Cell phone power management requires small regulators with fast response

Cell phone power management requires small regulators with fast response
The notion of power management means different things to different people. It can describe the software controlled turn-on and turn-off of different blocks of a complex system, like a cellular phone or notebook PC, to optimize the power consumption and battery life of the whole box. It can mean the collection of different voltage regulators that supply many different voltages these systems require, including their sequencing and supervisory functions. The term can also include battery charging, monitoring and fuel gauging. This paper will concentrate on the voltage regulators and their role within the cellular phone.

Voltage regulators are used in cellular phones for different reasons. A couple of phone generations ago, when the different sub-circuits were built on 0.5u-0.6u CMOS processes, everything could theoretically operate from the battery directly. However, it is easy to show that adding a regulator between a CMOS digital circuit, or a switched capacitor based mixed signal circuit, and the battery will significantly increase battery life. This is true even using regulators with relatively low efficiency, like linear regulators. The reason for this phenomenon is that the power consumption of a mostly digital CMOS circuit is proportional to the square of the supply voltage (i.e. the battery voltage if the circuit is operated from the battery directly), while if one adds a linear regulator in front of this circuit the power consumption will change only linearly with the battery voltage. The paradigm has shifted in the past 2 years. Most phones today operate on a single cell Li-Ion battery, which has a 4.2V maximum fully charged voltage. If the cellular phone manufacturer requires the phone to operate with removed battery and plugged in charger the maximum input voltage of the system can be higher, depending on how the battery charger is implemented. As the process evolution path moves down the main circuit components to 0.25u, 0.18u and below, it is clear that there has to be a voltage regulator between most of the phone circuits and the battery.

Initially the voltage regulator function was implemented using discrete low dropout linear regulators, LDOs. Today most phones are built using more integrated power management solutions, that include a large number of regulators, LDOs and switching regulators, battery chargers, sequencing circuitry, supervisory and house keeping circuitry.

The role of voltage regulators

The voltage regulators in cellular phones are used to accomplish different goals, aside form their basic task, which is to increase battery life:

1. To step down the voltage between the battery and the different sub-circuits that require lower supply voltage, or to step up the voltage for sub-circuits that need higher voltage than the battery (like SIM card, backlight LED, etc.). Occasionally a buck-boost type regulator is required to generate a voltage that is between the maximum and minimum value of the battery voltage.

2. To isolate the different subsystems from each other. This is important in the RF section, and also between digital and analog/mixed signal circuits. Using LDOs can be cheaper and more area efficient than adding the traditional LC isolation filters that are used in the supply lines of RF circuits.

3. To isolate sensitive circuitry from the transient voltage changes of the battery. This is especially relevant in GSM phones where the PA operates in a 217 Hz pulsed mode with 12.5% duty cycle. The high current of the PA, typically up to 1.6A, can cause a voltage transient of up to 0.5V due to the combined effect of the battery's ESR and protection circuitry. The PSSR of the voltage regulator significantly reduces the supply transient seen by the phone circuits.

Figure 1. Typical GSM handset diagram

Trade-offs in a GSM phone

Figure 1 shows the block diagram of a typical GSM phone. The different blocks show the different functions, but do not necessarily represent separate integrated circuits. The major functional blocks are the PA, the radio/RF section, the analog base band and digital base band units. One or more voltage regulators power all of these functional blocks. The supply voltage and current requirements of the different blocks, which correspond to the output requirements of the regulators, depend on the phone generation, the technology used and the system design of the phone.

Table 1. Typical power consumption in a GSM handset

Table 1 shows a typical power consumption map of a second generation GSM phone. The first two columns show the typical average current of the block during talk time and the typical supply voltage. The total typical current consumption of this example phone is 275mA during talk time. Let's assume here that all the voltage regulators are linear regulators, which means that the current draw from the battery is roughly the same as the current drawn from the regulators (plus the quiescent current that we will neglect here for simplicity) resulting in 153 minutes talk time using a 700mAH battery. This number may vary in real life circumstances, as the current consumption of the PA can be significantly less close to a base station. The last column shows the current consumption during stand by time, while the phone is on, waiting for a call. The total average current draw from the battery is 2.26mA.

This corresponds to a standby time of 310 hours, or 12.9 days. Of course if one mixes standby time and talk time, as it is the case in real life, then both the standby and talk times are much shorter, but this is part of the cellular phone specsmanship.

So far we assumed that all the voltage regulators are linear regulators, LDOs. This is true for most of the cellular phones that are produced today. However, linear regulators are not very efficient, especially when the difference between input and output voltage is large. The worst-case condition is a fully charged battery. The question naturally arises whether it makes sense to replace one or more of the inefficient linear regulators with highly efficient switching regulators. The increase in talk time would be the largest for the blocks with the highest current consumption. Looking at the blocks from this perspective the main consumer is the PA. However, let's leave this block for later consideration.

The second best candidate block for switching regulator is the digital base band and memory. They usually share one regulator. As the digital base band uses the finest geometry process in the phone, resulting in the lowest supply voltage (1.8V in the present example), this LDO has the lowest efficiency in the phone. If one wants to replace one LDO with a switcher in the phone it has to be the digital base band regulator. Let's see how much we can gain in talk time by doing this.

Let's assume that we use a state of the art switching regulator that operates at 94% efficiency. At the average battery voltage of 3.6V this translates to 13.3mA average consumption from the battery, instead of the 25mA with linear regulator. In real life the saving is less, because the switching regulator loss is predominantly I2Rdson loss, which does not average linearly over the GSM time slot. But, let's use this optimistic number for simplicity's sake. Leaving all the other LDOs in place, the total current consumption during talk time will go down to 263.3mA from the original 275mA. This corresponds to a talk time improvement of about 6 minutes, or 4%. Using typical low cost battery chargers, the charge-to-charge difference in the capacity of the fully charged battery can easily be +/-5%. This means that the gain in talk time would be hardly perceptible in real life circumstances.

On the other side of the equation, the component cost of the phone and the footprint of the regulator will significantly increase by changing this LDO to a switching regulator. First, silicon cost of a state of the art high efficiency high frequency switching regulator is significantly higher than an LDO's cost. Second, the switching regulator needs an inductor and some other small components (resistors and capacitors) that an LDO does not. Third, both the output capacitor and the input decoupling capacitor that the switching regulator needs are significantly higher in value, and correspondingly in cost and foot print, than the output and input capacitors for a state of the art LDO with the same current rating. To put some numbers on these differences, a state of the art LDO, used to power the digital base band circuitry, needs an output capacitor between 0.47uF and 2.2uF, depending on the silicon technology, and it needs an input decoupling capacitor of 1uF-2.2uF. A corresponding switching regulator would need a 22uF to 47uF output capacitor and a 22uF input decoupling capacitor. The difference in capacitor values is roughly and order of magnitude.

Let's check whether there is a more significant improvement in standby time when using this switching regulator. The current consumption of the digital base band and memory in stand by is 340uA, based on Table 1.The efficiency of a state of the art switching regulator at this low current is about 80%. Using this efficiency, the current consumption from the battery at 3.6V battery voltage is 212uA, representing a saving of 128uA, plus we save the 30uA quiescent current of one LDO, resulting in a total system standby current of 2.1mA (vs. the original 2.26mA). This would increase the standby time to 13.8 days from the original 12.9 days, an improvement of 20 hours, or 7.1%. Similarly to the improvement in talk time, the gain in standby time is not significant.

Figure 2. Optimal battery utilization

Looking ahead toward GPRS phones, it is difficult to obtain good data on how much the supply current of the digital base band will increase. On one hand, the phone uses more time slots to communicate, but on the other hand, it does not use the audio codec, which represents significant power saving. Based on the preliminary data we have the increase in talk time would be about 6% if one used a state of the art high efficiency high frequency switching regulator to power this block instead of an LDO. Looking further ahead, the coming EDGE generation will probably further increase the base band power consumption. If one does not take into account that this generation will be implemented on the next process generation (0.13u), with significantly better speed-power product than the present generation, it would seem that the cost to change from LDO to switching regulator will be justified. However, there is no data available on the power consumption of these new digital base band chips, so the cost-benefit ratio is not quite clear.

The only other block in the phone that has significant current consumption is the radio, with 32mA. However, this block is very sensitive to noise on its supply and is not a good candidate for using a switching regulator, due to the high ripple voltage on its output. Also, the low voltage difference between battery and radio supply voltage diminishes the efficiency advantage of the switcher over the LDO. The rest of the circuitry has a very low current consumption, and converting them to switchers would be counter-productive.

From the calculations above it seems like it would be very difficult to justify replacing the LDOs, which power the signal processing circuitry in cellular phones today, to switching regulators, as the cost-benefit ratio is very unfavorable. From the EDGE generation onward it probably will make sense to consider this change.

LDO issues

If one accepts the conclusion of the section above, it becomes very clear that a good LDO design is fundamental for optimal power management in a cellular phone. The LDOs powering the different sections need different specifications to gain the best cost-benefit ratio. In the following section we look at the different relevant LDO specifications and evaluating their importance in a cellular phone environment.

1. Accuracy

The supply voltage specifications of the digital and mixed signal sub-circuits have a tolerance of about +/-5%. However, the power consumption of a CMOS circuit, operating in a switching fashion, is proportional to the square of its supply voltage. So in order to extend battery life, these CMOS circuits should operate at the lowest possible supply voltage within its tolerance range. Operating close to the bottom of the supply voltage range is advantageous also for circuits with supply voltages close to the minimum discharged battery voltage, as it eases the drop out requirements for the LDO. This, in turn, will minimize silicon area and cost. The 2.8V supply voltage for most of the mixed signal blocks in Table 1 is a good example for this, as the 2.8V is very close to 3V, which is the typical minimum battery voltage that cellular phones with a single LiIon cell use. The lower this minimum voltage can be (down to 2.5V, which is set by the battery) the more energy can be used from the battery, albeit the difference between running down the battery to 3V and 2.5V is not very significant.

It is interesting to note that replacing all +/-3% accurate LDOs with +/-1% accurate ones would increase the talk time by the same amount as changing the digital base band LDO to a switching regulator, without the added complexity and added external components, at significantly lower total cost. A state of the art LDO can be +/-1.4% accurate over all line, load and temperature conditions.

2. Dropout Voltage

Dropout voltage is also an important parameter, as it limits how low one can discharge the battery. It can be designed to almost any value, in trade-off with the increase of silicon cost. Historically the importance of the dropout voltage has changed from phone generation to generation. This is due to the discrete nature of the number of battery cells that can be used and the discrete reduction of the supply voltages of the different blocks as they went down the process shrink pass. In the present generation phones the minimum battery voltage and the highest supply voltage are close, so the dropout voltage is important. Going forward, assuming that the battery will stay a single LiIon cell, and that the process shrink is continuing, pushing down the supply voltages, the dropout voltage will be less critical. If some low end cellular phones move to 2 cell NiMH batteries in the future, for cost reasons as some predict, low voltage ultra low dropout regulators will become very important. A state of the art LDO has a dropout voltage in the order of 1mV/mA.

3. Stability, size and type of the output capacitor

As LDOs employ grounded emitter/source output stages, their open loop output impedance is high. An output capacitor is needed in most cases to stabilize these types of circuits. This output capacitor is also needed to assure good load transient response, as the regulator loop can not instantly respond to high dI/dT load current transients. For this reason, LDOs that theoretically need no output capacitors, and are designed in with no output capacitors, are not practical in applications where response to load transients is important. This is the case in most cellular phone applications. However, the small output capacitor and tolerance to different types of output capacitors (with varying ESR) are important attributes of the LDO.

As an LDO needs to have high loop gain to have good load regulation, it has to have at least one high impedance node inside the loop, which creates a pole. This, together with the output pole (due to the output capacitor) form a second order system, that, with the unavoidable second order pole(s), pose a design challenge for the LDO designer. Historically the ESR of the (tantalum) output capacitor was utilized to ensure stability with good phase margin. However, the advent of MLCC (low ESR) capacitors pushes the ESR zeroes way beyond the loop's cross over frequency, LDO designers need to find new ways to stabilize the control loop.

Traditionally one can add a zero-pole pair to the loop at the right frequency so that the phase boost of this compensation ensures adequate phase margin. This solution usually yields designs that are stable only with one type of output capacitor and has limited stability zones, depending on the current, output capacitor size and ESR. This is limiting the ease of use of these regulators, as the output capacitor is very often a combination of the output capacitor required for stability and ceramic decoupling capacitors placed close to the circuits that the LDO feeds. Often it is difficult to know the exact size and type of the output capacitors that the application will use.

A better solution is to use a modified version of pole splitting, used in most operational amplifiers, to split the two poles of the LDO and gain a system with close to single pole roll off 1. Figure 3 shows the block diagram of an LDO using this technique, in case of a bipolar pass device. The high impedance internal gain node is the output of the gm amplifier. The Cc Miller capacitor splits this pole and the output pole to increase stability. The non-inverting buffer is a broad band circuit (its pole is far beyond the loop's cross over frequency) that drives the pass device with low impedance. This increases the bandwidth of the pass device and reduces the susceptibility of the circuit to line transients. The result is a circuit that is "anyCAP" i.e. stable with any type of practical output capacitor. This makes this design extremely easy to use.

PA power management

As we saw in Table 1, the biggest power consumer in a cell phone is the power amplifier (PA). It is a rather simple circuit, comprised of a grounded source amplifier (Silicon or GaAs) driving a choke and the antenna through a matching network. It can be driven in class A, AB or C mode. Class A is very linear but highly inefficient, while class C is efficient but has high distortion. In digital modulation schemes used in EDGE and 3G phones both the amplitude and the phase of the output is modulated. If the output stage could operate in a switching mode, carrying only the phase information, while the amplitude information would be added as a supply modulation, similar to class S audio amplifiers, the total efficiency of the PA system could be significantly increased, probably close to doubled 2. To implement this "envelope elimination and restoration" technique a high efficiency and very high bandwidth switching regulator is needed. This is the switching regulator that has the highest cost-benefit ratio in a cellular phone in the near future.

Figure 3. anyCAP LDO bloack diagram

References

*A version of this article was presented at the CommsDesign Conference (CDC) in San Jose last October

1 Paul Brokaw, "Low Dropout Regulators: New Directions, Trade-Offs, System Optimization", Power'96 Conference Proceeding, Santa Clara, California, October 1996

2 David K. Su, William J. McFarland, "An IC for Linearizing RF Power Amplifiers Using Envelop Elimination and Restoration", IEEE Journal of Solid-State Circuits, Vol 33, No. 12, Dec 1998.