page.title=Sensor stack
@jd:body

<!--
    Copyright 2014 The Android Open Source Project

    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.
    You may obtain a copy of the License at

        http://www.apache.org/licenses/LICENSE-2.0

    Unless required by applicable law or agreed to in writing, software
    distributed under the License is distributed on an "AS IS" BASIS,
    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    See the License for the specific language governing permissions and
    limitations under the License.
-->
<div id="qv-wrapper">
  <div id="qv">
    <h2>In this document</h2>
    <ol id="auto-toc">
    </ol>
  </div>
</div>

<p>The figure below represents the Android sensor stack. Each component
  communicates only with the components directly above and below it, though some
  sensors can bypass the sensor hub when it is present. Control flows from the
  applications down to the sensors, and data flows from the sensors up to the
  applications.</p>
<img src="images/sensor_layers.png" alt="Layers and owners of the Android sensor stack" />
<p class="img-caption"><strong>Figure 1.</strong> Layers of the Android sensor stack and their respective owners</p>

<h2 id="sdk">SDK</h2>
<p>Applications access sensors through the <a href="http://developer.android.com/reference/android/hardware/SensorManager.html">Sensors SDK (Software Development Kit) API</a>. The SDK contains functions to list available sensors and to register to a
  sensor.</p>
<p>When registering to a sensor, the application specifies its preferred sampling
  frequency and its latency requirements.</p>
<ul>
  <li> For example, an application might register to the default accelerometer,
    requesting events at 100Hz, and allowing events to be reported with a 1-second
    latency. </li>
  <li> The application will receive events from the accelerometer at a rate of at
    least 100Hz, and possibly delayed up to 1 second. </li>
</ul>
<p>See the <a href="index.html#targeted_at_developers">developer documentation</a> for more information on the SDK.</p>
<h2 id="framework">Framework</h2>
<p>The framework is in charge of linking the several applications to the <a href="hal-interface.html">HAL</a>. The HAL itself is single-client. Without this multiplexing happening at the
  framework level, only a single application could access each sensor at any
  given time.</p>
<ul>
  <li> When a first application registers to a sensor, the framework sends a request
    to the HAL to activate the sensor. </li>
  <li> When additional applications register to the same sensor, the framework takes
    into account requirements from each application and sends the updated requested
    parameters to the HAL.
    <ul>
      <li> The <a href="hal-interface.html#sampling_period_ns">sampling frequency</a> will be the maximum of the requested sampling frequencies, meaning some
        applications will receive events at a frequency higher than the one they
        requested. </li>
      <li> The <a href="hal-interface.html#max_report_latency_ns">maximum reporting latency</a> will be the minimum of the requested ones. If one application requests one
        sensor with a maximum reporting latency of 0, all applications will receive the
        events from this sensor in continuous mode even if some requested the sensor
        with a non-zero maximum reporting latency. See <a href="batching.html">Batching</a> for more details. </li>
    </ul>
  </li>
  <li> When the last application registered to one sensor unregisters from it, the
    frameworks sends a request to the HAL to deactivate the sensor so power is not
    consumed unnecessarily. </li>
</ul>
<h3 id="impact_of_multiplexing">Impact of multiplexing</h3>
<p>This need for a multiplexing layer in the framework explains some design
  decisions.</p>
<ul>
  <li> When an application requests a specific sampling frequency, there is no
    guarantee that events won’t arrive at a faster rate. If another application
    requested the same sensor at a faster rate, the first application will also
    receive them at the fast rate. </li>
  <li> The same lack of guarantee applies to the requested maximum reporting latency:
    applications might receive events with much less latency than they requested. </li>
  <li> Besides sampling frequency and maximum reporting latency, applications cannot
    configure sensor parameters.
    <ul>
      <li> For example, imagine a physical sensor that can function both in “high
        accuracy” and “low power” modes. </li>
      <li> Only one of those two modes can be used on an Android device, because
        otherwise, an application could request the high accuracy mode, and another one
        a low power mode; there would be no way for the framework to satisfy both
        applications. The framework must always be able to satisfy all its clients, so
        this is not an option. </li>
    </ul>
  </li>
  <li> There is no mechanism to send data down from the applications to the sensors or
    their drivers. This ensures one application cannot modify the behavior of the
    sensors, breaking other applications. </li>
</ul>
<h3 id="sensor_fusion">Sensor fusion</h3>
<p>The Android framework provides a default implementation for some composite
  sensors. When a <a href="sensor-types.html#gyroscope">gyroscope</a>, an <a href="sensor-types.html#accelerometer">accelerometer</a> and a <a href="sensor-types.html#magnetic_field_sensor">magnetometer</a> are present on a device, but no <a href="sensor-types.html#rotation_vector">rotation vector</a>, <a href="sensor-types.html#gravity">gravity</a> and <a href="sensor-types.html#linear_acceleration">linear acceleration</a> sensors are present, the framework implements those sensors so applications
  can still use them.</p>
<p>The default implementation does not have access to all the data that other
  implementations have access to, and it must run on the SoC, so it is not as
  accurate nor as power efficient as other implementations can be. As much as
  possible, device manufacturers should define their own fused sensors (rotation
  vector, gravity and linear acceleration, as well as newer composite sensors like
  the <a href="sensor-types.html#game_rotation_vector">game rotation vector</a>) rather than rely on this default implementation. Device manufacturers can
  also request sensor chip vendors to provide them with an implementation.</p>
<p>The default sensor fusion implementation is not being maintained and
  might cause devices relying on it to fail CTS.</p>
<h3 id="under_the_hood">Under the Hood</h3>
<p>This section is provided as background information for those maintaining the
  Android Open Source Project (AOSP) framework code. It is not relevant for
  hardware manufacturers.</p>
<h4 id="jni">JNI</h4>
<p>The framework uses a Java Native Interface (JNI) associated with <a href="http://developer.android.com/reference/android/hardware/package-summary.html">android.hardware</a> and located in the <code>frameworks/base/core/jni/</code> directory. This code calls the
  lower level native code to obtain access to the sensor hardware.</p>
<h4 id="native_framework">Native framework</h4>
<p>The native framework is defined in <code>frameworks/native/</code> and provides a native
  equivalent to the <a href="http://developer.android.com/reference/android/hardware/package-summary.html">android.hardware</a> package. The native framework calls the Binder IPC proxies to obtain access to
  sensor-specific services.</p>
<h4 id="binder_ipc">Binder IPC</h4>
<p>The Binder IPC proxies facilitate communication over process boundaries.</p>
<h2 id="hal">HAL</h2>
<p>The Sensors Hardware Abstraction Layer (HAL) API is the interface between the
  hardware drivers and the Android framework. It consists of one HAL interface
  sensors.h and one HAL implementation we refer to as sensors.cpp.</p>
<p>The interface is defined by Android and AOSP contributors, and the
  implementation is provided by the manufacturer of the device.</p>
<p>The sensor HAL interface is located in <code>hardware/libhardware/include/hardware</code>.
  See <a href="{@docRoot}devices/halref/sensors_8h.html">sensors.h</a> for additional details.</p>
<h3 id="release_cycle">Release cycle</h3>
<p>The HAL implementation specifies what version of the HAL interface it
  implements by setting <code>your_poll_device.common.version</code>. The existing HAL
  interface versions are defined in sensors.h, and functionality is tied to those
  versions.</p>
<p>The Android framework currently supports versions 1.0 and 1.3, but 1.0 will
  soon not be supported anymore. This documentation describes the behavior of version
  1.3, to which all devices should upgrade. For details on how to upgrade to
  1.3, see <a href="versioning.html">HAL version deprecation</a>.</p>
<h2 id="kernel_driver">Kernel driver</h2>
<p>The sensor drivers interact with the physical devices. In some cases, the HAL
  implementation and the drivers are the same software entity. In other cases,
  the hardware integrator requests sensor chip manufacturers to provide the
  drivers, but they are the ones writing the HAL implementation.</p>
<p>In all cases, HAL implementation and kernel drivers are the responsibility of
  the hardware manufacturers, and Android does not provide preferred approaches to
  write them.</p>
<h2 id="sensor_hub">Sensor hub</h2>
<p>The sensor stack of a device can optionally include a sensor hub, useful to
  perform some low-level computation at low power while the SoC can be in a
  suspend mode. For example, step counting or sensor fusion can be performed on
  those chips. It is also a good place to implement sensor batching, adding
  hardware FIFOs for the sensor events. See <a
href="batching.html">Batching</a> for more information.</p>
<p>How the sensor hub is materialized depends on the architecture. It is sometimes
  a separate chip, and sometimes included on the same chip as the SoC. Important
  characteristics of the sensor hub is that it should contain sufficient memory
  for batching and consume very little power to enable implementation of the low
  power Android sensors. Some sensor hubs contain a microcontroller for generic
  computation, and hardware accelerators to enable very low power computation for
  low power sensors.</p>
<p>How the sensor hub is architectured and how it communicates with the sensors
  and the SoC (I2C bus, SPI bus, …) is not specified by Android, but it should aim
  at minimizing overall power use.</p>
<p>One option that appears to have a significant impact on implementation
  simplicity is having two interrupt lines going from the sensor hub to the SoC:
  one for wake-up interrupts (for wake-up sensors), and the other for non-wake-up
  interrupts (for non-wake-up sensors).</p>
<h2 id="sensors">Sensors</h2>
<p>Those are the physical MEMs chips making the measurements. In many cases,
  several physical sensors are present on the same chip. For example, some chips
  include an accelerometer, a gyroscope and a magnetometer. (Such chips are often
  called 9-axis chips, as each sensor provides data over 3 axes.)</p>
<p>Some of those chips also contain some logic to perform usual computations such
  as motion detection, step detection and 9-axis sensor fusion.</p>
<p>Although the CDD power and accuracy requirements and recommendations target the
  Android sensor and not the physical sensors, those requirements impact the
  choice of physical sensors. For example, the accuracy requirement on the game
  rotation vector has implications on the required accuracy for the physical
  gyroscope. It is up to the device manufacturer to derive the requirements for
  physical sensors.</p>