1 <html devsite> 2 <head> 3 <title>Audio Latency Measurements</title> 4 <meta name="project_path" value="/_project.yaml" /> 5 <meta name="book_path" value="/_book.yaml" /> 6 </head> 7 <body> 8 <!-- 9 Copyright 2017 The Android Open Source Project 10 11 Licensed under the Apache License, Version 2.0 (the "License"); 12 you may not use this file except in compliance with the License. 13 You may obtain a copy of the License at 14 15 http://www.apache.org/licenses/LICENSE-2.0 16 17 Unless required by applicable law or agreed to in writing, software 18 distributed under the License is distributed on an "AS IS" BASIS, 19 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 20 See the License for the specific language governing permissions and 21 limitations under the License. 22 --> 23 24 25 <p><a href="http://en.wikipedia.org/wiki/Latency_%28engineering%29">Latency</a> 26 is an important system performance metric. While many kinds of 27 <a href="latency.html">audio latency</a> metrics exist, one useful and 28 well-understood metric is 29 <a href="latency_measure.html#measuringRoundTrip">round-trip latency</a>, 30 defined as the time it takes for an audio signal to enter the input of a mobile 31 device, be processed by an app running on the application processor, and exit 32 the output.</p> 33 34 <img src="images/round_trip_on_device.png" alt="Round-trip audio latency on 35 device" id="figure1" /> 36 <p class="img-caption"><strong>Figure 1.</strong> Round-trip audio latency on 37 device: T<sub>output</sub> - T<sub>input</sub></p> 38 39 <p>This page provides round-trip audio latency measurements for select 40 Nexus/Pixel devices and Android platform versions.</p> 41 42 <h2 id="why">Why we measure latency</h2> 43 44 <p>Google measures and reports latency so Android application developers have 45 the data they need to make informed decisions about available latency on actual 46 devices. By sharing latency numbers for select Nexus and Pixel devices, we hope 47 to encourage the entire Android community to measure, publish, and reduce 48 latency on <em>all</em> Android devices. Please join us in our commitment to 49 reducing audio latency!</p> 50 51 <h2 id="app">Application impact on latency</h2> 52 53 <p>Signal processing can add the following types of delay to latency:</p> 54 <ul> 55 <li><strong>Algorithmic</strong>. This delay is inherent and does not vary with 56 the CPU. An example is the delay added by a 57 <a href="http://en.wikipedia.org/wiki/Finite_impulse_response">finite impulse 58 response</a> (FIR) filter.</li> 59 <li><strong>Computational</strong>. This delay is related to the number of 60 required CPU cycles. For example, attenuation of a signal is usually done by a 61 multiplication operation that takes a varying number of cycles depending on the 62 CPU.</li> 63 </ul> 64 65 <h2 id="how">How we measure</h2> 66 67 <p>We made the measurements listed on this page using the 68 <a href="loopback.html">Dr. Rick O'Rang audio loopback dongle</a> and an 69 <a href="latency_measure.html#larsenTest">audio feedback (Larsen effect) 70 test</a>. Measurements assume the application signal processing adds zero 71 algorithmic delay and near-zero computational delay.</p> 72 73 <p>We measure round-trip latency via the headset connector for several reasons: 74 </p> 75 <img src="images/round_trip_via_headset_connector.png" alt="Round-trip latency 76 via headset connector" id="figure2" /> 77 <p class="img-caption"><strong>Figure 2.</strong> Round-trip latency via headset 78 connector: T<sub>output</sub> - T<sub>input</sub></p> 79 <ul> 80 <li>Important music applications (such as guitar and voice processing) use the 81 headset connector.</li> 82 <li>Measuring round-trip latency of the on-device microphone and speaker can be 83 cumbersome, as it is difficult to keep a feedback loop in open air from entering 84 uncontrolled oscillation.</li> 85 <li>On-device transducers are small and sacrifice frequency response to achieve 86 their small size. To compensate, digital signal processing is applied but 87 increases algorithmic delay for the on-device path.</li> 88 </ul> 89 90 <p>There are cases where on-device microphone and speaker latencies <em>do</em> 91 matter, but they are usually for one direction, not round-trip. Techniques for 92 measuring unidirectional latency are described at 93 <a href="latency_measure.html#measuringOutput">Measuring Output Latency</a> 94 and 95 <a href="latency_measure.html#measuringInput">Measuring Input Latency</a>.</p> 96 97 <h2 id="measurements">Example measurements</h2> 98 99 <p>The measurements listed below are specific to a 100 <a href="/source/build-numbers.html">build number</a>. Devices are listed in 101 approximate order of initial release and by platform version; you can also 102 <a href="#chart">view latencies in a chart</a>. The test application uses the 103 Android native audio API based on OpenSL ES.</p> 104 105 <table> 106 <tr> 107 <th>Model</th> 108 <th>Platform<br />version</th> 109 <th>Build<br />number</th> 110 <th>Sample rate<br />(Hz)</th> 111 <th>Buffer size<br />(frames)</th> 112 <th>Buffer size<br />(ms)</th> 113 <th>Round-trip<br />latency (ms)<br />± one buffer</th> 114 </tr> 115 116 <tr> 117 <td>Nexus One</td> 118 <td>2.3.6</td> 119 <td>GRK39F</td> 120 <td>44100</td> 121 <td>768</td> 122 <td>17.4</td> 123 <td>345</td> 124 </tr> 125 126 <tr> 127 <td>Nexus S</td> 128 <td>2.3.6</td> 129 <td>GRK39F</td> 130 <td>44100</td> 131 <td>1024</td> 132 <td>23.2</td> 133 <td>260</td> 134 </tr> 135 136 <tr> 137 <td>Nexus S</td> 138 <td>4.0.4</td> 139 <td>IMM76D</td> 140 <td>44100</td> 141 <td>1024</td> 142 <td>23.2</td> 143 <td>260</td> 144 </tr> 145 146 <tr> 147 <td>Nexus S</td> 148 <td>4.1.2</td> 149 <td>JZO54K</td> 150 <td>44100</td> 151 <td>880</td> 152 <td>20</td> 153 <td>210</td> 154 </tr> 155 156 <tr> 157 <td>Galaxy Nexus</td> 158 <td>4.0.1</td> 159 <td>ITL41D</td> 160 <td>44100</td> 161 <td>976</td> 162 <td>22.1</td> 163 <td>270</td> 164 </tr> 165 166 <tr> 167 <td>Galaxy Nexus</td> 168 <td>4.3</td> 169 <td>JWR66Y</td> 170 <td>44100</td> 171 <td>144</td> 172 <td>3.3</td> 173 <td>130</td> 174 </tr> 175 176 <tr> 177 <td>Nexus 4</td> 178 <td>4.2.2</td> 179 <td>JDQ39E</td> 180 <td>48000</td> 181 <td>240</td> 182 <td>5</td> 183 <td>195</td> 184 </tr> 185 186 <tr> 187 <td>Nexus 4</td> 188 <td>5.1</td> 189 <td>LMY47O</td> 190 <td>48000</td> 191 <td>240</td> 192 <td>5</td> 193 <td>58</td> 194 </tr> 195 196 <tr> 197 <td>Nexus 10</td> 198 <td>5.0.2</td> 199 <td>LRX22G</td> 200 <td>44100</td> 201 <td>256</td> 202 <td>5.8</td> 203 <td>36</td> 204 </tr> 205 206 <tr> 207 <td>Nexus 10</td> 208 <td>5.1</td> 209 <td>LMY47D</td> 210 <td>44100</td> 211 <td>256</td> 212 <td>5.8</td> 213 <td>35</td> 214 </tr> 215 216 <tr> 217 <td>Nexus 7<br />2013</td> 218 <td>4.3</td> 219 <td>JSR78D</td> 220 <td>48000</td> 221 <td>240</td> 222 <td>5</td> 223 <td>149</td> 224 </tr> 225 226 <tr> 227 <td>Nexus 7<br />2013</td> 228 <td>4.4</td> 229 <td>KRT16S</td> 230 <td>48000</td> 231 <td>240</td> 232 <td>5</td> 233 <td>85</td> 234 </tr> 235 236 <tr> 237 <td>Nexus 7<br />2013</td> 238 <td>5.0.2</td> 239 <td>LRX22G</td> 240 <td>48000</td> 241 <td>240</td> 242 <td>5</td> 243 <td>64</td> 244 </tr> 245 246 <tr> 247 <td>Nexus 7<br />2013</td> 248 <td>5.1</td> 249 <td>LMY47O</td> 250 <td>48000</td> 251 <td>240</td> 252 <td>5</td> 253 <td>55</td> 254 </tr> 255 256 <tr> 257 <td>Nexus 7<br />2013</td> 258 <td>6.0</td> 259 <td>MRA58K</td> 260 <td>48000</td> 261 <td>240</td> 262 <td>5</td> 263 <td>55</td> 264 </tr> 265 266 <tr> 267 <td>Nexus 5</td> 268 <td>4.4.4</td> 269 <td>KTU84P</td> 270 <td>48000</td> 271 <td>240</td> 272 <td>5</td> 273 <td>95</td> 274 </tr> 275 276 <tr> 277 <td>Nexus 5</td> 278 <td>5.0.0</td> 279 <td>LRX21O</td> 280 <td>48000</td> 281 <td>240</td> 282 <td>5</td> 283 <td>47</td> 284 </tr> 285 286 <tr> 287 <td>Nexus 5</td> 288 <td>5.1</td> 289 <td>LMY47I</td> 290 <td>48000</td> 291 <td>240</td> 292 <td>5</td> 293 <td>42</td> 294 </tr> 295 296 <tr> 297 <td>Nexus 5</td> 298 <td>6.0</td> 299 <td>MRA58K</td> 300 <td>48000</td> 301 <td>192</td> 302 <td>4</td> 303 <td>38</td> 304 </tr> 305 306 <tr> 307 <td>Nexus 9</td> 308 <td>5.0.0</td> 309 <td>LRX21L</td> 310 <td>48000</td> 311 <td>256</td> 312 <td>5.3</td> 313 <td>35</td> 314 </tr> 315 316 <tr> 317 <td>Nexus 9</td> 318 <td>5.0.1</td> 319 <td>LRX22C</td> 320 <td>48000</td> 321 <td>256</td> 322 <td>5.3</td> 323 <td>38</td> 324 </tr> 325 326 <tr> 327 <td>Nexus 9</td> 328 <td>5.1.1</td> 329 <td>LMY47X</td> 330 <td>48000</td> 331 <td>256</td> 332 <td>5.3</td> 333 <td>32</td> 334 </tr> 335 336 <tr> 337 <td>Nexus 9</td> 338 <td>6.0</td> 339 <td>MRA58K</td> 340 <td>48000</td> 341 <td>128</td> 342 <td>2.6</td> 343 <td>15</td> 344 </tr> 345 346 <tr> 347 <td>Nexus 6</td> 348 <td>5.0.1</td> 349 <td>LRX22C</td> 350 <td>48000</td> 351 <td>240</td> 352 <td>5</td> 353 <td>65</td> 354 </tr> 355 356 <tr> 357 <td>Nexus 6</td> 358 <td>5.1</td> 359 <td>LMY47I</td> 360 <td>48000</td> 361 <td>240</td> 362 <td>5</td> 363 <td>42</td> 364 </tr> 365 366 <tr> 367 <td>Nexus 6</td> 368 <td>6.0</td> 369 <td>MRA58K</td> 370 <td>48000</td> 371 <td>192</td> 372 <td>4</td> 373 <td>33</td> 374 </tr> 375 376 <tr> 377 <td>Nexus 5X</td> 378 <td>6.0</td> 379 <td>MDA89E</td> 380 <td>48000</td> 381 <td>192</td> 382 <td>4</td> 383 <td>18</td> 384 </tr> 385 386 <tr> 387 <td>Nexus 6P</td> 388 <td>6.0</td> 389 <td>MDA89D</td> 390 <td>48000</td> 391 <td>192</td> 392 <td>4</td> 393 <td>18</td> 394 </tr> 395 396 <tr> 397 <td>Pixel</td> 398 <td>7.1.2</td> 399 <td>NHG47L</td> 400 <td>48000</td> 401 <td>192</td> 402 <td>4</td> 403 <td>18</td> 404 </tr> 405 406 <tr> 407 <td>Pixel XL</td> 408 <td>7.1.2</td> 409 <td>NHG47L</td> 410 <td>48000</td> 411 <td>192</td> 412 <td>4</td> 413 <td>18</td> 414 </tr> 415 416 </table> 417 418 <a id=chart></a> 419 <img src="/devices/audio/images/round-trip-latencies.png" /> 420 <p class="img-caption"><strong>Figure 3.</strong> Round trip latencies.</p> 421 422 </body> 423 </html> 424
