Magnetic-field measurements hold the key to reducing dc/dc EMI

Performance-driven measurement instrumentation requires low-noise, high-bandwidth linear front-end circuits that combine with equally well-performing A/D converters and clocking (Figure 1). Designers work to quantize the measurement of interest into a digital signal early in the processing chain to keep out unwanted noise. Look at your favorite instrumentation Web sites. A brief scan of dc-measurement and ac-source-and-measurement instrumentation turns up instruments with dynamic ranges of 120 dB or more. Engineering for dynamic range is a search for the source of every spurious signal. High-performance-instrument designers must be aware of all of the potential noise sources—not just the usual culprits, such as power supplies and digital activity. As dynamic range exceeds 100 dB, engineering for high SNR leads to investigating the charge pump running in the FPGA, the thermal gradient that occurs when the processor starts and stops, and the magnetic coupling from the other instrument that someone set on top of your instrument. A significant part of the design is isolating precision analog circuits from internal and external electromagnetic activity.

Today's instrumentation-and-measurement industry is undergoing a transformation. After years of performance-based engineering, market forces are leading to new open architectures (Figure 2). With customers wanting the advantages of open architecture without performance compromise, the new environment brings new engineering challenges.

For example, consider the semiconductor-test industry, in which the test requirements of SOC (system-on-chip) ICs drive up the breadth of instrumentation that test systems must contain. Time to market and the cost of ownership of large ATE (automatic-test-equipment) systems have combined to define a critical need for an open-test-system architecture (Reference 1). The trend exceeds the bounds of just ATE, however. A growing need exists for high-performance, modular VXI and PXI instrumentation for production testing and characterization. Test-equipment architectures are opening up as a strategy to reduce cost through greater flexibility, which leads to higher efficiency, greater reuse, and lower barriers to competition among suppliers.

So where does this trend leave instrument-development teams? During the era of performance-driven design, development teams had control over a large part of the system architecture. That's not the case in open, card-modular architectures, such as VXI. In such architectures, the backplane interface and physical-packaging limits highly constrain design engineers. Engineers need to place more emphasis on environmental issues, such as cooling, power conversion, and EMI (electromagnetic interference). One of the more significant of these challenges is EMI from power-conversion components within the instrumentation system. A dc/dc converter within the system relieves a combination of space and power-supply constraints, but it also generates noise, which could be the factor that limits your spurious-free dynamic range. This scenario can occur whether this noise originates in the affected instrument or from a noisy neighbor whose design did not require the same attention to dynamic-range requirements.

Near-field, radiated EMI can create noise problems for sensitive instrumentation. Near fields contain both electric and magnetic fields in proportion to the impedance of the source (Reference 2). Low-impedance circuits—that is, low relative to the 377Ω impedance of free space, or air—emit predominantly magnetic fields, whereas high-impedance circuits emit predominantly electric fields. Coupling includes capacitive and mutually inductive coupling depending upon fields present and the configuration of the victim circuitry. Because circuit impedances in switch-mode power-supply circuits tend to be low and electric fields are relatively easy to shield, this article focuses on magnetic coupling Figure 3).

Faraday's Law leads to an understanding that the electromotive force—essentially voltage plus any resistive losses—in a circuit is proportional to the rate of change of the magnetic flux within the circuit. No voltage is induced if the rate of change is zero. Magnetic interference is an ac issue with a higher degree of coupling as frequency increases.

Magnetic flux, Φ_M, can be self-induced, as with the product of inductance and current, or mutually induced, as with the product of flux density and loop area (see sidebar "Magnetic circuits"). The relationship in the following equation is interesting: E=–dΦ_M/dt=d(LI)/dt=–d(BAcosθ)/dt, where I is current, B is magnetic-flux density, and L is inductance. The equation says that you can induce an error voltage into a circuit by changing any one or more of the parameters. A given percentage change in current, magnetic field, or inductance (loop area) produces the same effect on induced voltage. Therefore, the design practice of reducing loop areas in high-performance circuits to eliminate errors from conducted emissions also reduces errors from magnetic coupling.

Although high-performance design practice dictates that you minimize loop area and shield necessarily susceptible components, such as filter inductors, it's always better to stop noise emissions at their source. Toward that end, instrument designers would like to compare near-field performance in the time and the frequency domains before they select power converters for use in instruments. Magnetic-field specifications are not yet at this level of maturity, however, so device characterization is necessary.

For example, two similarly specified dc/dc converters have been characterized for magnetic-field emissions with a small loop antenna. Both converters are one-eighth-brick, wide-input-range devices using the same input voltage, 48V; output voltage, 5V; and load resistance, 4Ω. Both share the same conversion architecture—a fixed-ratio isolation stage following a regulation stage to support a 35 to 75V input range. These converters have two power magnetic sections within the design, but both run at the same frequency. Neither converter provides magnetic-emission data within the specification sheet.

In this characterization setup (Figure 4), a small pickup loop senses magnetic fields in the area above the dc/dc converters (see sidebar "Making your own magnetic-field-measurement probe"). The amplified-loop output connects to an oscilloscope and a spectrum analyzer. The magnetic emission of the dc/dc converters contains a broad band of energy from the fundamental switching frequency that reaches to 50 MHz or more. It is important to treat the measured signal's distribution network as a high-frequency transmission line. The signal runs past the high-impedance oscilloscope input with a BNC tee at the scope input. The line terminates at the spectrum-analyzer input.

To best represent a victim circuit in a neighboring slot, orient the probe tip in a plane parallel to the converter's pc-board substrate. Scan the surface of the board for maximum output; the strongest field is above the isolation transformer (second stage). The intent of the measurements, which take place about 0.65 in. above the top surface of the transformer, is to place the pickup loop 1 in. above the plane that would represent the surface of the motherboard if you mounted the converter in an appropriate through-hole design (Figure 5).

Viewing measurement results in the time domain, you can see the converter's fundamental frequency and ringing frequency, and you can get a sense of the magnetic-field intensity (Figure 6). These converters demonstrate a trade-off in the converter's magnetic design. The isolation transformer's leakage and magnetizing inductance mutually couples to the measurement probe. Brand X has a significantly lower magnetizing inductance in its isolation transformer, as the higher fundamental field in the measurement shows. Brand Y has lower leakage inductance and therefore higher ringing frequency. The ringing that occurs around the switching transient is the result of leakage inductance and the switch's parasitic capacitance.

You can make two observations based on the derivative relationship of the earlier equation: A square-wave response means that the magnetic flux is changing linearly. The magnetic component is operating in a linear region, and current is increasing linearly. In broad terms, the magnetic field for Brand X is an 18-µtesla p-p triangular wave (see sidebar "Converting voltage measurements to magnetic-field data").

Although lower leakage inductance is better for reduced emission, the higher frequency resonance couples greater peak voltage at the higher frequency (Figure 7). A closer observation of the switching transient provides some insight into the resonance within the dc/dc converter.

If your main concern is for spectral interference in an ac-source or -capture instrument, you may be more interested in the information the spectrum analyzer provides. Taking a broad look at Brand Y's magnetic-field spectrum, you can see the resonance near 10 MHz and the components peaking at 20 to 25 MHz (Figure 8). Table 1 summarizes the data from these samples.

Open-instrumentation architectures offer an important role for dc/dc converters. If the converters are not the sources of performance-limiting noise, they open a platform to a large set of applications. This article examines two similar converters using a high-bandwidth magnetic probe and finds different results. Because a system is only as quiet as its noisiest neighbor, anyone wishing to participate in open-instrument development should carefully evaluate to ensure a performance-compatible environment.