Embedded digital control facilitates adaptive power control strategies

Able to swap seamlessly between utility ac line and local battery backup power, the intermediate-bus architecture (IBA) that the telecoms industry has refined over decades to become cost-competitive with traditional offline-only sources is an evermore attractive proposition for system architects needing high availability in a wide range of industries. Continual development in areas such as intelligent power management strengthens the arguments in favor of the IBA approach and while such strategies have been possible for many years, the recent commercialization of digital power converters vastly simplifies their implementation. Contemporary economic and environmental sensibilities make potential energy savings impossible to ignore — and this particularly applies to demand-driven systems that experience significant load-level swings.

Evolution or continuous revolution?

Re-examining long-established practices and actively exploring new ones has never been more appropriate for a power-system designer. However, revolution is also a realistic description of the step-change in capability that recently available power converters, which exploit digital inner-loop control, deliver over analog counterparts that have decades of evolution behind them, making significant functionality or performance gains unlikely for such a well-refined platform.

To compete, the digital approach has to be better from initial product launches onwards, which is one reason it’s taken so long to successfully commercialize. Commodity mixed-signal processes that allow silicon architects to pack a measurement and control subsystem and communications interface alongside the digital PWM controller core at negligible additional cost now enable products that are electrically superior to analog designs. For instance, at 396 W with ±2% regulation Ericsson’s BMR453 almost doubles the power density delivered by tightly regulated analog quarter-brick intermediate-bus converters.

Moreover, the digital platform enables a raft of programmable functions that range from setting constants such as output voltage, sequencing delays and slew rates, and fault-condition thresholds in a one-time programming step to dynamically optimizing key parameters in a running system. With its SMBus hardware basis and a standard power-control command language, PMBus — an industry success story in its own right — makes it easy to explore and implement a level of control that’s unprecedented in analog converters and in turn enables compelling opportunities for system design improvements. Figure 1 shows the key functions within a representative digitally controlled buck converter:

Figure 1. A digitally controlled buck converter with power management and PMBus™ link

Able to adapt to line and load conditions in real time, digital inner-loop control mitigates losses using techniques that include adaptive dead-time control — that is, to vary the period between the power switches conducting to avoid shoot-through. Using a buck-converter example, the objective is to minimize the conduction period of the relatively lossy body diode in the lower sync FET. The efficiency improvement that results from optimizing this conventionally fixed parameter becomes greater with increasing downconversion ratios and higher switching speeds, and can be several percent higher. Relative to analog DC-DC converters that are typically most efficient between about 50 – 70% of full load, the result that figure 2 shows for the BMR453 is to widen the efficiency curve, which is almost flat from about 10% of full load upwards and achieves 96% or better in typical operation while also being relatively insensitive to input voltage levels. Significantly for power-converter designers, other family products of widely-varying power levels that employ common core designs reflect virtually identical efficiency characteristics that promise far greater scalability than delivered by analog circuitry.

Figure 2. Adaptive digital inner-loop control minimizes losses over wide-ranging conditions

Classic 48 VDC suits high-power systems

Reliable power is increasingly crucial as society’s reliance upon communications-system availability rises, and that data-centric infrastructures structurally consign any meaningful division between telecoms and datacomms to history. The generic model for assuring continuously available power in such systems combines an AC-DC front-end that sources utility-line power alongside a battery backup system and standby generator to supply a DC power-distribution bus; and it’s a well-proven arrangement that’s attractive to multiple industries. Keeping in mind that its roots lie in lead-acid accumulator technology that pre-dates AC-line supplies, the 48 VDC level that telecom-spec ETSI EN 300 132-2 defines as a service voltage of 40.5 – 57.0 VDC remains an excellent choice for a DC power-distribution bus for systems that may require kW+ power levels, either initially or as they evolve to meet increases in demand. Relative to say a 12 VDC distribution level that will struggle to service high-power applications, 48 VDC eases Ohmic-loss issues and reduces wiring and connector bulk.

Peripheral issues include the need to meet safety specification IEC/EN 60950-1, which is effectively a global prerequisite, which is eased by 60 VDC abnormal operation limit for 48 VDC systems. This consideration has cost-of-ownership implications that are easy to overlook. Providing that the AC-DC front-end furnishes double-insulation from the primary AC-line supply, almost any system merely requires an IBC to embody functional insulation that’s straightforward to construct. This approach is more energy-efficient than barriers that require greater separation between elements, such as transformer windings, as energy-transfer efficiency quickly degrades with increasing coupling distance.

It’s also worth remembering that the IBC’s isolation barrier rarely has any safety-related purpose, as systems almost invariably connect their input and output grounds at board level or at a remote protective-Earth reference point — short-circuiting the IBC’s isolation barrier as figure 3 shows. The “two-wire” option requires no isolation within the IBC, but most designers prefer the “three-wire” option as it offers more flexibility in EMC counter-measures. Most often, isolation within the IBC exists to protect the device from high voltages that transient currents may create in the wiring hierarchy due to events in external networks or switching between power sources. While isolated converters are intrinsically less efficient than non-isolated ones, market expectations and production economics result in vendors offering isolated IBCs with industry-standard 1500 VDC isolation that exceeds the requirements of the vast majority of applications other than truly niche environments — notably IEEE 802.3af-compliant outdoor Power-over-Ethernet links that demand 2250 VDC.

Figure 3. Practical systems most often short-circuit an IBC’s isolation barrier

Optimizing conversion efficiency for time-variant load conditions

As figure 4 illustrates, a typical IBA system board includes an IBC that transforms distribution-bus power to an intermediate-bus level that some number of point-of-load regulators (POLs) use to generate the final load voltages. Cascading DC-DC converters in this way may not be the most intuitive approach for maximizing efficiency or minimizing component count. The obvious alternative is to provide isolation and downconversion to a final level in one component, as the classic distributed power architecture (DPA) model demonstrates. Yet this arrangement can create more problems than it solves as multiple isolated DC-DC converters per board are intrinsically more complex, expensive, and less efficient than an IBC and an equivalent complement of non-isolated buck converters. Also, issues such as balancing conversion efficiency, unconditional load stability, and transient response performance make it difficult to manage the very large downconversion ratios that low-voltage silicon processes require in a single step.

Figure 4. An IBA system board with PMBus control for DC-DC converters in cascade

Yet the step down from 48 VDC to the optimum intermediate-bus level for the POLs remains difficult to specify. Despite the paradox that as semiconductor voltages fall to create greater downconversion issues for the POLs, the 12 VDC intermediate-bus level that has its origins in supplying legacy 12/5 VDC components now generally best suits boards that consume 150 W or more. Experience shows that below about 150 W, 12 VDC may not be most appropriate and that below about 75 W it’s not even a realistic option, leading vendors to offer IBCs with a range of preset output voltages such as the 9, 5, and 3.3 VDC options offered by the BMR453 and its eighth-brick derivative, the BMR454. Preset values are traditionally fine for systems that experience little load variation, but the implication for heavily demand-driven systems is obvious — for optimum efficiency, we need to adjust the intermediate-bus voltage as load demands change. 

How best to implement this strategy for a particular power system will depend upon factors such as its hardware configuration, the magnitude of the load-level changes that it experiences, and time-related demand patterns that are likely to be reasonably predictable. These factors do not have to be givens as smart algorithms can continually minimize energy losses in unpredictable circumstances without compromising supply stability, which is a key design objective. As a result, aggressive “bang-bang” control strategies are unlikely to match supervisory control software with a light-touch approach that always favors power availability over minimization, yet can respond to rapid increases in load demand.

A pragmatic approach to developing supervisory control algorithms starts with establishing a reference power loss value by recording the input and output voltage and current levels at the IBC and at each POL via PMBus commands as the system runs, possibly also monitoring and correlating each device’s on-chip temperature. In operation, application software computes power losses until reaching a threshold value that triggers an optimization cycle and trims the bus voltage to maximize efficiency. The routine always first checks for overcurrent risks and, if necessary, raises the bus voltage to safe levels.

Figure 5. System power losses depend upon intermediate-bus voltage and load demand

A breakpoint clearly exists where each combination of conditions converges, above or below which point adjustments to bus voltage deliver appreciable energy savings that are multiplied by similar savings in companion assemblies, and for communications infrastructures across multiple systems that work in parallel to provide adequate network capacity. Some people call this the “light-bulb effect” — saving a few Watts by selecting energy-saving luminaires makes consumers feel good about themselves and reduces their electricity charges, but the overall impact upon a society’s energy consumption is massive.

Development effort shifts to software

As with any other class of programmable device, from a user’s perspective the development effort that produces the physical hardware is invisible in comparison with any software development that’s necessary to apply the component. Accordingly, vendors invest in development environments that ease application challenges, and the quality of those environments is very often the final decider in an engineer’s part selection process. This model applies strongly to digital power converters, PMBus control, and power-management methods that remain unfamiliar to many designers.

More details appear on the company’s website www.ericsson.com/powermodules together with an archive of digital power conversion material that is useful as a general-purpose resource to anyone wishing to learn more about the technology and techniques for applying it.