December 4, 1997

Switching-regulator design
lowers noise to 100 µV

Jim Williams, Linear Technology Corp

Expending unconscionable amounts of bypass capacitors, ferrite beads, shields, Mumetal, and aspirin to ameliorate noise-induced effects is no longer the only way to tackle switching-regulator noise. Using a low-noise IC, you can now design switching converters with only 100 mV of noise.

Size, output-flexibility, and efficiency advantages have made switching regulators common in electronic apparatus. The emphasis on these attributes has resulted in circuitry requiring minimal board area and with 95% efficiency. Although these advantages are welcome, they necessitate compromising other parameters.

The switched-mode power delivery that permits high-efficiency conversion also creates wideband harmonic energy. This undesirable energy appears as radiated and conducted components, or "noise." Switching-regulator output noise is actually coherent, high-frequency residue that directly relates to the regulator's switching. Unfortunately, it is almost universal practice to refer to these parasitics as "noise."

The finite-storage capacity of a switching regulator's output filter components causes a slow, ramping output ripple, and the switching transitions cause quickly rising spikes (Figure 1a). Adequate filtering and linear postregulation can eliminate the ripple, but the wideband spikes remain (Figure 1b). These fast spikes' high-frequency content often corrupts associated circuitry, degrading performance or even disabling operation.

Noise gets into adjacent circuitry via conduction from the regulator-output lead, conduction back to the driving source ("reflected" noise), and radiation. These multiple transmission paths combine with the high-frequency content, making it difficult to suppress noise. One approach to ameliorate noise-induced effects is to use several bypass capacitors, ferrite beads, and shields and to use Mumetal. Alternative approaches involve synchronizing switching-regulator operation to the host system or implementing an "interrupt-driven" power supply by turning off switching during critical system operation. Another approach places critical system operations between switch cycles--running between electronic raindrops.

Minimize harmonics, noise

The most attractive design alternative to lowering switching noise is to use an inherently low-noise switching regulator because it eliminates noise concerns and maintains system flexibility. Such a regulator also doesn't force you to make the compromises of synchronized approaches.

The key to designing an inherently low-noise regulator is to minimize harmonic content in the switching transitions. Slowing the switching interval minimizes harmonics, although power dissipated during the transition causes some efficiency loss. Reducing switch-repetition rate can largely offset the losses, resulting in a reasonably efficient design with small magnetics and the desired low noise. Noise reduction by restricting harmonic generation is not new, but previous implementations were complex and narrowly applicable. You can use a monolithic approach, the Linear Technology LT1533, over a range of magnetics and applications (see box "A practical, low-noise monolithic regulator").

The 40-kHz, 5-to-12V converter in Figure 2a uses the low-noise LT1533 in a push-pull, "forward" configuration. The ratio of feedback resistors R₁ and R₂ produces a 12V output. A two-section LC filter provides high-ripple attenuation, although a one-section LC would also perform well. A noteworthy characteristic of this circuit is that the output filter--which filters the 40-kHz fundamental-related ripple--does not affect the high-frequency noise content, because the circuit itself develops little high-frequency energy.

The magnetics considerations for this circuit are simple. The regulator's symmetrical push-pull drive makes transformer behavior quite predictable. Thus, you can usually specify the transformer you need by indicating the operating frequency, power, and desired input/output voltages. The circuit in Figure 2a and related circuits can use a number of transformers with nominal input voltages of 5V and output powers of 1.5 to 10W.

Also, the inductors in Figure 2a and related circuits that follow have no special characteristics. This forward converter requires an inductor ahead of its filter capacitor, although additional LC filtering is optional. Other application circuits, such as a bipolar floating-output converter, need no output inductor unless heavily loaded, although you may still want to use LC sections for best possible ripple attenuation. In any case, inductor characteristics are unimportant. All of the low-noise circuits use Octa-Pak type toroidal core-based inductors (Coiltronics Inc, Boca Raton, FL.)

One inductor, L₁, used in the power-ground return (Pin 16) of the IC, compensates for the output-current control loop. In practice, L₁ can take several forms, including a length of pc-board trace; a small coil of wire; a ferrite bead; or a packaged, 22-nH inductor.

Figure 2b details circuit operation. Traces A and C are switching-transistor collector voltages; B and D are the respective transistor currents. Trace E is the test setup's output, representing circuit-output noise. Wideband spiking and ripple, just visible in the noise floor, are within 100 µV, even in a 100-MHz bandpass. This performance is even better than the photo shows. Removing all probes from the breadboard leaves only Trace E's coaxial connection and eliminates any possible ground-loop-induced error. The resultant trace shows 40-kHz ripple with about the same amplitude as in Figure 2b. Switching-related spikes, just faint outlines in the noise, decrease. Low-frequency noise is rarely a concern, although it is also well below 100 µV.

Other noteworthy signals in the circuit are the output of the first LC-filter section and the input current. At the first LC-filter section's output, you can clearly see the ripple's 40-kHz fundamental, but no wideband spikes are visible (Figure 3a). Expanding the time scale shows no observable high-frequency harmonics or spikes. The input current has a 40-kHz fundamental-related sinusoidal component (Figure 3b), and the driving source can easily handle these variations. Otherwise, this signal contains no high-frequency content.

Making these low-noise measurements requires specialized techniques with a precise test setup (see box "Specifying and measuring low-output noise"). Measurement technique, although not a way to obtain the lowest noise performance, must be trustworthy. You can lose hours chasing "circuit problems" that are manifestations of poor measurement techniques. Following the proper techniques prevents you from pursuing solutions to circuit noise that isn't really there.

System measurements are the real test

In the final analysis, the effect of switching-regulator output noise on the system is the ultimate test, and measuring the noise coupled into a 16-bit converter is a particularly stringent test. Crossplots of integral and differential nonlinearity (Figure 4) compare the result of powering an LTC1605 16-bit A/D converter using a bench supply with the result of powering the same converter with Figure 2's switching-regulator circuit. The residual error after taking the difference between the two corresponding differential and integral nonlinearity plots is within the test system's limit of error.

The results are impressive, but many underlying forces ultimately influence the final output noise. For example, theory suggests that simply setting the transition rate to low values achieves low noise. Practically, such an approach is workable but wastes power during transitions, which lowers efficiency. A compromise sets transition time at the fastest rate that also produces the desired noise performance. The LT1533's slew adjustments allow easy determination of this point.

The relationship between transition time and output noise for the low-noise switching-regulator circuit shows a greater than 5-to-1 noise reduction as switch transition time slows from 100 nsec (Figure 5a) to 1 µsec (Figure 5d). In Figure 5d, the noise is lower than the photo displays because the probing-induced error that monitoring the switch causes corrupts the measurement.

Trade off slew rate for efficiency

You also need to consider the relationship among noise, slew time, and efficiency. Significant noise reduction coincides with descending transition slew time until about 1.3 µsec (Figure 6); little additional noise benefit occurs beyond this point. A slew-time increase from 100 nsec to 1.3 µsec--the same region in which noise performance improves by a factor of five--results in a 6% loss in efficiency. Increasing the slew time beyond 1.3 µsec simply results in further efficiency losses without significant noise reduction, so operation in this region is undesirable.

In addition to choosing the slew rate, other factors help you achieve the lowest noise performance for your application. The filter capacitors you use should have low parasitic impedance. Sanyo OS-CON types are excellent in this regard and contribute to the performance levels that Figure 2b shows. Tantalum types are nearly as good. The input-supply bypass capacitor, which should reside directly at the transformer center tap, needs similarly good characteristics. Aluminum-electrolytic capacitors are unsuitable in low-noise circuits.

Some circuits may benefit from a 330 ohm, 1000-pF damper network across the transformer secondary if the lowest noise is necessary. Excursions of 20 to 30 µV can briefly appear during the switching interval when no energy is coming through the transformer. These events are so minuscule that they are barely measurable in the noise floor, but the damper eliminates them. Also, rigid adherence to low-noise measurement, layout, and breadboarding techniques are critical to low-noise performance. (An upcoming article details these techniques.)

Low-noise benefits

Based on Figure 2's basic circuit, you can design switching regulators with desirable characteristics, including low noise. These circuits include negative-output regulators, battery-powered regulators, floating-output regulators, low-quiescent-current regulators, high-voltage downconverters, and a 7500V isolated supply.

Transforming Figure 2a into a negative regulator is rivial because the LT1533 has a separate feedback input that directly accepts negative inputs. You simply feed back the output to the negative feedback pin (NFB) instead of the FB pin. A slight change in the feedback scale factor--changing R₁ to 9.6 kiloohms and R₂ to 2.4 kiloohms--is necessary because of the higher effective reference voltage. You also need to reverse the polarity of the output capacitors.

Running Figure 2's 5-to-12V converter from three 2.7 to 4V NiCd batteries to produce a 5 or 9V output requires only changing R₁ and R₂. For a 5V output, R₁=15 kiloohms and R₂=4.99 kiloohms. For a 9V output with 100 µV of noise--the electronic equivalent of a 9V battery--you need only to change R₂ to 3.48 kiloohms. All these changes result in the same low-noise performance as that in Figure 2.

Floating bipolar-output converter

Grounding the duty pin and biasing the FB input forces the IC into its 50%-duty-cycle mode. The circuit produces a full-wave-rectified output with respect to T₁'s secondary center tap, yielding bipolar outputs (Figure 7). Combining the forced 50% duty cycle with no feedback means that the outputs are unregulated and in proportion to T₁'s drive voltage. An output inductor is usually unnecessary. However, some inductance may be necessary at the highest output currents to limit inrush current; otherwise, the circuit may not start. For this circuit, you typically use optional linear regulators to provide regulation. With a linear regulator and output filter in use, noise is less than 100 µV. Also, with linear postregulation, the 40-kHz fundamental components are undetectable. Removing the optional output filter allows linear-regulator contributed noise and switching spikes to rise, but noise is still less than 300 µV p-p.

Low-quiescent-current regulator

The LT1533 has a quiescent current of about 6 mA, and the circuit of Figure 2a takes approximately 10 mA. You can reduce the no-load quiescent current, which increases in proportion to the load, to 100 µA by running an on-off control loop around the device (Figure 8). The control loop replaces the normal error amplifier, achieving regulation by switching the IC in and out of shutdown in accordance with loop demands. A comparator, the LTC1440, compares a scaled version of the output with its internal reference and biases the regulator's shutdown pin. Using the phase shift, or time delay, of the output LC components provides loop hysteresis. In a normal continuously closed loop, you must minimize and compensate for this phase shift. In this case, the phase shift promotes the desired hysteretic control characteristic. Local ac-positive feedback at the comparator ensures clean transitions. The loop's on-off control characteristic causes low-frequency output noise that relates to the LC tank ring. No wideband components are observable, and the noise is still within 100 µV.

Cascode design converts 24 to 5V

The LT1533's IC process limits collector breakdown to 30V. A complicating factor is that the transformer swings to twice the supply. Thus, 15V represents the maximum allowable input supply for Figure 2a's circuit. Many applications require higher voltage inputs, and you can use a cascoded output stage to achieve higher voltage capability. In a 24-to-5V converter (Figure 9), Q₁ and Q₂, which sit between the IC and the transformer, constitute a cascoded high-voltage stage. These transistors provide voltage gain and isolate the IC from their large collector voltage swings.

Normally, high-voltage cascodes simply provide voltage isolation. Cascoding the LT1533 presents special considerations because the circuit must accurately transmit, albeit at lower amplitude, the transformer's instantaneous voltage and current information to the IC. Otherwise, the regulator's slew-control loop does not function, dramatically increasing output noise. The ac-compensated resistor dividers that bias the base-drain junctions of Q₁ and Q₂ serve this purpose. RC gate-damper networks prevent transformer swings--which can couple into the IC via gate-channel capacitance--from corrupting the cascode's waveform transfer fidelity. The resultant cascode response is faithful in time and amplitude to the waveform swing at terminals 6 and 10 of the transformer, even with 100V swings. For output currents as high as 2A, noise is within a 400-µV peak.

7500V isolated low-noise supply

Extremely high-voltage isolation is often necessary for circuitry that must withstand high common-mode-voltage effects. Adding an isolation stage permits a fully floating, regulated output with a peak breakdown capability of 7500V (Figure 10). This circuit's operation and characteristics are similar to those in Figure 2a but add the benefit of the isolated output. The LT1431 shunt regulator compares a portion of the output with its internal reference and drives the optoisolator with the error signals. The optoisolator's collector output biases the LT1533's V_C pin, closing a feedback loop to regulate circuit output. The 0.22-µF capacitor stabilizes the loop.

Author's biography

Jim Williams, staff scientist at Linear Technology Corp (Milpitas, CA), specializes in analog-circuit and instrumentation design. He has served in similar capacities at National Semiconductor, Arthur D Little, and the Instrumentation Laboratory at the Massachusetts Institute of Technology (Cambridge, MA). A former student at Wayne State University (Detroit), Williams enjoys art, collecting antique scientific instruments, and restoring old Tektronix oscilloscopes.

| EDN Access | Feedback | Table of Contents |