materials.

Most materials list their permeability only at dc. This omission means that you often overlook frequency dependence when you choose a shielding material. As a rule, a material's permeability decreases with frequency. Also, materials with high dc permeabilities tend to lose more of their effective permeability as frequency increases. For example, Mumetal is no better than cold rolled steel at 100 kHz.

So although exotic materials such as Mumetal and Conetic have high initial permeability, they can quickly lose their effectiveness as frequency increases. Steel and nickel, on the other hand, have lower initial permeability but do not lose their overall effectiveness as fast. Therefore, either of these metals makes a good shield.

Shielding effectiveness

To determine a shield's overall effectiveness, you can calculate the absorption loss in decibels versus the shield's thickness as follows:

where f is the frequency of the emissions, t is the thickness of the shield in inches,  is the shield's relative conductivity, and µ_R is the shield's relative permeability. As an example, Table 3 shows the absorption loss versus frequency of a 0.76-mm-thick steel shield.

good ground scheme must exist to provide a path for currents that flow in the shield. The shield should be grounded at only one point, as multiple grounds will create a current loop through the shield. Without proper grounding, the noise current in the shield produces a magnetic field that couples to the circuitry that the shield is supposed to protect. Because it is supposed to protect the sensitive analog-measurement circuitry, the shield should be grounded to the digital ground plane. This connection ensures that currents flow through the less sensitive digital side of the design.

Because every circuit board has two sides, you can place shields on both sides. However, placing a shield on only one side of the board often works quite well; ground planes on and within the board complete the shield. At least one ground plane lies between the back of the board and the analog circuitry. This separate copper ground plane and the multilayer board's other layers shield the other side of the board. Although this shield may not be as robust as a nickel-plated steel shield, it still attenuates emissions before they reach the analog circuitry. However, unless clearances are a problem, you can always add a second shield for the back of the board if you are willing to pay for the additional protection. Nevertheless, shielding is only a small part of designing for EMI.

Shielding is a quick and easy way to reduce EMI. However, it is no substitute for proper circuit layout. If large loops exist in the layout, magnetic fields that reach the circuit can induce currents. These currents then couple to the signals as electrical noise. Therefore, to harden a design against EMI, the first step is proper layout.

Proper layout begins with minimizing current loops. Simply put, you should place components as close together as possible. If there is less metal between the components, there is less area for induced currents. A good layout also uses small traces and limited amounts of metal at connections to high-impedance components, such as operational amplifiers. Using ground planes, as described earlier, also helps to eliminate circuit loops. Although being picky about circuit loops and component placement can consume a lot of time, the effort often makes the difference between an instrument that can operate in a computer environment and one that can't.

Dealing with temperature

The temperature in a personal computer is generally 5 to 15°C greater than ambient. However, the addition of plug-in peripherals can reduce airflow and cause temperature gradients within the computer chassis. These gradients can affect sensitive measurements because circuit-component values can change with temperature. Relying on calibration circuitry and onboard references is the best way to deal with the temperature issue. Such techniques enable the instrument to recalibrate itself to compensate for environmental changes.

The simplest form of calibration circuitry is a D/A converter. When you tie it to the circuit's analog-input path, a D/A converter acts as a controllable source that can add voltage to correct for gain and offset errors. Calibration is also the point at which the software and hardware resources come together. During calibration, the software driver programmatically controls the D/A converter to generate appropriate correction values for the selected input range.

Before performing any adjustment, you must determine the amount of drift in the analog-input path. One way to determine this drift is by measuring a known reference value and determining the difference between the measurement result and the actual reference value. Any difference between the two is the drift in the analog-input path.

Using this approach requires that you include at least one onboard reference in the instrument. The onboard reference must be stable for long periods at constant temperature and must also exhibit low temperature drift. As a minimum, you need one reference with a value near the full-scale analog input. If cost is not an issue, you may want to include multiple references that match the circuit's measurement capabilities on each input range.

Because the reference is stable with respect to temperature, you can attribute any drift in the measured reference voltage to temperature drift in the analog-input circuitry. By using a ground reference and this full-scale reference, you can calculate the offset- and gain-error corrections for the analog-input path. The instrument can then apply these corrections via the calibration D/A converter. From the user's viewpoint, the internal calibration should be a simple software function call. In this manner, a user can correct for temperature variations at any time.

You can't exert total control over computer environment, but you can still design sensitive computer-based instruments. If you understand the possible sources of trouble, you can systematically overcome them to produce a highly accurate measurement system inside a standard PC.

Author info

Paul Packebush is a calibration engineer who develops calibration approaches and design strategies at National Instruments (NI) (Austin, TX). He has a BSEE and an MSEE from Texas A&M University (College Station, TX). During his three years at NI, he has developed automated-calibration and test systems. His hobbies include hunting with his black Labrador retriever and reading.PCI Local Bus Specification Revision 2.0, PCI Special Interest Group, 1992.

REFERENCE

1. Ott, HW, Noise Reduction Techniques in Electrical Systems, Second Edition, John Wiley & Sons, 1988.

2. Johnson, H, and M Graham, High-Speed Digital Design, A Handbook of Black Magic, Prentice Hall, 1993.