Blog: Do I Shut Off That Switching Supply?

Article By : Bill Schweber

There are different ways to attenuate power-rail and system noise during precision A/D conversion cycles and, as always, there are tradeoffs among the choices.

Mention “switching supply” and the first two instinctive, associated reactions are the terms “efficient” and “noisy.” Conversely, say “LDO” (low dropout regulator) and the opposite descriptive terms are used: “inefficient” and “quiet.” There’s no denying that those clichés are true but be careful and confirm them: as with most clichés, there are exceptions under some conditions and circumstances.

Certainly, even a low-noise switch-mode power supply acting as a DC/DC regulator is not as quiet as a linear regulator. Often, however, system-level considerations on dissipation and run time mandate a switcher, even though the design may need extremely low noise on some DC rails. These areas include sensitive, low-level  RF front ends as well as precision, high-resolution A/D conversions.

Fig 1: STOP mode operation for the TPS62840;  blue is the input signal for the STOP pin, magenta is the output voltage, and green is the inductor current, all taken with VIN = 3.6 V, VOUT = 1.8 V and IOUT = 10 mA in (a) PFM operation and (b) forced PWM operation. The measurements include COUT = 10 µF. (Image source: Texas Instruments)

For the RF front end, the duty cycle where low noise is need is 100%, and so it may make sense to use a local LDO in a topology where it operates from a noisier switcher. In addition, the rails for that front end will likely need filtering and careful PC trace layout or discrete power-wire routing as well to avoid noise pickup. In contrast, for the precision A/D conversion, the need for clean power is only during the conversion cycle itself and is thus intermittent, and usually at a fixed update rate.

Regardless of whether the conversions are done “on demand” or formal and periodic, one approach which has been used for many years is to shut the switcher off during the conversion cycle, and have the conversion circuitry “coast” on noise-free power provided by a capacitor. This certainly works but it requires solid analysis of converter power needs and conversion-cycle length to properly size the capacitor.

There are ICs which can ease implementation of this approach. For example, the Texas Instruments TPS62840 is a 1.8-V to 6.5-VIN, high-efficiency 750-mA step-down switcher with a “stop” function controlled by a package pin, and which eliminates switching noise while allowing the capacitor to provided needed power, Figure 1. It doesn’t deliver a lot of current, but that’s fine for the intended uses; it is discussed further in a recent TI blog “Advantages of the STOP function for low-noise data-acquisition applications.”

This is clearly an attractive solution, but it also requires care in the conversion-related software. The switcher must be stopped, the conversion initiated and completed, and the switching function restarted, all within the time window before the capacitor’s droop affects operation.

Equally important, appropriate software structuring and test is needed to be sure that a latent bug or high-priority interrupt does not delay either the stop or the un-stop directives. If this happens, there would be errors in the conversion due to noise or other hard-to-reproduce data, especially if the problem is intermittent. In effect, you are eliminating the noise by putting another burden on the software; or, perhaps you’ll feel it necessary  to include some sort of discrete, hardware-based timer circuit (and, no, the cheap 555-class timer is most likely not adequate here).

Fig 2: Comparison of radiated emissions for LT8614 and somewhat noisier LT8610 versus CISPR25 Class B radiated limits. (image source: Analog Devices)

Another approach is to use a very quiet switching regulator such as the LT8614 “Silent Switcher” from Analog Devices. This 42-V, 4-A synchronous step-down switcher easily exceeds the stringent CISPR25 standard for radiated emissions, Figure 2, and that is in addition to the low noise on the DC rail itself, which is below  10 mVP-P and independent of load current. The designer has to understand how much of the unwanted noise occurring on the conversion process is radiated from the DC/DC switcher, and how much is a function of noise seen as ripple on the DC output rail.

What is also interesting but not surprising about noise attenuation of these Analog Devices “silent switchers” is how they represent the reality of trying to “eke out” that last bit of performance improvement. In most cases, there is no single “magic bullet” which cuts the noise dramatically in one sharp stroke. Instead,  it’s more likely a methodical look at every known and perhaps not-yet understood noise source and how it can be knocked down, as detailed in the Analogue Dialogue article “Silent Switcher Devices Are Quiet and Simple” and LT Journal article “Silent Switcher Meets CISPR Class 5 Radiated Emissions While Maintaining High Conversion Efficiency.”

This route to minimizing error sources is not new, of course. An excellent exposition of it was in one of the first articles published in EDN (1976) by the late, much-missed analog-design genius Jim Williams, “This 30-ppm scale proves that analog designs aren’t dead yet.” His scale was designed to meet some very aggressive objectives: it had to be portable, low cost, resolve 0.01 pound in a 300.00 full-scale range (that’s the 30 parts per million), never require calibration or adjustment, and have an absolute accuracy within 0.02%.  To achieve this, he looked at first, second, and even third-order error sources, and methodically addressed them using a variety of tactics. Despite the article’s age (almost 50 years!) and the many changes in components and technologies, the underlying lessons it makes are still valid.

Have you had to deal with noisy DC rails or an operating environment where making precision A/D conversions and measurements was compromised? How did you handle it? Was there a single step that took care of most or all of the problem, or did you have to use a layered approach?

Leave a comment