5G-TSN joint scheduling method based on time delay budget and SMT

Abstract

The invention relates to a 5G-TSN joint scheduling method based on time delay budget and SMT, which belongs to the technical field of industrial communication, and comprises the following steps: establishing a 5G-TSN networking architecture; the SDNC obtains the information; the CNC determines a scheduling constraint condition, and the scheduling constraint condition is substituted into an SMT solver to solve a transmitting end TSN switch and a DS-TT gating list; the CNC1 calculates a CQI predicted value at a scheduling moment by using the CQI historical value; the SDNC maps the channel quality into an air interface transmission rate according to the mapping table of the CQI and the MCS and the mapping table of the TBS and the MCS; the SDNC calculates the transmission delay budget of the 5G system according to the maximum retransmission times corresponding to the set channel quality; the SDNC calculates NW-TT and a receiving end TSN switch gating list; the SDNC sends scheduling list configuration information to the CNC1 and the CNC2; and the CNC1 and the CNC2 send the gating list to the transmitting end TSN switch and the receiving end TSN switch.

Description

Technical Field

[0001] The present invention belongs to the field of industrial communication technology and relates to a 5G-TSN joint scheduling method based on delay budget and SMT. Background Art

[0002] Industrial communication technologies and systems are gaining more and more attention from industry and academia. Traditional industrial Ethernet protocols (such as PROFINET, EtherCAT, Sercos III, etc.) lack openness and interoperability. Time-Sensitive Networking (TSN) is a set of new standards proposed by the IEEE 802.1 working group. It provides deterministic bounded low latency, high reliability, and cross-domain flexibility for industrial data transmission. However, TSN has inherent problems with wired transmission, such as limited coverage, high maintenance costs, and complex installation. It cannot meet the flexibility required by emerging industrial use cases.

[0003] 5G provides flexible, ultra-reliable, low-latency communication services for future industrial applications. Time-sensitive communication (TSC) is introduced to provide performance guarantees for latency and reliability. Therefore, using 5G networks as an extension of TSN in the wireless field is a promising solution to meet the communication needs of industrial automation systems for high flexibility, adaptability, and scalability.

[0004] There are currently three main problems in 5G-TSN joint scheduling:

[0005] (1) Different data streams are allocated different wireless resources in 5G. How to process the data stream in the TSN domain based on the wireless resources of the data stream?

[0006] (2) There is a delay in UE reporting CQI (Channel Quality Indicator). When the CQI reporting delay exceeds the coherence time of the channel, the CQI at the reporting time will not match the channel quality at the time of scheduling.

[0007] (3) The uncontrollable impact of the time-varying characteristics of the 5G air interface on deterministic data transmission. The time-varying characteristics of the air interface may cause data packet disorder and data packet loss, resulting in delay jitter and incomplete data received by the receiving end.

[0008] At present, there is a relative lack of research on the joint scheduling algorithms and collaborative transmission mechanisms of 5G and TSN. The existing methods of 5G and TSN do not combine the characteristics of 5G and TSN themselves (data transmission protocol, real-time, reliability, low latency and other requirements). Directly using existing methods cannot meet the many requirements of 5G-TSN. Summary of the invention

[0009] In view of this, the object of the present invention is to provide a 5G-TSN joint scheduling method based on delay budget and SMT.

[0010] In order to achieve the above object, the present invention provides the following technical solutions:

[0011] A 5G-TSN joint scheduling method based on delay budget and SMT includes the following steps:

[0012] Step 1: Build a scenario based on the 5G-TSN networking architecture;

[0013] Step 2: The software-defined network controller SDNC obtains TSN network topology, data flow information, and 5G wireless channel information;

[0014] Step 3: The centralized network configuration CNC determines the scheduling constraints and brings them into the SMT solver to find the TSN switch and DS-TT gating list on the sender side;

[0015] Step 4: In the first centralized network configuration, CNC1 brings the CQI historical value into the CQI prediction model to obtain the CQI prediction value at the scheduling time;

[0016] Step 5: SDNC maps the channel quality to the air interface transmission rate according to the mapping table between CQI and modulation and coding strategy MCS and the mapping table between transport block size TBS and MCS;

[0017] Step 6: SDNC calculates the 5G system transmission delay budget based on the maximum number of retransmissions corresponding to the set channel quality;

[0018] Step 7: SDNC calculates NW-TT and the receiving-end TSN switch gating list based on the TSN switch gating list, 5G system transmission delay budget, and TSN wired link transmission delay;

[0019] Step 8: SDNC sends the scheduling list configuration information to the two centralized network configurations CNC1 and CNC2;

[0020] Step 9: CNC1 and CNC2 send the gating list to the sending end TSN switch and the receiving end TSN switch respectively.

[0021] Furthermore, the 5G-TSN networking architecture described in step 1 is as follows:

[0022] A TSN transmitter terminal is connected to a transmitter TSN switch and enters the 5G system through the device-side time-sensitive network converter DS-TT. The 5G system is connected to a receiver TSN switch through the network-side time-sensitive network converter NW-TT and finally connected to a TSN receiver terminal. The 5G-TSN joint scheduling calculation based on delay budget and SMT is performed at the software-defined network controller SDNC.

[0023] Furthermore, in step 2, when the transmitter wants to send a TSN stream to the receiver, the centralized network configuration CNC1 and CNC2 discover the network topology of TSN according to the link discovery protocol LLDP, and forward the network topology information of TSN to SDNC; the terminal device sends the sending cycle, data frame size, priority and delay requirement information of each data stream to the centralized user configurator through the user configuration protocol, and the centralized user configurator sends the data stream information to CNC1 and CNC2 through the user network interface, and CNC1 and CNC2 forward the data stream information to SDNC.

[0024] Furthermore, in step 3, CNC1 adds constraints to the SMT solver to constrain the value range of the variables so that they meet the limitations of network resources and the transmission quality requirements of data streams. The constraints are as follows:

[0025] 1) The TSN switch gating cycle Schedcycle is the lowest common multiple of the transmission cycles of all data flows passing through the TSN switch:

[0026]

[0027] Where F represents the set of data flows flowing through the TSN switch, T i Represents the data flow f i The period of LCM means taking the least common multiple of all elements in the brackets;

[0028] 2) The number of transmission time slots n of data stream i in the TSN switch gating cycle i It is equal to the TSN switch gating cycle divided by the data flow sending cycle:

[0029]

[0030] 3) Data flow f i The sending time of the mth data packet must be between the m-1th sending cycle and the mth sending cycle. The data flow f i The time difference between the sending times of two consecutive data packets is equal to the sending cycle:

[0031]

[0032]

[0033] 4) Data flow f i Each data packet must be received by DS-TT within the maximum transmission delay of the TSN domain on the sender side:

[0034]

[0035] The maximum transmission delay of the TSN domain on the sender side meets the following requirements:

[0036]

[0037] Among them, N i,priority The priority is greater than or equal to the data flow f i The number of data streams, l i is the data packet size of data stream fi, R TSN is the sending rate of the TSN switch, D pro is the processing delay of the TSN switch, R Wired is the wired link transmission rate;

[0038] 5) The transmission time slots in the gating cycle of the TSN switch cannot overlap, that is, the data flow f i The start time of a transmission time slot must be greater than or equal to the data flow f j The end time of a transmission time slot of or the end time is less than or equal to the data flow f j The start time of a transmission slot:

[0039] Each port can only have one message entering the link at a time, and the messages entering the same link cannot overlap in the time domain, that is, the data flow f i The sending time of the mth data packet must be greater than or equal to the data flow f j The time when the (m-1)th data packet arrives at the TSN switch:

[0040]

[0041]

[0042] 7) Data packets must be transmitted within the corresponding transmission time slot window, and the data flow f i The start time of the mth transmission time slot in the gating cycle of the TSN switch at the sending end cannot be greater than the time when the message starts to be transmitted, and the end time of the transmission time slot cannot be less than the time when the data packet ends transmission:

[0043]

[0044] 8) Data Flow f iWhen the data packet arrives at the sending TSN switch, the transmission time slot window must be within the corresponding scheduling period, and the data flow f i After the data packet arrives at the sending TSN switch, it can only choose the arrival time f i After the transmission time slot of the first data packet is determined, the transmission time slots of other data packets in the same data stream are also determined, that is, the following is satisfied:

[0045]

[0046]

[0047] CNC1 runs the SMT solver and tries all possible values ​​of the variables in the solution space in turn. If no feasible solution is found in the first round of solution, the information that the solution is not schedulable is returned. If a feasible solution is found, the maximum transmission delay of the TSN domain on the sender side of each data stream is reduced, the constraint conditions are modified, and the solution is performed again. This process is repeated until no feasible solution is found, and the result of the last solution is output. The maximum transmission delay of the TSN domain on the sender side is adjusted as follows:

[0048]

[0049] in is the data stream f i The j-th solved maximum transmission delay in the TSN domain on the sender side.

[0050] Further, in step 4, CNC1 brings the CQI historical value into the CQI prediction model to obtain the CQI prediction value at the scheduling time; the CQI prediction model uses a feedforward neural network, which is composed of an input layer, a hidden layer, and an output layer, and W1 and W2 are the connection weight matrices of the corresponding layers; the CQI historical values ​​of n moments before time t are selected as samples to train the CQI prediction model; the data stream f i The time when the mth data packet arrives at the UE As the scheduling moment on the 5G side.

[0051] Furthermore, the CQI prediction model training steps are as follows:

[0052] 1) The CQI historical values ​​of n moments before moment t are used as the input of the CQI prediction model, and its input matrix is:

[0053]

[0054] 2) Standardize the input data so that the processed data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1; standardize the input CQI historical value according to the following formula:

[0055]

[0056] Wherein, μ is the average value of the CQI historical values ​​input at n moments, and σ is the standard deviation of the CQI historical values ​​input at n moments;

[0057] 3) Initialize W1, W2, and the connection weight ω between every two neurons i，j The value is initialized to a random number between -1.0 and 1.0, and the bias θ of each neuron i It is also initialized to a random number between -1.0 and 1.0;

[0058] 4) Calculate the input and output of the first and second layer neurons based on the connection weights and the bias of the neurons, as shown in the following formula:

[0059]

[0060] The activation function uses the logistic function, and the neuron output is:

[0061]

[0062] 5) The output of the last neuron is the standardized CQI prediction value at the scheduling time, which is Denormalize it to get the CQI prediction value cqt at the scheduling time t，pr , the denormalization formula is as follows:

[0063]

[0064] 6) Standardize the actual CQI value at time t to obtain Calculate the error between the final output CQI prediction value and the actual CQI value. The error calculation of the last neuron (1+2n) is as follows:

[0065]

[0066] 7) The error obtained needs to be propagated from back to front. The error of the neuron j in the previous layer can be calculated by the errors of all neurons k in the next layer connected to it. The specific formula is:

[0067] E j =O j (1-O j )∑ k ω j，k E k

[0068] 8) Calculate the error of all neurons in the hidden layer, and then connect the weights ω i，j and the neuron bias θi Adjustment, the adjustment formula is as follows:

[0069] ω i，j =ω i，j +(l)O i E j

[0070] θ j =θ j +(l)E j

[0071] Where l is the learning rate, which is set to a constant between 0 and 1. After the adjustment is completed, if the final output error is less than the acceptable range or the number of iterations h reaches the threshold, the training ends; otherwise, the number of iterations h is increased by 1, and the training is turned to step 4) to continue using the current sample for training. The error acceptance range is determined by the following formula:

[0072]

[0073] E P =|cqi t -cqt t，pr |

[0074] E P <αE R

[0075] Among them, E R is the error caused by the CQI reporting delay, E P is the error between the actual CQI value and the predicted value. P <αE R The training ends when α is a constant between 0 and 1;

[0076] The moment when the first data packet of each data stream arrives at the UE is taken as its 5G side scheduling moment. The CQI historical values ​​of n moments before the scheduling moment are selected as the input of the CQI prediction model to predict the CQI at the scheduling moment. The CQI prediction value at the scheduling moment of the first data packet is used for scheduling the data packets after each data stream.

[0077] Further, in step 5, after CNC1 obtains the CQI prediction value at the scheduling time, it uploads it to SDNC, and SDNC obtains the data flow f according to the mapping table between CQI and MCS. i Modulation and coding strategy index MCSIndex i , and then according to the mapping table of TBS and MCS, get the data stream f i The transport block size TBS i ; The value assigned to data stream f is calculated by the following formula i The number of RBs (resource blocks) NRBi :

[0078]

[0079] The data flow f is calculated by the following formula i 5G ultra-reliable low-latency communication air interface data transmission rate

[0080]

[0081] Among them, B RB is the resource block bandwidth, P tx is the transmit power, N0 is the unilateral noise spectral density, and g is the path loss. All of them take fixed values, and only the impact of CQI is considered.

[0082] Furthermore, in step 6, the 5G system transmission delay budget is determined by the data packet size, air interface transmission rate, number of retransmissions, and retransmission delay, as shown in the following formula:

[0083]

[0084] K i The value of the data stream f i The CQI prediction value at the scheduling time is set as shown in the following formula:

[0085]

[0086] The beneficial effects of the present invention are as follows: the present invention is based on the traffic scheduling mechanism of SMT, and considers converting the TSN data stream time slot planning problem into a constraint satisfaction problem while meeting the 5G delay budget, and obtains the result of data stream time slot planning by using the SMT solver; the present invention is based on the 5G system transmission delay budget adjustment method, analyzes the relationship between different channel qualities and the maximum number of retransmissions, eliminates the time jitter caused by the transmission of data packets in the 5G system, and thus provides deterministic end-to-end delay guarantee for time-triggered service flows; the present invention is based on the neural network CQI prediction method, and predicts the current CQI value at the scheduling moment according to the CQI historical value obtained in the system, thereby reducing the impact of CQI reporting delay.

[0087] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0089] Figure 1 Provides 5G-TSN networking architecture;

[0090] Figure 2 Solve the flow chart for SMT;

[0091] Figure 3 is the CQI prediction model;

[0092] Figure 4 This is the network topology diagram that CNC has obtained. DETAILED DESCRIPTION

[0093] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0094] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0095] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0096] See also Figure 1 The present invention provides a 5G-TSN joint scheduling method based on delay budget and SMT. The method relies on the 5G-TSN networking architecture. The networking architecture is as follows Figure 1As shown. The network topology consists of: a TSN terminal (transmitter) is connected to a TSN transmitter switch, enters 5G through DS-TT (device-side TSN translator), 5G is connected to a TSN receiver switch through NW-TT (Network-side TSN translator), and finally connected to a TSN terminal (receiver). This method performs 5G-TSN joint scheduling calculation based on delay budget and SMT at SDNC (Software Defined Network Control). CNC1 and CNC2 (centralized network configuration) discover the network topology of TSN according to the Link Layer Discovery Protocol (LLDP) and forward the network topology information of TSN to SDNC. The terminal device sends the information such as the transmission period of each data stream, the size of the data frame, the priority and the delay requirement to the centralized user configurator through the user configuration protocol. The centralized user configurator sends the data stream information to CNC1 and CNC2 through the user network interface. CNC1 and CNC2 forward the data stream information to SDNC. In the 5G network, the UE (user equipment) communicates with the gNB (base station) through the physical layer interface and can upload CQI to the gNB. The gNB will record the information related to CQI in its log, and the gNB can upload the log to the SDNC. The SDNC organizes the collected network topology information and data flow information, where the network topology information includes TSN link information, TSN to 5G link information, 5G link information, and 5G to TSN link information; the data flow information includes end-to-end delay requirements and data flow priority.

[0097] This method mainly solves the transmission scheduling problem of time-triggered flow in the 5G-TSN networking architecture to ensure the certainty of transmission. The scheduling goal is to minimize the total transmission time.

[0098] The proposed 5G-TSN joint scheduling method based on delay budget and SMT includes the following three key research contents:

[0099] 1. Study the traffic scheduling mechanism based on SMT. Consider converting the TSN data stream time slot planning problem into a constraint satisfaction problem while meeting the 5G delay budget. Use the SMT solver to solve the problem and obtain the result of data stream time slot planning.

[0100] 2. Research the transmission delay budget adjustment method based on 5G system, analyze the relationship between different channel qualities and the maximum number of retransmissions, eliminate the time jitter caused by data packet transmission in 5G system, and thus provide deterministic end-to-end delay guarantee for time-triggered service flows.

[0101] 3. Research the CQI prediction method based on neural network, predict the current CQI value at the scheduling time according to the historical CQI value obtained in the system, and reduce the impact of CQI reporting delay.

[0102] This method divides the end-to-end transmission delay into two parts: TSN transmission delay and 5G transmission delay. Considering that each data stream needs to wait for higher priority data stream processing at the TSN switch at the sending end, the maximum transmission delay of the TSN domain at the sending end is obtained. The scheduling constraints are determined according to the maximum transmission delay of the TSN domain at the sending end, and the SMT (Satisfiability Modulo Theories) solver is introduced to obtain the gating list of the TSN switch at the sending end. For the CQI reporting delay, the neural network prediction CQI method is used to predict the current CQI value at the scheduling time based on the CQI historical value obtained in the system. In view of the disorder of data packet order and data packet loss caused by the time-varying characteristics of the 5G air interface, which leads to delay jitter and incomplete data received by the receiving end, the 5G system transmission delay budget is used to allocate the 5G system transmission delay budget according to the channel quality of the data stream, and then the DS-TT and the gating list of the TSN switch at the receiving end are calculated based on the gating list of the TSN switch at the sending end.

[0103] The implementation steps of the 5G-TSN joint scheduling method based on delay budget and SMT are as follows:

[0104] Each flow f i In sextuple (es s , eS d , T i , TS i , D i , l i , p i ) indicates that, among them, es i is the data stream f i The sending end, es i is the data stream f i The receiving end, T i is the data stream f i The sending cycle, TS i is the data stream f i The request is sent at the time D i is the data stream f i The maximum allowable end-to-end delay, l i is the data stream f i The packet size, p i is the data stream f i The priority of all data flows is F = {f1, f2, ..., f k}, k is the number of data streams.

[0105] Step 1: Build scenarios based on the 5G-TSN networking architecture;

[0106] Step 2: When the transmitter wants to send a TSN stream to the receiver, the two CNCs discover the TSN network topology based on LLDP and forward the TSN network topology information to the SDNC. The terminal device sends the information such as the sending period, data frame size, priority and latency requirements of each data stream to the centralized user configurator through the user configuration protocol. The centralized user configurator sends the data stream information to CNC1 and CNC2 through the user network interface. CNC1 and CNC2 then forward the data stream information to the SDNC.

[0107] Step 3: CNC1 adds constraints to the SMT solver to constrain the value range of the variables so that they meet the limitations of network resources and the transmission quality requirements of data streams. The specific constraints are as follows:

[0108] 1) The TSN switch gating cycle Schedcycle is the lowest common multiple of the transmission cycles of all data flows passing through the TSN switch, and needs to satisfy formula (1):

[0109]

[0110] Where F represents the set of data flows flowing through the TSN switch, T i Represents the data flow f i The period of LCM means taking the least common multiple of all elements in the brackets;

[0111] 2) The number of transmission time slots n of data stream i in the TSN switch gating cycle i It is equal to the TSN switch gating cycle divided by the data flow sending cycle, as shown in formula (2):

[0112]

[0113] 3) Data flow f i The sending time of the mth data packet must be between the (m-1)th sending cycle and the mth sending cycle. The data flow f i The time difference between the sending times of two consecutively sent data packets is equal to the sending cycle.

[0114]

[0115] 4) Data flow f i Each data packet must be received by DS-TT within the maximum transmission delay of the TSN domain on the sender side, and needs to satisfy formula (5):

[0116]

[0117] Consider the data stream f i At the switch, it is necessary to wait for all data flows with higher and equal priorities to be processed. The maximum transmission delay of the TSN domain on the sender side satisfies formula (6):

[0118]

[0119] Among them, N i,priority The priority is greater than or equal to the data flow f i The number of data streams, l i is the data stream f i Packet size, R TSN is the sending rate of the TSN switch, D pro is the processing delay of the TSN switch, R Wired is the wired link transmission rate. Assume that the sending rate of all TSN switches is consistent, the processing delay of all TSN switches is consistent, and the transmission rate of all wired links is consistent.

[0120] 5) The transmission time slots in the gating cycle of the TSN switch cannot overlap, that is, the data flow f i The start time of a transmission time slot must be greater than or equal to the data flow f j The end time of a transmission time slot of or the end time is less than or equal to the data flow f j The start time of a transmission time slot.

[0121]

[0122] 6) Each port can only have one message entering the link at a time, and the messages entering the same link cannot overlap in the time domain, that is, data flow f i The sending time of the mth data packet must be greater than or equal to the data flow f j The moment when the (m-1)th data packet arrives at the TSN switch must satisfy the following formula:

[0123]

[0124]

[0125] 7) Data packets must be transmitted within the corresponding transmission time slot window, so the data flow f i The start time of the mth transmission time slot in the gating cycle of the TSN switch at the sending end cannot be greater than the time when the message starts to be transmitted, and the end time of the transmission time slot cannot be less than the time when the data packet ends to be transmitted.

[0126]

[0127] 8) Data Flow f i When the data packet of the data stream arrives at the sending TSN switch, the transmission time slot window must be within the corresponding scheduling period. i After the data packet arrives at the sending end TSN switch, it can only choose the arrival time T i After the transmission time slot of the first data packet is determined, the transmission time slots of other data packets in the same data stream are also determined. That is, the following constraints must be met.

[0128]

[0129] CNC1 runs the SMT solver, which tries all possible values ​​of the variables in the solution space in turn. If a feasible solution is not found in the first round of solving, it returns unschedulable information. If a feasible solution is found, the maximum transmission delay of each data stream in the TSN domain on the sender side is reduced, the constraint conditions are modified, and the solution is performed again. This process is repeated until a feasible solution cannot be found, and the last solution result is output. The maximum transmission delay adjustment of the TSN domain on the sender side is shown in formula (13):

[0130]

[0131] The solution flow chart of the SMT solver is as follows: Figure 2 shown.

[0132] Step 4: CNC1 brings the CQI historical value into the CQI prediction model to obtain the CQI prediction value at the scheduling time. The CQI prediction model uses a feedforward neural network, which consists of an input layer (layer 0), a hidden layer (layer 1), and an output layer (layer 2). W1 and W2 are the connection weight matrices of the corresponding layers. The CQI prediction model is as follows: Figure 3 shown.

[0133] The data stream f i The time when the mth data packet arrives at the UE As the 5G side scheduling moment, due to the CQI reporting delay, the CQI used in actual scheduling is The value of the moment.

[0134] Before using the CQI prediction model to predict the CQI, the CQI prediction model needs to be trained first. Assuming that the actual CQI value at time t has been obtained, the CQI historical values ​​at n times before time t are selected as samples for training.

[0135] The CQI prediction model training process is as follows:

[0136] 1) The CQI historical values ​​of n moments before moment t are used as the input of the CQI prediction model, and its input matrix is:

[0137]

[0138] 2) Standardize the input data so that the processed data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1. Standardize the input CQI historical value according to the following formula:

[0139]

[0140] Wherein, μ is the average value of the CQI historical values ​​input at n moments, and σ is the standard deviation of the CQI historical values ​​input at n moments.

[0141] 3) Initialize W1, W2, and the connection weight ω between every two neurons i，j The value is initialized to a random number between -1.0 and 1.0, and the bias θ of each neuron i It is also initialized to a random number between -1.0 and 1.0.

[0142] 4) Calculate the input and output of the first and second layer neurons based on the connection weights and the bias of the neurons, as shown in the following formula:

[0143] S j =∑ i ω i，j O i +θ j (16)

[0144] The activation function uses the logistic function, and the neuron output is:

[0145]

[0146] 5) The output of the last neuron is the standardized CQI prediction value at the scheduling time, which is Denormalize it to get the CQI prediction value cqi at the scheduling time t,pr The denormalization formula is as follows:

[0147]

[0148] 6) Standardize the actual CQI value at time t to obtain The error between the final output CQI prediction value and the actual CQI value can be calculated. The error calculation of the last neuron (1+2n) is shown in formula (19):

[0149]

[0150] 7) The error obtained needs to be propagated from back to front. The error of the neuron j in the previous layer can be calculated by the errors of all neurons k in the next layer connected to it. The specific formula is:

[0151] E j =O j (1-O j )∑ k ω j，k E k (20)

[0152] 8) Calculate the error of all neurons in the hidden layer by the formula, and then connect the weights ω i，j and the neuron bias θ i Adjustment, the adjustment formula is as follows:

[0153] ω i，j =ω i，j +(l)O i E j (twenty one)

[0154] θ j =θ j +(l)E j (twenty two)

[0155] Where l is the learning rate, which is usually set to a constant between 0 and 1. An empirical rule is to set the learning rate to the reciprocal of the number of iterations h. After the adjustment is completed, if the final output error is less than the acceptable range or the number of iterations h reaches a certain threshold, the training ends; otherwise, the number of iterations h increases by 1, and the training is continued using the current sample in step 4). The error acceptance range is determined by the following formula:

[0156]

[0157] E P =|cqi t -cqt t，pr | (24)

[0158] E P <αE R (25)

[0159] Among them, E R is the error caused by the CQI reporting delay, E P is the error between the actual CQI value and the predicted value. When equation (25) is satisfied, the training is terminated, and α is a constant between 0 and 1.

[0160] The time when the first data packet of each data flow arrives at the UE is taken as its 5G side scheduling time. The CQI historical values ​​of n times before the scheduling time are selected as the input of the CQI prediction model to predict the CQI at the scheduling time. The CQI prediction value at the scheduling time of the first data packet is used for scheduling the data packets after each data flow. i For example, the first data packet At time t, the base station obtains the actual CQI value at the scheduling time. When the error between the actual value and the predicted value is too large, that is, it does not satisfy equation (25), the model is trained again.

[0161] Step 5: After CNC1 obtains the CQI prediction value at the scheduling time, it uploads it to SDNC. SDNC obtains the data flow f according to the mapping table between CQI and MCS. i Modulation and coding strategy index MCSIndex i , and then according to the mapping table of TBS and MCS, the transmission block size TBS of data stream fi is obtained i The value assigned to data stream f is calculated by formula (26): i The number of RBs (resource blocks) NRB i , and then use formula (27) to calculate the data flow f i 5G ultra-reliable low-latency communication air interface data transmission rate Among them, BRB is the resource block bandwidth, P tx is the transmit power, N0 is the unilateral noise spectral density, and g is the path loss. All of them take fixed values, and only the impact of CQI is considered.

[0162]

[0163]

[0164] Step 6: The transmission delay budget of the 5G system is determined by the data packet size, air interface transmission rate, number of retransmissions, retransmission delay, etc., as shown in formula (28).

[0165]

[0166] K i The value of the data stream f i The CQI prediction value at the scheduling time is set as shown in formula (29):

[0167]

[0168] Step 7: SDNC calculates NW-TT and the gating list of the TSN switch at the receiving end based on the gating list of the TSN switch at the sending end, the transmission delay budget of the 5G system, and the transmission delay of the TSN wired link;

[0169] Step 8: SDNC sends the scheduling list configuration information to CNC1 and CNC2;

[0170] Step 9: CNC1 and CNC2 encapsulate the calculation results into an XML file and use the NETCONF protocol to configure the XML-based gating scheduling table to the sending TSN switch and the receiving TSN switch respectively.

[0171] The definitions of terms in this protocol are shown in Table 1.

[0172] Table 1

[0173]

[0174]

[0175] Assume that CNC has obtained the network topology as follows Figure 4 As shown, there is a link transmission rate R Wired The transmission rate of the TSN switch is 1000Mbps, and the processing delay of the switch is D pro 4μs, resource block bandwidth B RB is 180kHz, the transmitting power is P tx The power consumption of the transmission line is 500mW, the unilateral noise spectral density N0 is -174dBm / Hz, and the path loss g is -100dB. There are two streams f1 and f2 to be sent. The sender of stream f1 is ES1, the receiver is ES3, the period is 300μs, the request to send time is 0, the maximum allowed delay is 3000μs, the data packet frame length is 512 bytes, and the priority is 7. f1 is expressed as (ES1, ES3, 300, 0, 3000, 512, 7). The sender of stream f2 is ES2, the receiver is ES4, the period is 600μs, the request to send time is 0, the maximum allowed delay is 3500μs, the data packet frame length is 512 bytes, and the priority is 7. f2 is expressed as (ES2, ES4, 600, 0, 3500, 512, 7).

[0176] 1) First, calculate the gate control cycle Schedcycle of all TSN switches according to formula (1)

[0177] Schedcycle=600μs

[0178] Then, the number of transmission time slots of the two data streams in the TSN switch gating cycle is calculated according to formula (2):

[0179]

[0180] According to formula (6), the maximum transmission delay of the two data streams in the TSN domain at the sending end is calculated:

[0181]

[0182] 2) Bring the information of the two data streams into the constraints:

[0183] Frame Constraints:

[0184]

[0185] Latency constraints:

[0186]

[0187] Conflicting constraints:

[0188]

[0189]

[0190] Time slot window constraints:

[0191]

[0192]

[0193] Substitute it into the SMT solver and the result is:

[0194] The sending time of the first data packet of data stream f1 is: 0us

[0195] The time when the first data packet of data stream f1 arrives at the TSN switch at the sending end is: 40.96us

[0196] The start time of the first time slot of data stream f1 at the sending end TSN switch is: 44.96us

[0197] The closing time of the first time slot of data stream f1 at the sending end TSN switch is: 49.056us

[0198] The sending time of the second data packet of data stream f1 at the sending end TSN switch is: 300us

[0199] The time when the second data packet of data stream f1 arrives at the TSN switch at the sending end is: 340.96us

[0200] The start time of the second time slot of data stream f1 at the sending end TSN switch is: 344.96us

[0201] The closing time of the second time slot of data stream f1 at the sending end TSN switch is: 349.056us

[0202] The sending time of the first data packet of data stream f2 is: 50us

[0203] The time when the first data packet of data stream f2 arrives at the TSN switch at the sending end is: 90.96us

[0204] The start time of the first time slot of data stream f2 at the sending end TSN switch is: 94.96us

[0205] The closing time of the first time slot of data stream f2 at the sending end TSN switch is: 99.056us

[0206] 3) Use the CQI prediction model to predict the CQI of data streams f1 and f2 at the scheduling time. Suppose the input matrix of f1 is:

[0207]

[0208] The input CQI historical values ​​are normalized to obtain a normalized input matrix:

[0209]

[0210] Initialize the connection weights ω between neurons i，j The values ​​of are shown in Table 2. The bias θ of each neuron i As shown in Table 3.

[0211] Table 2

[0212]

[0213] Table 3

[0214]

[0215] Calculate the input and output of the neurons in the first and second layers, as shown in Table 4.

[0216] Table 4

[0217]

[0218] The output of neuron 7 is the standardized CQI prediction value at the f1 scheduling time, which is de-standardized to obtain the CQI prediction value at the f1 scheduling time.

[0219]

[0220] Located in At the time when the base station obtains the actual CQI value of f1 scheduling, it is 9, and the error is:

[0221]

[0222]

[0223] α is set to 0.3, which satisfies the error acceptance range. As above, let the input matrix of f2 be:

[0224]

[0225] Then the predicted CQI value at the f2 scheduling time is:

[0226]

[0227] 3) According to the mapping table between CQI and MCS, the modulation and coding strategy indexes MCSIndex1 and MCSIndex2 of the data streams f1 and f2 are obtained, and then according to the mapping table between TBS and MCS, the transport block sizes TBS1 and TBS2 of the data streams f1 and f2 are obtained.

[0228] Table 5 is a mapping table between CQI and MCS, Table 6 is a mapping table between MCSIndex and MCS, and Table 7 is a mapping table between MCSIndex and TBS.

[0229] Table 5

[0230]

[0231] Table 6

[0232]

[0233]

[0234] Table 7

[0235]

[0236] From the table we can get:

[0237]

[0238] According to formula (26), the number of RBs NRB allocated to data streams f1 and f2 is calculated i .

[0239]

[0240] According to formula (27), the 5G ultra-reliable low-latency communication air interface data transmission rate of data streams f1 and f2 is calculated and

[0241]

[0242] The 5G system transmission delay budget of data streams f1 and f2 is calculated according to formula (28).

[0243]

[0244] SDNC calculates NW-TT and the gating list of the TSN switch at the receiving end based on the gating list of the TSN switch at the sending end, the transmission delay budget of the 5G system, and the transmission delay of the TSN wired link.

[0245] NW-TT:

[0246] The time when the first data packet of data stream f1 arrives at NW-TT is: 2181.776us

[0247] The start time of the first time slot of data stream f1 at NW-TT is: 2185.776us

[0248] The closing time of the first time slot of data stream f1 at NW-TT is: 2189.872us

[0249] The time when the second data packet of data stream f1 arrives at NW-TT is: 2481.776us

[0250] The start time of the second time slot of data stream f1 at NW-TT is: 2485.776us

[0251] The closing time of the second time slot of data stream f1 at NW-TT is: 2489.872us

[0252] The time when the first data packet of data stream f2 arrives at NW-TT is: 2190.816us

[0253] The start time of the first time slot of data stream f2 at NW-TT is: 2194.816us

[0254] The closing time of the first time slot of data stream f2 at NW-TT is: 2198.912us

[0255] TSN switch at the receiving end:

[0256] The time when the first data packet of data stream f1 arrives at the receiving end TSN switch is: 2230.832us

[0257] The start time of the first time slot of data stream f1 at the receiving end TSN switch is: 2234.832us

[0258] The closing time of the first time slot of data stream f1 at the receiving end TSN switch is: 2238.928us

[0259] The time when the second data packet of data stream f1 arrives at the TSN switch at the sending end is: 2530.832us

[0260] The start time of the second time slot of data stream f1 at the receiving end TSN switch is: 2534.832us

[0261] The closing time of the second time slot of data stream f1 at the receiving end TSN switch is: 2538.928us

[0262] The time when the first data packet of data stream f2 arrives at the TSN switch at the sending end is: 2239.872us

[0263] The start time of the first time slot of data stream f2 at the receiving end TSN switch is: 2243.872us

[0264] The closing time of the first time slot of data stream f2 at the receiving end TSN switch is: 2247.968us

[0265] A person skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, the steps of the method can be implemented. The storage medium, such as ROM / RAM, a disk, an optical disk, etc.

[0266] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A 5G-TSN joint scheduling method based on delay budget and SMT, characterized by: The following steps are involved: Step 1: Build a scenario based on the 5G-TSN networking architecture; Step 2: The software-defined network controller SDNC obtains TSN network topology, data flow information, and 5G wireless channel information; Step 3: The centralized network configuration CNC determines the scheduling constraints and brings them into the SMT solver to find the TSN switch and DS-TT gating list on the sender side; Step 4: In the first centralized network configuration, CNC1 brings the CQI historical value into the CQI prediction model to obtain the CQI prediction value at the scheduling time; Step 5: SDNC maps the channel quality to the air interface transmission rate according to the mapping table between CQI and modulation and coding strategy MCS and the mapping table between transport block size TBS and MCS; Step 6: SDNC calculates the 5G system transmission delay budget based on the maximum number of retransmissions corresponding to the set channel quality; Step 7: SDNC calculates NW-TT and the receiving-end TSN switch gating list based on the TSN switch gating list, 5G system transmission delay budget, and TSN wired link transmission delay; Step 8: SDNC sends the scheduling list configuration information to the two centralized network configurations CNC1 and CNC2; Step 9: CNC1 and CNC2 send the gating list to the sending end TSN switch and the receiving end TSN switch respectively.

2. The 5G-TSN joint scheduling method based on delay budget and SMT according to claim 1 is characterized in that: The 5G-TSN network architecture described in step 1 is as follows: A TSN transmitter terminal is connected to a transmitter TSN switch and enters the 5G system through the device-side time-sensitive network converter DS-TT. The 5G system is connected to a receiver TSN switch through the network-side time-sensitive network converter NW-TT and finally connected to a TSN receiver terminal. The 5G-TSN joint scheduling calculation based on delay budget and SMT is performed at the software-defined network controller SDNC.

3. The 5G-TSN joint scheduling method based on delay budget and SMT according to claim 1 is characterized in that: In step 2, when the transmitter wants to send a TSN stream to the receiver, the centralized network configuration CNC1 and CNC2 discover the network topology of TSN according to the link discovery protocol LLDP, and forward the network topology information of TSN to SDNC; the terminal device sends the sending cycle, data frame size, priority and delay requirement information of each data stream to the centralized user configurator through the user configuration protocol, and the centralized user configurator sends the data stream information to CNC1 and CNC2 through the user network interface, and CNC1 and CNC2 forward the data stream information to SDNC.

4. The 5G-TSN joint scheduling method based on delay budget and SMT according to claim 1 is characterized in that: In step 3, CNC1 adds constraints to the SMT solver to constrain the value range of the variables so that they meet the limitations of network resources and the transmission quality requirements of data streams. The constraints are as follows: 1) The TSN switch gating cycle Schedcycle is the lowest common multiple of the transmission cycles of all data flows passing through the TSN switch: Where F represents the set of data flows flowing through the TSN switch, T i Represents the data flow f i The period of LCM means taking the least common multiple of all elements in the brackets; 2) The number of transmission time slots n of data stream i in the TSN switch gating cycle i It is equal to the TSN switch gating cycle divided by the data flow sending cycle: 3) Data flow f i The sending time of the mth data packet must be between the m-1th sending cycle and the mth sending cycle. The data flow f i The time difference between the sending times of two consecutive data packets is equal to the sending cycle: 4) Data flow f i Each data packet must be received by DS-TT within the maximum transmission delay of the TSN domain on the sender side: The maximum transmission delay of the TSN domain on the sender side meets the following requirements: Among them, N i,priority The priority is greater than or equal to the data flow f i The number of data streams, l i is the data stream f i Packet size, R TSN is the sending rate of the TSN switch, D pro is the processing delay of the TSN switch, R Wired is the wired link transmission rate; 5) The transmission time slots in the gating cycle of the TSN switch cannot overlap, that is, the data flow f i The start time of a transmission time slot must be greater than or equal to the data flow f j The end time of a transmission time slot of or the end time is less than or equal to the data flow f j The start time of a transmission slot: Each port can only have one message entering the link at a time, and the messages entering the same link cannot overlap in the time domain, that is, the data flow f i The sending time of the mth data packet must be greater than or equal to the data flow f j The time when the (m-1)th data packet arrives at the TSN switch: 7) Data packets must be transmitted within the corresponding transmission time slot window, and the data flow f i The start time of the mth transmission time slot in the gating cycle of the TSN switch at the sending end cannot be greater than the time when the message starts to be transmitted, and the end time of the transmission time slot cannot be less than the time when the data packet ends transmission: 8) Data Flow f i When the data packet arrives at the sending TSN switch, the transmission time slot window must be within the corresponding scheduling period, and the data flow f i After the data packet arrives at the sending end TSN switch, it can only choose the arrival time T i After the transmission time slot of the first data packet is determined, the transmission time slots of other data packets in the same data stream are also determined, that is, the following is satisfied: CNC1 runs the SMT solver and tries all possible values ​​of the variables in the solution space in turn. If no feasible solution is found in the first round of solution, the information that the solution is not schedulable is returned. If a feasible solution is found, the maximum transmission delay of the TSN domain on the sender side of each data stream is reduced, the constraint conditions are modified, and the solution is performed again. This process is repeated until no feasible solution is found, and the result of the last solution is output. The maximum transmission delay of the TSN domain on the sender side is adjusted as follows: in is the data stream f i The j-th solved maximum transmission delay in the TSN domain on the sender side.

5. The 5G-TSN joint scheduling method based on delay budget and SMT according to claim 1 is characterized in that: In step 4, CNC1 brings the CQI historical value into the CQI prediction model to obtain the CQI prediction value at the scheduling time; the CQI prediction model uses a feedforward neural network, which consists of an input layer, a hidden layer, and an output layer. W1 and W2 are the connection weight matrices of the corresponding layers; the CQI historical values ​​of n moments before time t are selected as samples to train the CQI prediction model; the data stream f i The time when the mth data packet arrives at the UE As the scheduling moment on the 5G side.

6. The 5G-TSN joint scheduling method based on delay budget and SMT according to claim 5 is characterized in that: The steps for training the CQI prediction model are as follows: 1) The CQI historical values ​​of n moments before moment t are used as the input of the CQI prediction model, and its input matrix is: 2) Standardize the input data so that the processed data conforms to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1; The input CQI historical value is normalized according to the following formula: Wherein, μ is the average value of the CQI historical values ​​input at n moments, and σ is the standard deviation of the CQI historical values ​​input at n moments; 3) Initialize W1, W2, and the connection weight ω between every two neurons i,j The value is initialized to a random number between -1.0 and 1.0, and the bias θ of each neuron i It is also initialized to a random number between -1.0 and 1.0; 4) Calculate the input and output of the first and second layer neurons based on the connection weights and the bias of the neurons, as shown in the following formula: The activation function uses the logistic function, and the neuron output is: 5) The output of the last neuron is the standardized CQI prediction value at the scheduling time, which is Denormalize it to get the CQI prediction value cqi at the scheduling time t,pr , the denormalization formula is as follows: 6) Standardize the actual CQI value at time t to obtain Calculate the error between the final output CQI prediction value and the actual CQI value. The error calculation of the last neuron (1+2n) is as follows: 7) The error obtained needs to be propagated from back to front. The error of the neuron j in the previous layer can be calculated by the errors of all neurons k in the next layer connected to it. The specific formula is: AND j =The j (1-O j )∑ k ω j,k AND k 8) Calculate the error of all neurons in the hidden layer, and then connect the weights ω i,j With the neuron bias θ i Adjustment, the adjustment formula is as follows: oh i,j =ω i,j +(I)O i E j i j =θ j +(l)E j Where l is the learning rate, which is set to a constant between 0 and 1. After the adjustment is completed, if the final output error is less than the acceptable range or the number of iterations h reaches the threshold, the training ends; otherwise, the number of iterations h is increased by 1, and the training is turned to step 4) to continue using the current sample for training. The error acceptance range is determined by the following formula: E P =|cqi t -cqi t,pr | AND P <αE R Among them, E R is the error caused by the CQI reporting delay, E P is the error between the actual CQI value and the predicted value. P <αE R The training ends when α is a constant between 0 and 1; The moment when the first data packet of each data stream arrives at the UE is taken as its 5G side scheduling moment. The CQI historical values ​​of n moments before the scheduling moment are selected as the input of the CQI prediction model to predict the CQI at the scheduling moment. The CQI prediction value at the scheduling moment of the first data packet is used for scheduling the data packets after each data stream.

7. The 5G-TSN joint scheduling method based on delay budget and SMT according to claim 1, characterized in that: In step 5, after CNC1 obtains the CQI prediction value at the scheduling time, it uploads it to SDNC, and SDNC obtains the data flow f according to the mapping table between CQI and MCS. i Modulation and coding strategy index MCSIndex i , and then according to the mapping table between TBS and MCS, we get the data stream f i The transport block size TBS i ; The value assigned to data stream f is calculated by the following formula i The number of RBs (resource blocks) NRB i : The data flow f is calculated by the following formula i 5G ultra-reliable low-latency communication air interface data transmission rate Among them, B RB is the resource block bandwidth, P tx is the transmit power, N0 is the unilateral noise spectral density, and g is the path loss. All of them take fixed values, and only the impact of CQI is considered.

8. The 5G-TSN joint scheduling method based on delay budget and SMT according to claim 1, characterized in that: In step 6, the 5G system transmission delay budget is determined by the data packet size, air interface transmission rate, number of retransmissions, and retransmission delay, as shown in the following formula: K i The value of the data stream f i The CQI prediction value at the scheduling time is set as shown in the following formula:

Images