Power Profiles for Android

In this document

    Battery usage information is derived from battery usage statistics and power profile values.

    Battery Usage Statistics

    The framework automatically determines battery usage statistics by tracking how long device components spend in different states. As components (WiFi chipset, Cellular Radio, Bluetooth, GPS, Display, CPU) change states (OFF/ON, idle/full power, low/high brightness, etc.), the controlling service reports to the framework BatteryStats service, which collects information over time and stores it for use across reboots. The service doesn’t track battery current draw directly, but instead collects timing information that can be used to approximate battery consumption by different components.

    The framework gathers statistics using the following methods:

    • Push. Services aware of component changes push state changes to the BatteryStats service.
    • Pull. For components such as the CPU usage by apps, the framework automatically pulls the data at transition points (such as starting or stopping an activity) to take a snapshot.

    Resource consumption is associated with the application using the resource. When multiple applications simultaneously use a resource (such as wakelocks that prevent the system from suspending), the framework spreads consumption across those applications, although not necessarily equally.

    To avoid losing usage statistics for a shutdown event, which may indicate battery power consumption problems (i.e. shutdown occurs because the battery reached zero remaining capacity), the framework flashes statistics approximately every 30 minutes.

    Battery usage statistics are handled entirely by the framework and do not require OEM modifications.

    Power Profile Values

    Device manufacturers must provide a component power profile that defines the current consumption value for the component and the approximate the actual battery drain caused by the component over time. Within a power profile, power consumption is specified in milliamps (mA) of current draw at a nominal voltage and can be a fractional value specified in microamps (uA). The value should be the mA consumed at the battery and not a value applicable to a power rail that does not correspond to current consumed from the battery.

    For example, a display power profile specifies the mA of current required to keep the display on at minimum brightness and at maximum brightness. To determine the power cost (i.e the battery drained by the display component) of keeping the display on, the framework tracks the time spent at each brightness level, then multiplies those time intervals by an interpolated display brightness cost.

    The framework also multiplies the CPU time for each application by the mA required to run the CPU at a specific speed. This calculation establishes a comparative ranking of how much battery an application consumes by executing CPU code (time as the foreground app and total time including background activity are reported separately).

    Measuring Component Power

    You can determine individual component power consumption by comparing the current drawn by the device when the component is in the desired state (on, active, scanning, etc.) and when the component is off. Measure the average instantaneous current drawn on the device at a nominal voltage using an external power monitor, such as a bench power supply or specialized battery-monitoring tools (such as Monsoon Solution Inc. Power Monitor and Power Tool software).

    Note: Manufacturers often supply information about the current consumed by an individual component. Use this information if it accurately represents the current drawn from the device battery in practice. However, validate manufacturer-provided values before using those values in your device power profile.

    When measuring, ensure the device does not have a connection to an external charge source, such as a USB connection to a development host used when running Android Debug Bridge (adb). The device under test might draw current from the host, thus lowering measurements at the battery. Avoid USB On-The-Go (OTG) connections, as the OTG device might draw current from the device under test.

    Excluding the component being measured, the system should run at a constant level of power consumption to avoid inaccurate measurements caused by changes in other components. System activities that can introduce unwanted changes to power measurements include:

    • Cellular, Wi-Fi, and Bluetooth receive, transmit, or scanning activity. When not measuring cell radio power, set the device to airplane mode and enable Wi-Fi or Bluetooth as appropriate.
    • Screen on/off. Colors displayed while the screen is on can affect power draw on some screen technologies. Turn the screen off when measuring values for non-screen components.
    • System suspend/resume. A screen off state can trigger a system suspension, placing parts of the device in a low-power or off state. This can affect power consumption of the component being measured and introduce large variances in power readings as the system periodically resumes to send alarms, etc. For details, see Controlling System Suspend.
    • CPUs changing speed and entering/exiting low-power scheduler idle state. During normal operation, the system makes frequent adjustments to CPU speeds, the number of online CPU cores, and other system core states such as memory bus speed and voltages of power rails associated with CPUs and memory. During testing, these adjustments affect power measurements:
      • CPU speed scaling operations can reduce the amount of clock and voltage scaling of memory buses and other system core components.
      • Scheduling activity can affect the percentage of the time CPUs spend in low-power idle states. For details on preventing these adjustments from occurring during testing, see Controlling CPU Speeds.

    For example, Joe Droid wants to compute the screen.on value for a device. He enables airplane mode on the device, runs the device at a stable current state, holds the CPU speed constant , and uses a partial wakelock to prevent system suspend. Joe then turns the device screen off and takes a measurement (200mA). Next, Joe turns the device screen on at minimum brightness and takes another measurement (300mA). The screen.on value is 100mA (300 - 200).

    For components that don’t have a flat waveform of current consumption when active (such as cellular radio or Wi-Fi), measure the average current over time using a power monitoring tool.

    When using an external power source in place of the device battery, the system might experience problems due to an unconnected battery thermistor or integrated fuel gauge pins (i.e. an invalid reading for battery temperature or remaining battery capacity could shut down the kernel or Android system). Fake batteries can provide signals on thermistor or fuel gauge pins that mimic temperature and state of charge readings for a normal system, and may also provide convenient leads for connecting to external power supplies. Alternatively, you can modify the system to ignore the invalid data from the missing battery.

    Controlling System Suspend

    This section describes how to avoid system suspend state when you don’t want it to interfere with other measurements, and how to measure the power draw of system suspend state when you do want to measure it.

    Preventing System Suspend

    System suspend can introduce unwanted variance in power measurements and place system components in low-power states inappropriate for measuring active power use. To prevent the system from suspending while the screen is off, use a temporary partial wakelock. Using a USB cable, connect the device to a development host, then issue the following command:

    $ adb shell "echo temporary > /sys/power/wake_lock"
    

    While in wake_lock, the screen off state does not trigger a system suspend. (Remember to disconnect the USB cable from the device before measuring power consumption.)

    To remove the wakelock:

    $ adb shell "echo temporary > /sys/power/wake_unlock"
    

    Measuring System Suspend

    To measure the power draw during the system suspend state, measure the value of cpu.idle in the power profile. Before measuring:

    • Remove existing wakelocks (as described above).
    • Place the device in airplane mode to avoid concurrent activity by the cellular radio, which might run on a processor separate from the SoC portions controlled by the system suspend.
    • Ensure the system is in suspend state by:
      • Confirming current readings settle to a steady value. Readings should be within the expected range for the power consumption of the SoC suspend state plus the power consumption of system components that remain powered (such as the USB PHY).
      • Checking the system console output.
      • Watching for external indications of system status (such as an LED turning off when not in suspend).

    Controlling CPU Speeds

    Active CPUs can be brought online or put offline, have their clock speeds and associated voltages changed (possibly also affecting memory bus speeds and other system core power states), and can enter lower power idle states while in the kernel idle loop. When measuring different CPU power states for the power profile, avoid the power draw variance when measuring other parameters. The power profile assumes all CPUs have the same available speeds and power characteristics.

    While measuring CPU power, or while holding CPU power constant to make other measurements, keep the number of CPUs brought online constant (such as having one CPU online and the rest offline/hotplugged out). Keeping all CPUs except one in scheduling idle may product acceptable results. Stopping the Android framework with adb shell stop can reduce system scheduling activity.

    You must specify the available CPU speeds for your device in the power profile cpu.speeds entry. To get a list of available CPU speeds, run:

    adb shell cat /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state
    

    These speeds match the corresponding power measurements in value cpu.active.

    For platforms where number of cores brought online significantly affects power consumption, you might need to modify the cpufreq driver or governor for the platform. Most platforms support controlling CPU speed using the “userspace” cpufreq governor and using sysfs interfaces to set the speed. For example, to set speed for 200MHz on a system with only 1 CPU or all CPUs sharing a common cpufreq policy, use the system console or adb shell to run the following commands:

    echo userspace > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
    echo 200000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
    echo 200000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
    echo 200000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_setspeed
    cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq
    

    Note: The exact commands differ depending on the platform cpufreq implementation.

    These commands ensure the new speed is not outside the allowed bounds, set the new speed, then print the speed at which the CPU is actually running (for verification). If the current minimum speed prior to execution is higher than 200000, you might need to reverse the order of the first two lines, or execute the first line again to drop the minimum speed prior to setting the maximum speed.

    To measure current consumed by a CPU running at various speeds, use the system console place the CPU in a CPU-bound loop using the command:

    # while true; do true; done
    

    Take the measurement while the loop executes.

    Some devices can limit maximum CPU speed while performing thermal throttling due to a high temperature measurement (i.e. after running CPUs at high speeds for sustained periods). Watch for such limiting, either using the system console output when taking measurements or by checking the kernel log after measuring.

    For the cpu.active value, measure the power consumed when the system is not in suspend and not executing tasks. The CPU should be in a low-power scheduler idle loop , possibly executing an ARM Wait For Event instruction or in an SoC-specific low power state with a fast exit latency suitable for idle use. Your platform might have more than one idle state in use with differing levels of power consumption; choose a representative idle state for longer periods of scheduler idle (several milliseconds). Examine the power graph on your measurement equipment and choose samples where the CPU is at its lowest consumption, discarding higher samples where the CPU exited idle.

    Measuring Screen Power

    When measuring screen on power, ensure that other devices normally turned on when the screen is enabled are also on. For example, if the touchscreen and display backlight would normally be on when the screen is on, ensure these devices are on when you measure to get a realistic example of screen on power usage.

    Some display technologies vary in power consumption according to the colors displayed, causing power measurements to vary considerably depending on what is displayed on the screen at the time of measurement. When measuring, ensure the screen is displaying something that has power characteristics of a realistic screen. Aim between the extremes of an all-black screen (which consumes the lowest power for some technologies) and an all-white screen. A common choice is a view of a schedule in the calendar app, which has a mix of white background and non-white elements.

    Measure screen on power at minimum and maximum display/backlight brightness. To set minimum brightness:

    • Use the Android UI (not recommended). Set the Settings > Display Brightness slider to the minimum display brightness. However, the Android UI allows setting brightness only to a minimum of 10-20% of the possible panel/backlight brightness, and does not allow setting brightness so low that the screen might not be visible without great effort.
    • Use a sysfs file (recommended). If available, use a sysfs file to control panel brightness all the way down to the minimum brightness supported by the hardware.

    Additionally, if the platform sysfs file enables turning the LCD panel, backlight, and touchscreen on and off, use the file to take measurements with the screen on and off. Otherwise, set a partial wakelock so the system does not suspend, then turn on and off the screen with the power button.

    Measuring Wi-Fi Power

    Perform Wi-Fi measurements on a relatively quiet network. Avoid introducing additional work processing high volumes of broadcast traffic that is unrelated to the activity being measured.

    The wifi.on value measures the power consumed when Wi-Fi is enabled but not actively transmitting or receiving. This is often measured as the delta between the current draw in system suspend (sleep) state with Wi-Fi enabled vs. disabled.

    The wifi.scan value measures the power consumed during a Wi-Fi scan for access points. Applications can trigger Wi-Fi scans using the WifiManager class startScan()API. You can also open Settings > Wi-Fi, which performs access point scans every few seconds with an apparent jump in power consumption, but you must subtract screen power from these measurements.

    Note: Use a controlled setup (such as iperf) to generate network receive and transmit traffic.

    Measuring Device Power

    You can determine device power consumption for Android devices that include a battery fuel gauge such as a Summit SMB347 or Maxim MAX17050 (available on many Nexus devices). Use the in-system battery fuel gauge when external measurement equipment is not available or is inconvenient to connect to a device (such as in mobile usage).

    Measurements can include instantaneous current, remaining charge, battery capacity at test start and end, and more depending on the supported properties of the device (see below). For best results, perform device power measurements during long-running A/B tests that use the same device type with the same fuel gauge and same current sense resistor. Ensure the starting battery charge is the same for each device to avoid differing fuel gauge behavior at different points in the battery discharge curve.

    Even with identical test environments, measurements are not guaranteed to be of high absolute accuracy. However, most inaccuracies specific to the fuel gauge and sense resistor are consistent between test runs, making comparisons between identical devices useful. We recommend running multiple tests in different configurations to identify significant differences and relative power consumption between configurations.

    Reading Power Consumption

    To read power consumption data, insert calls to the API in your testing code.

    import android.os.BatteryManager;
    import android.os.ServiceManager;
    import android.content.Context;
    BatteryManager mBatteryManager =
    (BatteryManager)Context.getSystemService(Context.BATTERY_SERVICE);
    Long energy =
    mBatteryManager.getLongProperty(BatteryManager.BATTERY_PROPERTY_ENERGY_COUNTER);
    Slog.i(TAG, "Remaining energy = " + energy + "nWh");
    

    Available Properties

    Android supports the following battery fuel gauge properties:

    BATTERY_PROPERTY_CHARGE_COUNTER   Remaining battery capacity in microampere-hours
    BATTERY_PROPERTY_CURRENT_NOW      Instantaneous battery current in microamperes
    BATTERY_PROPERTY_CURRENT_AVERAGE  Average battery current in microamperes
    BATTERY_PROPERTY_CAPACITY         Remaining battery capacity as an integer percentage
    BATTERY_PROPERTY_ENERGY_COUNTER   Remaining energy in nanowatt-hours
    

    Most properties are read from kernel power_supply subsystem attributes of similar names. However, the exact properties, resolution of property values, and update frequency available for a specific device depend on:

    • Fuel gauge hardware, such as a Summit SMB347 or Maxim MAX17050.
    • Fuel gauge-to-system connection, such as the value of external current sense resistors.
    • Fuel gauge chip software configuration, such as values chosen for average current computation intervals in the kernel driver.

    For details, see the properties available for Nexus devices.

    Maxim Fuel Gauge

    When determining battery state-of-charge over a long period of time, the Maxim fuel gauge (MAX17050, BC15) corrects for coulomb-counter offset measurements. For measurements made over a short period of time (such as power consumption metering tests), the fuel gauge does not make corrections, making the offset the primary source of error when current measurements are too small (although no amount of time can eliminate the offset error completely).

    For a typical 10mOhm sense resistor design, the offset current should be better than 1.5mA, meaning any measurement is +/-1.5mA (PCBoard layout can also affect this variation). For example, when measuring a large current (200mA) you can expect the following:

    • 2mA (1% gain error of 200mA due to fuel gauge gain error)
    • +2mA (1% gain error of 200mA due to sense resistor error)
    • +1.5mA (current sense offset error from fuel gauge)

    The total error is 5.5mA (2.75%). Compare this to a medium current (50mA) where the same error percentages give a total error of 7% or to a small current (15mA) where +/-1.5mA gives a total error of 10%.

    For best results, we recommend measuring greater than 20mA. Gain measurement errors are systematic and repeatable, enabling you to test a device in multiple modes and get clean relative measurements (with exceptions for the 1.5mA offset).

    For +/-100uA relative measurements, required measurement time depends on:

    • ADC sampling noise. The MAX17050 with its normal factory configuration produces +/-1.5mA sample-to-sample variation due to noise, with each sample delivered at 175.8ms. You can expect a rough +/-100uA for a 1 minute test window and a clean 3-sigma noise less than 100uA (or 1-sigma noise at 33uA) for a 6 minute test window.
    • Sample Aliasing because of load variation. Variation exaggerates errors, so for samples with variation inherent in the loading, consider using a longer test window.

    Supported Nexus Devices

    Nexus 5
    Model Nexus 5
    Fuel Gauge Maxim MAX17048 fuel gauge (ModelGauge™, no coulomb counter)
    Properties BATTERY_PROPERTY_CAPACITY
    Measurements The fuel gauge does not support any measurements other than battery State Of Charge to a resolution of %/256 (1/256th of a percent of full battery capacity).
    Nexus 6
    Model Nexus 6
    Fuel Gauge Maxim MAX17050 fuel gauge (a coulomb counter with Maxim ModelGauge™ adjustments), and a 10mohm current sense resistor.
    Properties BATTERY_PROPERTY_CAPACITY
    BATTERY_PROPERTY_CURRENT_NOW
    BATTERY_PROPERTY_CURRENT_AVERAGE
    BATTERY_PROPERTY_CHARGE_COUNTER
    BATTERY_PROPERTY_ENERGY_COUNTER
    Measurements CURRENT_NOW resolution 156.25uA, update period is 175.8ms.
    CURRENT_AVERAGE resolution 156.25uA, update period configurable 0.7s - 6.4h, default 11.25 secs.
    CHARGE_COUNTER (accumulated current, non-extended precision) resolution is 500uAh (raw coulomb counter read, not adjusted by fuel gauge for coulomb counter offset, plus inputs from the ModelGauge m3 algorithm including empty compensation).
    CHARGE_COUNTER_EXT (extended precision in kernel) resolution 8nAh.
    ENERGY_COUNTER is CHARGE_COUNTER_EXT at nominal voltage of 3.7V.
    Nexus 9
    Model Nexus 9
    Fuel Gauge Maxim MAX17050 fuel gauge (a coulomb counter with Maxim ModelGauge™ adjustments), and a 10mohm current sense resistor.
    Properties BATTERY_PROPERTY_CAPACITY
    BATTERY_PROPERTY_CURRENT_NOW
    BATTERY_PROPERTY_CURRENT_AVERAGE
    BATTERY_PROPERTY_CHARGE_COUNTER
    BATTERY_PROPERTY_ENERGY_COUNTER
    Measurements CURRENT_NOW resolution 156.25uA, update period is 175.8ms.
    CURRENT_AVERAGE resolution 156.25uA, update period configurable 0.7s - 6.4h, default 11.25 secs.
    CHARGE_COUNTER (accumulated current, non-extended precision) resolution is 500uAh.
    CHARGE_COUNTER_EXT (extended precision in kernel) resolution 8nAh.
    ENERGY_COUNTER is CHARGE_COUNTER_EXT at nominal voltage of 3.7V.
    Accumulated current update period 175.8ms.
    ADC sampled at 175ms quantization with a 4ms sample period. Can adjust duty cycle.
    Nexus 10
    Model Nexus 10
    Fuel Gauge Dallas Semiconductor DS2784 fuel gauge (a coulomb counter), with a 10mohm current sense resistor.
    Properties BATTERY_PROPERTY_CAPACITY
    BATTERY_PROPERTY_CURRENT_NOW
    BATTERY_PROPERTY_CURRENT_AVERAGE
    BATTERY_PROPERTY_CHARGE_COUNTER
    BATTERY_PROPERTY_ENERGY_COUNTER
    Measurements Current measurement (instantaneous and average) resolution is 156.3uA.
    CURRENT_NOW instantaneous current update period is 3.5 seconds.
    CURRENT_AVERAGE update period is 28 seconds (not configurable).
    CHARGE_COUNTER (accumulated current, non-extended precision) resolution is 625uAh.
    CHARGE_COUNTER_EXT (extended precision in kernel) resolution is 144nAh.
    ENERGY_COUNTER is CHARGE_COUNTER_EXT at nominal voltage of 3.7V.
    Update period for all is 3.5 seconds.

    Viewing Battery Usage Data

    The dumpsys batterystats command generates interesting statistical data about battery usage on a device, organized by unique application ID. You can view a history of battery-related events such as mobile radio state, Wi-Fi and Bluetooth power states, and wakelock reasons.

    Statistics include:

    • History of battery-related events
    • Global statistics for the device
    • Approximate power use per UID and system component
    • System UID aggregated statistics

    Use the output of the dumpsys command with the Battery Historian tool to generate HTML visualizations of power-related events from logs.

    Power Values

    Name Description Example Value Notes
    none Nothing 0
    screen.on Additional power used when screen is turned on at minimum brightness. 200mA Includes touch controller and display backlight. At 0 brightness, not the Android minimum which tends to be 10 or 20%.
    screen.full Additional power used when screen is at maximum brightness, compared to screen at minimum brightness. 100mA-300mA A fraction of this value (based on screen brightness) is added to the screen.on value to compute the power usage of the screen.
    bluetooth.active Additional power used when playing audio through bluetooth A2DP. 14mA
    bluetooth.on Additional power used when bluetooth is turned on but idle. 1.4mA
    wifi.on Additional power used when Wi-Fi is turned on but not receiving, transmitting, or scanning. 2mA
    wifi.active Additional power used when transmitting or receiving over Wi-Fi. 31mA
    wifi.scan Additional power used when Wi-Fi is scanning for access points. 100mA
    dsp.audio Additional power used when audio decoding/encoding via DSP. 14.1mA Reserved for future use.
    dsp.video Additional power used when video decoding via DSP. 54mA Reserved for future use.
    gps.on Additional power used when GPS is acquiring a signal. 50mA
    radio.active Additional power used when cellular radio is transmitting/receiving. 100mA-300mA
    radio.scanning Additional power used when cellular radio is paging the tower. 1.2mA
    radio.on Additional power used when the cellular radio is on. Multi-value entry, one per signal strength (no signal, weak, moderate, strong). 1.2mA Some radios boost power when they search for a cell tower and do not detect a signal. These numbers could all be the same or decreasing with increasing signal strength. If you provide only one value, the same value will be used for all strengths. If you provide 2 values, the first will be for no-signal and the second for all other strengths, and so on.
    cpu.speeds Multi-value entry that lists each possible CPU speed in KHz. 125000, 250000, 500000, 1000000, 1500000 The number and order of entries must correspond to the mA entries in cpu.active.
    cpu.idle Total power drawn by the system when CPUs (and the SoC) are in system suspend state. 3mA
    cpu.awake Additional power used when CPUs are in scheduling idle state (kernel idle loop); system is not in system suspend state. 50mA
    cpu.active Additional power used by CPUs when running at different speeds. 100, 120, 140, 160, 200 Set the max speed in the kernel to each of the allowed speeds and peg the CPU at that speed. The number of entries here correspond to the number of entries in cpu.speeds, in the same order.
    battery.capacity The total battery capacity in mAh. 3000mAh

    The power_profile.xml file is placed in an overlay in device///frameworks/base/core/res/res/xml/power_profile.xml

    Sample file

    <!-- Most values are the incremental current used by a feature, in mA (measured at
    nominal voltage). OEMs must measure and provide actual values before shipping a device.
    Example real-world values are given, but are dependent on the platform
    and can vary significantly, so should be measured on the shipping platform with a power meter.
    -->
    0
    200
    160
    10
    <!-- Bluetooth stereo audio playback 10.0 mA -->
    1.3
    0.5
    30
    100
    12
    50
    50
    75
    1.1
    <!-- Strength 0 to BINS-1 (4) -->
    1.1
    
    <!-- Different CPU speeds as reported in
    /sys/devices/system/cpu/cpu0/cpufreq/stats/time_in_state -->
    
    250000  
    500000  
    750000  
    1000000 
    1200000 
    
    <!-- Power consumption when CPU is idle -->
    3.0
    50.1
    <!-- Power consumption at different speeds -->
    
    100 <!-- 250 MHz -->
    120 <!-- 500 MHz -->
    140 <!-- 750 MHz -->
    155 <!-- 1   GHz -->
    175 <!-- 1.2 GHz -->
    
    <!-- This is the battery capacity in mAh -->
    3000
    <!-- Battery capacity is 3000 mAH (at 3.6 Volts) -->