Here, the emphasis is firmly on increasing power density while maintaining tight regulation with the least possible conversion loss -- objectives that traditional dc-dc converters using analogue control loops have generally found to be mutually exclusive. But in an industry first, Ericsson's series of BMR453 isolated quarter-brick dc-dc converters use a digital control platform that increases power density from a typical converter's 300W level to 400W, at the same time maintaining +/-2% regulation and a typical 96% conversion efficiency for virtually all of the converter's useful output power range -- an approximately 2% efficiency improvement over the company's previous generation of analogue converters.

Freewheeling diode

So, what is a digital power-control platform, and how does it differ from its analogue equivalent? Taking a conventional non-isolated buck (step-down) point-of-load converter as an example, both approaches require a series switch that commutates the input voltage, an inductor that stores and releases energy into a smoothing capacitor and a "freewheeling" diode that completes the current-flow loop, leading to the familiar arrangement shown in Figure 1. The circuit's operation is conceptually simple: the control block closes SW1 for a period T_on, current builds linearly in L1, and the voltage on Cout rises. When this voltage reaches a target value, the control block opens SW1 for a period T_off and the stored energy within L1 discharges into the load, returning through D1. To regulate the output voltage, the control block monitors the output voltage and adjusts a pulse-width-modulation (PWM) waveform of duty cycle D to SW1. The value of D is proportional to the ratio of the output to the input voltage. That is, D = Vout/Vin = T_on/T_on + T_off. The inductor and output capacitor also act as a filter, removing unwanted artefacts that the switching process introduces. In any practical converter that sources more than a few Watts, multiple enhancements and adjustments to control strategies apply. For instance, D1 is normally replaced with a MOSFET in parallel with a Schottky diode, with the control block turning on this composite SW2 switch when SW1 is off. This "synchronous rectification" arrangement reduces the forward conduction losses in D1 to the level of the MOSFET's on resistance, but requires careful timing that's a compromise between ensuring that the two MOSFETs are never on together -- which would trigger mutually-assured destruction -- and minimising the time that the Schottky diode alone conducts. The period between these switching events is normally known as dead-time.

Reference voltage

Similarly, analogue control schemes for the control block are typically a compromise between performance and complexity. Possibly the most familiar scheme is voltage-mode feedback, when an error amplifier monitors Vout and a reference voltage to generate a control voltage signal. A comparator then compares this signal with a reference ramp waveform to vary the PWM waveform's duty cycle in response to changing input voltages and output load factors (See Figure 2). Alternatively, it's possible to derive the ramp waveform directly from the current that flows within the inductor, again applying this voltage to one input of a comparator whose other input is tied to a reference voltage. This hysteretic current-mode control approach requires few components and also simplifies feedback-loop compensation by effectively removing the inductor from the picture. But to maximise performance over a representative range of input and output conditions, most contemporary analogue buck converters combine current and voltage feedback. Again, the ramp to the PWM comparator derives from the current that flows within the inductor, with the comparator's other input being the error amplifier's control voltage signal. In practice, a typical current-mode controller that operates with duty cycles of around 50% or more invariably requires an additional compensation ramp to avoid instability. This is just one of myriad complexities that dc-dc converter designers face, and which apply more or less equally for analogue or digital control architectures. Nevertheless, careful design allows Ericsson's engineers to routinely achieve conversion efficiencies of 94% using analogue control techniques.
To achieve comparable or better performance, a digital control architecture must simulate the actions of its analogue equivalent (See Figure 3). In this instance, a digital PWM generator creates the waveform that controls the MOSFET switches in response to control signals that a digital error amplifier supplies. The inputs to this amplifier comprise a digital reference value and the results from an analogue-to-digital converter (ADC) that continuously converts the output voltage into digital form. Arguably the smartest part of the system is the digital proportional-integral-differential (PID) filter that stabilises the feedback path.

Ability to adjust

In conjunction with firmware, this digital-signal-processor (DSP) based control block is also able to adaptively fine-tune the converter's responses to changes in input and output conditions on a cycle-by-cycle basis. This ability is in stark contrast to fixed parameters such as time constants that discrete components dictate within an analogue control system. A digital converter's ability to adjust, for instance, the dead time between SW1 and SW2 in Figure 1 can contribute as much as another 1% efficiency to the performance of an already very good analogue design, without in any way compromising other figures of merit, such as transient response time and output noise levels. Furthermore, and unlike analogue components, digital systems do not suffer from time and temperature-related drifts that compromise performance. And where such variations unavoidably occur outside of the digital loop -- such as within the MOSFET switches, the inductor and filter capacitors -- it's typically possible to characterise and compensate for them digitally. Today, IC solutions are increasingly available that hugely simplify this task, leaving the power-supply designer free to concentrate on optimising the application. Multiple additional benefits follow from adopting digital control, such as slashing component count and improving reliability. But from the user's viewpoint, the digital core offers easy integration with digital power management schemes that require substantial additional circuitry to realise within an analogue converter. Comparatively speaking, features such as monitors, alarms, sequencers and digital communications such as PMBus come more or less for free.

Carefully constructed

Ericsson has already demonstrated the effectiveness of all-digital architectures within isolated and non-isolated point of load regulators. A carefully constructed digital power-control system outperforms its analogue equivalent on every count -- conversion efficiency, regulation, including noise and transient response, component count, reliability and build cost -- while offering the ease of control that accompanies digital implementation. The company's BMR453 series now extends the digital control concept to tackle the isolated intermediate bus quarter-brick converter that supplies the point-of-load converters (See Figure 4). At the heart of these new converters lies a custom control IC that Ericsson developed in partnership with DSP experts Texas Instruments. Offering 1500V input-to-output isolation that meets EN60950 requirements, the BMR453 converts telecom-standard 36 to 75VDC distribution rails to output voltages that are adjustable via the PMBus interface from 8.5 to 13.5VDC. Set to the normal 12V level and capable of delivering 33A, this represents a power level of some 400W -- an approximately 33% increase in power density over the best previously available quarter-brick converters in this market segment. Moreover, the efficiency performance is truly impressive for an isolated converter, being nominally flat from around 10 to 100% of output loading as Figure 5 shows. You can connect multiple BMR453s in parallel for more power, when a synchronisation facility ensures that each converter operates at exactly the same frequency. This feature simplifies filter design to quash conducted and radiated emissions, when a simple external LC filter enables the converter to meet class-B requirements of EN 55022, CISPR 22 and FCC part 15J. The module's operating frequency is 140kHz by default, but is adjustable via the PMBus interface. Other features include remote sensing, configurable overtemperature, overcurrent and overvoltage protection, and power-good indication.

Usable with any standard two-wire I2C or SMBus hardware, the revision 1.1-compliant PMBus interface also allows users to set numerous operating characteristics, including soft-start ramp times and voltage margining thresholds. A system controller can interrogate the module to extract a wealth of data such as input and output voltages, output current, internal junction temperatures, switching frequency and duty cycle. For evaluation and development, Ericsson's CCM software provides a graphical user interface that allows you to "see inside" the running converter. Available now, the module is available with an optional baseplate that simplifies thermal management in many applications. Figure 6 shows the baseplate together with the relatively few components and clean layout on either side of the circuit board that contribute to a calculated mean-time-to-failure of 1.1 million hours. An evaluation kit is available with a board, operating manual and a CD that includes the GUI and cabling.

Hall A4.260