Voltage Conversion Strategy
For minimum power
consumption, it is important for the system designer to understand the voltage
and current needs of the load. Typically, some portion of the power overhead
(i.e., power consumed by the converter from the source but not delivered to the
load) of the converter is fixed, regardless of power delivered to the load, and
some portion is variable, and is directly proportional (to a first
approximation) to the power delivered to the load at any given time. The fixed
portion of the overhead in most designs is commensurate with the maximum current
capability of the converter. Operating a converter significantly below its
maximum design value, therefore, represents a waste of power because the same
output power could be delivered by a converter with a lower maximum power output
capability and, therefore, lower fixed overhead. Importantly, as the power
consumed by the load falls, the efficiency of the total system drops, because a
greater fraction of total consumed power is consumed by the converter, instead
of the load. Reducing the power consumption of the load becomes an effort with
diminishing returns, for, although the absolute value of load power saved is
directly reflected in a reduction in input power, the percentage improvement in
input power becomes less and less as the load power becomes comparable with the
converter overhead. In the limit, the minimum power the system can consume is
that of the converter itself. This means, for example, that no matter how
low-power the standby circuits (e.g., oscillators, timers) are in a
battery-powered network node, the maximum lifetime of the node ultimately is
determined by the overhead of its power converter. It is for this reason that it
is desirable to avoid the use of power converters completely, by matching the
current and voltage requirements of the load with those available from the
source. When this is done, one gets the full benefit one expects from a
reduction in load power consumption; a 40 percent power reduction of the load
(achieved, say, by shrinking the size of the circuits in a smaller-geometry IC
process), results in a 40 percent reduction in power consumed from the source
and a 40 percent increase in battery life.
Should the use of a converter be necessary, however, a method to
ameliorate its fixed overhead power consumption is to operate it in "burst mode"
(i.e., cycling it on and off). When the converter is off, the load is supplied
from the filter capacitor, the value of which usually is made as large as
possible to extend the "off" periods. When the output voltage reaches a low
threshold, the converter is restarted, increasing the voltage until an upper
threshold is reached, when the converter is turned off again. Use of burst mode
can be considered either a method of regulation (especially if no other form of
regulation is present), or a strategy for power reduction (especially if a
second type of regulation is present when the converter is active). It is
especially useful for lightly loaded converters, which may otherwise operate
very inefficiently due to their relatively
large power overhead, and for loads (such as standard cell digital logic) that
are insensitive to relatively large variations in their supply voltage. Burst
mode is often employed during the sleep periods of wireless snesor network
communication protocols, during which the converter load may consist of a single
clock oscillator and counter, used to count down the time until the next active
period.
The power consumption of wireless sensor network nodes typically
varies significantly over time; it is very low while the node is asleep, rises
several orders of magnitude for a brief period of transmission and/or reception,
then returns to its low value as the node returns to sleep. For example, a node
may have a constant 30-μA current drain while asleep,
but have an additional 400-μs-long pulse of 20 mA
every 2 seconds when the receiver is active. Supplying this load at all times
from a single converter, capable of handling the maximum power sunk by the load
at all times, would imply that for the vast majority of time, the converter was
operating in a very inefficient mode; a converter capable of sourcing 20 mA may
have a fixed overhead current of 20 μA. The average
current drain of such a system would be
A much more efficient system would be to employ two converters — a
first converter (perhaps an inductive switching converter) providing power only
when the node is active, followed by a second converter (perhaps an integratable
capacitive switching converter) providing power during sleep mode. The first
converter could be the ssme one used in the first design, and would be optimized
to supply the 20-mA active current. The second converter would be designed for
optimum efficiency supplying standby current drain values (e.g., an overhead of
3 μA while supplying 30 μA). The average current drain of this design is
The second design has reduced the average power consumption of the
system by 31 percent.
The design of such multiple-converter systems results in
additional system complexity; for example, the switching of the converters must
not produce any momentary spikes or glitches in the supply, and the warm-up
behavior of the second converter must be well-characterized so that it can be
ready to supply the load of the active network node when required. A
multiconverter design can be significantly more efficient than the
single-converter approach, however, and can greatly extend the battery life of a
network node. Burst mode may also be used in some applications to create a
low-output mode from a high-output converter, with reasonably good efficiency, and thereby get the performance of
two converters with only one set of components.
This quirk of converter behavior also must be considered in the
design of application-flexible wireless sensor network nodes. When designing the
power converter system for a network node for which the end application, and,
therefore, the sensors and actuators that may be used, is not known, it is
tempting to place a single power converter in the design, capable of handling
the worst-case envisioned power consumption by the load. Unfortunately, this
saddles the typical application with less-than-optimum overall power
consumption. One solution to this problem is to offer multiple converters to
supply the node transducers and ask the user (or OEM supplier) to select the
proper one. This solution is expensive, due to the duplication of converters and
their associated die and board area, chip pin-outs, and external components. A
more economically attractive solution is to design a single, programmable
converter, the maximum power capability of which may be adjusted via an external
port. The use of such a converter is not a panacea, however; its design is
nontrivial, and often only a limited range of power capabilities is possible
without changing external components.
The problem of matching power converter efficiency with a
time-varying load affects all supplied circuits. To optimize system power
consumption, the network node designer must understand not just the maximum
current required by all loads, but how each load's current requirements vary
with time. The analysis of each load's current requirements can become complex,
especially when multiple independent loads are driven from the same converter
(e.g., an LCD driver and standby timing circuits), or when the outputs of
multiple converters are associated (e.g., an LCD driver and its backlight). A
thorough analysis that covers all potential states of operation should be
performed to ensure that no case is overlooked and that the transitions between
states are well understood.
In addition to these factors, the required temperature coefficient
of the converter output must also be considered in a low-power design. For
example, a wireless sensor network node may be required to operate over the
industrial temperature range of -20°C to +85°C. A CMOS digital circuit in the
network node may meet its speed requirements with a 1.0 V supply at a
temperature of –20°C, but require a 1.2-V supply to achieve the same speed at
+85°C. A converter with a temperature coefficient of zero ppm/C that supplies
this circuit would require an output voltage of 1.2 V to ensure operation over
the entire operating temperature range. This represents a waste of power over
most of the temperature range, when a lower supply voltage would be sufficient.
A better design uses a converter with a positive temperature coefficient that
matches that of the load; such a converter would produce 1.2 V at +85°C, but a
lesser voltage at lower temperatures, reducing the average power consumed by the system. Other loads,
such as liquid crystal displays, can require a supply with a negative
temperature coefficient. Matching the temperature coefficient of supplies with
the needs of loads is important to minimize system power consumption, but this
can be a complex task when multiple loads are present. In the extreme, one may
see designs in which two or more regulators have the same nominal (room
temperature) voltage, but different temperature coefficients.
When multiple source voltages are required, generating them by
multiple switching converters can require a large number of components. It is
often more practical to generate multiple source voltages by employing a single
switching converter to generate a relatively high voltage, then using several
linear regulators to create the required voltages from the switching converter
output. This scheme is especially useful when the unregulated switching
converter output voltage itself is used to supply a load, when the currents
supplied by the lower voltage supplies are small in comparison to the current
used by that load, and when the desired linear regulator output voltage is not
too far from that of the switching converter output voltage. It must always be
kept in mind that the current supplied by the linear regulators is very
expensive in terms of the current drawn from the primary source (e.g., a
battery) used to produce it, due to their derivation from the relatively high
switching converter output voltage. This system can be much more economical to
produce, however, and occupy much less physical area than an alternative design
employing multiple switching converters.
Although in principle this concept could be extended,
producing lower output voltages by placing linear voltage regulators as the
loads of other linear regulators (with higher output voltages), each voltage
converter has a power consumption of its own. The efficiency of the system is
equal to the multiplication of the efficiencies of each of the converters, which
make such "daisy-chaining" of regulators inefficient when more than a very few
are used. It is best when the minimum number of voltage conversions is
made.
Sections
Syndication
