Latency

Introduction

The definition of latency depends on the system being observed and the nature of stimulation; from a general point of view, latency is a time delay between the stimulation and the response of the system being observed.

Just as every physical system, the player presents a latency as well, mainly generated by the internal rendering engine; in most cases, this rendering latency is not human-noticeable. However, the rendering latency does have a noticeable effect in case of interactive content and real-time streaming media; in these cases you might need to reduce the rendering latency for faster response to interactive events or to lower the delay in relation to the streaming source. Note that there are also cases when the rendering latency should not be reduced.

Rendering latency

In order to render the content, the HMP must first perform a multitude of tasks, like executing the SVG & JavaScript code, decoding the media resources, preparing the output frames etc., each of these tasks taking variable amounts of time. To ensure a smooth output at the frame rate configured under display settings, the HMP needs to buffer the output frames in advance, which results in a rendering latency.

The HMP buffering model is similar to a queue where the frame buffers are filled up by the rendering engine and emptied as the content is displayed on the screen on the other side. The buffering duration is set by default to upper limit of 1500 milliseconds. When the content is complex and the HMP detects that buffering doesn't keep up with the frame outputting, it will either lower the frame rate or drop some frames.

In most cases, the rendering latency is not noticeable and the default settings should be kept to prevent frames being dropped. However, the rendering latency does have a noticeable effect in case of interactive content and when playing real-time streaming media. In these cases you might need to reduce the rendering latency for faster response to interactive events or to lower the delay in relation to the streaming source. Note that there are also cases when the rendering latency should not be reduced.

How to reduce the rendering latency

Advanced Applications > Interactivity page

Display Settings page > Interactivity tab

The two settings that control the rendering latency can be found on:

HMP400/W, HMP350, HMP300: "Advanced Applications" > Interactivity page.
HMP200, HMP130, HMP100: "Display Settings" page > "Interactivity" tab, under "Interactivity Settings" section.

Maximum rendering latency

The HMP rendering latency can be reduced by changing the value of the "Maximum rendering latency" option from 1500ms to 1000ms, 500ms or 250ms.

Reducing the maximum rendering latency should be done with care as the available buffer will be smaller. A lower latency means less overhead for the player to render more complex content and more likely to result in dropped frames.

Testing can be done by gradually reducing the latency while checking the info.log file to make sure that the player is not dropping any frames (factor 1-1, Nb Pic) and the maximum rendering time of any frame is within the allowed overhead (note the maximum time "Max: XXXms" value).

Interactivity boost

To improve interactivity even further, the HMP can temporarily ignore the value of the "Maximum rendering latency" option and reduce its rendering latency to only 60ms if the "Limit / Reduce latency to 60ms when events are received" option is enabled (which by default it is). In this case, the following happens:

The first interactive event is processed after the "regular" latency period.
After an interactive event is detected, the HMP reduces its rendering latency to 60ms, for the following interactive events to be processed very fast.
If no interactive event is received within the following 90 seconds, the HMP reverts the rendering latency back to the "regular" value.

Ultra-low latency

Starting with 4.0.2 release, is it possible to configure the HMP400/W, HMP350 and HMP300 devices into a ultra-low latency mode. This is done by setting the value of the "Maximum rendering latency" option to ultra-low values, such as: 200ms, 100ms, or 50ms, from the advanced mode of the player.

Furthermore, this can be combined with a negative value for the spx:buffering attribute to achieve very low streaming latency on the HMP400/W and HMP350.

When not to reduce the rendering latency

When using audio in your project, the "Reduce latency to 60ms when events are received" option must be disabled. Moreover, the "Maximum rendering latency" option should be set to at least 500ms.
When the content is very complex the "Reduce latency to 60ms when events are received" option should be disabled because temporarily reducing the latency to such a low duration might have a negative impact over the quality of the output. Also, the "Maximum rendering latency" option can be lowered only if there are no performance issues (see explanation above).
When some interactive events are skipped (i.e. the first interactive event might appear to be "lost") the "Reduce latency to 60ms when events are received" option should be disabled.

Streaming latency

For the special case of streaming, the latency is computed as following:

Total latency = Encoding latency + Network latency +  HMP stream decoding latency + HMP rendering latency + Screen latency

Encoding latency
This is the time required by the streaming hardware to prepare and send the stream packets, usually around one second. It cannot be controlled by the HMP; some encoders might allow the change of the latency value.

Network latency
This is the time for the network packets to travel from the streaming source to the HMP. This cannot be controlled by the HMP, but is usually negligible when the streaming server is on the same network as the HMP.

Stream buffering latency
This is the time required by the HMP to buffer the incoming network packets and is needed to avoid image freezing due to network packets not being received. The behavior is different depending on the firmware version, as following:
- Starting with firmware 4.0.0, the buffering delay depends on the source type:
  - For RTP sources, the HMP is buffering 750 ms of video and 350 ms of audio by default - this buffer can be increased (or reduced starting with firmware 4.0.2) using the spx:buffering svg attribute.
  - For MPEG2 transport stream sources, the buffering delay is extracted from the embedded DTS and PTS timestamps of each stream, and is thus under control of the streaming source.
- For firmware versions lower than 4.x, the HMP is buffering at least 10 frames of video (equal to 333ms @ 30fps or 400ms at 25fps). Note that large variations in the bitrate of the streaming media will determine an increase of the buffering up to 20 frames.

Screen latency
This is induced by the refresh frequency of the screen and shouldn't be more than 30-40 ms. In some cases, image processing options might contribute to the the screen latency - disabling them would help in such cases.

Note:

For a typical scenario and using the default values, the you can expect a total latency of around 3 seconds:
Total latency = 1s + 0s + 0.75s + 1.5s + 0.04s = 3.3s

How to reduce the streaming latency

Using the default values works for most cases, but in some, a low-latency might be required (for instance when security cameras are employed) - for that, follow these steps:

Check if the encoding latency can be reduced - some devices might have a low-latency option which can be enabled.
If the streaming server is not on the same network as the HMP, try to put them within the same network (even better, isolate that from the rest of the network).
Reduce the maximum rendering latency of the HMP down to 1000ms, 500ms or 250ms (when audio is not used), taking into account the complexity of the content.