src/devices/audio_usb.jd

page.title=USB Digital Audio
@jd:body

<div id="qv-wrapper">
  <div id="qv">
    <h2>In this document</h2>
    <ol id="auto-toc">
    </ol>
  </div>
</div>

<p>
This article reviews Android support for USB digital audio and related
USB-based protocols.
</p>

<h3 id="audience">Audience</h3>

<p>
The target audience of this article is Android device OEMs, SoC vendors,
USB audio peripheral suppliers, advanced audio application developers,
and others seeking detailed understanding of USB digital audio internals on Android.
</p>

<p>
End users should see the <a href="https://support.google.com/android/">Help Center</a> instead.
Though this article is not oriented towards end users,
certain audiophile consumers may find portions of interest.
</p>

<h2 id="overview">Overview of USB</h2>

<p>
Universal Serial Bus (USB) is informally described in the Wikipedia article
<a href="http://en.wikipedia.org/wiki/USB">USB</a>,
and is formally defined by the standards published by the
<a href="http://www.usb.org/">USB Implementers Forum, Inc</a>.
For convenience, we summarize the key USB concepts here,
but the standards are the authoritative reference.
</p>

<h3 id="terminology">Basic concepts and terminology</h3>

<p>
USB is a <a href="http://en.wikipedia.org/wiki/Bus_(computing)">bus</a>
with a single initiator of data transfer operations, called the <i>host</i>.
The host communicates with
<a href="http://en.wikipedia.org/wiki/Peripheral">peripherals</a> via the bus.
</p>

<p>
<b>Note:</b> the terms <i>device</i> or <i>accessory</i> are common synonyms for
<i>peripheral</i>.  We avoid those terms here, as they could be confused with
Android <a href="http://en.wikipedia.org/wiki/Mobile_device">device</a>
or the Android-specific concept called
<a href="http://developer.android.com/guide/topics/connectivity/usb/accessory.html">accessory mode</a>.
</p>

<p>
A critical host role is <i>enumeration</i>:
the process of detecting which peripherals are connected to the bus,
and querying their properties expressed via <i>descriptors</i>.
</p>

<p>
A peripheral may be one physical object
but actually implement multiple logical <i>functions</i>.
For example, a webcam peripheral could have both a camera function and a
microphone audio function.
</p>

<p>
Each peripheral function has an <i>interface</i> that
defines the protocol to communicate with that function.
</p>

<p>
The host communicates with a peripheral over a
<a href="http://en.wikipedia.org/wiki/Stream_(computing)">pipe</a>
to an <a href="http://en.wikipedia.org/wiki/Communication_endpoint">endpoint</a>,
a data source or sink
provided by one of the peripheral's functions.
</p>

<p>
There are two kinds of pipes: <i>message</i> and <i>stream</i>.
A message pipe is used for bi-directional control and status.
A stream pipe is used for uni-directional data transfer.
</p>

<p>
The host initiates all data transfers,
hence the terms <i>input</i> and <i>output</i> are expressed relative to the host.
An input operation transfers data from the peripheral to the host,
while an output operation transfers data from the host to the peripheral.
</p>

<p>
There are three major data transfer modes:
<i>interrupt</i>, <i>bulk</i>, and <i>isochronous</i>.
Isochronous mode will be discussed further in the context of audio.
</p>

<p>
The peripheral may have <i>terminals</i> that connect to the outside world,
beyond the peripheral itself.  In this way, the peripheral serves
to translate between USB protocol and "real world" signals.
The terminals are logical objects of the function.
</p>

<h2 id="androidModes">Android USB modes</h2>

<h3 id="developmentMode">Development mode</h3>

<p>
<i>Development mode</i> has been present since the initial release of Android.
The Android device appears as a USB peripheral
to a host PC running a desktop operating system such as Linux,
Mac OS X, or Windows.  The only visible peripheral function is either
<a href="http://en.wikipedia.org/wiki/Android_software_development#Fastboot">Android fastboot</a>
or
<a href="http://developer.android.com/tools/help/adb.html">Android Debug Bridge (adb)</a>.
The fastboot and adb protocols are layered over USB bulk data transfer mode.
</p>

<h3 id="hostMode">Host mode</h3>

<p>
<i>Host mode</i> is introduced in Android 3.1 (API level 12).
</p>

<p>
As the Android device must act as host, and most Android devices include
a micro-USB connector that does not directly permit host operation,
an on-the-go (<a href="http://en.wikipedia.org/wiki/USB_On-The-Go">OTG</a>) adapter
such as this is usually required:
</p>

<img src="audio/images/otg.jpg" style="image-orientation: 90deg;" height="50%" width="50%" alt="OTG">

<p>
An Android device might not provide sufficient power to operate a
particular peripheral, depending on how much power the peripheral needs,
and how much the Android device is capable of supplying.  Even if
adequate power is available, the Android device battery charge may
be significantly shortened.  For these situations, use a powered
<a href="http://en.wikipedia.org/wiki/USB_hub">hub</a> such as this:
</p>

<img src="audio/images/hub.jpg" alt="Powered hub">

<h3 id="accessoryMode">Accessory mode</h3>

<p>
<i>Accessory mode</i> was introduced in Android 3.1 (API level 12) and back-ported to Android 2.3.4.
In this mode, the Android device operates as a USB peripheral,
under the control of another device such as a dock that serves as host.
The difference between development mode and accessory mode
is that additional USB functions are visible to the host, beyond adb.
The Android device begins in development mode and then
transitions to accessory mode via a re-negotiation process.
</p>

<p>
Accessory mode was extended with additional features in Android 4.1,
in particular audio described below.
</p>

<h2 id="audioClass">USB audio</h2>

<h3 id="class">USB classes</h3>

<p>
Each peripheral function has an associated <i>device class</i> document
that specifies the standard protocol for that function.
This enables <i>class compliant</i> hosts and peripheral functions
to inter-operate, without detailed knowledge of each other's workings.
Class compliance is critical if the host and peripheral are provided by
different entities.
</p>

<p>
The term <i>driverless</i> is a common synonym for <i>class compliant</i>,
indicating that it is possible to use the standard features of such a
peripheral without requiring an operating-system specific
<a href="http://en.wikipedia.org/wiki/Device_driver">driver</a> to be installed.
One can assume that a peripheral advertised as "no driver needed"
for major desktop operating systems
will be class compliant, though there may be exceptions.
</p>

<h3 id="audioClass">USB audio class</h3>

<p>
Here we concern ourselves only with peripherals that implement
audio functions, and thus adhere to the audio device class.  There are two
editions of the USB audio class specification: class 1 (UAC1) and 2 (UAC2).
</p>

<h3 id="otherClasses">Comparison with other classes</h3>

<p>
USB includes many other device classes, some of which may be confused
with the audio class.  The
<a href="http://en.wikipedia.org/wiki/USB_mass_storage_device_class">mass storage class</a>
(MSC) is used for
sector-oriented access to media, while
<a href="http://en.wikipedia.org/wiki/Media_Transfer_Protocol">Media Transfer Protocol</a>
(MTP) is for full file access to media.
Both MSC and MTP may be used for transferring audio files,
but only USB audio class is suitable for real-time streaming.
</p>

<h3 id="audioTerminals">Audio terminals</h3>

<p>
The terminals of an audio peripheral are typically analog.
The analog signal presented at the peripheral's input terminal is converted to digital by an
<a href="http://en.wikipedia.org/wiki/Analog-to-digital_converter">analog-to-digital converter</a>
(ADC),
and is carried over USB protocol to be consumed by
the host.  The ADC is a data <i>source</i>
for the host.  Similarly, the host sends a
digital audio signal over USB protocol to the peripheral, where a
<a href="http://en.wikipedia.org/wiki/Digital-to-analog_converter">digital-to-analog converter</a>
(DAC)
converts and presents to an analog output terminal.
The DAC is a <i>sink</i> for the host.
</p>

<h3 id="channels">Channels</h3>

<p>
A peripheral with audio function can include a source terminal, sink terminal, or both.
Each direction may have one channel (<i>mono</i>), two channels
(<i>stereo</i>), or more.
Peripherals with more than two channels are called <i>multichannel</i>.
It is common to interpret a stereo stream as consisting of
<i>left</i> and <i>right</i> channels, and by extension to interpret a multichannel stream as having
spatial locations corresponding to each channel.  However, it is also quite appropriate
(especially for USB audio more so than
<a href="http://en.wikipedia.org/wiki/HDMI">HDMI</a>)
to not assign any particular
standard spatial meaning to each channel.  In this case, it is up to the
application and user to define how each channel is used.
For example, a four-channel USB input stream might have the first three
channels attached to various microphones within a room, and the final
channel receiving input from an AM radio.
</p>

<h3 id="isochronous">Isochronous transfer mode</h3>

<p>
USB audio uses isochronous transfer mode for its real-time characteristics,
at the expense of error recovery.
In isochronous mode, bandwidth is guaranteed, and data transmission
errors are detected using a cyclic redundancy check (CRC).  But there is
no packet acknowledgement or re-transmission in the event of error.
</p>

<p>
Isochronous transmissions occur each Start Of Frame (SOF) period.
The SOF period is one millisecond for full-speed, and 125 microseconds for
high-speed.  Each full-speed frame carries up to 1023 bytes of payload,
and a high-speed frame carries up to 1024 bytes.  Putting these together,
we calculate the maximum transfer rate as 1,023,000 or 8,192,000 bytes
per second.  This sets a theoretical upper limit on the combined audio
sample rate, channel count, and bit depth.  The practical limit is lower.
</p>

<p>
Within isochronous mode, there are three sub-modes:
</p>

<ul>
<li>Adaptive</li>
<li>Asynchronous</li>
<li>Synchronous</li>
</ul>

<p>
In adaptive sub-mode, the peripheral sink or source adapts to a potentially varying sample rate
of the host.
</p>

<p>
In asynchronous (also called implicit feedback) sub-mode,
the sink or source determines the sample rate, and the host accomodates.
The primary theoretical advantage of asynchronous sub-mode is that the source
or sink USB clock is physically and electrically closer to (and indeed may
be the same as, or derived from) the clock that drives the DAC or ADC.
This proximity means that asynchronous sub-mode should be less susceptible
to clock jitter.  In addition, the clock used by the DAC or ADC may be
designed for higher accuracy and lower drift than the host clock.
</p>

<p>
In synchronous sub-mode, a fixed number of bytes is transferred each SOF period.
The audio sample rate is effectively derived from the USB clock.
Synchronous sub-mode is not commonly used with audio because both
host and peripheral are at the mercy of the USB clock.
</p>

<p>
The table below summarizes the isochronous sub-modes:
</p>

<table>
<tr>
  <th>Sub-mode</th>
  <th>Byte count<br \>per packet</th>
  <th>Sample rate<br \>determined by</th>
  <th>Used for audio</th>
</tr>
<tr>
  <td>adaptive</td>
  <td>variable</td>
  <td>host</td>
  <td>yes</td>
</tr>
<tr>
  <td>asynchronous</td>
  <td>variable</td>
  <td>peripheral</td>
  <td>yes</td>
</tr>
<tr>
  <td>synchronous</td>
  <td>fixed</td>
  <td>USB clock</td>
  <td>no</td>
</tr>
</table>

<p>
In practice, the sub-mode does of course matter, but other factors
should also be considered.
</p>

<h2 id="androidSupport">Android support for USB audio class</h2>

<h3 id="developmentAudio">Development mode</h3>

<p>
USB audio is not supported in development mode.
</p>

<h3 id="hostAudio">Host mode</h3>

<p>
Android 5.0 (API level 21) and above supports a subset of USB audio class 1 (UAC1) features:
</p>

<ul>
<li>The Android device must act as host</li>
<li>The audio format must be PCM (interface type I)</li>
<li>The bit depth must be 16-bits, 24-bits, or 32-bits where
24 bits of useful audio data are left-justified within the most significant
bits of the 32-bit word</li>
<li>The sample rate must be either 48, 44.1, 32, 24, 22.05, 16, 12, 11.025, or 8 kHz</li>
<li>The channel count must be 1 (mono) or 2 (stereo)</li>
</ul>

<p>
Perusal of the Android framework source code may show additional code
beyond the minimum needed to support these features.  But this code
has not been validated, so more advanced features are not yet claimed.
</p>

<h3 id="accessoryAudio">Accessory mode</h3>

<p>
Android 4.1 (API level 16) added limited support for audio playback to the host.
While in accessory mode, Android automatically routes its audio output to USB.
That is, the Android device serves as a data source to the host, for example a dock.
</p>

<p>
Accessory mode audio has these features:
</p>

<ul>
<li>
The Android device must be controlled by a knowledgeable host that
can first transition the Android device from development mode to accessory mode,
and then the host must transfer audio data from the appropriate endpoint.
Thus the Android device does not appear "driverless" to the host.
</li>
<li>The direction must be <i>input</i>, expressed relative to the host</li>
<li>The audio format must be 16-bit PCM</li>
<li>The sample rate must be 44.1 kHz</li>
<li>The channel count must be 2 (stereo)</li>
</ul>

<p>
Accessory mode audio has not been widely adopted,
and is not currently recommended for new designs.
</p>

<h2 id="applications">Applications of USB digital audio</h2>

<p>
As the name indicates, the USB digital audio signal is represented
by a <a href="http://en.wikipedia.org/wiki/Digital_data">digital</a> data stream
rather than the <a href="http://en.wikipedia.org/wiki/Analog_signal">analog</a>
signal used by the common TRS mini
<a href=" http://en.wikipedia.org/wiki/Phone_connector_(audio)">headset connector</a>.
Eventually any digital signal must be converted to analog before it can be heard.
There are tradeoffs in choosing where to place that conversion.
</p>

<h3 id="comparison">A tale of two DACs</h3>

<p>
In the example diagram below, we compare two designs.  First we have a
mobile device with Application Processor (AP), on-board DAC, amplifier,
and analog TRS connector attached to headphones.  We also consider a
mobile device with USB connected to external USB DAC and amplifier,
also with headphones.
</p>

<img src="audio/images/dac.png" alt="DAC comparison">

<p>
Which design is better?  The answer depends on your needs.
Each has advantages and disadvantages.
<b>Note:</b> this is an artificial comparison, since
a real Android device would probably have both options available.
</p>

<p>
The first design A is simpler, less expensive, uses less power,
and will be a more reliable design assuming otherwise equally reliable components.
However, there are usually audio quality tradeoffs vs. other requirements.
For example, if this is a mass-market device, it may be designed to fit
the needs of the general consumer, not for the audiophile.
</p>

<p>
In the second design, the external audio peripheral C can be designed for
higher audio quality and greater power output without impacting the cost of
the basic mass market Android device B.  Yes, it is a more expensive design,
but the cost is absorbed only by those who want it.
</p>

<p>
Mobile devices are notorious for having high-density
circuit boards, which can result in more opportunities for
<a href="http://en.wikipedia.org/wiki/Crosstalk_(electronics)">crosstalk</a>
that degrades adjacent analog signals.  Digital communication is less susceptible to
<a href="http://en.wikipedia.org/wiki/Noise_(electronics)">noise</a>,
so moving the DAC from the Android device A to an external circuit board
C allows the final analog stages to be physically and electrically
isolated from the dense and noisy circuit board, resulting in higher fidelity audio.
</p>

<p>
On the other hand,
the second design is more complex, and with added complexity come more
opportunities for things to fail.  There is also additional latency
from the USB controllers.
</p>

<h3 id="applications">Applications</h3>

<p>
Typical USB host mode audio applications include:
</p>

<ul>
<li>music listening</li>
<li>telephony</li>
<li>instant messaging and voice chat</li>
<li>recording</li>
</ul>

<p>
For all of these applications, Android detects a compatible USB digital
audio peripheral, and automatically routes audio playback and capture
appropriately, based on the audio policy rules.
Stereo content is played on the first two channels of the peripheral.
</p>

<p>
There are no APIs specific to USB digital audio.
For advanced usage, the automatic routing may interfere with applications
that are USB-aware.  For such applications, disable automatic routing
via the corresponding control in the Media section of
<a href="http://developer.android.com/tools/index.html">Settings / Developer Options</a>.
</p>

<h2 id="compatibility">Implementing USB audio</h2>

<h3 id="recommendationsPeripheral">Recommendations for audio peripheral vendors</h3>

<p>
In order to inter-operate with Android devices, audio peripheral vendors should:
</p>

<ul>
<li>design for audio class compliance;
currently Android targets class 1, but it is wise to plan for class 2</li>
<li>avoid <a href="http://en.wiktionary.org/wiki/quirk">quirks</a>
<li>test for inter-operability with reference and popular Android devices</li>
<li>clearly document supported features, audio class compliance, power requirements, etc.
so that consumers can make informed decisions</li>
</ul>

<h3 id="recommendationsAndroid">Recommendations for Android device OEMs and SoC vendors</h3>

<p>
In order to support USB digital audio, device OEMs and SoC vendors should:
</p>

<ul>
<li>enable all kernel features needed: USB host mode, USB audio, isochronous transfer mode</li>
<li>keep up-to-date with recent kernel releases and patches;
despite the noble goal of class compliance, there are extant audio peripherals
with <a href="http://en.wiktionary.org/wiki/quirk">quirks</a>,
and recent kernels have workarounds for such quirks
</li>
<li>enable USB audio policy as described below</li>
<li>test for inter-operability with common USB audio peripherals</li>
</ul>

<h3 id="enable">How to enable USB audio policy</h3>

<p>
To enable USB audio, add an entry to the
audio policy configuration file.  This is typically
located here:
<pre>device/oem/codename/audio_policy.conf</pre>
The pathname component "oem" should be replaced by the name
of the OEM who manufactures the Android device,
and "codename" should be replaced by the device code name.
</p>

<p>
An example entry is shown here:
</p>

<pre>
audio_hw_modules {
  ...
  usb {
    outputs {
      usb_accessory {
        sampling_rates 44100
        channel_masks AUDIO_CHANNEL_OUT_STEREO
        formats AUDIO_FORMAT_PCM_16_BIT
        devices AUDIO_DEVICE_OUT_USB_ACCESSORY
      }
      usb_device {
        sampling_rates dynamic
        channel_masks dynamic
        formats dynamic
        devices AUDIO_DEVICE_OUT_USB_DEVICE
      }
    }
    inputs {
      usb_device {
        sampling_rates dynamic
        channel_masks AUDIO_CHANNEL_IN_STEREO
        formats AUDIO_FORMAT_PCM_16_BIT
        devices AUDIO_DEVICE_IN_USB_DEVICE
      }
    }
  }
  ...
}
</pre>

<h3 id="sourceCode">Source code</h3>

<p>
The audio Hardware Abstraction Layer (HAL)
implementation for USB audio is located here:
<pre>hardware/libhardware/modules/usbaudio/</pre>
The USB audio HAL relies heavily on
<i>tinyalsa</i>, described at <a href="audio_terminology.html">Audio Terminology</a>.
Though USB audio relies on isochronous transfers,
this is abstracted away by the ALSA implementation.
So the USB audio HAL and tinyalsa do not need to concern
themselves with this part of USB protocol.
</p>