Base station system aisg protocol stack implementation method based on monte carlo tree search

Abstract

The application discloses a base station system AISG protocol stack implementation method based on Monte Carlo tree search and belongs to the field of communication technologies.The base station system comprises a radio frequency remote unit, a baseband processing unit and a remote electrically adjustable antenna, is connected through an optical fiber between the baseband unit (BBU) and the radio remote unit (RRU), is connected through an RS485 coaxial cable line between the RRU and the remote electrically adjustable antenna (RET), transmits and receives messages through HDLC frames internally, and supports a multi-RET cascading scene; after the RRU is powered on, a scan event is sent to the RET device connected to the RRU, and all scanned RET device information is reported to the BBU; the application adopts a fast scanning method based on Monte Carlo tree search, combines the generality of random simulation and the accuracy of tree search, and MCTS executes an algorithm for growing the adaptive search space topology structure of an asymmetric tree, the algorithm more frequently accesses more interesting nodes, and focuses the search time on more relevant nodes.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and specifically to an AISG protocol stack implementation method for a base station system based on Monte Carlo tree search. Background Technology

[0002] With the continuous expansion of mobile communication network systems, the requirements for scanning speed of electrically tunable antenna devices are becoming increasingly stringent. The AISG (Antenna Interface Stand Group) protocol is a standard protocol for achieving interoperability between base station antennas and base station systems. In the AISG standard protocol, device scanning identification is the foundation and prerequisite for communication between the Remote Electrical Tilt Control System (RCS) and the Antenna Line Device (ALD). However, the current AISG standard protocol does not mention the device scanning algorithm and does not provide guidance on the specific methods involved in the algorithm. To address this deficiency in the AISG standard protocol, this study conducts in-depth research on the properties and implementation methods of the ALD scanning algorithm, leading to many meaningful conclusions.

[0003] Real-time control of electrically tunable antenna devices by the RRU is achieved through the AISG standard protocol. The RRU can identify all addressless devices on the interface through device scan messages. The RRU scans devices by sending device scan frames (XID, Exchange Identification), using a unique matching code and mask. When an addressless electrically tunable device receives a scan frame from the RRU, the device performs a bitwise AND operation between its own UID (Unique ID) and the mask in the device's scan frame, and then compares it with the device's UID. If a match is found, the device responds to the RRU's scan frame message; if the match is unsuccessful, it does not respond. If multiple electrically tunable devices match successfully, the RRU will assign different addresses to the devices sequentially for differentiation.

[0004] If a scanning event occurs, the electrically tunable antenna needs to expand the scan code using relevant strategies to narrow the scanning range until a unique ALD is scanned and identified, thus obtaining the correctly matching ALD. The binary tree-based scanning algorithm uses a method of increasing the number of scan bits to narrow the search range. However, the higher the probability of the matching code generated when a collision occurs matching the ALD ID, the fewer the collision period, but the more the idle period. Both the increased collision period and the increased idle period significantly increase the scanning time. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and realize an AISG protocol stack implementation method for base station systems based on the Monte Carlo tree search algorithm. It combines the generality of random simulation with the accuracy of tree search, executing an asymmetric tree algorithm that adapts to the growth of the search space topology. This algorithm visits more interesting nodes more frequently and focuses search time on more relevant nodes. By adaptively selecting nodes with higher probability and expanding the search range accordingly, the algorithm iterates continuously to enable faster device scanning.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] A method for implementing the AISG protocol stack in a base station system based on Monte Carlo tree search is disclosed. The base station system includes a baseband processing unit (BBU), a remote radio unit (RRU), and a remote electrically tunable antenna (RET). The BBU and RRU are connected via optical fiber using the Common Radio Interface (CPRI) protocol. The RRU and RET are connected via RS485 twisted-pair cables, and the RETs are also connected via RS485 to each other, forming a cascaded configuration. The method is characterized by the following steps:

[0008] Step 1: Ensure that the fiber optic connection between the BBU and RRU is normal and that the CPRI link is synchronized;

[0009] Step 2: After power-on, the RRU sends an AISG protocol stack scan event once, where the scanning algorithm is based on the Monte Carlo tree search algorithm;

[0010] Step 3: After the scan is completed, the RRU sends a start-up event. After receiving the start-up event from the RRU, the BBU sends a power-on switch message to the RRU.

[0011] Step 4: After receiving the power switch open message from the BBU, the RRU turns on the power switch and replies with a switch open result message to the BBU;

[0012] Step 5: Configure ADD-RET in the BBU and send a message to the RRU;

[0013] Step 6: The RRU replies with an ADD-RET message to the BBU;

[0014] Step 7: The BBU sends a calibration CALIBRATION message to the RRU;

[0015] Step 8: The RRU performs the calibration CALIBRATION procedure and replies with a message to the BBU.

[0016] The scanning algorithm uses the Monte Carlo tree search algorithm, which is superior to the traditional binary tree algorithm. It adaptively selects nodes with higher probability and expands the search range accordingly, continuously iterating to scan for devices more quickly.

[0017] Among them, the AISG protocol stack scanning algorithm based on Monte Carlo tree search is a decision algorithm, and each simulation consists of the following four steps:

[0018] Selection: Starting from the root node, select a node that urgently needs to be expanded;

[0019] Extension: Add a new child node to the tree, and add it to the node that is best reached during the selection process;

[0020] Simulation: Perform a simulation by selecting actions or strategies until a result or predefined state is reached;

[0021] Backpropagation: After determining the value of the newly added node, the remaining tree must be updated. Therefore, a backpropagation process is performed, which propagates backward from the new node to the root node; during this process, the number of simulations stored in each node increases.

[0022] Step 1. Select:

[0023]

[0024] Among them, V τ This represents the average value of state Si, meaning this function is called recursively until the search tree ends. Starting from the initial node S0, a choice must be made between the following two nodes (assuming there are only two nodes). The selection criterion is the value of UCB(Si). The node that maximizes UCB(S1) is selected as the next node. Initially, UCB1(S1) = UCB1(S2) = ∞, and S1 is selected sequentially.

[0025] Step 2. Expand nodes:

[0026] One or more leaf nodes will be created. Determine if the current node S1 (Current Node) is a leaf node, where a leaf node is one that has not been expanded.

[0027] Step 3. Simulation:

[0028] Next, according to the flowchart, it is necessary to determine whether the coefficient of node S1 being visited is 0. If it is 0, then Rollout needs to be performed (Rollout is essentially taking random actions in each of the following steps until the stopping point).

[0029] Step 4. Backpropagation:

[0030] That is, the value obtained from Rollout is used to update the T and N values ​​of each node on the path. (Then the Rollout result is deleted. The idea of ​​MCTS is to start from S0 and continuously iterate, updating the node values ​​until a certain number of iterations or time is reached.)

[0031] 5. If a certain number of iterations or time has not been reached, continue from the root node to perform steps 1-4.

[0032] Compared to the minimax method, this strategy assumes your opponent is performing at their best and then adjusts the strategy accordingly to maximize gains. Simply put, given a state, you want to find a device that performs a bitwise AND operation between its own UID and the mask in the device's scan frame; if a match is found, the process ends.

[0033] Compared with the prior art, the technical solution proposed in this invention has the following technical effects:

[0034] 1. This invention can effectively reduce the number of scans, greatly improve the efficiency of scanning equipment, and provide a reference for guiding algorithm optimization.

[0035] 2. This invention supports automatic adaptation of AISG protocol stack mode and RS485 tunnel transparent transmission mode, and supports application scenarios of one-to-many and many-to-many cross-control antennas of base stations, and has good scalability.

[0036] 3. This invention combines the generality of random simulation with the accuracy of tree search, and executes an algorithm for an asymmetric tree that adapts to the growth of the search space topology. This algorithm visits more interesting nodes more frequently and focuses the search time on more relevant nodes. The algorithm adaptively selects nodes with higher probability and expands the search range accordingly, iterating continuously to enable it to scan for devices more quickly. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of a base station system based on Monte Carlo tree search according to an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the implementation method of the AISG protocol stack of the base station system based on Monte Carlo tree search according to an embodiment of the present invention.

[0039] Figure 3 This is a schematic diagram of the Monte Carlo scanning algorithm based on an embodiment of the present invention.

[0040] Figure 4 This describes the implementation process of the Monte Carlo tree search algorithm in an embodiment of the present invention.

[0041] Figure 5This is a flowchart of the CRC calculation process according to an embodiment of the present invention. Detailed Implementation

[0042] The technical solutions of the present invention will now be described more completely with reference to the accompanying drawings.

[0043] like Figure 1 As shown, the Radio Remote Unit (RRU) and the Baseband Processing Unit (BBU) are connected via optical fiber and use the Common Radio Interface (CPRI) protocol. The RRU and the Remote Electrically Adjustable Antenna (RET) are connected via RS485 twisted-pair cable, and the RETs can be connected to each other via RS485 to form a cascaded mode.

[0044] The BBU and RRU communicate using a proprietary communication protocol to define the message format. After CPRI fiber synchronization between the BBU and RRU, all communication is conducted via this proprietary protocol once the power is turned on. The RRU and RET communicate via the standard antenna interface standard group AISG2.0 protocol, which is the standard communication protocol that must be followed between the RRU and the electronically controlled equipment.

[0045] like Figure 2 As shown, after power-on, the BBU and RRU synchronize via fiber optic cable. The RRU first performs a scan event based on the Monte Carlo tree search algorithm, as follows: Figure 4 As shown, the Monte Carlo tree search algorithm consists of four steps: node selection, node expansion, simulation, and backpropagation. This scanning algorithm has the advantages of high efficiency and good scalability. After the RRU scan is completed, it sends a start-up event to the BBU, notifying the BBU that the remote radio unit is ready for the relevant service process. After receiving the start-up event from the RRU, the first thing the BBU does is send a power-on switch message to the RRU. Upon receiving this message, the RRU turns on the power supply voltage and prepares to execute the AISG service process. Next, the BBU sends an ADD-RET configuration message to the RRU. Upon receiving this message, the RRU replies with an ADD-RET configuration message to the BBU. The BBU then sends a calibration message to the RRU. Upon receiving this message, the RRU sends a message to the RET device. The RET device executes the calibration process and, upon completion, returns the calibration result to the RRU. The RRU then reports the calibration result to the BBU.

[0046] The Monte Carlo algorithm is based on probability and mathematical statistics theory, with the Central Limit Theorem and the Law of Large Numbers as its main theoretical foundations. The Central Limit Theorem studies the conditions under which the limiting distribution of the sum of random variables is normally distributed, and applies the favorable properties of the normal distribution to solve practical problems. Let ξ1, ξ2, ..., ξ... n Let E[ξ] be a sequence of independently distributed random variables. k]=μ<∞,D[ξ k ]=σ 2 If K = 1, 2, ..., n, then we have Its equivalent form is The law of large numbers is a series of limiting laws in probability theory used to describe the stability of the average outcome of a large number of random phenomena. In the Monte Carlo algorithm, Kolmogorov's law of large numbers is used, which applies to sequences of random variables that are identically distributed. Let ξ1, ξ2, ..., ξ... n Let E[ξ] be a sequence of independently distributed random variables. k If μ < ∞, K = 1, 2, ..., n, then we have Obviously, if ξ1,ξ2,...,ξ n Since it is an abstraction derived from the same population, the sample mean can be determined by the law of large numbers. When n is large, it converges to the population mean μ with probability 1.

[0047] like Figure 3 As shown, the scanning algorithm based on Monte Carlo tree search has the following specific implementation steps: 1. The RRU first sends a disconnect message to the ESC device, as shown below: 7E 01 53 81 76 7E

[0048] As shown above, all numbers are displayed in hexadecimal, starting and ending with 0x7E. 0x01 represents the RET device address. In a Single RET environment, only one ESC device is supported. In a cascaded environment, multiple ESC devices can be supported. If multiple RET devices are supported, the disconnect message can be increased accordingly. 0x53 represents control bits, and 0x81 and 0x76 represent Cyclic Redundancy Check (CRC). The CRC uses a lookup table algorithm, such as... Figure 5 As shown:

[0049] Wherein, CRC is the final verification result, len is the length of the data to be verified, and aisg_hdlc_crc_tab_array is the table data to be queried, and finally the CRC is calculated.

[0050] 2. The RRU sends a reset message to the ESC device: 7E FF BF 81F0 02 07 00A4 F3 7E

[0051] As shown above, this is a reset message sent from RRU to BBU. It starts and ends with 0x7E, 0xFF is the broadcast address, 0xBF is the control bit, 0x81 is FI, 0xF0 is GI, and 0x02, 0x07 and 0x00 are fixed data with a length of 0x05.

[0052] 3. The RRU broadcasts 10 scans of XID frames to the ESC device: 7E FF BF 81F0 08 01 02FF FF 0302FF FF 0C 43 7E

[0053] As shown above, the messages begin and end with 0x7E, 0xFF is the device's broadcast address, 0xBF is the control bit, 0x81 is FI, 0xF0 is GI, 0x08 is 2*n+4, where n is the data length, PI is 1, PL is 0x02, PV is 0xFF 0xFF, PI is 3, PL is 0x02, UID is 0xFF 0xFF, and CRC is 0x0C0x43.

[0054] 4. RRU broadcasts xid frames based on the Monte Carlo tree search algorithm.

[0055] 7E FF BF 81F0 2A 01 13 00 00 00 00 00 00 00 00 00 00 00 00 00 00 0000 00 00 00 03 13 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 80778B 7E

[0056] As shown above, this HDLC frame message begins and ends with 0x7E, 0xFF is the broadcast address, 0xBF is the control bit, 0x81 is FI, 0xF0 is GI, 0x2A is GL, GL = 2*n+4, so n is 19, UID is 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00, PI is 0x03, PL is 0x13, bit mask is 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 80, and CRC is 0x77 0x8B.

[0057] The condition for a device to respond to a scan frame is that the result of bitwise ANDing the device's UID with the MASK in the scan frame matches the last two bits of the device's UID. In this case, the device will respond to the scan's XID frame.

[0058] Building upon this, a Monte Carlo tree search scanning algorithm was added. The purpose of this algorithm is to calculate a reasonable UID and MASK, enabling a more efficient scanning method, and then broadcasting it to the RET device. The specific implementation process of the Monte Carlo tree search algorithm is as follows:

[0059] Step 1: Select:

[0060] In this process, the MCTS algorithm uses a specific strategy to traverse the current tree starting from the root node. This strategy uses an evaluation function to optimize the selection of the node with the highest estimated value. MCTS uses the Upper Confidence Bound (UCB) formula applied to the tree as a strategy for traversing the tree, balancing the trade-offs between exploration and development. During the tree traversal, a node is selected based on some parameters that return the maximum value. The parameters are characterized by the formula typically used for this purpose, as shown below:

[0061]

[0062] Si represents the value of node i, x i Let represent the empirical average value of node i, C be a constant, and t represent the total number of simulations. When traversing a tree during the selection process, the child node that returns the maximum value from the above equation will be selected. During traversal, once a child node that is also a leaf node is found, MCTS jumps to the unfolding step.

[0063] During the selection phase, there are three possibilities for the node being inspected:

[0064] First, all possible actions at this node have been expanded.

[0065] Second, the actions available at this node have not yet been expanded.

[0066] Third, the node scan has been completed and the scan rules have been met.

[0067] If all possible actions have been expanded, then use the UCB formula to calculate the UCB value of all child nodes of the node, find the child node with the largest value and continue to check, iterating downwards repeatedly.

[0068] If there are still child nodes that have not been expanded, then this node is the target node N of this iteration, and find the actions that N has not yet been expanded, and then execute step 2.

[0069] If the node being checked is a node that has already ended, then proceed directly to step 4 from that node;

[0070] The number of visits to each node being checked increments at this stage; after repeated iterations, a node is found at the bottom of the search tree to continue the subsequent steps.

[0071] After the selection phase, find the single node N that most urgently needs to be expanded, and an action A that has not yet been expanded. Create a new node Nn in the search tree as a new child node of N. The state of Nn is the state of node N after performing action A. In order to give Nn an initial score, start the game from Nn and let it proceed randomly until a scan ending is obtained. This ending will be used as Nn's initial score. Use success / failure as the score, which is only 1 or 0.

[0072] After the simulation of Nn ends, its parent node N and all nodes on the path from the root node to N will add their own cumulative scores based on the results of this simulation.

[0073] Each iteration expands the search tree, and the size of the search tree increases with the number of iterations. When a certain number of iterations or time has elapsed, the process ends, and the child node under the root node is selected as the result of this decision.

[0074] Step 2: Extension:

[0075] During this process, a new child node is added to the tree, and then added to the node that is best reached during the selection process.

[0076] Step 3: Simulation:

[0077] In this process, simulations are performed by selecting actions or strategies until a result or predefined state is reached. Step 4: Backpropagation:

[0078] After determining the value of the newly added node, the remaining tree must be updated. Therefore, a backpropagation process is performed, moving from the new node back to the root node. During this process, the number of simulations stored in each node increases. Furthermore, if the simulation result of the new node wins, the number of wins also increases. These steps can be performed using... Figure 4 The steps are used to understand this intuitively.

[0079] Step 5. The RRU sends an address allocation frame to the RRU.

[0080] 7E FF BF 81F0 18 01 13 54 59 30 30 30 30 30 32 33 30 32 31 36 30 303934 42 31 02 01 02FE 76 7E

[0081] As shown above, the addresses start and end with 0x7E respectively. 0xFF is the broadcast address, 0xBF is the control bit, 0x81 is FI, 0xF0 is GI, 0x18 is the parameter length plus 5, 0x01 is PI, 0x13 is PL, and the UID is 0x54. The values ​​are 0x59, 0x30, 0x30, 0x30, 0x32, 0x33, 0x30, 0x32, 0x31, 0x36, 0x30, 0x30, 0x39, 0x34, 0x42, 0x31. 0x02 is PI, 0x01 is PL, 0x02 is PV (the assigned address), and 0xFE and 0x76 are CRC checksums.

[0082] After the address allocation HDLC frame is sent to the ESC device, the ESC device replies with an address allocation response frame to the RRU, and the RRU scanning process ends.

[0083] Obviously, the above embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A method for implementing the AISG protocol stack of a base station system based on Monte Carlo tree search, wherein the base station system includes a baseband processing unit (BBU), a remote radio unit (RRU), and a remote electrically tunable antenna (RET); the BBU and RRU are connected via optical fiber using the Common Radio Interface (CPRI) protocol, the RRU and RET are connected via RS485 twisted-pair cables, and the RETs are connected to each other via RS485, forming a cascaded configuration; characterized in that... Includes the following steps: Step 1: Ensure that the fiber optic connection between the BBU and RRU is normal and that the CPRI link is synchronized; Step 2: After power-on, the RRU sends an AISG protocol stack scan event once, where the scanning algorithm is based on the Monte Carlo tree search algorithm; Step 3: After the scan is completed, the RRU sends a start-up event. After receiving the start-up event from the RRU, the BBU sends a power-on switch message to the RRU. Step 4: After receiving the power switch open message from the BBU, the RRU turns on the power switch and replies with a switch open result message to the BBU; Step 5: Configure ADD-RET in the BBU and send a message to the RRU; Step 6: The RRU replies with an ADD-RET message to the BBU; Step 7: The BBU sends a calibration CALIBRATION message to the RRU; Step 8: The RRU performs the calibration process and replies to the BBU with a message; In step 2, the AISG protocol stack scanning algorithm based on Monte Carlo tree search is a decision algorithm. Each simulation consists of four steps, including: Selection: Starting from the root node, select a node that urgently needs to be expanded; Extension: Add a new child node to the tree, and add it to the node that is best reached during the selection process; Simulation: Perform a simulation by selecting actions or strategies until a result or predefined state is reached; Backpropagation: After determining the value of the newly added node, the remaining part of the tree must be updated. Therefore, a backpropagation process is performed, which moves from the new node back to the root node. During this process, the number of simulations stored in each node increases. The selection process specifically involves the following: In this process, the Monte Carlo Tree Search algorithm traverses the current tree starting from the root node using a specific strategy. This strategy employs an evaluation function to optimize the selection of the node with the highest estimated value. The Monte Carlo Tree Search algorithm uses the upper confidence bound (UCB) formula applied to the tree as the strategy for traversing the tree, balancing the trade-offs between exploration and development. During the tree traversal, a node is selected based on the parameter that returns the maximum value. The formula for the parameter is shown below: ； This represents the value of node i. Let represent the empirical average value of node i, C be a constant, and t represent the total number of simulations. When traversing a tree during the selection process, the child node that returns the maximum value from the above formula will be selected. During the traversal, once a child node is found, which is also a leaf node, the Monte Carlo tree search algorithm jumps to the expansion step. Through continuous iteration, the scanning algorithm continues until the scanned UID & RET UID sent by the RRU are the same as the last two digits of the RET device's UID. The RET device then responds with a scan frame, indicating that the device has been scanned. At this point, the scanning algorithm ends.

2. The method for implementing the AISG protocol stack of a base station system based on Monte Carlo tree search according to claim 1, characterized in that, During the selection phase, there are three possibilities for the node being inspected: First, all possible actions at this node have been expanded. Second, the actions available at this node have not yet been expanded. Third, the node scan has been completed and the scan rules have been met.

3. The method for implementing the AISG protocol stack of a base station system based on Monte Carlo tree search according to claim 2, characterized in that: If all possible actions have been expanded, then use the UCB formula to calculate the UCB value of all child nodes of the node, find the child node with the largest value and continue to check, iterating downwards repeatedly. If there are still child nodes that have not been expanded, then this node is the target node N of this iteration, and find the actions that N has not yet been expanded, and then execute step 2. If the node being checked is a node that has already ended, then proceed directly to step 4 from that node; The number of visits to each node being checked increments at this stage; after repeated iterations, a node is found at the bottom of the search tree to continue the subsequent steps.

4. The method for implementing the AISG protocol stack of a base station system based on Monte Carlo tree search according to claim 3, characterized in that: After the selection phase, find the single node N that most urgently needs to be expanded, and an action A that has not yet been expanded. Create a new node Nn in the search tree as a new child node of N. The state of Nn is the state of node N after it has performed action A. In order to give Nn an initial score, start the game from Nn and let it run randomly until a scan ending is obtained. This ending will be used as Nn's initial score. Use success / failure as the rating, with only 1 or 0.

5. The method for implementing the AISG protocol stack of a base station system based on Monte Carlo tree search according to claim 4, characterized in that: After the simulation of Nn ends, its parent node N and all nodes on the path from the root node to N will add their own cumulative scores based on the results of this simulation. Each iteration expands the search tree, and the size of the search tree increases with the number of iterations. When a certain number of iterations or time has elapsed, the process ends, and the child node under the root node is selected as the result of this decision.

Images