[go: up one dir, main page]

CN112911408A - Intelligent video code rate adjustment and bandwidth allocation method based on deep learning - Google Patents

Intelligent video code rate adjustment and bandwidth allocation method based on deep learning Download PDF

Info

Publication number
CN112911408A
CN112911408A CN202110097764.7A CN202110097764A CN112911408A CN 112911408 A CN112911408 A CN 112911408A CN 202110097764 A CN202110097764 A CN 202110097764A CN 112911408 A CN112911408 A CN 112911408A
Authority
CN
China
Prior art keywords
user
code rate
bandwidth
allocation
video
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110097764.7A
Other languages
Chinese (zh)
Other versions
CN112911408B (en
Inventor
宋彤雨
任婧
胡文昱
谈雪彬
王雄
徐世中
王晟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202110097764.7A priority Critical patent/CN112911408B/en
Publication of CN112911408A publication Critical patent/CN112911408A/en
Application granted granted Critical
Publication of CN112911408B publication Critical patent/CN112911408B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/60—Network structure or processes for video distribution between server and client or between remote clients; Control signalling between clients, server and network components; Transmission of management data between server and client, e.g. sending from server to client commands for recording incoming content stream; Communication details between server and client 
    • H04N21/63—Control signaling related to video distribution between client, server and network components; Network processes for video distribution between server and clients or between remote clients, e.g. transmitting basic layer and enhancement layers over different transmission paths, setting up a peer-to-peer communication via Internet between remote STB's; Communication protocols; Addressing
    • H04N21/647—Control signaling between network components and server or clients; Network processes for video distribution between server and clients, e.g. controlling the quality of the video stream, by dropping packets, protecting content from unauthorised alteration within the network, monitoring of network load, bridging between two different networks, e.g. between IP and wireless
    • H04N21/64723—Monitoring of network processes or resources, e.g. monitoring of network load
    • H04N21/64738—Monitoring network characteristics, e.g. bandwidth, congestion level
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/50—Queue scheduling
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N7/00—Television systems
    • H04N7/18—Closed-circuit television [CCTV] systems, i.e. systems in which the video signal is not broadcast

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Neurology (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computer Security & Cryptography (AREA)
  • Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)

Abstract

本发明公开了一种基于深度学习的智能视频码率调整及带宽分配方法，构建码率分配神经网络，其输入为包括可用带宽归一化值、用户缓存视频时长归一化值和用户活跃状态的状态向量，其输出为每个用户的码率分配方案向量；设置若干用户状态和可用带宽场景以获取训练样本，对码率分配神经网络进行训练；在每个决策时隙到来时，获取状态向量输入至训练好的码率分配神经网络，得到码率分配方案，再根据码率分配方案和当前可用带宽确定每个用户所占用的带宽，完成带宽分配。本发明利用神经网络推断过程执行速度快的特点，快速求解码率和带宽分配方案，提高时效及准确性。

Figure 202110097764

The invention discloses an intelligent video bit rate adjustment and bandwidth allocation method based on deep learning. A bit rate allocation neural network is constructed, and the input includes a normalized value of available bandwidth, a normalized value of user cached video duration and a user's active state. The output is the bit rate allocation scheme vector of each user; set several user status and available bandwidth scenarios to obtain training samples, and train the bit rate allocation neural network; when each decision time slot arrives, obtain the status The vector is input to the trained bit rate allocation neural network to obtain the bit rate allocation scheme, and then the bandwidth occupied by each user is determined according to the bit rate allocation scheme and the current available bandwidth to complete the bandwidth allocation. The present invention utilizes the characteristics of fast execution speed of the neural network inference process to quickly obtain the decoding rate and bandwidth allocation scheme, thereby improving the timeliness and accuracy.

Figure 202110097764

Description

Intelligent video code rate adjustment and bandwidth allocation method based on deep learning
Technical Field
The invention belongs to the technical field of communication, and particularly relates to an intelligent video code rate adjustment and bandwidth allocation method based on deep learning.
Background
In recent years, the traffic volume of video streams has increased dramatically and has occupied a major portion of the overall internet traffic. To adapt to Dynamic network conditions and provide better quality of service, most video service providers have deployed Dynamic Adaptive Streaming over HTTP (DASH) schemes based on HTTP. In the DASH scheme, a video file is divided into a plurality of segments of a fixed segment length (in seconds). Each segment of video may be encoded into a different copy using a different bit rate, and each bit rate corresponds to a particular video frame rate and resolution. The client can adaptively request the bit rate (different frame rate and resolution) of the next segment based on the estimated current network conditions.
However, with widespread deployment of DASH and rapid growth in video demand, it is becoming more common for multiple DASH streams to compete for bandwidth resources on the same bottleneck link. Under the condition that multiple DASH streams compete for the same bottleneck link, the existing adaptive code rate adjustment algorithm has the following disadvantages: 1) each client independently makes a bit rate self-adaptive decision without knowing the existence of other clients, and the cooperation among the clients is lacked; 2) the client can only detect the physical bandwidth at the application layer and lacks accurate network bandwidth information. This results in: 1) the user experience quality is low, for example, the video playing times are increased; 2) the quality of service experience gap between different users is large.
One possible solution is to deploy a centrally controlled rate and bandwidth allocation scheme at the bottleneck link controller. At present, scholars perform mathematical modeling on the method, and the mathematical model is as follows:
Figure RE-GDA0003017361750000021
Subject to:
Figure RE-GDA0003017361750000022
Figure RE-GDA0003017361750000023
Figure RE-GDA0003017361750000024
Figure RE-GDA0003017361750000025
Figure RE-GDA0003017361750000026
Figure RE-GDA0003017361750000027
wherein A, B, C, a is a coefficient adjustable according to actual conditions, ri,kCode rate, omega, allocated to the k-th time slot user ii,kBandwidth, β, allocated to the k-th time slot user ii,kIndicating the lowest bandwidth required by the k time slot user i to guarantee that the video is not blocked, bi,kThe video duration in the buffer memory for the kth time slot user i, pi is a unit decision time slot, WkIs the available bandwidth of the k-th slot.
Figure RE-GDA0003017361750000028
The model respectively represents the contribution of the standard deviation of the code rate, the residual cache and all the user code rates to the user service quality, ensures that a user can watch the video service with the code rate as high as possible in the next unit time slot, the video of all the users can not be blocked, more video caches are provided, and the fairness of the service quality among the users is higher.
However, at present, the problem is proved to be an NP difficult problem, the solution can not be completed in polynomial time, and the user service requirement sensitive to the time delay at the present stage can not be adapted. Therefore, a solution that can solve this problem quickly and efficiently is highly desirable.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide an intelligent video code rate adjustment and bandwidth allocation method based on deep learning, and the method utilizes the characteristic of high execution speed of a neural network inference process to quickly solve a code rate and bandwidth allocation scheme and improve the timeliness and the accuracy.
In order to achieve the above object, the method for adjusting the code rate and allocating the bandwidth of the intelligent video based on the deep learning of the present invention comprises the following steps:
s1: recording the maximum number of users which can simultaneously request DASH service flow and are controlled by a bottleneck link controller accessed to a core network as N, setting M user selectable code rate gears according to actual conditions, wherein the code rate is higher when the serial number of the code rate gear is larger;
s2: constructing a code rate distribution neural network, comprising an input layer, a full-connection hidden layer and an output layer, wherein the input layer comprises 2N +1 neural units, and the input is a 2N +1 dimensional state vector I ═ W, b1,s1,b2,s2,…,bN,sN) Wherein W represents a normalized value of available bandwidth for access of the bottleneck link controller to the core network,
Figure RE-GDA0003017361750000031
snrespectively representing the time length normalization value and the active state s of the cached video of the nth user n1 indicates that the user is active and needs to perform rate allocation, s n0 means user inactivity, not requiredCode rate allocation is carried out, wherein N is 1,2, …, N; the output layer comprises N independent output layers, each independent output layer comprises M +1 nerve units, and the output vector of the nth independent output layer is On＝(rn,0,rn,1,…,rn,M) Wherein r isn,mThe probability of selecting the mth gear code rate in the next time slot of the nth user is represented, and the value range is [0,1 ]]M-0, 1, …, M-0 indicates no code rate is allocated;
setting D user states and available bandwidth scenes, normalizing the user cache video time length and the available bandwidth according to the maximum value and the minimum value of the user cache video time length and the maximum value and the minimum value of the available bandwidth, and determining corresponding state vectors
Figure RE-GDA0003017361750000032
Wherein D is 1,2, …, D, WdA normalized value representing the available bandwidth for the bottleneck link controller to access the core network in scenario d,
Figure RE-GDA0003017361750000033
respectively representing a normalized value and an active state of the nth user cache video time length in the d scene; obtaining the code rate selected by each user in the next time slot in the scene according to theoretical analysis, and allocating the code rate to the scheme vector
Figure RE-GDA0003017361750000034
The corresponding element in the group is set to be 1, the other elements are set to be 0,
Figure RE-GDA0003017361750000035
indicating that the nth user selects the mth gear rate in the next time slot in the d-th scenario,
Figure RE-GDA0003017361750000036
indicating that the nth user does not select the mth gear code rate in the next time slot in the d scenes;
state vector of each scene
Figure RE-GDA0003017361750000037
As input, a code rate allocation scheme vector for N users
Figure RE-GDA0003017361750000038
As expected output, training a code rate distribution neural network;
s3: when each decision time slot comes, obtaining the number N' of DASH service users controlled by the current bottleneck link controller and the available bandwidth W of the bottleneck link controller accessed to the core network*Normalizing the time length and the available bandwidth of the user cache video according to the maximum value and the minimum value of the time length of the user cache video and the maximum value and the minimum value of the available bandwidth to obtain a normalized value b of the time length of the user cache videon′And normalizing the value W by the available bandwidth, and simultaneously determining the activity state s of each usern′N '═ 1,2, …, N'; constructing a current state vector I ═ (W, b)1,s1,b2,s2,…,bN,sN) When N is less than or equal to N', bn＝bn′，sn＝sn′Otherwise bn＝0，sn＝0；
S4: inputting the current state vector n' obtained in the step S3 into the code rate allocation neural network trained in the step S102 to obtain a code rate allocation scheme vector ω of each user in the next time slotn′For the current n' users, selecting the code rate gear corresponding to the probability maximum value in the code rate distribution scheme vector as the code rate gear R of the next time slot of the usern′，βn′Obtaining a code rate distribution scheme;
s5: and calculating the bandwidth allocated to each user according to the current available bandwidth and the code rate of the current n' users determined by the code rate allocation scheme, and completing bandwidth allocation.
The invention relates to an intelligent video code rate adjustment and bandwidth allocation method based on deep learning, wherein a code rate allocation neural network is constructed, the input of the neural network is a state vector comprising an available bandwidth normalization value, a user cache video time normalization value and a user active state, and the output of the neural network is a code rate allocation scheme vector of each user; setting a plurality of user states and available bandwidth scenes to obtain training samples, and training a code rate distribution neural network; and when each decision time slot arrives, acquiring a state vector, inputting the state vector to the trained code rate distribution neural network to obtain a code rate distribution scheme, and determining the bandwidth occupied by each user according to the code rate distribution scheme and the current available bandwidth to finish bandwidth distribution.
The invention has the following technical effects:
1) compared with the traditional solution, the invention designs a central control DASH service scheme capable of acquiring global user information and network state information, and the scheme can know the global state to adjust the code rate selection of all users, so that the service quality obtained by each user is fair under the condition of ensuring that the video playing of all users is not blocked;
2) the invention solves the problem of code rate allocation by applying a deep learning technology, and trains a neural network in advance by using a previously obtained optimal solution example as a data set. In actual implementation, a trained neural network is utilized, and a code rate distribution result close to an optimal solution can be obtained only within millisecond time;
3) the invention designs a cross-layer optimized DASH service scheme, and simultaneously regulates and controls the video code rate of an application layer and the bandwidth of a network layer, so that the service quality of a user is greatly improved, the smooth playing of a video can be ensured, and the video code rate and the high service quality fairness as high as possible can be obtained.
Drawings
FIG. 1 is a flowchart of an embodiment of an intelligent video bitrate adjustment and bandwidth allocation method based on deep learning according to the present invention;
FIG. 2 is a block diagram of a rate allocation neural network in accordance with the present invention;
FIG. 3 is a diagram illustrating an application scenario of intelligent rate adjustment and bandwidth allocation in this embodiment;
fig. 4 is a schematic diagram of information interaction between the central controller and the user in the present embodiment.
Detailed Description
The following description of the embodiments of the present invention is provided in order to better understand the present invention for those skilled in the art with reference to the accompanying drawings. It is to be expressly noted that in the following description, a detailed description of known functions and designs will be omitted when it may obscure the subject matter of the present invention.
Examples
Fig. 1 is a flowchart of an embodiment of an intelligent video bitrate adjustment and bandwidth allocation method based on deep learning according to the present invention. As shown in fig. 1, the method for adjusting the code rate and allocating the bandwidth of the intelligent video based on the deep learning of the present invention comprises the following specific steps:
s101: acquiring system data:
the maximum number of users which can simultaneously request DASH service flows and are controlled by a bottleneck link controller accessed to a core network is recorded as N, M user selectable code rate gears are set according to actual conditions, and the code rate is higher when the sequence number of the code rate gears is larger.
S102: constructing and training a code rate distribution neural network:
in order to rapidly and accurately allocate the code rate, the invention needs to construct a code rate allocation neural network. FIG. 2 is a block diagram of a rate allocation neural network in the present invention. As shown in fig. 2, the rate allocation neural network in the present invention includes an input layer, a fully-connected hidden layer and an output layer, wherein the input layer includes 2N +1 neural units, and the input is 2N +1 dimensional state vector I ═ W, b1,s1,b2,s2,…,bN,sN) Where W represents the available bandwidth normalization value for bottleneck link controller access to the core network, bn、snRespectively representing the time length normalization value and the active state s of the cached video of the nth user n1 indicates that the user is active and needs to perform rate allocation, s n0 means that the user is not active and no rate allocation is needed, N is 1,2, …, N; the output layer comprises N independent output layers, each independent output layer comprises M +1 nerve units, and the output vector of the nth independent output layer is On＝(rn,0,rn,1,…,rn,M) Wherein r isn,mThe probability of selecting the mth gear code rate in the next time slot of the nth user is represented, and the value range is [0 ],1]M-0, 1, …, M-0 indicates no code rate is allocated. The number of the nerve units of the fully-connected hidden layer can be set according to actual needs.
Setting D user states and available bandwidth scenes, normalizing the user cache video time length and the available bandwidth according to the maximum value and the minimum value of the user cache video time length and the maximum value and the minimum value of the available bandwidth, and determining corresponding state vectors
Figure RE-GDA0003017361750000051
Wherein D is 1,2, …, D, WdA normalized value representing the available bandwidth for the bottleneck link controller to access the core network in scenario d,
Figure RE-GDA0003017361750000052
respectively representing a normalized value and an active state of the nth user cache video time length in the d scene; obtaining the code rate selected by each user in the next time slot in the scene according to theoretical analysis, and allocating the code rate to the scheme vector
Figure RE-GDA0003017361750000061
The corresponding element in the group is set to be 1, the other elements are set to be 0,
Figure RE-GDA0003017361750000062
indicating that the nth user selects the mth gear rate in the next time slot in the d-th scene,
Figure RE-GDA0003017361750000063
indicating that the nth user does not select the mth gear rate in the next time slot in the d-th scene.
State vector of each scene
Figure RE-GDA0003017361750000064
As input, a code rate allocation scheme vector for N users
Figure RE-GDA0003017361750000065
As the desired output, a code rate assignment neural network is trained.
S103: acquiring the current user state:
when each decision time slot comes, obtaining the number N' of DASH service users controlled by the current bottleneck link controller and the available bandwidth W of the bottleneck link controller accessed to the core network*The available bandwidth can directly grab the actual available bandwidth of the network end or call the available bandwidth estimator to acquire the actual available bandwidth. Normalizing the time length and the available bandwidth of the user cache video according to the maximum value and the minimum value of the time length of the user cache video and the maximum value and the minimum value of the available bandwidth to obtain a normalized value b of the time length of the user cache videon′And normalizing the value W by the available bandwidth, and simultaneously determining the activity state s of each usern′N 'is 1,2, …, N'. Constructing a current state vector I ═ (W, b)1,s1,b2,s2,…,bN,sN) When N is less than or equal to N', bn＝bn′，sn＝sn′Otherwise bn＝0，sn＝0。
For the active state of the user, the present embodiment provides a method for determining the active state of the user:
if the user leaves, the user DASH service flow request is completely completed, or the time length of the video cached by the user exceeds a decision time slot, which indicates that the user does not need bandwidth downloading and does not have video blocking in the next time slot, the states of the two types of users are set to be 0, which indicates that the user is not active, the next time slot does not need to allocate the code rate, the states of the other users are set to be 1, which indicates that the user is active, and the next time slot needs to allocate the code rate.
S104: determining a code rate allocation scheme:
the current state vector I obtained in step S103 is set to (W, b)1,s1,b2,s2,…,bN,sN) Inputting the code rate to the code rate distribution neural network trained in the step S102 to obtain a code rate distribution scheme vector O of each user in the next time slotn＝(rn,0,rn,1,…,rn,M) For the current N' users, selecting the code rate gear corresponding to the probability maximum value in the code rate allocation scheme vector as the code rate of the next time slot of the userGear Rn′， R n′0,1, …, M, resulting in a rate allocation scheme.
The rate allocation scheme obtained by the rate allocation neural network has certain probability of the following three conditions: the next time slot with the user does not need to allocate code rate, but is allocated with the code rate; allocating code rate to the next time slot of a user, but not allocating; the overall allocation code rate of all users is too high, so that the total bandwidth allocation is insufficient, and the users watch the video in the next time slot. In order to avoid these situations, feasibility check can be performed on the code rate allocation scheme, and the code rate allocation result is finely adjusted to meet the scene service requirement, and the specific method is as follows:
1) for the nth 'user of the current N' DASH service users, if the user activity state s isn′0, and a code rate step R assigned thereton′Not equal to 0, then let Rn′When the code rate is not allocated, the code rate is adjusted to be 0;
2) for the nth 'user of the current N' DASH service users, if the user activity state s isn′1, and the code rate gear position R allocated to itn′When the value is equal to 0, then let R n′1, namely adjusting to allocate the lowest gear code rate;
3) after the two adjustments are completed, the following formula is used to calculate the minimum total bandwidth W' required for ensuring that all the user videos are not blocked under the current rate allocation scheme:
Figure RE-GDA0003017361750000071
wherein r isn′Representing gear R according to code raten′The determined code rate, pi, represents a decision time slot for code rate adjustment and bandwidth allocation.
If the minimum total bandwidth W' is less than or equal to the current available bandwidth W*If not, the current available bandwidth cannot be ensured to be not blocked by each user video under the code rate allocation scheme, so that the video is not blocked by each user, and the video is not blocked by each user videoReducing the code rate of one random user in the users with the highest code rate by one gear, calculating the minimum total bandwidth W 'again, and repeating the processing process until W' is less than W*。
S105: and (3) bandwidth allocation:
and calculating the bandwidth allocated to each user according to the current available bandwidth and the code rates of the current N' users determined by the code rate allocation scheme, and completing bandwidth allocation.
In this embodiment, bandwidth allocation is performed based on a linear programming model, and the specific method is as follows:
according to the current available bandwidth and code rate allocation scheme, constructing the following linear programming model:
Figure RE-GDA0003017361750000072
subject to:
βn'-ωn'≤0
Figure RE-GDA0003017361750000073
wherein S isn′The buffer duration of the nth' user is represented as the contribution of the user service quality, and the calculation formula is as follows:
Figure RE-GDA0003017361750000081
ωn′indicates the bandwidth allocation result of the nth' user, rn′Code rate, beta, representing the nth usern′Indicating that the minimum bandwidth allocation is required to ensure that the nth' user video is not played.
And solving by using a basic linear programming solving method to obtain the bandwidth allocation result of each user.
And after the code rate distribution and the bandwidth distribution are finished, transmitting the code rate distribution result and the bandwidth distribution result to the user side, and performing the next round of decision when the next decision time slot comes.
The invention can be applied to a scenario where multiple DASH services compete for the same bottleneck link, and common scenarios include:
firstly, a home network: a plurality of family members in a family may watch videos by using terminals such as a personal computer, a smart phone, a tablet personal computer and a smart television, and due to the randomness of behaviors and different lengths of videos, the number of frequent requests changes, and at the moment, intelligent code rate selection and bandwidth allocation are particularly important, so that the scheme of the invention can be used;
II, ultra-dense cellular network: the intelligent devices of users in the same cell are connected with the core network through the base station to obtain the DASH video, and the high dynamic property of the number of the users and the difference of the characteristics of the video need reasonable resource allocation, so the scheme of the invention can be used.
In order to better illustrate the technical solution of the present invention, the present invention is illustrated by a specific example.
Fig. 3 is a schematic diagram of an application scenario of intelligent rate adjustment and bandwidth allocation in this embodiment. As shown in fig. 3, in this embodiment, the application scenario of intelligent rate adjustment and bandwidth allocation is a home network scenario, the bottleneck link controller is a central controller, and the maximum number N of users controlled by the central controller is 10, that is, the system can accommodate up to 10 people simultaneously requesting DASH service flows, and has a variable bandwidth W of 2.5Mbps to 10Mbps, and the selectable rate steps are 5: 350Kbps, 700Kbps, 1Mbps, 1.5Mbps, and 2.5 Mbps. Setting the decision time slot pi to 5 seconds, setting the maximum value of the video caching time length of the user for normalization to 5 seconds (namely, the normalization values of the video caching time length exceeding 5 seconds are all 1), and setting the minimum value to 0 second. The present invention can be implemented in a central controller, or can be implemented by using a separate decision module, and the present embodiment deploys a functional module for implementing the present invention in the central controller.
From the above information, it can be seen that the input layer of the rate allocation neural network in the present embodiment includes 21 neural units, and the input is 21-dimensional state vector I ═ (W, b)1,s1,b2,s2,…,b10,s10) The output layer comprises 10 independent output layers, each independent output layer comprises 6 nervesA unit, the output vector of the nth independent output layer is On＝(rn,0,rn,1,…,rn,5). The fully-connected hidden layer comprises 2 layers, the width of each layer is 128, the hidden layer activation function is ReLU (rectified Linear Unit), the network update uses RMSprop (root Mean Square Prop) optimizer, and the learning rate is 1 multiplied by 10-4The sum of the cross-entropies of all output layer data and label data is used as a loss function, and the L2 norm is added to prevent overfitting. Setting a plurality of user states and available bandwidth scenes to obtain corresponding state vectors, solving an optimal solution of code rate allocation by using mixed integer programming to obtain code rate allocation scheme vectors serving as marking data, and training a code rate allocation neural network.
When each decision time slot comes, the central controller obtains the number of the currently controlled users and the available bandwidth of the central controller accessing the core network. Assuming that the number of users currently requesting DASH traffic flow is 5, the length of the state vector is 1+2 × 5 — 11, and then 0 is added to make up the 21-dimensional state vector. And inputting the state vector into a trained code rate distribution neural network to obtain a code rate distribution scheme. And then carrying out feasibility check on the code rate distribution scheme, and adjusting to obtain a final code rate distribution scheme. And then obtaining a bandwidth allocation result based on the linear programming model of the embodiment. And the central controller transmits the code rate distribution scheme and the bandwidth distribution result to each user and network equipment.
Fig. 4 is a schematic diagram of information interaction between the central controller and the user in the present embodiment. As shown in fig. 4, the interaction of the central controller and the user client can be summarized as the following steps:
step 1: the central controller firstly sends an information collection request to all access user terminals which provide DASH service flow requests so as to obtain the video cache state and the active state of the user at the moment;
step 2: after receiving an information collection request of a central controller, each user client requesting a DASH service flow transmits information such as the caching duration of a currently played video, whether the video is cached completely and the like to the central controller;
and step 3: after the central controller collects the user information, the central controller captures the actual available total bandwidth at the gateway, and then the code rate and bandwidth allocation in the next decision time slot of each user are obtained by adopting the quick decision of the invention according to the collected information;
and 4, step 4: the central controller distributes the code rate in the decided user next decision time slot to the corresponding user client and writes the bandwidth distribution result into the gateway controller;
and 5: the user client receives the code rate selection scheme issued by the central controller, applies the code rate, waits for receiving the video and sends confirmation information to the central controller;
step 6: and (4) the central controller receives the user confirmation information, waits for the next decision time slot and repeats the operation from the step 1.
Although illustrative embodiments of the present invention have been described above to facilitate the understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, and various changes may be made apparent to those skilled in the art as long as they are within the spirit and scope of the present invention as defined and defined by the appended claims, and all matters of the invention which utilize the inventive concepts are protected.

Claims (4)

1. An intelligent video code rate adjustment and bandwidth allocation method based on deep learning is characterized by comprising the following steps:
s1: recording the maximum number of users which can simultaneously request DASH service flow and are controlled by a bottleneck link controller accessed to a core network as N, setting M user selectable code rate gears according to actual conditions, wherein the code rate is higher when the serial number of the code rate gear is larger;
s2: constructing a code rate distribution neural network, comprising an input layer, a full-connection hidden layer and an output layer, wherein the input layer comprises 2N +1 neural units, and the input is a 2N +1 dimensional state vector I ═ W, b1,s1,b2,s2,…,bN,sN) Wherein W represents a normalized value of available bandwidth for access of the bottleneck link controller to the core network,bn、snrespectively representing the time length normalization value and the active state s of the cached video of the nth usern1 indicates that the user is active and needs to perform rate allocation, sn0 means that the user is not active and no rate allocation is needed, N is 1,2, …, N; the output layer comprises N independent output layers, each independent output layer comprises M +1 nerve units, and the code rate allocation scheme vector output by the nth independent output layer is On＝(rn,0,rn,1,…,rn,M) Wherein r isn,mThe probability of selecting the mth gear code rate in the next time slot of the nth user is represented, and the value range is [0,1 ]]M-0, 1, …, M-0 indicates no code rate is allocated;
setting D user states and available bandwidth scenes, normalizing the user cache video time length and the available bandwidth according to the maximum value and the minimum value of the user cache video time length and the maximum value and the minimum value of the available bandwidth, and determining corresponding state vectors
Figure FDA0002914616280000011
Wherein D is 1,2, …, D, WdA normalized value representing the available bandwidth for the bottleneck link controller to access the core network in scenario d,
Figure FDA0002914616280000012
respectively representing a normalized value and an active state of the nth user cache video time length in the d scene; obtaining the code rate selected by each user in the next time slot in the scene according to theoretical analysis, and allocating the code rate to the scheme vector
Figure FDA0002914616280000013
The corresponding element in the group is set to be 1, the other elements are set to be 0,
Figure FDA0002914616280000014
indicating that the nth user selects the mth gear rate in the next time slot in the d-th scene,
Figure FDA0002914616280000015
indicating that the nth user does not select the mth gear code rate in the next time slot in the d scenes;
state vector of each scene
Figure FDA0002914616280000016
As input, a code rate allocation scheme vector for N users
Figure FDA0002914616280000017
As expected output, training a code rate distribution neural network;
s3: when each decision time slot comes, obtaining the number N' of DASH service users controlled by the current bottleneck link controller and the available bandwidth W of the bottleneck link controller accessed to the core network*Normalizing the time length and the available bandwidth of the user cache video according to the maximum value and the minimum value of the time length of the user cache video and the maximum value and the minimum value of the available bandwidth to obtain a normalized value b of the time length of the user cache videon′And normalizing the value W by the available bandwidth, and simultaneously determining the activity state s of each usern′N '═ 1,2, …, N'; constructing a current state vector I ═ (W, b)1,s1,b2,s2,…,bN,sN) When N is less than or equal to N', bn＝bn′，sn＝sn′Otherwise bn＝0，sn＝0；
S4: setting the current state vector I obtained in step S3 to (W, b)1,s1,b2,s2,…,bN,sN) Inputting the code rate distribution neural network trained in the step S2 to obtain a code rate distribution scheme vector O of each user in the next time slotn＝(rn,0,rn,1,…,rn,M) For the current N' users, selecting the code rate gear corresponding to the probability maximum value in the code rate distribution scheme vector as the code rate gear R of the next time slot of the usern′，Rn′Obtaining a code rate allocation scheme, wherein the code rate allocation scheme is 0,1, … and M;
s5: and calculating the bandwidth allocated to each user according to the current available bandwidth and the code rates of the current N' users determined by the code rate allocation scheme, and completing bandwidth allocation.
2. The intelligent video rate adjustment and bandwidth allocation method according to claim 1, wherein the user active status in step S3 is determined by the following method:
if the user leaves, the user DASH service flow request is completely completed, or the video caching duration of the user exceeds a decision time slot, which indicates that the video jamming can not occur in the next time slot of the user without bandwidth downloading, the active states of the two types of users are set to be 0, and the active states of other users are set to be 1.
3. The intelligent video bitrate adjustment and bandwidth allocation method according to claim 1, wherein the step S4 further includes performing feasibility check on the bitrate allocation scheme, and the specific method is as follows:
1) for the nth 'user of the current N' DASH service users, if the user activity state s isn′0, and a code rate step R assigned thereton′Not equal to 0, then let Rn′When the code rate is not allocated, the code rate is adjusted to be 0;
2) for the nth 'user of the current N' DASH service users, if the user activity state s isn′1, and the code rate gear position R allocated to itn′When the value is equal to 0, then let Rn′1, namely adjusting to allocate the lowest gear code rate;
3) after the two adjustments are completed, the following formula is used to calculate the minimum total bandwidth W' required for ensuring that all the user videos are not played in a pause mode under the current code rate allocation scheme:
Figure FDA0002914616280000021
wherein r isn′Representing gear R according to code raten′The determined code rate, pi, represents the decision time slot of code rate adjustment and bandwidth allocation;
if it is minimumThe total bandwidth W' is less than or equal to the current available bandwidth W*If not, reducing the code rate of one random user with the highest code rate by one gear, calculating the minimum total bandwidth W again, and repeating the above processing until W' is less than or equal to W*。
4. The intelligent video rate adjustment and bandwidth allocation method according to claim 1, wherein the bandwidth allocation in step S5 is performed based on a linear programming model, and the specific method is as follows:
according to the current available bandwidth and code rate allocation scheme, constructing the following linear programming model:
Figure FDA0002914616280000031
subject to:
βn'-ωn'≤0
Figure FDA0002914616280000032
wherein S isn′The buffer duration of the nth' user is represented as the contribution of the user service quality, and the calculation formula is as follows:
Figure FDA0002914616280000033
ωn′indicates the bandwidth allocation result of the nth' user, rn′Code rate, beta, representing the nth usern′Indicating that the minimum bandwidth allocation is required to ensure that the nth' user video is not played.
And solving by using a basic linear programming solving method to obtain the bandwidth allocation result of each user.
CN202110097764.7A 2021-01-25 2021-01-25 Intelligent video code rate adjustment and bandwidth allocation method based on deep learning Expired - Fee Related CN112911408B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110097764.7A CN112911408B (en) 2021-01-25 2021-01-25 Intelligent video code rate adjustment and bandwidth allocation method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110097764.7A CN112911408B (en) 2021-01-25 2021-01-25 Intelligent video code rate adjustment and bandwidth allocation method based on deep learning

Publications (2)

Publication Number Publication Date
CN112911408A true CN112911408A (en) 2021-06-04
CN112911408B CN112911408B (en) 2022-03-25

Family

ID=76119528

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110097764.7A Expired - Fee Related CN112911408B (en) 2021-01-25 2021-01-25 Intelligent video code rate adjustment and bandwidth allocation method based on deep learning

Country Status (1)

Country Link
CN (1) CN112911408B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114039916A (en) * 2021-10-21 2022-02-11 北京邮电大学 Deep hybrid model flow control method, device and storage medium for real-time video quality optimization
CN114040257A (en) * 2021-11-26 2022-02-11 深圳大学 A kind of adaptive video streaming transmission and playback method, device, equipment and storage medium
CN116864153A (en) * 2023-07-13 2023-10-10 中世康恺科技有限公司 Regional remote diagnosis system
CN119729056A (en) * 2024-12-20 2025-03-28 中国科学技术大学 Operator bandwidth allocation method based on network video traffic QoE identification

Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110090922A1 (en) * 2008-06-30 2011-04-21 Qi Wang Method, apparatus, and system for adjusting multimedia encoding rate
CN104159110A (en) * 2014-04-02 2014-11-19 上海大学 Video transmission bandwidth optimal allocation method based on Nash Bargaining Theory
US20150104062A1 (en) * 2013-10-14 2015-04-16 National Taipei University Of Technology Probabilistic neural network based moving object detection method and an apparatus using the same
US20170026713A1 (en) * 2015-03-26 2017-01-26 Carnegie Mellon University System and Method for Dynamic Adaptive Video Streaming Using Model Predictive Control
CN107371028A (en) * 2017-08-22 2017-11-21 南京惟初信息科技有限公司 A kind of high-quality video coding method for adapting to bandwidth
CN108063961A (en) * 2017-12-22 2018-05-22 北京联合网视文化传播有限公司 A kind of self-adaption code rate video transmission method and system based on intensified learning
US20180262439A1 (en) * 2017-03-08 2018-09-13 Arris Enterprises Llc Excess bitrate distribution based on quality gain in sabr server
CN108833350A (en) * 2018-05-16 2018-11-16 武汉烽火凯卓科技有限公司 A kind of data transmission method suitable for multiserver adaptive stream media system
CN108965949A (en) * 2018-07-27 2018-12-07 清华大学 Meet the code rate adaptive approach of user individual experience in a kind of video traffic
CN109982118A (en) * 2019-03-27 2019-07-05 北京奇艺世纪科技有限公司 A kind of video code rate self-adapting regulation method, device and electronic equipment
US20200099733A1 (en) * 2018-09-26 2020-03-26 Vmware, Inc. System and method for widescale adaptive bitrate selection
CN111031387A (en) * 2019-11-21 2020-04-17 南京大学 A method for controlling the video encoding flow rate of a monitoring video sending end
CN111447511A (en) * 2020-03-09 2020-07-24 重庆邮电大学 A Bandwidth Allocation Method with User Perceived Quality of Experience
CN111669617A (en) * 2020-04-07 2020-09-15 鹏城实验室 A transmission method of live video stream based on intelligent edge
CN111901642A (en) * 2020-07-31 2020-11-06 成都云格致力科技有限公司 Real-time video bit rate adaptive control method and system based on reinforcement learning

Patent Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110090922A1 (en) * 2008-06-30 2011-04-21 Qi Wang Method, apparatus, and system for adjusting multimedia encoding rate
US20150104062A1 (en) * 2013-10-14 2015-04-16 National Taipei University Of Technology Probabilistic neural network based moving object detection method and an apparatus using the same
CN104159110A (en) * 2014-04-02 2014-11-19 上海大学 Video transmission bandwidth optimal allocation method based on Nash Bargaining Theory
US20170026713A1 (en) * 2015-03-26 2017-01-26 Carnegie Mellon University System and Method for Dynamic Adaptive Video Streaming Using Model Predictive Control
US20180262439A1 (en) * 2017-03-08 2018-09-13 Arris Enterprises Llc Excess bitrate distribution based on quality gain in sabr server
CN107371028A (en) * 2017-08-22 2017-11-21 南京惟初信息科技有限公司 A kind of high-quality video coding method for adapting to bandwidth
CN108063961A (en) * 2017-12-22 2018-05-22 北京联合网视文化传播有限公司 A kind of self-adaption code rate video transmission method and system based on intensified learning
CN108833350A (en) * 2018-05-16 2018-11-16 武汉烽火凯卓科技有限公司 A kind of data transmission method suitable for multiserver adaptive stream media system
CN108965949A (en) * 2018-07-27 2018-12-07 清华大学 Meet the code rate adaptive approach of user individual experience in a kind of video traffic
US20200099733A1 (en) * 2018-09-26 2020-03-26 Vmware, Inc. System and method for widescale adaptive bitrate selection
CN109982118A (en) * 2019-03-27 2019-07-05 北京奇艺世纪科技有限公司 A kind of video code rate self-adapting regulation method, device and electronic equipment
CN111031387A (en) * 2019-11-21 2020-04-17 南京大学 A method for controlling the video encoding flow rate of a monitoring video sending end
CN111447511A (en) * 2020-03-09 2020-07-24 重庆邮电大学 A Bandwidth Allocation Method with User Perceived Quality of Experience
CN111669617A (en) * 2020-04-07 2020-09-15 鹏城实验室 A transmission method of live video stream based on intelligent edge
CN111901642A (en) * 2020-07-31 2020-11-06 成都云格致力科技有限公司 Real-time video bit rate adaptive control method and system based on reinforcement learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
薄磊: "《基于机器学习的图像视频自适应高效传输》", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114039916A (en) * 2021-10-21 2022-02-11 北京邮电大学 Deep hybrid model flow control method, device and storage medium for real-time video quality optimization
CN114040257A (en) * 2021-11-26 2022-02-11 深圳大学 A kind of adaptive video streaming transmission and playback method, device, equipment and storage medium
CN114040257B (en) * 2021-11-26 2023-06-13 深圳大学 Self-adaptive video stream transmission playing method, device, equipment and storage medium
CN116864153A (en) * 2023-07-13 2023-10-10 中世康恺科技有限公司 Regional remote diagnosis system
CN116864153B (en) * 2023-07-13 2024-04-26 中世康恺科技有限公司 Regional remote diagnosis system
CN119729056A (en) * 2024-12-20 2025-03-28 中国科学技术大学 Operator bandwidth allocation method based on network video traffic QoE identification

Also Published As

Publication number Publication date
CN112911408B (en) 2022-03-25

Similar Documents

Publication Publication Date Title
CN112911408B (en) Intelligent video code rate adjustment and bandwidth allocation method based on deep learning
Fu et al. Soft actor–critic DRL for live transcoding and streaming in vehicular fog-computing-enabled IoV
Zhang et al. Deep learning for wireless coded caching with unknown and time-variant content popularity
WO2021123139A1 (en) Systems and methods for enhanced feedback for cascaded federated machine learning
CN110049357B (en) A kind of bandwidth estimation method, apparatus, equipment and storage medium
CN118175356A (en) Video transmission method, device, equipment and storage medium
Hou et al. Admission control and scheduling for QoS guarantees for variable-bit-rate applications on wireless channels
CN119211585B (en) Private domain live broadcast content distribution and visitor interaction method and system based on edge calculation
CN114518956A (en) Load balancing method and device based on queue overhead under edge streaming media
CN114885422A (en) Dynamic edge computing unloading method based on hybrid access mode in ultra-dense network
CN108833995B (en) Transmission method of self-adaptive streaming media in wireless network environment
CN112153702B (en) Local area network bandwidth resource allocation method, storage device and equipment
CN116248603B (en) Traffic scheduling method, device, equipment and storage medium
Mu et al. AMIS: Edge computing based adaptive mobile video streaming
CN115314944A (en) A collaborative caching method for Internet of Vehicles based on perception of social relations of mobile vehicles
CN104918077A (en) Video transmission method, device and system
CN110913239A (en) A refined video cache update method for mobile edge computing
CN117793803A (en) Multi-user wireless channel resource allocation optimization method based on deep reinforcement learning
Kazemian et al. Neuro-Fuzzy approach to video transmission over ZigBee
CN114945097A (en) Video stream processing method and device
Yuan et al. MUABR: Multi-user adaptive bitrate algorithm based multi-agent deep reinforcement learning
KR102709561B1 (en) Scalable video coding apparatus and method for real-time video streaming
CN115460401B (en) A method for optimizing video bit rate allocation
CN117560696A (en) A network slice selection method for in-vehicle network based on experience replay and deep Q learning
CN114091649B (en) A data processing method, device and equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220325