Machine learning for adaptive bitrate selection

Patent metadata

Field	Value
Publication number	US-11800167-B2
Application number	US-202117515225-A
Country	US
Kind code	B2
Filing date	Oct 29, 2021
Priority date	Oct 29, 2021
Publication date	Oct 24, 2023
Grant date	Oct 24, 2023

Abstract

Disclosed herein are system, apparatus, article of manufacture, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for modifying one or more parameters of a data streaming bitrate selection algorithm based on machine learning. An example embodiment operates by training and operating a first machine learning model to predict a sustainable network bandwidth. A second machine learning model is trained to receive the sustainable network bandwidth and predict a likelihood that this network bandwidth will not empty a data buffer of streaming data. A bitrate is selected based on the likelihood being below a threshold percentage, such as 50%.

First claim

What is claimed is:

1. A method performed by a system having at least a processor and a memory therein, wherein the method comprises:
receiving, by a client device, a data streaming request;
predicting, by a speed predictive machine learning model and based on one or more streaming parameters, a current sustainable network bandwidth, wherein the speed predictive machine learning model is trained using a training data set comprising a history of the one or more streaming parameters;
predicting, by a rebuffer predictive machine learning model and based on the current sustainable network bandwidth, a buffer level of a data buffer, a chunk duration, and N available discrete bitrates, a candidate bitrate at which a likelihood of rebuffering of the data buffer occurs less than a threshold percentage;
selecting, based on the candidate bitrate, a download bitrate to complete the data streaming request; and
downloading, by the client device, streaming data at the download bitrate.

2. The method of claim 1 , wherein the predicting the candidate bitrate further comprises:
selecting the current sustainable network bandwidth as a first instance of the candidate bitrate.

3. The method of claim 2 , wherein the predicting the candidate bitrate further comprises:
predicting that the first instance of the candidate bitrate would produce the rebuffering of the data buffer; and
selecting, from the N available discrete bitrates, a second instance of the candidate bitrate.

4. The method of claim 3 , wherein the selecting, from the N available discrete bitrates, the second instance of the candidate bitrate further comprises:
selecting a next lower bitrate from the N available discrete bitrates.

5. The method of claim 1 , wherein the predicting the candidate bitrate further comprises:
calculating the likelihood of rebuffering of the data buffer.

6. The method of claim 5 , wherein the threshold percentage is 50%.

7. The method of claim 5 , further comprising:
in response to the likelihood of rebuffering of the data buffer being over the threshold percentage, selecting a next lower available bitrate until the likelihood of rebuffering of the data buffer is under the threshold percentage.

8. The method of claim 1 , wherein the one or more streaming parameters further comprise any of:
N previous bitrates, where N represents a number of most recent bitrates in a history of previous download bitrates;
an average speed of the N previous bitrates;
a standard deviation of the N previous bitrates; and
the chunk duration.

9. A system comprising:
a memory; and
at least one processor coupled to the memory and configured to perform operations comprising:
receiving a data streaming request;
predicting, by a speed predictive machine learning model and based on one or more of a plurality of streaming parameters, a current sustainable network bandwidth, wherein the speed predictive machine learning model is trained using a training data set comprising a history of the one or more streaming parameters;
predicting, by a rebuffer predictive machine learning model and based on the current sustainable network bandwidth, a buffer level of a data buffer, a chunk duration, and N available discrete bitrates, a candidate bitrate at which a likelihood of rebuffering of the data buffer occurs less than a threshold percentage;
selecting, based on the candidate bitrate, a download bitrate to complete the streaming request; and
downloading streaming data at the download bitrate.

10. The system of claim 9 , wherein the predicting the candidate bitrate further comprises:
selecting the current sustainable network bandwidth as a first instance of the candidate bitrate.

11. The system of claim 10 , wherein the predicting the candidate bitrate further comprises:
predicting that the first instance of the candidate bitrate would produce the rebuffering of the data buffer; and
selecting, from the N available discrete bitrates, a second instance of the candidate bitrate.

12. The system of claim 11 , wherein the selecting, from the N available discrete bitrates, the second instance of the candidate bitrate further comprises:
selecting a next lower bitrate from the N available discrete bitrates.

13. The system of claim 9 , wherein the predicting the candidate bitrate further comprises:
calculating the likelihood of rebuffering of the data buffer.

14. The system of claim 13 , further configured to perform operations comprising:
in response to the likelihood of rebuffering of the data buffer being over the threshold percentage, selecting a next lower available bitrate until the likelihood of rebuffering of the data buffer is under the threshold percentage.

15. The system of claim 9 , wherein the one or more streaming parameters further comprise any of:
N previous bitrates, where N represents a number of most recent bitrates in a history of previous download bitrates;
an average speed of the N previous bitrates;
a standard deviation of the N previous bitrates; and
the chunk duration.

16. A non-transitory computer-readable medium having instructions stored thereon that, when executed by at least one computing device, cause the at least one computing device to perform operations comprising:
receiving a data streaming request;
predicting, by a speed predictive machine learning model and based on one or more of a plurality of streaming parameters, a current sustainable network bandwidth, wherein the speed predictive machine learning model is trained using a training data set comprising a history of the one or more streaming parameters;
predicting, by a rebuffer predictive machine learning model and based on the one or more streaming parameters and the current sustainable network bandwidth, a buffer level of a data buffer, a chunk duration, and N available discrete bitrates, a candidate bitrate at which a likelihood of rebuffering of the data buffer occurs less than a threshold percentage;
selecting, based on the candidate bitrate, a download bitrate to complete the streaming request; and
downloading streaming data at the download bitrate.

17. The non-transitory computer-readable medium of claim 16 , wherein the predicting the candidate bitrate further comprises:
selecting the current sustainable network bandwidth as a first instance of the candidate bitrate.

18. The non-transitory computer-readable medium of claim 17 , wherein the predicting the candidate bitrate further comprises:
predicting that the first instance of the candidate bitrate would produce the rebuffering of the data buffer; and
selecting, from the N available discrete bitrates, a second instance of the candidate bitrate.

19. The non-transitory computer-readable medium of claim 18 , wherein the selecting, from the N available discrete bitrates, the second instance of the candidate bitrate further comprises:
selecting a next lower bitrate from the N available discrete bitrates.

20. The non-transitory computer-readable medium of claim 16 , wherein the operations further comprise:
in response to the likelihood of rebuffering of the data buffer being over the threshold percentage, selecting a next lower available bitrate until the likelihood of rebuffering of the data buffer is under the threshold percentage.

21. The method of claim 1 , further comprising:
retraining the speed predictive machine learning model based on the current sustainable network bandwidth; and
retraining the rebuffer predictive machine learning model based on th

Assignees

• Roku Inc

Inventors

• Paliwal Amit
• Marsavin Andrey
• Vaidya Govind
• Michiels Wim
• Logan Beth Teresa
• Han Zheng
• Oza Tapan
• Raghavan Vijay Anand

Classifications

• H04N21/2662 Primary

  Controlling the complexity of the video stream, e.g. by scaling the resolution or bitrate of the video stream based on the client capabilities · CPC title

• G06N20/00

  Machine learning · CPC title

• H04N21/2401

  Monitoring of the client buffer · CPC title

• H04N21/251

  Learning process for intelligent management, e.g. learning user preferences for recommending movies (details of learning user preferences for the retrieval of video data in a video database G06F16/739; computer systems using learning methods G06N3/08) · CPC title

• H04N21/44004 Primary

  involving video buffer management, e.g. video decoder buffer or video display buffer · CPC title