Network resource dynamic allocation method based on network slices

Abstract

The invention relates to the technical field of network resource dynamic allocation, and discloses a network resource dynamic allocation method based on network slices, which comprises the following steps of: determining the size and remaining time of a data packet to be sent, constructing an autoregression model by utilizing a flow autocorrelation coefficient, and predicting each slice stream in a prediction window length in combination with Kalman filtering, obtaining a mean value and a variance of each slice stream; establishing opportunity constraints and a target function based on the mean value and the variance of each slice stream in combination with the failure probability upper limit; on the basis of the opportunity constraint and the objective function, calculating a resource allocation proportion of each slice stream in combination with the minimum resource reservation proportion; and based on the resource allocation proportion of each slice stream, in combination with the initial contention window value, the transmission time upper limit, the number of resource units and the number of prior antennas, respectively mapping the resource allocation proportion of each slice stream into contention window value transmission time and resource units of each slice stream.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic allocation of network resources, and more specifically, to a method for dynamic allocation of network resources based on network slicing. Background Art

[0002] In a home local area network environment, game streaming services (such as SteamLink) impose bidirectional hard real-time closed-loop constraints on network transmission:

[0003] Downlink: One frame of 1080p / 60fps high-bitrate video must be delivered every 16.67ms, otherwise the picture will be dropped. Uplink: Controller button feedback must be delivered once every 20ms, otherwise the player will perceive delay.

[0004] The problem with game streaming services lies in the strong coupling of traffic burstiness and latency sensitivity, the burst traffic of I frames in video streams and the instantaneous pulse transmission requirements of control signals. Existing technologies mainly rely on average bandwidth allocation, static priority scheduling, or simple traffic shaping strategies. For example, the 802.11 EDCA mechanism prioritizes different services through fixed contention windows and transmission times, but such static strategies can only achieve long-term resource smoothing and cannot perceive the remaining available time slots within each scheduling cycle. When there are concurrent services such as arbitrary downloads and video conferencing in a home local area network environment that preempt the channel, the lack of dynamic response capabilities to real-time traffic fluctuations will cause the probability of frame expiration failure of game streaming to increase sharply, and the end-to-end delay of control signals will be significant, seriously affecting the user experience.

[0005] The existing technology has the following problems:

[0006] 1. Traffic characteristics are not fully utilized. The mean-variance statistical characteristics of video streams and control signals are not combined, making it impossible to reserve resources through forward-looking predictions, resulting in queue congestion during bursts.

[0007] 2. Decoupling of physical layer parameters. Network slicing technology mostly stays at the logical layer isolation, without deep linkage with local network physical layer parameters (such as OFDMA resource unit RU and contention window CW). Resource allocation instructions are difficult to convert into hardware-executable scheduling strategies.

[0008] 3. There is an imbalance between fairness and real-time performance. Static priority can easily lead to high-priority services monopolizing resources for a long time (such as video streams occupying all resource units (RUs)). Alternatively, low-priority services cannot receive timely services due to fixed ratio allocation, making it impossible to adapt to the dynamic changes in service load in game streaming scenarios. Summary of the Invention

[0009] The present invention provides a method for dynamic allocation of network resources based on network slicing, which solves the technical problems raised in the background technology.

[0010] The present invention provides a network resource dynamic allocation method based on network slicing, which is applied to resource allocation of video slice streams, instruction slice streams, and background slice streams, including:

[0011] Step 1: Initialize the following parameters:

[0012] Current period t, prediction window of length G, upper limit of failure probability δ vid , δ ctl , δ be , initial contention window value Transmission time limit TXOP max , the number of resource units RU, the minimum resource reservation ratio κ, the traffic autocorrelation coefficient ρ and the number of antennas M;

[0013] Step 2: Determine the size and remaining time of the data packet to be sent, use the traffic autocorrelation coefficient ρ to build an autoregressive model and combine it with Kalman filtering to predict each slice flow within the prediction window length H, and obtain the mean μ of each slice flow vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 ;

[0014] Step 3: Based on the mean and variance of each slice flow, combined with the upper limit of failure probability δ vid , δ ctl , δ be Establish chance constraints and objective functions;

[0015] Step 4: Based on the opportunity constraint and the objective function, combined with the minimum resource reservation ratio κ, calculate the resource allocation ratio x of each slice flow vid 、x ctl 、x be ;

[0016] Step 5: Based on the resource allocation ratio of each slice flow and combined with the initial contention window value Transmission time limit TXOP max , the number of resource units RU and the a priori number of antennas M, and map the resource allocation ratio of each slice flow to the contention window value CW of each slice flow vid 、CW ctl 、CW be TXOP vid TXOP ctl TXOP be and resource units RU vid , RUctl , RU be ;

[0017] Step 6: Allocate resources for the video slice stream, the instruction slice stream, and the background slice stream according to the contention window value, transmission time, and resource unit of each slice stream.

[0018] Furthermore, the size and remaining time of the data packet to be sent are determined, and the autoregressive model is constructed using the traffic autocorrelation coefficient ρ and combined with Kalman filtering to predict each slice flow within the prediction window length H, and the mean μ of each slice flow is obtained. vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 ,as follows:

[0019] In the current cycle t, the size and remaining time of the data packets to be sent are collected, and the observed load of each slice flow is calculated respectively;

[0020] The instantaneous load and the flow autocorrelation coefficient ρ are combined to construct a first-order autoregressive model to make a priori prediction of the load in period T+1 and obtain the prior load.

[0021] The Kalman filter is used to fuse the prior load and the observed load, and the mean μ of each slice flow is output within the prediction window. vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 .

[0022] Furthermore, based on the mean and variance of each slice flow, combined with the upper limit of failure probability δ vid , δ ctl , δ be Establish the chance constraints and objective function, including:

[0023] The upper limit of failure probability δ vid , δ ctl , δ be The upper limit B of the number of bits transmitted per cycle max Combine to form a confidence bound;

[0024] The mean and variance of each slice flow and the required resource allocation ratio of each slice flow are used to construct a traffic risk function as the first optimization item;

[0025] Based on the resource allocation ratio to be requested for each slice flow, a nonlinear penalty is applied to the resource allocation ratio of each slice flow to construct an entropy regularization function as the second optimization item;

[0026] The confidence bound is used as a chance constraint, and the first optimization term and the second optimization term are combined to form the objective function.

[0027] Furthermore, based on the opportunity constraint and the objective function, combined with the minimum resource reservation ratio κ, the resource allocation ratio x of each slice flow is calculated. vid 、x ctl 、x be ,include:

[0028] Based on the combination of opportunity constraints and objective function, and combined with the minimum resource reservation ratio κ, the resource allocation ratio x is constructed respectively. vid 、x ctl 、x be The lower limit of the feasible region;

[0029] Based on the first-order derivative and second-order derivative of the objective function, the resource allocation ratio of each slice flow is determined within the lower limit of the feasible region through the interior point Newton iteration algorithm.

[0030] Furthermore, the contention window value CW of each slice flow vid 、CW ctl 、CW be ,include:

[0031] By dividing the resource allocation ratio of each slice flow with the initial contention window value Combined with the a priori number of antennas M, the contention window value of each slice flow is calculated as follows:

[0032]

[0033] Where, i∈{vid,ctl,be}, CW i represents the contention window value of the i-th slice flow, x i Indicates the resource allocation ratio of the i-th slice flow, Indicates rounding up to even integers.

[0034] Furthermore, the transmission time TXOP of each slice flow vid TXOP ctl TXOP be ,include:

[0035] Allocate the resource ratio of each slice flow to x vid 、x ctl 、x be Transmission time upper limit TXOP max Multiply them together to get the transmission time TXOP of each slice flow vidTXOP ctl TXOP be .

[0036] Furthermore, the resource unit RU of each slice stream vid , RU ctl , RU be ,include:

[0037] Allocate the resource ratio of each slice flow to x vid 、x ctl 、x be Multiply by the number of resource units RU and round down to get the resource unit RU of each slice stream vid , RU ctl , RU be .

[0038] Furthermore, resource allocation of the video slice stream, the instruction slice stream, and the background slice stream is performed according to the contention window value, the transmission time, and the resource unit of each slice stream, including:

[0039] The contention window value, transmission time and resource unit of each slice stream are written into the corresponding media access control layer register respectively, a joint scheduling table for the video slice stream, instruction slice stream and background slice stream is constructed, and resource allocation is performed.

[0040] The beneficial effects of the present invention lie in: by introducing the traffic prediction results and failure probability upper bounds of different service slices into the optimization model, a resource allocation mechanism with statistical confidence guarantee is constructed. Combined with the dual-objective optimization method of risk term and entropy regularization term, a resource allocation ratio that satisfies the constraints is dynamically generated, thereby achieving dynamic allocation of network resources. Compared with traditional scheduling methods based solely on average rate or static weights, the present invention can effectively prevent frame loss and control lag in high-interference, high-concurrency home local area network environments, significantly improving the interactive fluency and screen coherence of users in cross-device games. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flowchart of a method for dynamic allocation of network resources based on network slicing of the present invention. DETAILED DESCRIPTION

[0042] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0043] like Figure 1 As shown, a network resource dynamic allocation method based on network slicing is applied to resource allocation of video slice streams, instruction slice streams, and background slice streams, including:

[0044] It should be noted that game streaming involves mixed transmission of multiple services. For example, when running game streaming, remote control, and file downloads simultaneously on home WiFi, network traffic is divided into three types of slice flows with independent functions and differentiated requirements:

[0045] Video Ingestion (VID) streaming; Services such as HD game streaming (such as SteamLink) and video conferencing require stable bandwidth and low packet loss, but tolerate a certain amount of latency. Traffic characteristics include I-frame bursts (peak bitrate 3–5 times the average bitrate), high mean I-frame traffic, and large variance.

[0046] Command Slice Stream (CTL); Service type: Remote control commands (game controller input), requiring ultra-low latency and ultra-high reliability, with small data packets but requiring instant response. Traffic characteristics: Low average traffic (pulse-like), latency-sensitive, and prioritized channel resources.

[0047] Background Slice Flow (BE); Service Type: Non-real-time services such as file downloads and device firmware upgrades require basic resource guarantees (minimum resource reservation ratio κ), are allowed lower priority, and can be transmitted within idle resources. Traffic Characteristics: Traffic has a medium mean, large variance, and is insensitive to latency.

[0048] Step 1: Initialize the following parameters:

[0049] Current period t, prediction window of length G, upper limit of failure probability δ vid , δ ctl , δ be , initial contention window value Transmission time limit TXOP max , the number of resource units RU, the minimum resource reservation ratio κ, the traffic autocorrelation coefficient ρ and the number of antennas M;

[0050] In detail, the parameters are explained as follows:

[0051] The prediction window length is G, which is set by the system designer according to business requirements. G=nH, where n is an integer and H represents the cycle length, which is preferably 10ms.

[0052] Upper limit of failure probability δ vid , δ ctl , δ be , set by the system designer according to business requirements, the upper limit of the failure probability δ vid , δ ctl , δbe They represent the maximum failure (such as delay exceeding limit) probability allowed for “video slice stream VID”, “control slice stream CTL” and “background slice stream BE” in one cycle. For example, δ vid =0.5% means that the video service allows 0.5% of the frames to exceed the delay limit within the prediction window.

[0053] Initial contention window value Usually taken from the 802.11EDCA standard, the preferred default value is 15, the initial contention window value Indicates the minimum contention window size (in timeslots) used by each slice flow when initially sending a packet under the EDCA mechanism.

[0054] Transmission time limit TXOP max , as specified by the WiFi standard for home LAN, the MaxTXOP parameter can be queried in the router firmware, and the upper limit of the transmission time TXOP max Indicates the maximum continuous transmission time (in milliseconds) allowed for a slice flow after it obtains a channel in the EDCA mechanism. If this time is exceeded, the channel must be released to allow other slice flows to have a chance to compete.

[0055] The resource unit number (RU) indicates the total number of minimum resource units available in a frame of OFDMA physical resources. For example, nine 106-tone RUs can be allocated in an 80 MHz bandwidth, with each RU corresponding to a subcarrier cluster. The number of RUs available in the current bandwidth mode can be queried in the router firmware. This can be obtained through hostapd or the WiFi / 6 / SDK interface.

[0056] The minimum resource reservation ratio κ is set by the system designer according to business requirements. The minimum resource reservation ratio κ represents the lower bound of the ratio that each slice flow must at least retain in the number of resource units RU.

[0057] The flow autocorrelation coefficient ρ represents the degree of autocorrelation between each slice flow in consecutive cycles and is used to capture the inertial characteristics of flow changes in the autoregressive model. Its value ranges from 0 to 1, with values ​​closer to 1 indicating a stronger dependence of the current cycle t on the previous cycle.

[0058] The flow autocorrelation coefficient ρ is obtained as follows:

[0059] For example, statistical analysis can be performed using historical traffic logs, including summarizing the observed load of video slice streams over multiple consecutive periods. The observed load is obtained by:

[0060] Data collection: The system divides all packets to be sent into several slices i∈{vid,ctl,be}. Each packet belongs to a slice and has the following properties:

[0061] b i,j represents the size of the jth packet belonging to slice i (in bits);

[0062] represents the arrival time of the jth packet belonging to slice i (in ms);

[0063] Indicates the time (in ms) when the jth packet belonging to slice i is sent.

[0064] At the end of the t-th period (t+1)H, the set of all packets that have not been sent in this period for slice i is recorded as:

[0065]

[0066] Set P i,t Contains the jth packet that has not been sent at time (t+1)H.

[0067] Data processing: For slice i, the observed load L at time (t+1)H i,t for:

[0068]

[0069] According to the first-order autoregressive model formula L i,t =ρ×L i,t-1 +ε t ; Use the least squares method to fit and estimate the flow autocorrelation coefficient ρ; where L i,t represents the observed load of video slice i in the tth period, ε t represents the random error term in the tth period, L i,t-1 represents the observed load of video slice i in the t1th period;

[0070] The number of antennas M indicates the number of concurrent antennas that the wireless device can use for game streaming services. The number of concurrent antennas of the current device can be queried in the router firmware.

[0071] Step 2: Determine the size and remaining time of the data packet to be sent, use the traffic autocorrelation coefficient ρ to build an autoregressive model and combine it with Kalman filtering to predict each slice flow within the prediction window length H, and obtain the mean μ of each slice flow vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 ;

[0072] In one embodiment of the present invention, the size and remaining time of the data packet to be sent are determined, and an autoregressive model is constructed using the traffic autocorrelation coefficient ρ and combined with Kalman filtering to predict each slice flow within the prediction window length H to obtain the mean μ of each slice flow. vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 ,as follows:

[0073] In the current cycle t, the size and remaining time of the data packets to be sent are collected, and the observed load of each slice flow is calculated respectively;

[0074] The instantaneous load and the flow autocorrelation coefficient ρ are combined to construct a first-order autoregressive model to make a priori prediction of the load in period T+1 and obtain the prior load.

[0075] The Kalman filter is used to fuse the prior load and the observed load, and the mean μ of each slice flow is output within the prediction window. vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 .

[0076] In detail, the traffic in the game streaming scenario has the following characteristics:

[0077] Video frame size and control command data fluctuate significantly over time, but maintain high consistency. Based on the autocorrelation coefficient between historical load and traffic, traffic flows are somewhat predictable in the short term.

[0078] In order to ensure that the probability of triggering excessive non-delivery at a certain time in the future is controllable, it is necessary to accurately estimate the mean + variance so that the resource allocation ratio that meets the upper limit of the failure probability of each flow can be obtained in the subsequent opportunity constraints.

[0079] Since the autoregressive model itself has limited ability to track sudden noise, it is necessary to use the a priori and posterior mechanism of the Kalman filter to fuse the actual observed load just after the end of each basic cycle with the a priori predicted value, correct the error and obtain a more reliable posterior estimate.

[0080] In one embodiment of the present invention, step 2 is specifically as follows:

[0081] Step 211: In the current cycle t (length H), scan the data packets that have not been sent one by one and record the data packet size b i,j and the remaining time slot s from the current time to the deadline of the packet i,j ;

[0082] In a basic cycle, it is necessary to clearly list the data packets that have not been sent by each slice flow, as well as their sizes and deadlines, so that the observed load can be obtained by summing up the sizes of the unsent packets. i,j Less than or equal to 0, it can be regarded as an overdue package with a past deadline; if s i,j If it is less than H but greater than 0, it means that the data packet needs to be sent in the current or next cycle and must be included in the observed load calculation. i,j With s i,j , in order to filter out all the packets that need to be sent in the next cycle and calculate the observed load in that cycle.

[0083] In step 212, the deadline of each slice stream is defined as:

[0084] For a frame in a video slice VID, the deadline is fixed at 16.67ms after the Hth millisecond from the start of the current cycle;

[0085] For an instruction in the control slice CTL, the deadline is fixed at 20ms after the start of this cycle;

[0086] Background slice BE does not set a deadline to ensure that the background slice flow does not pollute the real-time prediction indicators. It is only served when there are remaining resources after the video and control streams are optimized.

[0087] Step 213, calculate the observed load L i,t :

[0088]

[0089] By comparing the arrival time with the deadline, the data packets that must be delivered before the end of the current cycle are filtered out and their sizes are summed up to form the observed load L i,t Observation load L i,t It can immediately reflect the amount of data that has been accumulated but not delivered at the end of the business cycle.

[0090] Step 221, using the observed load sequence {L i,t}, t = 0, 1, 2, ..., and the traffic autocorrelation coefficient ρ, the load of period t + 1 is estimated a priori as follows:

[0091]

[0092] in, represents the a priori load of period t+1;

[0093] Through the first-order autoregressive model, the observed load of the current period t is mapped to the preliminary forecast of period t+1, which provides a reasonable reference value at the beginning of period t+1;

[0094] Step 222, after the prior prediction is obtained After entering the cycle t+1, the observed load L is calculated again immediately i,t+1 ;

[0095] Step 223: Use Kalman filtering to convert the prior load and observed load L i,t+1 Perform fusion to obtain the posterior load estimate of period t+1 And the corresponding estimated variance P i,t+1|t+1 ;

[0096] Kalman filter fusion includes:

[0097] Prior stage:

[0098]

[0099] P i,t+1|t =ρ 2 P i,t|t +Q

[0100] Kalman gain calculation:

[0101]

[0102] Posterior update:

[0103]

[0104] P i,t+1|t+1 =(1-K i,t+1 )P i,t+1|t

[0105] in, represents the prior load of slice i for period t+1, L i,t+1 represents the observed load of slice i in period t+1, P i,t|t represents the posterior covariance of slice i in the current cycle t, Q represents the process noise covariance, which is an empirical value, R represents the observation noise covariance, which is a measured value, and K i,t+1 represents the Kalman gain, represents the posterior load estimate of slice i at period t+1, P i,t+1|t+1 represents the posterior covariance of slice i at period t+1.

[0106] P i,t+1|t By ρ2 P i,t|t +Q gives the prior error covariance, which helps the Kalman filter calculate the Kalman gain K in the posterior update phase i,t+1 , to balance the weight distribution of prior load and observed load;

[0107] Step 231, set a sliding time window of the same length G;

[0108] Step 232: After the current cycle t ends, the latest a posteriori load estimate and the estimated covariance P i,t|t Add to the sliding time window; if the sliding time window exceeds the length G, the earliest group is discarded and P i,t|t ;

[0109] Step 233, summarize the included and P i,t|t , calculate the average load mean μ of slice i in the window i and load variance σ i 2 ,as follows:

[0110]

[0111] Where N represents the number of cycles contained in the sliding time window,

[0112] Specifically, due to the short-term bursts in real network traffic, directly using a posteriori load estimates to construct opportunistic constraints can cause significant fluctuations in allocation ratios within a cycle. Using a sliding window of G milliseconds for mean and variance statistics effectively smooths cycle-level noise, making resource allocation more stable.

[0113] In one embodiment of the present invention, based on the mean and variance of each slice flow, combined with the upper limit of failure probability δ vid , δ ctl , δ be Establish the chance constraints and objective function, including:

[0114] The upper limit of failure probability δ vid , δ ctl , δ be The upper limit B of the number of bits transmitted per cycle max Combine to form a confidence bound;

[0115] The mean and variance of each slice flow and the required resource allocation ratio of each slice flow are used to construct a traffic risk function as the first optimization item;

[0116] Based on the resource allocation ratio to be requested for each slice flow, a nonlinear penalty is applied to the resource allocation ratio of each slice flow to construct an entropy regularization function as the second optimization item;

[0117] The confidence bound is used as a chance constraint, and the first optimization term and the second optimization term are combined to form the objective function.

[0118] In one embodiment of the present invention, step 3 is specifically as follows:

[0119] Step 311: Obtain the mean and variance of each slice flow, as well as the upper limit of the failure probability δ i And the upper limit of the transmission bit amount B max , B max =V max H, V max Indicates the maximum transmission rate of the local area network;

[0120] Step 312: Based on the normal distribution, the probability of traffic exceeding the allocated resources is ≤ the upper limit of the failure probability δ i , converted into linear inequality constraints (confidence bounds), as follows:

[0121]

[0122] in, Represents the quantile of the standard normal distribution, represented by δ i Determine, for example, δ i When it is 0.5%, It is 2.576.

[0123] Step 321, combining the mean μ i , variance σ i Ratio to unrequested resources x i ,Quantify the slice overload risk: sum the risks of each slice to form the traffic risk function as follows:

[0124]

[0125] Among them, the molecule μ i 2 +σ i 2 Reflects the total energy of the slice traffic, characterizes the transmission demand, and the denominator is the resource allocation ratio x i , reflecting the traffic pressure borne by unit resources. The smaller the ratio, the greater the risk penalty.

[0126] Step 322: Based on the resource allocation ratio x i Construct an entropy function to impose penalties on extreme allocations as follows:

[0127]

[0128] Where λ represents the regularization coefficient.

[0129] The entropy function ensures that resource allocation avoids extreme skew, while guaranteeing the VID and CTL of key slices and retaining basic service capabilities for background slices BE;

[0130] Step 331 integrates the confidence bound constraint, traffic risk function, entropy regularization function, and resource allocation ratio constraint (the sum of resource allocation ratios is 1, and the minimum retention ratio κ) as follows:

[0131] Objective function:

[0132] Opportunity constraints:

[0133] Specifically, the opportunity constraint is based on the central limit theorem, traffic approximates a normal distribution, quantiles quantify risk, and reliability analysis based on probability theory ensures that resource allocation meets the probabilistic needs of the business. The numerator of the risk function is the second-order moment, reflecting the energy of the traffic, and the denominator reflects resource scarcity. The function is convex, ensuring the existence of a global optimal solution and suitable for fast solution of real-time systems. The entropy function is convex, and even when superimposed with the risk function, it remains convex, ensuring efficient solution of the optimization problem.

[0134] Step 4: Based on the opportunity constraint and the objective function, combined with the minimum resource reservation ratio κ, calculate the resource allocation ratio x of each slice flow vid 、x ctl 、x be ;

[0135] In one embodiment of the present invention, based on the opportunity constraint and the objective function, combined with the minimum resource reservation ratio κ, the resource allocation ratio x of each slice flow is calculated. vid 、x ctl 、x be ,include:

[0136] Based on the combination of opportunity constraints and objective function, and combined with the minimum resource reservation ratio κ, the resource allocation ratio x is constructed respectively. vid 、x ctl 、x be The lower limit of the feasible region;

[0137] Based on the first-order derivative and second-order derivative of the objective function, the resource allocation ratio of each slice flow is determined within the lower limit of the feasible region through the interior point Newton iteration algorithm.

[0138] In one embodiment of the present invention, step 4 is specifically as follows:

[0139] Step 411, determining the opportunity constraint, the objective function and the minimum resource reservation ratio κ;

[0140] Step 412: define the slice resource allocation vector: x = (x vid ,x ctl ,x be )

[0141] Step 413: limit x according to the minimum resource reservation ratio κ. i ≥κ, and the sum of the three is equal to 1, forming a feasible region Δ k ;

[0142] Step 421, for the current iteration point x (n) , calculate the first-order derivative and the second-order derivative

[0143] The objective function is a convex function, and the calculation of the gradient (first-order derivative) and the Hessian matrix (second-order derivative) provides a quadratic approximation basis for the Newton method to determine the curvature of the iteration direction, accelerate convergence, and ensure global optimization.

[0144] Step 422, according to the first-order derivative and the second-order derivative Solve for Newton direction d (n) ,as follows:

[0145]

[0146] The Newton direction is the direction of the minimum point of the quadratic approximation of the objective function. It converges very quickly near the optimal solution to determine the optimal search direction of the current iteration point and quickly approach the global optimal solution.

[0147] Step 423, in the feasible region Δ k Perform line search on the Newton direction to find the maximum step size α (n) , while satisfying: x (n) +α (n) d (n) ∈Δ k as well as

[0148] Step 424, update the next iteration point:

[0149] x (n+1) =x (n) +α (n) d (n)

[0150] Step 425, if If it is less than or equal to the preset threshold, the optimal solution is output; otherwise, n←n+1 is set and the process returns to step 421 to continue iteration.

[0151] Step 5: Based on the resource allocation ratio of each slice flow and combined with the initial contention window value Transmission time limit TXOP max , the number of resource units RU and the a priori number of antennas M, and map the resource allocation ratio of each slice flow to the contention window value CW of each slice flow vid 、CW ctl 、CW be TXOP vid TXOP ctl TXOP be and resource units RU vid , RU ctl , RU be ;

[0152] In one embodiment of the present invention, the contention window value CW of each slice flow is vid 、CW ctl 、CW be ,include:

[0153] By dividing the resource allocation ratio of each slice flow with the initial contention window value Combined with the a priori number of antennas M, the contention window value of each slice flow is calculated as follows:

[0154]

[0155] Where, i∈{vid,ctl,be}, CW i represents the contention window value of the i-th slice flow, x i Indicates the resource allocation ratio of the i-th slice flow, Indicates rounding up to even integers.

[0156] Specifically, the channel access priority of each slice stream in the home WiFi game streaming scenario is dynamically adjusted. The core of this is to convert the resource proportion of each slice stream into a corresponding contention window size. This allows streams with a higher proportion and stricter latency requirements to have a shorter backoff waiting time when competing for the channel, thereby significantly improving their real-time transmission capabilities.

[0157] This is the baseline value of the initial contention window that the system sets for all slice flows during initialization. For example, some Wi-Fi devices set the minimum backoff window for each flow to 15 slots in their default configuration. This ensures that even in extreme cases where the slice ratio and antenna number are the same, excessive congestion or short backoffs are avoided.

[0158] x i Indicates the resource allocation ratio required for the i-th slice in the current cycle. In other words, if x vidA value of 0.5 means that within a 10ms period, the video slice stream should theoretically occupy 50% of the schedulable airtime.

[0159] In WiFi6 (802.11ax) and subsequent protocols that support multi-spatial stream parallel transmission, a device can use M antennas to send data to different services in parallel. Parallel transmission means that when a device has M antennas, the spectrum resources or air interface resources available at the same time increase by M times. In order to maintain a balance between the resource allocation ratio and the backoff window size, x must be mapped. i Multiply by M. In other words, if M is 2, theoretically, after a single channel is acquired, the device can use two antennas to send packets in parallel. Therefore, the denominator of the formula uses x i ×M, when the number of parallel antennas increases, the required backoff window size is correspondingly enlarged, ensuring a balanced distribution inversely proportional to the actual parallel capacity.

[0160] IEEE802.11 EDCA has strict requirements on the contention window size: CW must be an integer multiple of 2. Mapping formula The result is often an arbitrary real number or a multiple of a non-2 integer value, so it must be rounded up to ensure that the returned CW i An even number.

[0161] When the video stream needs to maintain a high frame rate of 60fps, it often obtains a higher x in the optimization model. vid . Mapped CW vid It will be significantly smaller than other slice streams, so that the video stream can get more coherent preemption opportunities during the competition phase, thereby reducing frame loss and jitter. If a sudden control command flow (such as a player key press) occurs in the network, its x ctl It will also rise rapidly, CW ctl Rapidly shrinking allows control instructions to seize the channel and ensure ultra-low latency feedback.

[0162] In one embodiment of the present invention, the transmission time TXOP of each slice flow is vid TXOP ctl TXOP be ,include:

[0163] Allocate the resource ratio of each slice flow to x vid 、x ctl 、x be Transmission time upper limit TXOP max Multiply them together to get the transmission time TXOP of each slice flow vid TXOP ctl TXOP be .

[0164] It should be noted that in the home LAN scenario, in order to ensure that the video frames and control instructions of the game streaming service can be transmitted within the strict delay constraints, the solution uses early traffic prediction and optimization calculation to obtain the ideal network resource allocation ratio x for each slice flow in the next cycle. vid 、x ctl 、x be . Resource allocation ratio x vid 、x ctl 、x be Essentially, it represents the percentage of airtime that each slice flow should occupy. However, the underlying scheduling of WiFiMAC does not directly use the resource allocation ratio x vid 、x ctl 、x be Instead, TXOP (Transmission Opportunity) is used to control the maximum duration that a flow is allowed to occupy the channel continuously after successfully competing for the channel. Therefore, a mapping step is required to convert the above resource allocation ratio x vid 、x ctl 、x be Convert it into specific transmission time to drive the hardware to dynamically allocate channel resources according to the optimization results.

[0165] The conversion calculation formula is as follows:

[0166] TXOP i =x i ×TXOP max

[0167] TXOP i Indicates the specific transmission duration sent to the i-th slice stream;

[0168] If the video stream should ideally occupy 40% of the airtime in a cycle, then when the hardware is granted access to the channel, the video stream connection should be given a one-time continuous channel occupation time equal to 40% of the continuous duration in the cycle. max It represents the maximum continuous usage time allowed by the firmware, which is x i With TXOP max Multiply them together to get a continuous packet transmission duration that is proportional to the Airtime ratio of the slice flow. For example, if TXOP max 4ms, x vid is 0.5, then TXOP vid That is, after a video stream wins the channel, it can continuously send packets for 2ms, and then release them to allow other slice streams to compete again.

[0169] Traditional EDCA solutions typically assign a fixed TXOP to each Access Category, for example, 3ms for voice ACs, 5ms for video ACs, and larger TXOPs for data or background ACs. However, this static allocation cannot dynamically adjust to actual network load and service urgency. If network interference or a sudden increase in demand for a particular flow occurs, the fixed TXOP may fail to support large flows or overuse small flows.

[0170] Therefore, within the TXOP duration, the sliced ​​stream can continuously send multiple data frames without having to participate in channel competition in the middle, reducing the backoff waiting delay between each frame, which is beneficial to reducing the inter-frame delay jitter. Because the TXOP itself can only be configured in the hardware register in integer milliseconds, and the resource ratio x i It is usually a floating point number. After multiplication, the product can be rounded down to an integer millisecond with an error within ±0.5ms. This is sufficient accuracy for a 10ms period and does not significantly affect the latency guarantee.

[0171] In one embodiment of the present invention, the resource unit RU of each slice stream vid , RU ctl , RU be ,include:

[0172] Allocate the resource ratio of each slice flow to x vid 、x ctl 、x be Multiply by the number of resource units RU and round down to get the resource unit RU of each slice stream vid , RU ctl , RU be .

[0173] In detail, in order to enable wireless hardware to allocate resources according to the ratio x i When allocating the minimum resource unit (RU), these abstract proportions must be mapped to an integer number of RUs so that they can be directly written into the WiFi6 / 802.11ax MAC scheduling module.

[0174] in,

[0175] RU represents the total number of resource units available for allocation under a given channel bandwidth (e.g., 80 MHz);

[0176] Indicates rounding down to the maximum integer not greater than the input value, ensuring that the final RU i An integer that does not exceed the total available.

[0177] In the WiFi6 / 11ax environment, for each scheduling cycle, the MAC layer can divide the channel into several equal-width subcarrier clusters through OFDMA (Orthogonal Frequency Division Multiple Access), each cluster is called a resource unit (RU). At any time, the hardware can only allocate several adjacent or non-adjacent subcarrier clusters to a site or a slice flow at an integer number of RUs. The system obtains the resource allocation ratio of each slice flow x i , and it is not possible to allocate 3.2 RUs to the video slice stream through hardware, because the number of RUs must be an integer. Therefore, it is necessary to allocate the resource ratio of each slice stream to x i Mapped to an integer RU, and it must be ensured that the total number of RUs after all slices are allocated will not be greater than the total number of RUs available in the channel.

[0178] Step 6: Allocate resources for the video slice stream, the instruction slice stream, and the background slice stream according to the contention window value, transmission time, and resource unit of each slice stream.

[0179] In one embodiment of the present invention, resource allocation of the video slice stream, the instruction slice stream, and the background slice stream is performed based on the contention window value, the transmission time, and the resource unit of each slice stream, including:

[0180] The contention window value, transmission time and resource unit of each slice stream are written into the corresponding media access control layer register respectively, a joint scheduling table for the video slice stream, instruction slice stream and background slice stream is constructed, and resource allocation is performed.

[0181] By writing hardware-level parameters and dynamic scheduling, the resource allocation ratio is x i This implementation enables end-to-end resource assurance for multiple service slices in a home Wi-Fi environment. Without relying on dedicated hardware, firmware-level algorithm optimization significantly improves the stability and resource utilization of hard real-time services, providing an efficient and reliable solution for game streaming.

[0182] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A network resource dynamic allocation method based on network slicing, applied to resource allocation of video slice streams, instruction slice streams, and background slice streams, characterized in that: include: Step 1: Initialize the following parameters: Current period t, prediction window of length G, upper limit of failure probability δ vid , δ ctl , δ be , initial contention window value Transmission time limit TXOP max , the number of resource units RU, the minimum resource reservation ratio κ, the traffic autocorrelation coefficient ρ and the number of antennas M; Step 2: Determine the size and remaining time of the data packet to be sent, use the traffic autocorrelation coefficient ρ to build an autoregressive model and combine it with Kalman filtering to predict each slice flow within the prediction window length H, and obtain the mean μ of each slice flow vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 ; Step 3: Based on the mean and variance of each slice flow, combined with the upper limit of failure probability δ vid , δ ctl , δ be Establish chance constraints and objective functions; Step 4: Based on the opportunity constraint and the objective function, combined with the minimum resource reservation ratio κ, calculate the resource allocation ratio x of each slice flow vid 、x ctl 、x be ; Step 5: Based on the resource allocation ratio of each slice flow and combined with the initial contention window value Transmission time limit TXOP max , the number of resource units RU and the a priori number of antennas M, and map the resource allocation ratio of each slice flow to the contention window value CW of each slice flow vid 、CW ctl 、CW be TXOP vid TXOP ctl TXOP be and resource units RU vid , RU ctl , RU be ; Step 6: Allocate resources for the video slice stream, the instruction slice stream, and the background slice stream according to the contention window value, transmission time, and resource unit of each slice stream.

2. The method for dynamic allocation of network resources based on network slicing according to claim 1, characterized in that: Determine the size and remaining time of the data packet to be sent, use the traffic autocorrelation coefficient ρ to build an autoregressive model and combine it with Kalman filtering to predict each slice flow within the prediction window length H, and obtain the mean μ of each slice flow vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 ,as follows: In the current cycle t, the size and remaining time of the data packets to be sent are collected, and the observed load of each slice flow is calculated respectively; The instantaneous load and the flow autocorrelation coefficient ρ are combined to construct a first-order autoregressive model to make a priori prediction of the load in period T+1 and obtain the prior load. The Kalman filter is used to fuse the prior load and the observed load, and the mean μ of each slice flow is output within the prediction window. vid 、μ ctl 、μ be and variance σ vid 2 , σ ctl 2 , σ be 2 .

3. The method for dynamic allocation of network resources based on network slicing according to claim 2, characterized in that: Based on the mean and variance of each slice flow, combined with the upper limit of failure probability δ vid , δ ctl , δ be Establish the chance constraints and the objective function, include: The upper limit of failure probability δ vid , δ ctl , δ be The upper limit B of the number of bits transmitted per cycle max Combine to form a confidence bound; The mean and variance of each slice flow and the required resource allocation ratio of each slice flow are used to construct a traffic risk function as the first optimization item; Based on the resource allocation ratio to be requested for each slice flow, a nonlinear penalty is applied to the resource allocation ratio of each slice flow to construct an entropy regularization function as the second optimization item; The confidence bound is used as a chance constraint, and the first optimization term and the second optimization term are combined to form the objective function.

4. The method for dynamic allocation of network resources based on network slicing according to claim 3, characterized in that: Based on the opportunity constraint and the objective function, combined with the minimum resource reservation ratio κ, the resource allocation ratio x of each slice flow is calculated vid 、x ctl 、x be ,include: Based on the combination of opportunity constraints and objective function, and combined with the minimum resource reservation ratio κ, the resource allocation ratio x is constructed respectively. vid 、x ctl 、x be The lower limit of the feasible region; Based on the first-order derivative and second-order derivative of the objective function, the resource allocation ratio of each slice flow is determined within the lower limit of the feasible region through the interior point Newton iteration algorithm.

5. The method for dynamic allocation of network resources based on network slicing according to claim 4, characterized in that: Contention window value CW of each slice flow vid 、CW ctl 、CW be ,include: By dividing the resource allocation ratio of each slice flow with the initial contention window value Combined with the a priori number of antennas M, the contention window value of each slice flow is calculated as follows: Where, i∈{vid,ctl,be}, CW i represents the contention window value of the i-th slice flow, x i Indicates the resource allocation ratio of the i-th slice flow, Indicates rounding up to even integers.

6. The method for dynamic allocation of network resources based on network slicing according to claim 5, characterized in that: The transmission time TXOP of each slice flow vid TXOP ctl TXOP be ,include: Allocate the resource ratio of each slice flow to x vid 、x ctl 、x be Transmission time upper limit TXOP max Multiply them to get the transmission time TXOP of each slice flow vid TXOP ctl TXOP be .

7. The method for dynamic allocation of network resources based on network slicing according to claim 6, characterized in that: Resource unit RU of each slice stream vid , RU ctl , RU be ,include: Allocate the resource ratio of each slice flow to x vid 、x ctl 、x be Multiply by the number of resource units RU and round down to get the resource unit RU of each slice stream vid , RU ctl , RU be .

8. The method for dynamic allocation of network resources based on network slicing according to claim 7, characterized in that: Resource allocation for the video slice stream, instruction slice stream, and background slice stream is performed based on the contention window value, transmission time, and resource unit of each slice stream, including: The contention window value, transmission time and resource unit of each slice stream are written into the corresponding media access control layer register respectively, a joint scheduling table for the video slice stream, instruction slice stream and background slice stream is constructed, and resource allocation is performed.

Images