docs/html/ndk/guides/audio/sample-rates.jd - platform/frameworks/base - Gitiles

 page.title=Sample Rates
 @jd:body

 <div id="qv-wrapper">
     <div id="qv">
       <h2>On this page</h2>

       <ol>
         <li><a href="#best">Best Practices for Sampling and Resampling</a></li>
         <li><a href="#info">For More Information</a></li>
       </ol>
     </div>
   </div>

 <a class="notice-developers-video" href="https://www.youtube.com/watch?v=6Dl6BdrA-sQ">
 <div>
     <h3>Video</h3>
     <p>Sample Rates: Why Can't We All Just Agree?</p>
 </div>
 </a>

 <p>As of Android 5.0 (Lollipop), the audio resamplers are now entirely based
 on FIR filters derived from a Kaiser windowed-sinc function. The Kaiser windowed-sinc
 offers the following properties:
 <ul>
     <li>It is straightforward to calculate for its design parameters (stopband
  ripple, transition bandwidth, cutoff frequency, filter length).</li>
 <li>It is nearly optimal for reduction of stopband energy compared to overall
 energy.</li>
 </ul>
 See P.P. Vaidyanathan, <a class="external-link"
 href="https://drive.google.com/file/d/0B7tBh7YQV0DGak9peDhwaUhqY2c/view">
 <i>Multirate Systems and Filter Banks</i></a>, p. 50 for discussions of the
 Kaiser Window and its optimality and relationship to Prolate Spheroidal
 Windows.</p>

 <p>The design parameters are automatically computed based on internal
 quality determination and the sampling ratios desired. Based on the
 design parameters, the windowed-sinc filter is generated.  For music use,
 the resampler for 44.1 to 48 kHz and vice versa is generated at a higher
 quality than for arbitrary frequency conversion.</p>

 <p>The audio resamplers provide increased quality, as well as speed
 to achieve that quality. But resamplers can introduce small amounts
 of passband ripple and aliasing harmonic noise, and they can cause some high
 frequency loss in the transition band, so avoid using them unnecessarily.</p>

 <h2 id="best">Best Practices for Sampling and Resampling</h2>
 <p>This section describes some best practices to help you avoid problems with sampling rates.</p>
 <h3>Choose the sampling rate to fit the device</h3>

 <p>In general, it is best to choose the sampling rate to fit the device,
 typically 44.1 kHz or 48 kHz.  Use of a sample rate greater than
 48 kHz will typically result in decreased quality because a resampler must be
 used to play back the file.</p>

 <h3>Use simple resampling ratios (fixed versus interpolated polyphases)</h3>

 <p>The resampler operates in one of the following modes:</p>
 <ul>
     <li>Fixed polyphase mode. The filter coefficients for each polyphase are precomputed.</li>
     <li>Interpolated polyphase mode. The filter coefficients for each polyphase must
 be interpolated from the nearest two precomputed polyphases.</li>
 </ul>
 <p>The resampler is fastest in fixed polyphase mode, when the ratio of input
 rate over output rate L/M (taking out the greatest common divisor)
 has M less than 256.  For example, for 44,100 to 48,000 conversion, L = 147,
 M = 160.</p>

 <p>In fixed polyphase mode, the sampling rate is locked for as
 many samples converted and does not change.  In interpolated polyphase
 mode, the sampling rate is approximate. The drift is generally on the
 order of one sample over a few hours of playback on a 48-kHz device.
 This is not usually a concern because approximation error is much less than
 frequency error of internal quartz oscillators, thermal drift, or jitter
  (typically tens of ppm).</p>

 <p>Choose simple-ratio sampling rates such as 24 kHz (1:2) and 32 kHz (2:3) when playing back
  on a 48-kHz device, even though other sampling
 rates and ratios may be permitted through AudioTrack.</p>

 <h3>Use upsampling rather than downsampling when changing sample rates</h3>

 <p>Sampling rates can be changed on the fly. The granularity of
 such change is based on the internal buffering (typically a few hundred
 samples), not on a sample-by-sample basis. This can be used for effects.</p>

 <p>Do not dynamically change sampling rates when
 downsampling. When changing sample rates after an audio track is
 created, differences of around 5 to 10 percent from the original rate may
 trigger a filter recomputation when downsampling (to properly suppress
 aliasing). This can consume computing resources and may cause an audible click
 if the filter is replaced in real time.</p>

 <h3>Limit downsampling to no more than 6:1</h3>

 <p>Downsampling is typically triggered by hardware device requirements. When the
  Sample Rate converter is used for downsampling,
 try to limit the downsampling ratio to no more than 6:1 for good aliasing
 suppression (for example, no greater downsample than 48,000 to 8,000). The filter
 lengths adjust to match the downsampling ratio, but you sacrifice more
 transition bandwidth at higher downsampling ratios to avoid excessively
 increasing the filter length. There are no similar aliasing concerns for
 upsampling.  Note that some parts of the audio pipeline
 may prevent downsampling greater than 2:1.</p>

 <h3 id="latency">If you are concerned about latency, do not resample</h3>

 <p>Resampling prevents the track from being placed in the FastMixer
 path, which means that significantly higher latency occurs due to the additional,
  larger buffer in the ordinary Mixer path. Furthermore,
  there is an implicit delay from the filter length of the resampler,
  though this is typically on the order of one millisecond or less,
  which is not as large as the additional buffering for the ordinary Mixer path
  (typically 20 milliseconds).</p>

 <h2 id="info">For More Information</h2>
 <p>This section lists some additional resources about sampling and resampling.</p>

 <h3>Sample rates</h3>

 <p>
 <a href="http://en.wikipedia.org/wiki/Sampling_%28signal_processing%29" class="external-link" >
 Sampling (signal processing)</a> at Wikipedia.</p>

 <h3>Resampling</h3>

 <p><a href="http://en.wikipedia.org/wiki/Sample_rate_conversion" class="external-link" >
 Sample rate conversion</a> at Wikipedia.</p>

 <p><a href="http://source.android.com/devices/audio/src.html" class="external-link" >
 Sample Rate Conversion</a> at source.android.com.</p>

 <h3>The high bit-depth and high kHz controversy</h3>

 <p><a href="http://people.xiph.org/~xiphmont/demo/neil-young.html" class="external-link" >
 24/192 Music Downloads ... and why they make no sense</a>
 by Christopher "Monty" Montgomery of Xiph.Org.</p>

 <p><a href="https://www.youtube.com/watch?v=cIQ9IXSUzuM" class="external-link" >
 D/A and A/D | Digital Show and Tell</a>
 video by Christopher "Monty" Montgomery of Xiph.Org.</p>

 <p><a href="http://www.trustmeimascientist.com/2013/02/04/the-science-of-sample-rates-when-higher-is-better-and-when-it-isnt/" class="external-link">
 The Science of Sample Rates (When Higher Is Better - And When It Isn't)</a>.</p>

 <p><a href="http://www.image-line.com/support/FLHelp/html/app_audio.htm" class="external-link" >
 Audio Myths &amp; DAW Wars</a></p>

 <p><a href="http://forums.stevehoffman.tv/threads/192khz-24bit-vs-96khz-24bit-debate-interesting-revelation.317660/" class="external-link">
 192kHz/24bit vs. 96kHz/24bit "debate"- Interesting revelation</a></p>
	page.title=Sample Rates
	@jd:body

	<div id="qv-wrapper">
	<div id="qv">
	<h2>On this page</h2>

	<ol>
	<li><a href="#best">Best Practices for Sampling and Resampling</a></li>
	<li><a href="#info">For More Information</a></li>
	</ol>
	</div>
	</div>

	<a class="notice-developers-video" href="https://www.youtube.com/watch?v=6Dl6BdrA-sQ">
	<div>
	<h3>Video</h3>
	<p>Sample Rates: Why Can't We All Just Agree?</p>
	</div>
	</a>

	<p>As of Android 5.0 (Lollipop), the audio resamplers are now entirely based
	on FIR filters derived from a Kaiser windowed-sinc function. The Kaiser windowed-sinc
	offers the following properties:
	<ul>
	<li>It is straightforward to calculate for its design parameters (stopband
	ripple, transition bandwidth, cutoff frequency, filter length).</li>
	<li>It is nearly optimal for reduction of stopband energy compared to overall
	energy.</li>
	</ul>
	See P.P. Vaidyanathan, <a class="external-link"
	href="https://drive.google.com/file/d/0B7tBh7YQV0DGak9peDhwaUhqY2c/view">
	<i>Multirate Systems and Filter Banks</i></a>, p. 50 for discussions of the
	Kaiser Window and its optimality and relationship to Prolate Spheroidal
	Windows.</p>

	<p>The design parameters are automatically computed based on internal
	quality determination and the sampling ratios desired. Based on the
	design parameters, the windowed-sinc filter is generated. For music use,
	the resampler for 44.1 to 48 kHz and vice versa is generated at a higher
	quality than for arbitrary frequency conversion.</p>

	<p>The audio resamplers provide increased quality, as well as speed
	to achieve that quality. But resamplers can introduce small amounts
	of passband ripple and aliasing harmonic noise, and they can cause some high
	frequency loss in the transition band, so avoid using them unnecessarily.</p>

	<h2 id="best">Best Practices for Sampling and Resampling</h2>
	<p>This section describes some best practices to help you avoid problems with sampling rates.</p>
	<h3>Choose the sampling rate to fit the device</h3>

	<p>In general, it is best to choose the sampling rate to fit the device,
	typically 44.1 kHz or 48 kHz. Use of a sample rate greater than
	48 kHz will typically result in decreased quality because a resampler must be
	used to play back the file.</p>

	<h3>Use simple resampling ratios (fixed versus interpolated polyphases)</h3>

	<p>The resampler operates in one of the following modes:</p>
	<ul>
	<li>Fixed polyphase mode. The filter coefficients for each polyphase are precomputed.</li>
	<li>Interpolated polyphase mode. The filter coefficients for each polyphase must
	be interpolated from the nearest two precomputed polyphases.</li>
	</ul>
	<p>The resampler is fastest in fixed polyphase mode, when the ratio of input
	rate over output rate L/M (taking out the greatest common divisor)
	has M less than 256. For example, for 44,100 to 48,000 conversion, L = 147,
	M = 160.</p>

	<p>In fixed polyphase mode, the sampling rate is locked for as
	many samples converted and does not change. In interpolated polyphase
	mode, the sampling rate is approximate. The drift is generally on the
	order of one sample over a few hours of playback on a 48-kHz device.
	This is not usually a concern because approximation error is much less than
	frequency error of internal quartz oscillators, thermal drift, or jitter
	(typically tens of ppm).</p>

	<p>Choose simple-ratio sampling rates such as 24 kHz (1:2) and 32 kHz (2:3) when playing back
	on a 48-kHz device, even though other sampling
	rates and ratios may be permitted through AudioTrack.</p>

	<h3>Use upsampling rather than downsampling when changing sample rates</h3>

	<p>Sampling rates can be changed on the fly. The granularity of
	such change is based on the internal buffering (typically a few hundred
	samples), not on a sample-by-sample basis. This can be used for effects.</p>

	<p>Do not dynamically change sampling rates when
	downsampling. When changing sample rates after an audio track is
	created, differences of around 5 to 10 percent from the original rate may
	trigger a filter recomputation when downsampling (to properly suppress
	aliasing). This can consume computing resources and may cause an audible click
	if the filter is replaced in real time.</p>

	<h3>Limit downsampling to no more than 6:1</h3>

	<p>Downsampling is typically triggered by hardware device requirements. When the
	Sample Rate converter is used for downsampling,
	try to limit the downsampling ratio to no more than 6:1 for good aliasing
	suppression (for example, no greater downsample than 48,000 to 8,000). The filter
	lengths adjust to match the downsampling ratio, but you sacrifice more
	transition bandwidth at higher downsampling ratios to avoid excessively
	increasing the filter length. There are no similar aliasing concerns for
	upsampling. Note that some parts of the audio pipeline
	may prevent downsampling greater than 2:1.</p>

	<h3 id="latency">If you are concerned about latency, do not resample</h3>

	<p>Resampling prevents the track from being placed in the FastMixer
	path, which means that significantly higher latency occurs due to the additional,
	larger buffer in the ordinary Mixer path. Furthermore,
	there is an implicit delay from the filter length of the resampler,
	though this is typically on the order of one millisecond or less,
	which is not as large as the additional buffering for the ordinary Mixer path
	(typically 20 milliseconds).</p>

	<h2 id="info">For More Information</h2>
	<p>This section lists some additional resources about sampling and resampling.</p>

	<h3>Sample rates</h3>

	<p>
	<a href="http://en.wikipedia.org/wiki/Sampling_%28signal_processing%29" class="external-link" >
	Sampling (signal processing)</a> at Wikipedia.</p>

	<h3>Resampling</h3>

	<p><a href="http://en.wikipedia.org/wiki/Sample_rate_conversion" class="external-link" >
	Sample rate conversion</a> at Wikipedia.</p>

	<p><a href="http://source.android.com/devices/audio/src.html" class="external-link" >
	Sample Rate Conversion</a> at source.android.com.</p>

	<h3>The high bit-depth and high kHz controversy</h3>

	<p><a href="http://people.xiph.org/~xiphmont/demo/neil-young.html" class="external-link" >
	24/192 Music Downloads ... and why they make no sense</a>
	by Christopher "Monty" Montgomery of Xiph.Org.</p>

	<p><a href="https://www.youtube.com/watch?v=cIQ9IXSUzuM" class="external-link" >
	D/A and A/D \| Digital Show and Tell</a>
	video by Christopher "Monty" Montgomery of Xiph.Org.</p>

	<p><a href="http://www.trustmeimascientist.com/2013/02/04/the-science-of-sample-rates-when-higher-is-better-and-when-it-isnt/" class="external-link">
	The Science of Sample Rates (When Higher Is Better - And When It Isn't)</a>.</p>

	<p><a href="http://www.image-line.com/support/FLHelp/html/app_audio.htm" class="external-link" >
	Audio Myths & DAW Wars</a></p>

	<p><a href="http://forums.stevehoffman.tv/threads/192khz-24bit-vs-96khz-24bit-debate-interesting-revelation.317660/" class="external-link">
	192kHz/24bit vs. 96kHz/24bit "debate"- Interesting revelation</a></p>