Apparatus, methods, and procedures for mobility robust optimization in wireless communication networks

Abstract

A technical solution is proposed to improve the mobility experience by appropriately adjusting the cell individual offset (CIO) and / or trigger time (TTT) parameters in a wireless communication network. More specifically, information about the number of handover (HO) attempts made between a source network node and each of its neighboring network nodes within a predefined time period and the number of occurrences of each type of HO problem during the HO attempts is collected as a data vector. Then, for each "source network node ‑ neighboring network node" pair, two data sub-vectors are obtained by aggregating the numbers in the data vector at the carrier level and the inter-node level, respectively. Afterwards, an ensemble of two machine learning models uses these two data sub-vectors to predict whether the TTT and / or CIO parameters need to be changed.

Description

Technical Field

[0001] The present disclosure relates generally to the field of wireless communications, and in particular to techniques for optimizing mobility robustness in wireless communication networks by properly tuning mobility parameters such as cell individual offset (CIO) and time to trigger (TTT). Background Art

[0002] Mobility Robustness Optimization (MRO) is a key use case for User Equipment (UE) mobility and plays a vital role in providing stable and reliable connectivity to mobile UEs. In addition, due to the increasing network complexity in 4G and 5G technologies, traffic patterns often change during the day, so mobility parameters need to be constantly adjusted to ensure a seamless connectivity experience for UEs.

[0003] The main purpose of MRO is to reduce radio link failure (RLF) and unnecessary HO (also known as Ping-Pong (PP) HO). When moving from one cell to another, the UE may be affected by RLF and PP HO, which may lead to a poor inter-cell HO experience. This may be due to improper settings of mobility parameters, such as CIO and TTT parameters.

[0004] Therefore, there is a need for a technical solution that will allow the CIO and TTT parameters to be adjusted according to the changes in RLF and PP HO trends caused by time-varying traffic patterns. Summary of the invention

[0005] This Summary introduces a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features of the disclosure, nor is it intended to be used to limit the scope of the disclosure.

[0006] The object of the present disclosure is to provide a technical solution that allows improving the mobility experience by properly tuning CIO and / or TTT parameters in a wireless communication network.

[0007] The above objects are achieved by the features of the independent claims in the attached claims. Further embodiments and examples are apparent from the dependent claims, the detailed description and the drawings.

[0008] According to a first aspect, a network entity in a wireless communication network is provided. The network entity includes at least one processor and at least one memory. The at least one memory stores instructions, which, when executed by the at least one processor, cause the network entity to operate at least as follows. First, the network entity receives a data vector, the data vector including: (i) the number of HO attempts performed between a source network node and at least one neighboring network node within a time period, and (ii) the number of at least one type of HO problems that occurred between the source network node and the at least one neighboring network node during the HO attempt. Then, the network entity performs the following operations on each neighboring network node of the at least one neighboring network node. The network entity obtains a first data subvector by aggregating those numbers (i) and (ii) from the data vector that are mutually correlated at the level of the carriers used by the source network node and the neighboring network node. Next, the network entity uses a first machine learning (ML) model to predict whether a TTT parameter needs to be increased, decreased, or kept unchanged. The first ML model is configured to use the first data subvector as input data. By appropriately adjusting the TTT parameter (ie, deciding whether to increase, decrease, or keep it unchanged), the network entity can reduce the number of occurrences of one or more HO problems (eg, RLF and / or PP HO) in the wireless communication network, thereby improving the overall mobility experience.

[0009] In an exemplary embodiment of the first aspect, the at least one type of HO problem comprises at least one of: too late HO, too early type 1 HO, too early type 2 HO and PP HO. By using the number of occurrences of one or more of these types of HO problems in the first ML model, effective TTT parameter change prediction can be provided.

[0010] In an exemplary embodiment of the first aspect, the network entity is further caused to supplement the data vector with at least one of the following values:

[0011] RLF_RATIO=(TE1+TE2+TL) / (HOatt+TL),

[0012] TF_TS_diff = abs[((TE1+TE2)-TL) / (TE+TL)], and

[0013] PP_RATIO=PP / HOatt,

[0014] Wherein TE1 is the number of occurrences of premature type 1 HO between the source network node and each of the at least one neighboring network node, TE2 is the number of occurrences of premature type 2 HO between the source network node and each of the at least one neighboring network node, TE=TE1+TE2, TL is the number of occurrences of too late HO between the source network node and each of the at least one neighboring network node, PP is the number of occurrences of PP HO between the source network node and each of the at least one neighboring network node, and HOatt is the number of HO attempts made between the source network node and each of the at least one neighboring network node. By using these additional values ​​in the first ML model, the efficiency of TTT parameter change prediction can be improved.

[0015] In an exemplary embodiment of the first aspect, the first ML model is configured as a dual deep Q network (DDQN). By using DDQN, the efficiency of TTT parameter change prediction can be improved.

[0016] In an exemplary embodiment of the first aspect, the network entity is further enabled to perform the following operations for each neighboring network node of at least one neighboring network node. The network entity obtains a second data sub-vector by aggregating those numbers (i) and (ii) in the data vector that are mutually associated at the inter-node level for a pair of the source network node and the neighboring network node. If the prediction result of the first ML model is that the TTT parameter needs to remain unchanged, the network entity predicts whether the CIO parameter needs to be increased, decreased, or remained unchanged by using the second ML model. The second ML model is configured to use the prediction results of the first ML model and the second data sub-vector as input data. Therefore, the CIO parameter change prediction can be triggered in response to a "no change" result for the TTT parameter prediction. Similarly, by appropriately tuning the CIO parameter (i.e., deciding whether to increase, decrease, or remain unchanged), the network entity can reduce the number of occurrences of one or more HO problems (e.g., RLF and / or PP HO) in the wireless communication network, thereby improving the overall mobility experience.

[0017] In an exemplary embodiment of the first aspect, the second ML model is also configured as DDQN. By using DDQN, the efficiency of CIO parameter change prediction can be improved.

[0018] In an exemplary embodiment of the first aspect, the network entity is further caused to perform the following operations. The network entity receives another data vector, which includes: (i) the number of HO attempts performed between the source network node and at least one neighboring network node in another time period, and (ii) the number of at least one type of HO problems that occurred between the source network node and the at least one neighboring network node during the HO attempts. Then, the network entity calculates the difference between the data vector and the number (ii) of the other data vector. Next, the network entity calculates a reward for the data vector based on whether the difference is greater than, less than, or equal to a threshold. Thereafter, the network entity trains at least one of the first ML model and the second ML model using the reward. By using such a reward in the training phase of the first ML model and the second ML model, their prediction accuracy can be improved.

[0019] According to a second aspect, a method for operating a network entity in a wireless communication network is provided. The method starts with the step of receiving a data vector, the data vector comprising: (i) the number of HO attempts performed between a source network node and at least one neighboring network node within a time period, and (ii) the number of at least one type of HO problems that occurred between the source network node and at least one neighboring network node during the HO attempts. The method then proceeds to the following steps performed for each neighboring network node of the at least one neighboring network node:

[0020] - obtaining a first data sub-vector by aggregating those numbers (i) and (ii) from the data vector that are correlated with each other at the level of the carriers used by the source network node and the neighboring network nodes; and

[0021] - predicting whether the TTT parameter needs to be increased, decreased, or kept unchanged by using a first ML model, the first ML model being configured to use the first data sub-vector as input data.

[0022] By properly tuning TTT parameters (i.e., deciding whether to increase, decrease, or keep them unchanged), the number of one or more types of HO problems (e.g., RLF and / or PP HO) occurring in a wireless communication network can be reduced, thereby improving the overall mobility experience.

[0023] In an exemplary embodiment of the second aspect, the at least one type of HO problem comprises at least one of: too late HO, too early type 1 HO, too early type 2 HO, and ping-pong HO. By using the number of occurrences of one or more of these types of HO problems in the first ML model, effective TTT parameter change prediction can be provided.

[0024] In an exemplary embodiment of the second aspect, the method further comprises the step of supplementing the data vector with at least one of the following values:

[0025] RLF_RATIO=(TE1+TE2+TL) / (HOatt+TL),

[0026] TF_TS_diff = abs[((TE1+TE2)-TL) / (TE+TL)], and

[0027] PP_RATIO=PP / HOatt,

[0028] Wherein TE1 is the number of occurrences of premature type 1 HO between the source network node and each of the at least one neighboring network node, TE2 is the number of occurrences of premature type 2 HO between the source network node and each of the at least one neighboring network node, TE=TE1+TE2, TL is the number of occurrences of too late HO between the source network node and each of the at least one neighboring network node, PP is the number of occurrences of PP HO between the source network node and each of the at least one neighboring network node, and HOatt is the number of HO attempts made between the source network node and each of the at least one neighboring network node. By using these additional values ​​in the first ML model, the efficiency of TTT parameter change prediction can be improved.

[0029] In an exemplary embodiment of the second aspect, the first ML model is configured as DDQN. By using DDQN, the efficiency of TTT parameter change prediction can be improved.

[0030] In an exemplary embodiment of the second aspect, the method further comprises a step performed for each neighboring network node of at least one neighboring network node. A second data sub-vector is obtained by aggregating those numbers (i) and (ii) from the data vector that are mutually associated at the inter-node level for a pair of source network nodes and neighboring network nodes. If the prediction result of the first ML model is that the TTT parameter needs to remain unchanged, the second ML model is used to predict whether the CIO parameter needs to be increased, decreased or remained unchanged. The second ML model is configured to use the prediction results of the first ML model and the second data sub-vector as input data. Therefore, in response to a "no change" result for the TTT parameter prediction, a CIO parameter change prediction can be triggered. Similarly, by appropriately tuning the CIO parameter (i.e., deciding whether to increase, decrease or keep it unchanged), the network entity can reduce the number of one or more types of HO problems (e.g., RLF and / or PP HO) occurring in the wireless communication network, thereby improving the overall mobility experience.

[0031] In an exemplary embodiment of the second aspect, the second ML model is also configured as DDQN. By using DDQN, the efficiency of CIO parameter change prediction can be improved.

[0032] In an exemplary embodiment of the second aspect, the method further comprises the following steps. Another data vector is received, which includes: (i) the number of HO attempts performed between the source network node and at least one neighboring network node within another time period, and (ii) the number of at least one type of HO problems that occurred between the source network node and at least one neighboring network node during the HO attempts. Then, a difference between the data vector and the number (ii) of the another data vector is calculated. Next, a reward for the data vector is calculated based on whether the difference is greater than, less than or equal to a threshold. Thereafter, at least one of the first ML model and the second ML model is trained by using the reward. By using such a reward in the training phase of the first ML model and the second ML model, their prediction accuracy can be improved.

[0033] According to a third aspect, a computer program product is provided. The computer program product comprises a computer-readable storage medium storing computer code. The computer code is executed by at least one processor, so that the at least one processor performs the method according to the second aspect. By using such a computer program product, the implementation of the method according to the second aspect in any network entity, such as the network entity according to the first aspect, can be simplified.

[0034] According to a fourth aspect, a network entity in a wireless communication network is provided. The network entity comprises means for receiving a data vector, the data vector comprising: (i) a number of HO attempts performed between a source network node and at least one neighboring network node within a time period, and (ii) a number of at least one type of HO problems occurring between the source network node and the at least one neighboring network node during the HO attempts. The network entity further comprises one or more means for performing the following operations for each neighboring network node of the at least one neighboring network node:

[0035] - obtaining a first data sub-vector by aggregating those numbers (i) and (ii) from the data vector that are correlated with each other at the level of the carriers used by the source network node and the neighboring network nodes; and

[0036] - predicting whether the TTT parameter needs to be increased, decreased, or kept unchanged by using a first ML model, the first ML model being configured to use the first data sub-vector as input data.

[0037] By appropriately tuning TTT parameters (i.e., deciding whether to increase, decrease, or keep them unchanged), network entities can reduce the number of one or more types of HO problems (e.g., RLF and / or PP HO) that occur in wireless communication networks, thereby improving the overall mobility experience.

[0038] Other features and advantages of the present disclosure will be apparent upon reading the following detailed description and viewing the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The present disclosure is explained below with reference to the accompanying drawings, in which:

[0040] Figure 1 schematically illustrates a premature type 1 handover (HO) event in a wireless communication network;

[0041] Figure 2 schematically illustrates a premature type 2HO event in a wireless communication network;

[0042] Figure 3 schematically illustrates a late HO event in a wireless communication network;

[0043] Figure 4 schematically illustrates a Ping Pong (PP) HO event in a wireless communication network;

[0044] Figure 5 A block diagram of a network entity according to an example embodiment is shown;

[0045] Figure 6 A method for operating according to a first exemplary embodiment is shown. Figure 5 A flowchart of a method for a network entity;

[0046] Figure 7 Shows Figure 6 A possible tabular structure of the data vectors used in the method;

[0047] Figure 8 Schematically explains how to calculate the reward by comparing the data vectors of adjacent time periods T0 and T1;

[0048] Fig. 9 A method for operating according to a second exemplary embodiment is shown. Figure 5 Flowchart of a method of a network entity

[0049] Fig.10 Shown is a Figure 6 Results of Method Flowchart of a method for updating a Time to Trigger (TTT) parameter. DETAILED DESCRIPTION

[0050] Various embodiments of the present disclosure are further described in more detail with reference to the accompanying drawings. However, the present disclosure can be embodied in many other forms and should not be construed as limited to any specific structure or function discussed in the following description. On the contrary, these embodiments are provided to make the description of the present disclosure detailed and complete.

[0051] Based on the detailed description, it is obvious to those skilled in the art that the scope of the present disclosure includes any embodiment disclosed herein, regardless of whether the embodiment is implemented independently or in conjunction with any other embodiment of the present disclosure. For example, the apparatus and method disclosed herein can be implemented in practice by using any number of embodiments provided herein. In addition, it should be understood that any embodiment of the present disclosure can be implemented using one or more elements set forth in the appended claims.

[0052] Unless otherwise stated, any embodiment listed herein as an "example embodiment" is not necessarily to be construed as preferred or having advantages over other embodiments.

[0053] Although numerical terms, such as "first," "second," etc., may be used herein to describe various embodiments, elements, or features, it should be understood that these embodiments, elements, or features should not be limited by the numerical terms. The numerical terms are only used herein to distinguish one embodiment, element, or feature from another embodiment, element, or feature. For example, the first ML model discussed below may be referred to as a second ML model, and vice versa, without departing from the teachings of the present disclosure.

[0054] According to the example embodiments disclosed herein, a user equipment (UE) may refer to an electronic computing device configured to perform wireless communication. UE may be implemented as a mobile station, a mobile terminal, a mobile user unit, a mobile phone, a cellular phone, a smart phone, a cordless phone, a personal digital assistant (PDA), a wireless communication device, a desktop computer, a laptop computer, a tablet computer, a gaming device, a netbook, a smartbook, an ultrabook, a medical mobile device or device, a biometric sensor, a wearable device (e.g., a smart watch, smart glasses, a smart wristband, etc.), an entertainment device (e.g., an audio player, a video player, etc.), a vehicle component or sensor (e.g., a driving assistance system), a smart meter / sensor, an unmanned vehicle (e.g., an industrial robot, a quadcopter, etc.) and its components (e.g., an autonomous vehicle computer), an industrial manufacturing device, a global positioning system (GPS) device, an Internet of Things (IoT) device, an industrial Internet of Things (IIoT) device, a machine type communication (MTC) device, a group of large-scale Internet of Things (MIoT) or large-scale MTC (mMTC) devices / sensors, or any other suitable mobile device configured to support wireless communication. In some embodiments, a UE may refer to at least two juxtaposed and interconnected UEs defined in this way.

[0055] As used in the example embodiments disclosed herein, a network node may refer to a fixed point of communication / communication node of a UE in a specific wireless communication network. More specifically, a network node may be used to connect a UE to a data network (DN) through a core network (CN), and may be referred to as a base transceiver station (BTS) in 2G communication technology, NodeB in 3G communication technology, and evolved NodeB (eNodeB or eNB) in 4G communication technology, and gNB in ​​5G new radio (NR) communication technology. A network node may serve different cells, such as macro cells, micro cells, micro cells, femto cells, and / or other types of cells. A macro cell may cover a relatively large geographic area (e.g., a radius of at least several kilometers). A micro cell may cover a geographic area with a radius of less than two kilometers. A micro cell may cover a relatively small geographic area, such as an office, a shopping mall, a train station, a stock exchange, etc. A femto cell may cover an even smaller geographic area (e.g., a home). Accordingly, a network node serving a macro cell may be referred to as a macro node, a network node serving a micro cell may be referred to as a micro node, and the like.

[0056] According to the example embodiments disclosed herein, a wireless communication network, in which UE and network nodes communicate with each other, may refer to a cellular or mobile network, a wireless local area network (WLAN), a wireless personal area network (WPAN), a wireless wide area network (WWAN), a satellite communication (SATCOM) system, or any other type of wireless communication network. Each of these types of wireless communication networks supports wireless communication according to one or more communication protocol standards. For example, a cellular network may operate according to a global system for mobile communications (GSM) standard, a code division multiple access (CDMA) standard, a wideband code division multiple access (WCDM) standard, a time division multiple access (TDMA) standard, or any other communication protocol standard, a WLAN may operate according to one or more versions of the IEEE 802.11 standard, a WPAN may operate according to an infrared data association (IrDA), wireless USB, Bluetooth, or ZigBee standard, and a WWAN may operate according to a worldwide interoperability for microwave access (WiMAX) standard.

[0057] When a UE in a wireless communication network (e.g., a 5G network) moves from a source cell to a neighboring cell, the source network node (e.g., a gNB) serving the source cell may initiate a handover (HO) for the UE to a neighboring network node (e.g., another gNB) serving the neighboring cell. In other words, the source network node makes a HO attempt in this case. If multiple UEs are about to leave the source cell within a given time period, the source network node will make multiple HO attempts. Not all of the multiple HO attempts may be successful; some of them may fail due to different HO problems. Therefore, the number of HO attempts made by the source network node can be considered as the sum of the number of successful HO attempts and the number of unsuccessful (failed) HO attempts. It should be noted that the ratio of the number of unsuccessful HO attempts to the number of HO attempts is called the HO failure (HOF) ratio.

[0058] In the example embodiments disclosed herein, a HO problem may refer to any event that occurs only when the HO process (during a HO attempt) cannot be completed in a usual and optimal manner (i.e., as defined by the 3rd Generation Partnership Project (3GPP) for different communication technologies (e.g., 4G and 5G)).

[0059] For example, a HO problem may be related to a Radio Link Failure (RLF) which occurs when the radio link connection is lost for a predetermined time. RLF events can be divided into the following 3 types: Too Late (TL) HO, Too Early Type 1 (TE1) HO, and Too Early Type 2 (TE2) HO. If the mobility parameters are set such that the UE does not report HO measurements on time, the UE may lose the connection with the source network node before the HO procedure is initiated. This is known as a TL HO event, where the UE attempts to re-establish the connection with another (neighboring) network node after the RLF detection timer expires. On the other hand, if the mobility parameters are set to trigger the HO procedure very early, the RLF may occur shortly after the target network node HO. This is known as a TE1 HO event, where the UE attempts to re-establish the connection with the source network node after the RLF detection timer expires. Even if the HO procedure is triggered at the right time, an incorrectly defined Cell Individual Offset (CIO) parameter may cause the UE to switch to the wrong network node, followed by RLF and a re-establishment request to a network node other than the wrong network node or the source network node. This is known as HO to the wrong cell or TE2 HO.

[0060] HO problems may also involve unnecessary HO (e.g., so-called Ping-Pong (PP) HO), which does not cause RLF but increases the amount of control data for the UE in an unnecessary manner. In other words, in the case of PP HO, the UE switches back and forth between two neighboring cells for a short time.

[0061] Figure 1Schematically illustrates a TE1 HO event in the wireless communication network 100. Figure 1 As shown, network 100 includes source network node 102 and neighboring network node 104. Source network node 102 serves source cell 106, and neighboring network node 104 serves neighboring cell 108. Assume that UE 110 (e.g., a mobile phone) moves to an overlapping area of ​​cells 106 and 108, where UE 110 experiences a HO attempt to neighboring network node 104, but faces RLF (see Figure 1 RLF may be caused by, for example, a too low value of a Time to Trigger (TTT) parameter. The TTT parameter is the time period that must satisfy the HO criteria before the HO itself is executed. In response to the RLF, the UE 110 attempts to reestablish a connection with the source network node 102 (see Figure 2 S2 in ).

[0062] Figure 2 Schematically illustrates a TE2 HO event in a wireless communication network 200. Figure 2 As shown, network 200 includes a source network node 202 and two neighboring network nodes 204, 206. Source network node 202 serves source cell 208, and neighboring network nodes 204, 206 serve neighboring cells 210, 212, respectively. Assume that UE 214 (e.g., a mobile phone) moves in an overlapping area of ​​cells 208-212, where UE 214 experiences a HO attempt to neighboring network node 206 but faces RLF (see Figure 2 The RLF may be caused by incorrectly selecting the neighboring network node 206 as the target network node, for example due to signal overlap at the cell edge. In response to the RLF, the UE 214 attempts to establish a connection with the neighboring network node 204 (see Figure 2 S2 in ).

[0063] Figure 3 The TL HO event in the wireless communication network 300 is schematically shown. Figure 3 As shown, network 300 includes a source network node 302 and a neighboring network node 304, wherein source network node 302 serves source cell 306 and neighboring network node 304 serves neighboring cell 308. Assume that UE 310 (e.g., a mobile phone) moves to an overlapping area of ​​cells 306 and 308, wherein UE 310 faces RLF, i.e., UE 310 loses connection with source network node 302 before initiating a HO procedure for neighboring network node 304 (see Figure 3 In response to the RLF, the UE 310 attempts to reestablish a connection with the source network node 304 (see S1 in FIG. 1 ). The RLF may be caused, for example, by a too high value of the TTT parameter. Figure 3 S2 in ).

[0064] Figure 4 Schematically illustrates a PP HO event in a wireless communication network 400, such as Figure 4 As shown, the network 400 includes a source network node 402 and a neighboring network node 404, wherein the source network node 402 serves a source cell 406 and the neighboring network node 404 serves a neighboring cell 408. Assume that a UE 410 (e.g., a mobile phone) moves in an overlapping area of ​​cells 406 and 408 and triggers multiple unnecessary (PP) HOs between the source network node 402 and the neighboring network node 404 in a short period of time. The PP HO may occur due to different factors, such as specific UE behavior, radio frequency conditions in the overlapping area, inappropriate network parameterization, etc.

[0065] It should be noted that Figure 1-Figure 4 The number of occurrences of each adverse event shown in can be minimized by appropriately adjusting the CIO and TTT parameters. More specifically, Figure 1 , Figure 2 and Figure 4 The number of occurrences of each of the TE1, TE2, and PP HO events shown in can be minimized by increasing the TTT parameter and decreasing the CIO parameter, respectively. Figure 3 The TL HO events shown in FIG. 1 can be minimized by reducing the TTT parameter and increasing the CIO parameter.

[0066] The example embodiments disclosed herein provide a technical solution that allows the mobility experience to be improved by appropriately adjusting the CIO and / or TTT parameters in a wireless communication network. More specifically, information about the number of HO attempts made between a source network node and each of its neighboring network nodes within a predefined time period and the number of occurrences of each type of HO problem (e.g., TE1 HO, TE2 HO, TL HO, PP HO) during the HO attempt is collected as a data vector. Then, for each neighboring network node, a first data sub-vector is obtained by aggregating those numbers from the data vector that are mutually associated at the carrier level used by the source network node and the neighboring network node, and a second data sub-vector is obtained by aggregating those numbers from the data vector that are mutually associated at the node level for the pairing of the source network node with the neighboring network node. Thereafter, a first machine learning (ML) model uses the first data sub-vector to predict whether the TTT parameter needs to be increased, decreased, or kept unchanged. If a "no change" result is predicted for the TTT parameter, this causes the second ML model to perform CIO parameter change prediction by using the results of the first ML model and the second data sub-vector as input data.

[0067] It should be noted that the CIO parameter is defined separately for each neighbor relationship and can be expressed as a source node / cell-neighboring node / cell pair (i.e., at the inter-node or inter-cell level). For the TTT parameter, it is a cell-level parameter defined separately for each target Evolved Universal Terrestrial Radio Access (E-UTRA) Absolute Radio Frequency Channel Number (EARFCN) belonging to the source cell. For example, if the source network node has (target) neighboring network nodes across 3 carriers, the source network node will have 3 TTT values, one for each target EARFCN. Each unique relationship between a source network node or source cell and a target EARFCN can be expressed as a source node-carrier or source cell-carrier pair (i.e., at the carrier level used by the source and neighboring network nodes). If the source and target EARFCNs are the same, a co-frequency neighbor scenario occurs; if they are different, a co-frequency neighbor scenario is involved. In view of this, it can be considered that the first data sub-vector is obtained by aggregating data (i.e., number) from the data vector for each source cell-neighboring cell pair, and the second data sub-vector is obtained by aggregating data (i.e., number) from the data vector for each source cell-carrier pair.

[0068] Figure 5 A block diagram of a network entity 500 according to an example embodiment is shown. The network entity 500 is intended to communicate with one or more network nodes (e.g., gNBs) in any of the wireless communication networks described above. The network entity 500 may be implemented as part of a near real-time radio intelligent controller (near RT RIC), but the present disclosure is not limited to this implementation example of the network entity 500. Figure 5 As shown, the network entity 500 includes a processor 502, a memory 504, and a transceiver 506. The memory 504 stores processor-executable instructions 508, which, when executed by the processor 502, cause the processor 502 to perform aspects of the present disclosure, which will be described in more detail below. It should be noted that Figure 5 The number, arrangement, and interconnection of the construction elements constituting the network entity 500 shown in the figure are not intended to be any limitation of the present disclosure, but are merely intended to provide a general idea of ​​how the construction elements are implemented within the network entity 500. For example, the processor 502 may be replaced by several processors, and the memory 504 may be replaced by several removable and / or fixed storage devices, depending on the specific application. In addition, in some embodiments, the transceiver 506 may be implemented as two separate devices, one for receiving operations and the other for sending operations. Regardless of its implementation, the transceiver 506 is intended to be able to perform different operations required for data reception and transmission, such as signal modulation / demodulation, encoding / decoding, etc. In other embodiments, the transceiver 506 may be part of the processor 502 itself.

[0069] Processor 502 can be implemented as a CPU, a general-purpose processor, a single-purpose processor, a microcontroller, a microprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a complex programmable logic device, etc. It should also be noted that processor 502 can be implemented as any combination of one or more of the above. As an example, processor 502 can be a combination of two or more microprocessors.

[0070] The memory 504 may be implemented as a classic non-volatile or volatile memory used in modern electronic computing machines. As examples, non-volatile memory may include read-only memory (ROM), ferroelectric random access memory (RAM), programmable ROM (PROM), electrically erasable PROM (EEPROM), solid-state drive (SSD), flash memory, magnetic disk storage (such as hard disk drives and tapes), optical disk storage (such as CDs, DVDs, and Blu-ray discs), etc. As for volatile memory, examples thereof include dynamic RAM, synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), static RAM, etc.

[0071] The processor-executable instructions 508 stored in the memory 504 may be configured to cause the processor 502 to perform computer-executable program code of aspects of the present disclosure. The computer-executable program code for performing the operations or steps of aspects of the present disclosure may be written in any combination of one or more programming languages, such as Java, C++, etc. In some examples, the computer-executable program code may be in the form of a high-level language or in a pre-compiled form and generated on the fly by an interpreter (also pre-stored in the memory 504).

[0072] Figure 6 A flow chart of a method 600 for operating a network entity 500 according to a first exemplary embodiment is shown.

[0073] The method 600 starts at step S602, where the processor 502 receives a data vector, the data vector including: (i) the number of HO attempts performed between the source network node and at least one neighboring network node in a (last) time period, and (ii) the number of at least one type of HO problems that occurred between the source network node and at least one neighboring network node during the HO attempts. As mentioned above, the types of HO problems may include but are not limited to TE1HO, TE2HO, TLHO, and PPHO. Each network node may be equipped with a counter specific to the type of HO problem, and the counter value may be reported to the network entity 500 periodically (every "n" durations, where n is an optimization cycle period), so that the processor 502 can track the trend change of each type of HO problem by comparing the current data vector with the past data vector, which will be described in detail below.

[0074] In some embodiments, the processor 502 may supplement the data vector in or after step S602 with one or more of the following values ​​(super features):

[0075] RLF_RATIO=(TE1+TE2+TL) / (HOatt+TL),

[0076] TF_TS_diff = abs[((TE1+TE2)-TL) / (TE+TL)], and

[0077] PP_RATIO=PP / HOatt,

[0078] wherein TE1 is the number of occurrences of premature TE1 HO between the source network node and each of the at least one neighboring network node, TE2 is the number of occurrences of TE2 HO between the source network node and each of the at least one neighboring network node, TE=TE1+TE2, TL is the number of occurrences of TL HO between the source network node and each of the at least one neighboring network node, PP is the number of occurrences of PP HO between the source network node and each of the at least one neighboring network node, and HOatt is the number of HO attempts made between the source network node and each of the at least one neighboring network node.

[0079] Figure 7 A possible table structure of the data vector used in method 600 is shown as Figure 7 As shown, in addition to the features TE1, TE2, TL, PP and HOatt, the data vector received at time T0 may also include the above-mentioned super features RLF_RATIO, TF_TS_diff (in modulus) and PP_RATIO, as well as the source cell ID (i.e., Cell_A), the target neighbor cell ID (i.e., Cell_B), the duplex scheme (i.e., frequency division duplex (FDD) / time division duplex (TDD)), the source cell EARFCN, the target neighbor cell EARFCN, TF_KPI_RATIO (i.e., the ratio of the number of occurrences of TE1 HO and TE2 HO to the total number of RLFs) and TS_KPI_RATIO (i.e., the ratio of the number of occurrences of TL HO to the total number of RLFs). The last two super functions are defined as follows:

[0080] TF_KPI_RATIO=(TE1+TE2) / (TE1+TE2+TL),

[0081] TS_KPI_RATION=TL / (TE1+TE2+TL).

[0082] Figure 7The last column ("CIO / TTT") in the table structure of the data vector is intended to be used for the results of the CIO / TTT parameter change prediction (i.e., "increase", "decrease", "no change"). It should be noted that when the TTT parameter predicts a "no change" result, the CIO parameter change prediction is triggered, which will be discussed in more detail below.

[0083] After step S602 , the method proceeds to steps S604 and S606 . According to the performance of the processor 502 , the above steps are executed sequentially or in parallel for each adjacent network node.

[0084] At step S604, the processor 502 obtains a first data subvector by aggregating those numbers from the data vector that are associated with each other at the carrier level used by the source network node and the corresponding neighboring network node (e.g., those numbers of occurrences of TE1 HO, TE2 HO, TL HO, PP HO, or any combination thereof. In other words, the number required for the TTT parameter change prediction from the data vector must be aggregated for each source cell-carrier pairing.

[0085] Next, in step S606 , the processor 502 uses the first ML model to predict whether the TTT parameter needs to be changed (ie, whether it should be increased, decreased, or kept unchanged). The first ML model is configured to use the first data sub-vector as input data.

[0086] In one embodiment, the method 600 includes an additional step, where the processor 502 performs a CIO parameter change prediction for each neighboring network node. More specifically, in response to the "no change" (equivalent to "remain unchanged") result for the TTT parameter prediction in step S606, the CIO parameter change prediction is performed. To this end, the processor 502 obtains a second data sub-vector by aggregating those numbers from the data vector that are mutually associated at the inter-node level for the source network node and the corresponding neighboring network node pairing (for example, those numbers of occurrences of TE1 HO, TE2 HO, TL HO, PP HO or any combination thereof). In other words, the number of CIO parameter change predictions required from the data vector - target cell (neighbor relationship) pairs must be aggregated for each source cell. Thereafter, the processor 502 uses a second ML model to predict whether the CIO parameter needs to be changed (i.e., whether it should be increased, decreased or remained unchanged). The second ML model is configured to use the prediction results of the first ML model and the second data sub-vector as input data.

[0087] The first and second ML models should be pre-trained. Their training can be performed as follows: the features of the data vector (i.e., the above-mentioned counter values ​​or numbers) can be collected for at least 2 days, and then the features can be aggregated according to the optimization period duration and related to the CIO and TTT parameters to obtain the first and second data sub-vectors. The first and second ML models can use the corresponding aggregated features of the data vector to make CIO and TTT parameter change predictions (i.e., "decrease", "increase" or "no change") . Each of the first and second ML models can be configured as a dual deep Q network (DDQN) or any other reinforcement learning model. Each of the first and second ML models can be integrated with a custom reward function, which analyzes the reward gain obtained based on the number of occurrences of each type of HO problem (e.g., RLF and PP HO) for each parameter change suggestion, and drives model retraining and inference. In the inference mode, each of the first and second models considers the corresponding one of the first and second data sub-vectors, and based on it, the ML model recommends the best action to take on the CIO / TTT parameters (i.e., "decrease", "increase" or "no change"). From a model retraining perspective, each of the first and second ML models looks at the changes in the counter values ​​(e.g., TE1, TE2, TL, and PP values) and CIO / TTT parameters noted above across subsequent optimization cycles and learns patterns in the reward gains obtained across each change. Thus, for the first and second data sub-vectors, the first and second ML models evaluate all possible outcomes relative to the predicted reward gains and recommend CIO and TTT parameter changes that achieve the maximum reward.

[0088] Figure 8 Schematically explains how to calculate the reward by comparing the data vectors of adjacent time periods T0 and T1. It should be noted that the data vectors do not necessarily correspond to adjacent time periods; in some other embodiments, they may simply correspond to different time periods. Figure 8 In the example, W1, W2, and W3 are weights assigned to the three super features RLF_RATIO, PP_RATIO, and TF_TS_diff for the purpose of calculating the reward. Based on the comparison of the two data vectors, the overall reward is calculated as follows. Each super feature RLF_RATIO, PP_RATIO, and TF_TS_diff obtained at T1 is compared with the corresponding value at T0. The calculation of individual feature rewards is based on, for example, the following conditions:

[0089] - Condition 1: If T1 value < (T0 value + threshold), assign a positive reward (R1), for example +2;

[0090] - Condition 2: If the T1 value is "within" (T0 value + / - threshold), assign a positive reward (R2), e.g. +1;

[0091] - Condition 3: If T1 value > (T0 value + threshold), assign a negative reward (R3), such as -2;

[0092] The threshold can be set to a small value, such as 0.02.

[0093] The above conditions are checked for all 3 low features, and based on the individual feature rewards, the feature weights are used to obtain a compound reward score, for example, as follows: W1·R1+W2·R2+W3·R3.

[0094] Fig. 9 A method for operating according to a second exemplary embodiment is shown. Figure 5 Flowchart of method 900 of a network entity of a network entity. In general, method 900 corresponds to method 600 performed using a second ML model. Method 900 starts at step S902, which is similar to step S606 of method 600. In the following step S904, processor 502 obtains a first and a second data subvector, as described above. In other words, step S904 corresponds to a combination of step S604 of method 600 and a step consisting of obtaining a second data subvector. If necessary and depending on its performance, processor 502 can obtain the first and the second data subvector in parallel. After step S904, method 900 proceeds to step S906, which is similar to step S606 of method 600. If the first ML model predicts in step S906 that the TTT parameter of one or more neighboring cells does not need to be changed, the prediction result is provided to the second ML model as additional input data for the next step S908 of CIO parameter change prediction. In general, TTT and CIO parameter changes have similar effects on each type of HO problem (e.g., RLF and PPHO) at the source cell-carrier and source cell-neighboring cell levels, respectively. Sufficiently, the first step of prediction involves evaluating neighboring cells for TTT-based variations, i.e., each source cell-carrier pair with a dominant failure class. For a specific cell-carrier pairing with a mixed failure class scenario, the first ML model does not recommend any TTT optimization and the second step of prediction is triggered by using the second ML model.

[0095] As Fig. 9 As described, the reasoning for each of the first and second ML models can be provided to a post-processing component (not shown) in the form of a predicted category for any of the CIO and TTT parameters, i.e., "increase / decrease / no change". Now, the complexity lies in applying each predicted change to the relevant parameter. For the CIO parameter, its change predicted by the second ML model can be made for a specific source cell-neighbor cell pair and can be directly applied to that specific pair. For the TTT parameter, it is set for each source cell-carrier pair and has separate parameters defined for co-frequency / inter-frequency neighbor cells, as shown below:

[0096] Parameter name abbreviation A3 Trigger Time a3TimeToTrigger A3 trigger time RSRP frequency a3TimeToTriggerRsrplnterFreq A3 trigger time RSRQ frequency difference a3TimeToTriggerRsrqInterFreq A5 trigger time a5TimeToTrigger A5 trigger time frequency a5TimeToTriggerInterFreq RSRQ A5 frequency trigger time a5TimeToTriggerRsrqInterFreq

[0097] To set the TTT parameter value, for each corresponding source cell, a unique target carrier frequency group is determined and reasoning is performed for each source cell-carrier combination. If the first ML model recommends a change for a specific source cell-carrier pair, the corresponding TTT parameter applicable to that source cell-carrier pair (same frequency / different frequency) is modified based on the predicted category.

[0098] Fig.10 A flow chart of a method 1000 for updating TTT parameters based on the results of method 600 according to an exemplary embodiment is shown. Similar to methods 600 and 900, method 1000 can be performed by a near-RT RIC. Method 1000 begins at step S1002, where a specific source cell-carrier pair is selected (in other words, a specific EARFCN is selected). Next, method 1000 proceeds to step S1004, where the TTT inference obtained by using method 600 or 900 is checked. If it is predicted that the TTT parameter needs to be increased / decreased, method 1000 proceeds to step S1006, where it is checked whether an inter-frequency or co-frequency scenario is being considered. If an inter-frequency scenario occurs, method 1000 proceeds to step S1008, where it is recommended to modify the following parameters: A3timetotrigger and A5timetotrigger. If a co-frequency scenario occurs, method 1000 proceeds to step S1010, where a type of measurement quality is checked. If the measurement quality is represented by reference signal received power (RSRP), the method proceeds to step S1012, where it is suggested to modify the following parameters: A3timetotriggerRSRPinter-freq and A5timetotriggerInterfreq. If the measurement quality is represented by RSRP and reference signal received quality (RSRQ), the method proceeds to step S1014, where it is suggested to modify the following parameters: A3timetotriggerRSRQinter-freq, A3timetotriggerRSRPinter-freq and A5timetotriggerInterfreq. If the measurement quality is represented by RSRQ only, the method proceeds to step S1016, where it is suggested to modify the following parameters: A3timetotriggerRSRQinter-freq. Referring back to step S1004, if the predicted TTT parameters need to remain unchanged, method 1000 proceeds to step S1018, where the predicted TTT parameters are modified by using method 900 (using Fig. 9 The two ML models shown in the figure obtain CIO reasoning. In the next step S1020, it is recommended to modify the following parameters: CellIndOffNeigh.

[0099] It should be noted that each step or operation of methods 600, 900 and 1000, or any combination of steps or operations, can be implemented by various means, such as hardware, firmware and / or software. As an example, one or more of the above steps or operations can be implemented by processor executable instructions, data structures, program modules and other suitable data representations. In addition, the processor executable instructions embodying the above steps or operations can be stored on a corresponding data carrier and executed by the processor 502. The data carrier can be implemented as any computer readable storage medium, configured to be readable by the at least one processor to execute the processor executable instructions. Such computer readable storage media may include volatile and non-volatile media, removable and non-volatile media removable media. As an example and not limitation, computer readable media include media implemented in any method or technology suitable for storing information. In more detail, practical examples of computer readable media include, but are not limited to, information delivery media, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD), holographic media or other optical disk storage, magnetic tape, cassette, disk storage and other magnetic storage devices.

[0100] Although example embodiments of the disclosure are described herein, it should be noted that any of the various changes and modifications may be made in the embodiments of the disclosure without departing from the scope of legal protection defined by the appended claims. In the appended claims, the word "comprising" does not exclude other elements or operations, and the indefinite article "a" or "an" does not exclude a plurality. The fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

Claims

1. A network entity (500) in a wireless communication network, comprising: at least one processor (502); as well as At least one memory (504) storing instructions, the instructions (508) when executed by the at least one processor (502) causing the network entity (500) to at least: receiving (S602) a data vector comprising: (i) a number of handover (HO) attempts performed between a source network node (102; 202; 302; 402) and at least one neighboring network node (104; 204, 206; 304; 404) during a time period, and (ii) a number of occurrences of at least one type of HO problem between the source network node (102; 202; 302; 402) and the at least one neighboring network node (104; 204, 206; 304; 404) during the HO attempts; For each of the at least one neighboring network node (104; 204, 206; 304; 404): obtaining a first data subvector by aggregating those numbers (i) and (ii) from the data vector that are correlated at the level of carriers used by the source network node (102; 202; 302; 402) and the neighboring network nodes (104; 204, 206; 304; 404); and predicting ( S606 ) whether a time to trigger (TTT) parameter needs to be increased, decreased, or kept unchanged by using a first machine learning (ML) model, the first ML model being configured to use the first data sub-vector as input data; The method is characterized in that the network entity (500) is further configured to: for each of the at least one neighboring network node (104; 204, 206; 304; 404): obtaining (S904) a second data sub-vector by aggregating those numbers (i) and (ii) from the data vector that are mutually associated at an inter-node level for a pair of the source network node (102; 202; 302; 402) and the neighboring network node (104; 204, 206; 304; 404); and If the prediction result of the first ML model is that the TTT parameter needs to remain unchanged, a second ML model is used to predict (S908) whether a cell individual offset (CIO) parameter needs to be increased, decreased, or remained unchanged, and the second ML model is configured to use the prediction result of the first ML model and the second data sub-vector as input data.

2. The network entity (500) according to claim 1, wherein the at least one type of HO problem comprises at least one of the following: Too late HO; Premature type 1HO; Premature type 2HO; and Ping Pong (PP) HO.

3. The network entity (500) of claim 2, wherein the network entity (500) is further caused to supplement the data vector with at least one of the following values: RLF_RATIO=(TE1+TE2+TL) / (HOatt+TL), TF_TS_diff = abs[((TE1+TE2)-TL) / (TE+TL)], and PP_RATIO=PP / HOatt, in TE1 is the number of occurrences of premature type 1 HO between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404), TE2 is the number of occurrences of premature type 2 HO between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404), TE=TE1+TE2, TL is the number of occurrences of too late HO between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404), PP is the number of occurrences of PP HO between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404), and HOatt is the number of HO attempts made between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404).

4. The network entity (500) according to any one of claims 1 to 3, wherein the first ML model is configured as a Dual Deep Q Network (DDQN).

5. The network entity (500) according to any one of claims 1 to 3, wherein the second ML model is configured as a DDQN.

6. The network entity according to any one of claims 1 to 3, wherein the network entity (500) is further configured to: Another data vector is received, comprising: (i) at the source network node (102; 202; 302; 402) and the at least one neighboring network node (104; 204, 206; 304; 404), and (ii) a number of occurrences of the at least one type of HO problem between the source network node (102; 202; 302; 402) and the at least one neighboring network node (104; 204, 206; 304; 404) during the HO attempts; calculating a difference between said data vector and said number (ii) of said further data vector; calculating a reward for the data vector based on whether the difference is greater than, less than, or equal to a threshold; as well as At least one of the first ML model and the second ML model is trained by using the reward.

7. A method (600) for operating a network entity (500) in a wireless communication network, comprising: receiving (S602) a data vector comprising: (i) a number of handover (HO) attempts performed between a source network node (102; 202; 302; 402) and at least one neighboring network node (104; 204, 206; 304; 404) during a time period, and (ii) a number of occurrences of at least one type of HO problem between the source network node (102; 202; 302; 402) and the at least one neighboring network node (104; 204, 206; 304; 404) during the HO attempts; For each of the at least one neighboring network node (104; 204, 206; 304; 404): obtaining (S604) a first data sub-vector by aggregating those numbers (i) and (ii) from the data vector that are mutually correlated at the level of the carriers used by the source network node (102; 202; 302; 402) and the neighboring network nodes (104; 204, 206; 304; 404); and predicting (S606) whether a time to trigger (TTT) parameter needs to be increased, decreased, or kept unchanged by using a first machine learning (ML) model, the first ML model being configured to use the first data sub-vector as input data, Characterized in that the method (600) further comprises, for each of the at least one neighboring network node (104; 204, 206; 304; 404): obtaining (S904) a second data sub-vector by aggregating those numbers (i) and (ii) from the data vector that are mutually associated at an inter-node level for a pair of the source network node (102; 202; 302; 402) and the neighboring network node (104; 204, 206; 304; 404); and If the prediction result of the first ML model is that the TTT parameter needs to remain unchanged, a second ML model is used to predict (S908) whether a cell individual offset (CIO) parameter needs to be increased, decreased, or remained unchanged, and the second ML model is configured to use the prediction result of the first ML model and the second data sub-vector as input data.

8. The method (600) of claim 7, wherein the at least one type of HO problem comprises at least one of the following: Too late HO; Premature type 1HO; Premature type 2HO; and Ping Pong (PP) HO.

9. The method according to claim 8, further comprising: The data vector is supplemented with at least one of the following values: RLF_RATIO=(TE1+TE2+TL) / (HOatt+TL), TF_TS_diff = abs[((TE1+TE2)-TL) / (TE+TL)], and PP_RATIO=PP / HOatt, in TE1 is the number of occurrences of premature type 1 HO between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404), TE2 is the number of occurrences of premature type 2 HO between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404), TE=TE1+TE2, TL is the number of occurrences of too late HO between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404), PP is the number of occurrences of PP HO between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404), and HOatt is the number of HO attempts made between the source network node (102; 202; 302; 402) and each of the at least one neighboring network node (104; 204, 206; 304; 404).

10. The method according to any one of claims 7 to 9, further comprising: receiving another data vector comprising: (i) a number of HO attempts performed between the source network node (102; 202; 302; 402) and the at least one neighboring network node (104; 204, 206; 304; 404) during another time period, and (ii) a number of occurrences of the at least one type of HO problem between the source network node (102; 202; 302; 402) and the at least one neighboring network node (104; 204, 206; 304; 404) during the HO attempts; calculating a difference between said data vector and said number (ii) of said further data vector; calculating a reward for the data vector based on whether the difference is greater than, less than, or equal to a threshold; and At least one of the first ML model and the second ML model is trained by using the reward.

11. A computer program product comprising a computer-readable storage medium, wherein the computer-readable storage medium stores computer code, which, when executed by the at least one processor (502) of the network entity (500) according to any one of claims 1 to 6, causes the at least one processor (502) to perform the method (600) according to any one of claims 7 to 10.

Images