4 The latency is the time it takes for a sample captured at timestamp 0 to reach the
5 sink. This time is measured against the clock in the pipeline. For pipelines
6 where the only elements that synchronize against the clock are the sinks, the
7 latency is always 0 since no other element is delaying the buffer.
9 For pipelines with live sources, a latency is introduced, mostly because of the
10 way a live source works. Consider an audio source, it will start capturing the
11 first sample at time 0. If the source pushes buffers with 44100 samples at a
12 time at 44100Hz it will have collected the buffer at second 1.
13 Since the timestamp of the buffer is 0 and the time of the clock is now >= 1
14 second, the sink will drop this buffer because it is too late.
15 Without any latency compensation in the sink, all buffers will be dropped.
17 The situation becomes more complex in the presence of:
19 - 2 live sources connected to 2 live sinks with different latencies
20 * audio/video capture with synchronized live preview.
21 * added latencies due to effects (delays, resamplers...)
22 - 1 live source connected to 2 live sinks
24 * RTP, with added latencies because of jitter buffers.
25 - mixed live source and non-live source scenarios.
26 * synchronized audio capture with non-live playback. (overdubs,..)
27 - clock slaving in the sinks due to the live sources providing their own
30 To perform the needed latency corrections in the above scenarios, we must
31 develop an algorithm to calculate a global latency for the pipeline. The
32 algorithm must be extensible so that it can optimize the latency at runtime.
33 It must also be possible to disable or tune the algorithm based on specific
34 application needs (required minimal latency).
37 Pipelines without latency compensation
38 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
40 We show some examples to demonstrate the problem of latency in typical
45 An audio capture/playback pipeline.
47 asrc: audio source, provides a clock
48 asink audio sink, provides a clock
50 .--------------------------.
52 | .------. .-------. |
53 | | asrc | | asink | |
55 | '------' '-------' |
56 '--------------------------'
59 asink: NULL->READY: probes device, returns SUCCESS
60 asrc: NULL->READY: probes device, returns SUCCESS
63 asink: READY:->PAUSED open device, returns ASYNC
64 asrc: READY->PAUSED: open device, returns NO_PREROLL
66 * Since the source is a live source, it will only produce data in the
67 PLAYING state. To note this fact, it returns NO_PREROLL from the state change
69 * This sink returns ASYNC because it can only complete the state change to
70 PAUSED when it receives the first buffer.
72 At this point the pipeline is not processing data and the clock is not
73 running. Unless a new action is performed on the pipeline, this situation will
77 asrc clock selected because it is the most upstream clock provider. asink can
78 only provide a clock when it received the first buffer and configured the
79 device with the samplerate in the caps.
81 asink: PAUSED:->PLAYING, sets pending state to PLAYING, returns ASYNC becaus
82 it is not prerolled. The sink will commit state to
83 PLAYING when it prerolls.
84 asrc: PAUSED->PLAYING: starts pushing buffers.
86 * since the sink is still performing a state change from READY -> PAUSED, it
87 remains ASYNC. The pending state will be set to PLAYING.
88 * The clock starts running as soon as all the elements have been set to
90 * the source is a live source with a latency. Since it is synchronized with
91 the clock, it will produce a buffer with timestamp 0 and duration D after
92 time D, ie. it will only be able to produce the last sample of the buffer
93 (with timestamp D) at time D. This latency depends on the size of the
95 * the sink will receive the buffer with timestamp 0 at time >= D. At this
96 point the buffer is too late already and might be dropped. This state of
97 constantly dropping data will not change unless a constant latency
98 correction is added to the incoming buffer timestamps.
100 The problem is due to the fact that the sink is set to (pending) PLAYING
101 without being prerolled, which only happens in live pipelines.
105 An audio/video capture/playback pipeline. We capture both audio and video and
106 have them played back synchronized again.
108 asrc: audio source, provides a clock
109 asink audio sink, provides a clock
113 .--------------------------.
115 | .------. .-------. |
116 | | asrc | | asink | |
118 | '------' '-------' |
119 | .------. .-------. |
120 | | vsrc | | vsink | |
122 | '------' '-------' |
123 '--------------------------'
125 The state changes happen in the same way as example 1. Both sinks end up with
126 pending state of PLAYING and a return value of ASYNC until they receive the
129 For audio and video to be played in sync, both sinks must compensate for the
130 latency of its source but must also use exactly the same latency correction.
132 Suppose asrc has a latency of 20ms and vsrc a latency of 33ms, the total
133 latency in the pipeline has to be at least 33ms. This also means that the
134 pipeline must have at least a 33 - 20 = 13ms buffering on the audio stream or
135 else the audio src will underrun while the audiosink waits for the previous
140 An example of the combination of a non-live (file) and a live source (vsrc)
141 connected to live sinks (vsink, sink).
143 .--------------------------.
145 | .------. .-------. |
146 | | file | | sink | |
148 | '------' '-------' |
149 | .------. .-------. |
150 | | vsrc | | vsink | |
152 | '------' '-------' |
153 '--------------------------'
155 The state changes happen in the same way as example 1. Except sink will be
156 able to preroll (commit its state to PAUSED).
158 In this case sink will have no latency but vsink will. The total latency
159 should be that of vsink.
161 Note that because of the presence of a live source (vsrc), the pipeline can be
162 set to playing before sink is able to preroll. Without compensation for the
163 live source, this might lead to synchronisation problems because the latency
164 should be configured in the element before it can go to PLAYING.
169 An example of the combination of a non-live and a live source. The non-live
170 source is connected to a live sink and the live source to a non-live sink.
172 .--------------------------.
174 | .------. .-------. |
175 | | file | | sink | |
177 | '------' '-------' |
178 | .------. .-------. |
179 | | vsrc | | files | |
181 | '------' '-------' |
182 '--------------------------'
184 The state changes happen in the same way as example 3. Sink will be
185 able to preroll (commit its state to PAUSED). files will not be able to
188 sink will have no latency since it is not connected to a live source. files
189 does not do synchronisation so it does not care about latency.
191 The total latency in the pipeline is 0. The vsrc captures in sync with the
194 As in example 3, sink can only be set to PLAYING after it successfully
201 A Sink is never set to PLAYING before it is prerolled. In order to do this, the
202 pipeline (at the GstBin level) keeps track of all
203 elements that require preroll (the ones that return ASYNC from the state
204 change). These elements posted a ASYNC_START message without a matching
207 The pipeline will not change the state of the elements that are still doing an
210 When an ASYNC element prerolls, it commits its state to PAUSED and posts an
211 ASYNC_DONE message. The pipeline notices this ASYNC_DONE message and matches it
212 with the ASYNC_START message it cached for the corresponding element.
214 When all ASYNC_START messages are matched with an ASYNC_DONE message, the
215 pipeline proceeds with setting the elements to the final state again.
217 The base time of the element was already set by the pipeline when it changed the
218 NO_PREROLL element to PLAYING. This operation has to be performed in the
219 separate async state change thread (like the one currently used for going from
220 PAUSED->PLAYING in a non-live pipeline).
226 Latency calculation and compensation is performed before the pipeline proceeds to
229 When the pipeline collected all ASYNC_DONE messages it can calculate the global
232 - perform a latency query on all sinks.
233 - latency = MAX (all min latencies)
234 - if MIN (all max latencies) < latency we have an impossible situation and we
235 must generate an error indicating that this pipeline cannot be played. This
236 usually means that there is not enough buffering in some chain of the
237 pipeline. A queue can be added to those chains.
239 The sinks gather this information with a LATENCY query upstream. Intermediate
240 elements pass the query upstream and add the amount of latency they add to the
248 MIN (20, 40) = 20 < 33 -> impossible
255 MIN (50, 40) = 40 >= 33 -> latency = 33
257 The latency is set on the pipeline by sending a LATENCY event to the sinks
258 in the pipeline. This event configures the total latency on the sinks. The
259 sink forwards this LATENCY event upstream so that intermediate elements can
260 configure themselves as well.
262 After this step, the pipeline continues setting the pending state on its
265 A sink adds the latency value, received in the LATENCY event, to
266 the times used for synchronizing against the clock. This will effectively
267 delay the rendering of the buffer with the required latency. Since this delay is
268 the same for all sinks, all sinks will render data relatively synchronised.
271 Flushing a playing pipeline
272 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
274 We can implement resynchronisation after an uncontrolled FLUSH in (part of) a
275 pipeline in the same way. Indeed, when a flush is performed on
276 a PLAYING live element, a new base time must be distributed to this element.
278 A flush in a pipeline can happen in the following cases:
280 - flushing seek in the pipeline
281 - performed by the application on the pipeline
282 - performed by the application on an element
283 - flush preformed by an element
284 - after receiving a navigation event (DVD, ...)
286 When a playing sink is flushed by a FLUSH_START event, an ASYNC_START message is
287 posted by the element. As part of the message, the fact that the element got
288 flushed is included. The element also goes to a pending PAUSED state and has to
289 be set to the PLAYING state again later.
291 The ASYNC_START message is kept by the parent bin. When the element prerolls,
292 it posts an ASYNC_DONE message.
294 When all ASYNC_START messages are matched with an ASYNC_DONE message, the bin
295 will capture a new base_time from the clock and will bring all the sinks back to
296 PLAYING after setting the new base time on them. It's also possible
297 to perform additional latency calculations and adjustments before doing this.
300 Dynamically adjusting latency
301 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
303 An element that want to change the latency in the pipeline can do this by
304 posting a LATENCY message on the bus. This message instructs the pipeline to:
306 - query the latency in the pipeline (which might now have changed) with a
308 - redistribute a new global latency to all elements with a LATENCY event.
310 A use case where the latency in a pipeline can change could be a network element
311 that observes an increased inter packet arrival jitter or excessive packet loss
312 and decides to increase its internal buffering (and thus the latency). The
313 element must post a LATENCY message and perform the additional latency
314 adjustments when it receives the LATENCY event from the downstream peer element.
316 In a similar way can the latency be decreased when network conditions are
319 Latency adjustments will introduce glitches in playback in the sinks and must
320 only be performed in special conditions.