Chunk-based prediction adaptation logic
A multimedia player downloads chunks (parts of the segment file) during the download of a segment of a stream of segments of a low-latency stream. The first chunks of a segment may be downloaded while the segment is still being written to the CDN server. A chunk-based prediction adaptation logic uses throughput measurements on a chunk instead of a segment and specifically looks at bursts in a sliding window. This data is used to build a prediction of future throughput by applying linear adaptive filter algorithms that may rely on recursive least squares. This adaptation logic leads to very accurate bandwidth predictions and as a consequence, better user experience, compared to existing adaptation algorithms.
1. A method for adaptively selecting a bitrate level for streaming a low latency multimedia stream, the multimedia stream comprising a plurality of segments and each segment comprising a plurality of chunks of video, the method comprising:
computing an average throughput based on throughput measurements for a plurality of prior successful chunk downloads;
based on the average throughput, estimating a future bandwidth of a network connection between a media player and a server;
selecting a bitrate level based on the average throughput and the estimated future bandwidth, the bitrate level selected from a plurality of available bitrate levels defined in a manifest file for the low latency multimedia stream; and
selecting a network address from the manifest file, the network address corresponding to a stream encoded at the selected bitrate level,
wherein computing the average throughput comprises recursively computing a gain vector and an inverse correlation matrix of throughput estimation values, and wherein the gain vector and the inverse correlation matrix are recursively computed using a forgetting factor pre-programmed to a given value configured to reduce a weight of throughput estimates.
2. The method of claim 1 , further comprising initializing a filter tabs vector with a length, the future bandwidth of the network being based in part on the filter tabs vector.
3. The method of claim 2 , further comprising setting an initial inverse correlation matrix according to an initial input variance estimate parameter and the length of the filter tabs vector.
4. The method of claim 1 , wherein the bitrate level is further selected based on the playback buffer occupancy.
5. The method of claim 1 , further comprising maintaining a buffer occupancy within a range between a minimum threshold and a maximum threshold predetermined for CMAF low latency.
6. A client device, comprising:
a multimedia player comprising a processor and a memory, the multimedia player configured to implement:
a buffer configured to buffer a media content,
an adaptation logic configured to control a bitrate at which the media content is retrieved,
a decoder configured to reproduce the media content; and
a playback interface,
wherein the adaptation logic is configured to implement a throughput estimator, a bandwidth forecasting module, and an ABR controller,
wherein the throughput estimator is configured to compute an average throughput based on throughput measurements for a plurality of prior successful chunk downloads, wherein computing the average throughput comprises recursively computing a gain vector and an inverse correlation matrix of throughput estimation values, and wherein the gain vector and the inverse correlation matrix are recursively computed using a forgetting factor pre-programmed to a given value configured to reduce a weight of throughput estimates,
wherein the bandwidth forecasting module is configured to estimate a future bandwidth of a network connection between the multimedia player and a server based on the average throughput, and
wherein the ABR controller is configured to select a bitrate level based on the average throughput and the estimated future bandwidth and a network address from a manifest file, the bitrate level selected from a plurality of available bitrate levels defined in the manifest file for a low latency multimedia stream, the network address corresponding to a stream encoded at the selected bitrate level.
7. The device of claim 6 , wherein the adaptation logic sets the bitrate based on a buffer fill status of the buffer.
8. The device of claim 6 , wherein the adaptation logic sets the bitrate based on a measure of a retrieval throughput at which stream data is retrieved.
9. The device of claim 6 , wherein the adaptation logic sets the bitrate based on a hybrid approach using a buffer fill status of the buffer and a measure of a retrieval throughput at which stream data is retrieved.
10. The device of claim 6 , wherein the adaptation logic further implements an oscillation compensator.
11. The device of claim 6 , wherein the adaptation logic further implements a buffer controller and the ABR controller is configured to monitor a playback buffer occupancy using the buffer controller.
12. The device of claim 6 , wherein the ABR controller is configured to output a bitrate at which streaming of the media content is controlled.
13. The device of claim 6 , further comprising a logger configured to record a status of the multimedia player.
14. The device of claim 6 , wherein the ABR controller is configured to implement a throughput-based approach using an estimated throughput.
15. The device of claim 6 , wherein the ABR controller is configured to implement a buffer-based approach, using a playback buffer occupancy.
16. The device of claim 6 , wherein the ABR controller is configured to implement a hybrid approach using an estimated throughput and a playback buffer occupancy.
17. The device of claim 6 , wherein the bandwidth forecasting module comprises an adaptive filter configured to implement a recursive least squares approach.
18. The device of claim 6 , wherein the bandwidth forecasting module is configured to implement a deep reinforcement learning approach.
19. The device of claim 6 , wherein the bandwidth forecasting module is configured to implement an autoregressive integrated moving average approach.