WO2022250454A1

WO2022250454A1 - Method and system for managing robust header compression (rohc) in a wireless network

Info

Publication number: WO2022250454A1
Application number: PCT/KR2022/007431
Authority: WO
Inventors: Veerabhadrappa Murigeppa GADAG; Chintan JOBANPUTRA; Siva Kumar MUMMADI; Swaraj Kumar; Vishal Murgai
Original assignee: Samsung Electronics Co., Ltd.
Priority date: 2021-05-27
Filing date: 2022-05-25
Publication date: 2022-12-01

Abstract

The present invention provides a system and method for managing Robust Header Compression (RoHC). The system and method includes monitoring over a time period a plurality of network characteristics and a plurality of key performance indicator (KPI) parameters and then generating an AI model having co-relation between the KPI's parameters and the network characteristics. The method further comprises analysing a current network characteristic with reference to the co-relation between the KPI's parameters and the network characteristics in the AI model. The method further includes performing by the AI model the enablement or the disablement the Robust header compression (RoHC), based on a result of the analysis, when the current network characteristics are below a threshold value as a result of the analysis.

Description

METHOD AND SYSTEM FOR MANAGING ROBUST HEADER COMPRESSION (ROHC) IN A WIRELESS NETWORK

The present disclosure, in general, relates for managing robust header compression (RoHC) in a wireless communication network. More particularly, the disclosure relates to enablement or disablement of RoHC by using reinforcement learning (RL) mechanism.

Figure 1 depicts an example a state-of-the-art network implemented in 5G and beyond. A user using a VoLTE/VoNR network 101 for communication often complains about mute problem during a call, in case of a network with high bit error rate. In particular, the user experiences a complete silence during audio conversation for a few seconds. This leads to an annoying experience to the user. The main reason, for an occurrence of such phenomenon is due to a sub optimal radio conditions in the network 101 that leads to maximize the packets discard. A Robust Header Compression (RoHC) is a key contributor to such aforesaid phenomenon.

The Robust Header Compression (ROHC) is a standardized method to compress the IP, UDP, RTP and TCP headers of IP packets. The protocol suite of a Next generation Radio Access Networks (RAN) consist of multiple protocol headers including Real-time Transport Protocol (RTP), User Datagram Protocol (UDP), Internet Protocol (IP) etc. These protocol headers are placed on top of an actual payload. Further, most of fields in these headers usually does not change for a given packet stream.

The current standardized RoHC does not provide any mechanism to evaluate the actual radio conditions and it is enabled by default. However, enabling RoHC by default in VoNR/VoLTE might lead to inefficient usage of bandwidth when radio conditions are suboptimal. Thus, the present solution lacks providing a methodology to know whether enabling RoHC is beneficial or not. This leads to suboptimal radio resource utilization.

Figure 2a and 2b depicts an RoHC operation according to the state of the art. As can be seen at figure 2a a transmitter 201 end the RoHC compressor compressed the header thereby transmitting the packet with compressed header to a receiver 203. The RoHC protocol helps in avoiding the repeated transmission of the redundant header fields, which leads to efficient utilization of radio resources. According to a current mechanism, RoHC sender sends initialization packets to receiver 203 in order for the receiver 203 to set the context for decompressing subsequent packets. If context setting packets are lost due to suboptimal radio conditions, then the receiver 203 will not be able to set the context to decompress the subsequent packets. The suboptimal radio conditions may occur when the UE 105 is present in densely populated areas. Further, enabling the RoHC by default in VoNR/VoLTE, M2M, V2X might lead to inefficient usage of bandwidth when radio conditions are suboptimal.

As shown in the figure 2b, the current ROHC Compressor starts with the Initialization (IR) packets at the compressor 201-1 at the transmitter 201. These IR packets are used by the decompressor to set context for decompressing the subsequent packets. The RoHC compressor 201-1 starts in Initialization and Refresh (IR) State of Unidirectional (U) Mode where in it sends IR packets to decompressor 203-1 at the receiver 203. The decompressor 203-1 starts with "no context", receives IR packet and initializes context.

Since there is no feedback for error recovery in U Mode, ROHC uses the timers to lower the compression state and periodically refreshes the context to recover from the error if any. If RoHC context "initialization packets" are lost due to suboptimal radio conditions, the decompressor 203-1 will not be able to set the context to decompress the subsequent packets. As depicted in figure for the "Init Timeout period".

Thus, there is a need a way to determine whether to enable RoHC or not in a current radio condition. Thus, there is a need for a solution that overcomes the above deficiencies.

The present invention provides a system and method for managing Robust Header Compression (RoHC). The system and method includes monitoring over a time period a plurality of network characteristics and a plurality of key performance indicator (KPI) parameters and then generating an AI model having co-relation between the KPI's parameters and the network characteristics. The method further comprises analysing a current network characteristic with reference to the co-relation between the KPI's parameters and the network characteristics in the AI model. The method further includes performing by the AI model the enablement or the disablement the Robust header compression (RoHC), based on a result of the analysis.

To further clarify advantages and features of the present invention, a more particular description of the invention will be rendered by reference to specific embodiments thereof, which is illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. The invention will be described and explained with additional specificity and detail with the accompanying drawings.

The invention can provide a method for avoiding the packet loss due to RoHC state machine optimizations in radio access networks.

These and other features, aspects, and advantages of the disclosure will become better understood when the following detailed description is read with reference to the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:

Figure 1 depicts an example a state-of-the-art network implemented in 5G and beyond.

Figure 2a and 2b depicts an RoHC operation according to the state of the art.

Figure 3 illustrates a network node implemented in 5G and beyond system 100, according to an embodiment of the present disclosure.

Figure 4 illustrates a flow diagram for managing Robust Header Compression (RoHC), according to an embodiment of the present disclosure.

Figure 5 illustrates a training and deployment of the RL mechanism in the EVAL engine, according to an embodiment of the present disclosure.

Figure 6 illustrates a training phase of the RL mechanism in the EVAL engine, according to an embodiment of the present disclosure.

Figure 7 illustrates a flow chart depicting the working of a deep Q learning mechanism implemented in the EVAL engine for the RL mechanism, according to an embodiment of the present disclosure.

Figure 8 illustrates a Deep Q learning the flow of the model, according to an embodiment of the present disclosure.

Figure 9 illustrates a deployment phase of the RL mechanism in the EVAL engine, according to an embodiment of the present disclosure.

Figure 10 illustrates an example architecture of the neural network used in the RL agent, according to an embodiment of the present disclosure.

Figure 11 illustrates a curve depicting a variation of reward with training iterations, according to an embodiment of the present disclosure.

Figure 12 depicts curve showing a variation of KPI list size with training iterations, according to an embodiment of the present disclosure.

Figure 13 illustrates another exemplary diagram of a network node, according to an embodiment of the present disclosure.

Figure 14 illustrates a diagram illustrating the configuration of a terminal in a wireless communication system according to an embodiment of the present disclosure.

Further, skilled artisans will appreciate that elements in the drawings are illustrated for simplicity and may not have been necessarily been drawn to scale. For example, the flow charts illustrate the method in terms of the most prominent steps involved to help to improve understanding of aspects of the disclosure. Furthermore, in terms of the construction of the device, one or more components of the device may have been represented in the drawings by conventional symbols, and the drawings may show only those specific details that are pertinent to understanding the embodiments of the disclosure so as not to obscure the drawings with details that will be readily apparent to those of ordinary skill in the art having benefit of the description herein.

For the purpose of promoting an understanding of the principles of the invention, reference will now be made to the embodiment illustrated in the drawings and specific language will be used to describe the same. It will nevertheless be understood that no limitation of the scope of the invention is thereby intended, such alterations and further modifications in the illustrated system, and such further applications of the principles of the invention as illustrated therein being contemplated as would normally occur to one skilled in the art to which the invention relates.

It will be understood by those skilled in the art that the foregoing general description and the following detailed description are explanatory of the invention and are not intended to be restrictive thereof.

Reference throughout this specification to "an aspect", "another aspect" or similar language means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the disclosure. Thus, appearances of the phrase "in an embodiment", "in another embodiment" and similar language throughout this specification may, but do not necessarily, all refer to the same embodiment.

The terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process or method that comprises a list of steps does not include only those steps but may include other steps not expressly listed or inherent to such process or method. Similarly, one or more devices or sub-systems or elements or structures or components proceeded by "comprises... a" does not, without more constraints, preclude the existence of other devices or other sub-systems or other elements or other structures or other components or additional devices or additional sub-systems or additional elements or additional structures or additional components.

Figure 3 illustrates a network node 101 implemented in 5G and beyond system 100, according to an embodiment of the present disclosure. Further, the reference numerals were kept the same wherever applicable throughout the disclosure for ease of explanation.

According to an embodiment, a network node 101 is implemented with a EVAL engine 301 as depicted in figure 3. The network node 101 is a system implemented in a 5G and beyond system 100. The network node 101 may be alternatively referred to as a system without deviating from the scope of the disclosure. According to an embodiment, the system 101 may be configured to determine whether to enable RoHC or not based on a current radio channel condition. The radio channel condition may alternatively be referred to as network characteristics throughout the description without deviating from the scope of the disclosure. Further, the system 101 may be implemented in any radio access network device.

The EVAL engine 301 may be deployed in a virtual central unit (vCU)/an open radio access network (o-RAN) 103 of the network node 101. The EVAL engine 301 is implemented with a reinforcement learning model that runs a reinforcement learning (RL) mechanism using the various key performance indicators (KPIs) 303 as an input vectors data set. Further, the EVAL engine 301 may be alternatively referred to as a reinforcement learning (RL) engine without deviating from the scope of the disclosure. Furthermore, the various KPIs 303 which is used as an input vectors data set are a pre-classified set of KPI parameters. According to an embodiment, the decision logic in EVAL engine 301 decides whether to run RoHC ON or OFF.

According to an embodiment, the main concept is to avoid large packet loss due to RoHC for the VoLTE or VoNR calls when radio channel conditions are suboptimal. The RL-based mechanism at the radio access network device may be configured to dynamically enable or disable RoHC in the suboptimal radio condition. A deep RL approach is utilized to decide on the ROHC on/off depending on the KPIs. The EVAL engine 301 may be able to learn a correct policy for accurately enabling RoHC on the basis of a feedback reward mechanism. Further, the EVAL engine 301 is trained using the deep Q learning method and a dense neural network to correlate the state of KPIs to action for enabling/disabling RoHC. The detailed working of the same will be explained in the forthcoming paragraphs

Figure 4 illustrates a flow diagram for managing Robust Header Compression (RoHC), according to an embodiment of the present disclosure. Figure 4 shows a method 400 that is implemented at the network node 101. The processors 107 included in the network node 101 performs the method 400. Further, the explanation of method 400 will be made with reference to figure 3.

Initially, at step 401, method 400 monitors one or more network characteristics and one or more key performance indicator (KPI) parameters over a time period. As an example, the KPI's may include, but not limited to, average DL MCS, average UL MCS, average VoLTE CQI, average TA, average DL PRB Utilization, average UL PRB Utilization, UL Interference Power, Load Histogram, ERAB Success Rate, Session Setup Success Rate, UL Residual BLER, Average Active UE-QCI, DLReceivedSubband10CQI8, DLReceivedSubband10CQI5, SrsSilencedUpPTSRssi13Tot, NBIoT_ContentionResolutionForAccess. Further, the network characteristics may arise due to UE in a densely populated region and the like. According to an embodiment, information about the KPI's and the various network characteristics associated with the radio conditions may be provided by network 100.

Thereafter, the monitored KPI parameters and network characteristics are fed to the EVAL engine 301 for generating the AI model that decides whether to enable or disable the RoCH. At step 403, the method 400 includes generating an AI model having co-relation between the KPI's parameters and the network characteristics. In an implementation, a huge number KPI's are fed to the EVAL engine 301. The RL model uses deep Q learning for training one or more KPI's by performing several iterations. In each iteration, the EVAL engine 301 was trained.

Figure 5 illustrates a training and deployment of the RL mechanism in the EVAL engine, according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the RL mechanism is divided into an EVAL engine deployment phase 500-1 and EVAL engine training phase 500-3. In the EVAL engine training phase 500-3, the EVAL engine 301 may be fed with the pre-classified data 501. A result associated with the pre-classified set of KPI parameters obtained as an output of the training phase is used to calibrate the RL Engine. This result is utilized at the deployment phase 500-1 for generating the AI model dynamically. The generated AI model at the RL interference 511 is defined with an action of RoHCon 513 or RoHCoff 515.

Figure 6 illustrates a training phase of the RL mechanism in the EVAL engine, according to an embodiment of the present disclosure. Figure 6 will be explained with reference to figure 5. According to an embodiment, a rewards agent 503 is the sole decision-maker and learner. The reward agent 503 takes the action to enable ROHC based on the KPI list 501. The reward agent’s 503 decision is based on a policy that is learned over time with the help of reinforcement learning.

According to an embodiment, the state i.e the KPI list of the network characteristics is determined by the training sample 501. As shown in figure 6, each training sample xt is a RoHC KPI list containing the parameters that are important in determining whether RoHC should be on or off. In each training cycle of the RL algorithm, a KPI list 501 is provided to the reward agent 503 at step 601. This KPI list 501 is obtained from a dataset that is pre-stored may be in memory.

According to an embodiment, the KPI list 501 may keep on changes according to the requirements of the network. The reward agent 503 may be capable to handle such change. The reward agent 503 on evaluating the state (KPI list) may decide on the action

t to be taken (RoHC on/off ) as shown in block 603. The action taken

t may be given by the expression 1 below.

---- (1)

According to an embodiment, a reward rt is feedback from the EVAL engine output 301 by which success or failure of the reward agent’s 503 actions may be computed. The agent's decision is compared with the decision in the training dataset. If agent's action (

t) is equal to the dataset label (yt), then reward (rt) is positive. If action and label are not equal then the reward (rt) is negative. The reward calculation is based on the accuracy of the prediction of the enablement or the disablement of the RoHC and based on an effect of the enablement or the disablement of the RoCH, the reward is rewarded with a positive reward and a negative reward. The reward (rt) is mathematically expressed by equation 2. Thus, learning the policy through the RL engine over a time period for deciding RoHC to be enabled or disabled is based on the prediction of the enablement or the disablement of the RoHC and the reward received. The reward agent 503 learns a policy for deciding RoHC to be on/off is given by equation 3. Thus, the result as shown in equation 3 associated with the pre-classified set of KPI parameters obtained as an output of the training phase is used to calibrate the RL Engine.

---- (2)

----- (3)

Thus, the main objective of the reward agent 503 is to find a suitable RoHC action model which would increase the total cumulative reward of the reward agent 503. It learns via interaction and feedback. Thus, the prediction for the enablement or the disablement of the RoHC by the RL engine is based on the feedback for maximum reward as the output of the EVAL engine 301. Thus, a correlation between the fed KPI list i.e the state, the corresponding action, a maximum reward becomes a basis for determining enablement or disablement of the RoCH. As an example, the total reward may also be referred as Q-value.

The reward agent 503 keeps on learning continuously in an interactive environment (KPI list is continuously given) from its own actions and experiences. Thus, the policy, the reward, and the list of KPI parameters are updated according to the accuracy of the prediction of the enablement or the disablement of the RoHC. Further, the policy of the agent depends on Q value. Till now the explanation has been made for the training phase and reward feedback mechanism. Further, the explanation will be made for establishing a correlation between the KPI and the network characteristics for geneartig the AI model.

Figure 7 illustrates a flow chart depicting the working of a deep Q learning mechanism implemented in the EVAL engine for the RL mechanism, according to an embodiment of the present disclosure. The reward agent 501 may perform a sequence of actions that may eventually generate the maximum total reward. This total reward is also called the Q-value. As an example, equation 4 at block 709 states that the Q-value yielded from being at state　St　and performing an action 　is the immediate reward rt plus the highest Q-value possible from the next state　st+1. Adjusting the value of gamma will diminish or increase the contribution of future rewards.

---- (4)

According to an embodiment, over the time the Q values in the Q table will converge to the optimal policy. Thereafter, the reward (Q1, Q2쪋 Q4) to each of the corresponding correlated plurality of network characteristics with the plurality of KPI parameters are assigned. After, assigning the reward, a log 701 i.e. a Q table comprising the correlated plurality of network characteristics with the plurality of KPI parameters and the assigned reward thereof is generated. An example, the Q table is used by the agent to make the decision of RoHC on/off based on the KPI list. Thereafter, a sub plurality of the KPIs is shortlisted from the plurality of the KPI's parameters for a performance of an action defined by the enablement or disablement of the RoHC based on the correlation.

Figure 8a illustrates a Deep Q learning the flow of the model, according to an embodiment of the present disclosure. As the KPI list includes a huge number of KPIs. It is extremely difficult for a Q table to converge at an optimal solution. Therefore, a deep RL is used for obtaining a converged list of KPI. In deep RL, the RL agent learns a policy πθ which is learned using a neural network, where θ is the model trainable parameter as shown in Figure 8b. As an example, deep RL algorithms like deep Q learning, Actor critic algorithm, etc can be used. In case a neural network is utilized for generating the AI model, it takes the KPIs as input and predicts the action (RoHC on/off). The neural network can be dense neural network (DNN), convolution neural network (CNN and recurrent neural network (RNN). Thus, depending on the KPI list size, datatype and availability of data, an appropriate neural network architecture or AI model may be generated which is utilised in the deployment phase.

According to the further embodiment, after a fixed interval of time, the reward agent 501 may analyze the efficacy of the KPIs in the shortlisted KPI list. Thus, the KPIs which are not useful in making the decision is removed. Further, a new KPIs list may also be added. This way the reward agent 503 may be able to explore new KPIs.

Continuing with the method 400, at step 405, the system 101 analyses a current network characteristic with reference to the co-relation between the KPI's parameters and the network characteristics in the AI model. Thereafter, the generated AI model is deployed for analysing the current network characteristics thereby performing enablement or disablement of the RoHC. Figure 9 illustrates a deployment phase of the RL mechanism in the EVAL engine, according to an embodiment of the present disclosure. The EVAL engine 301 is fed with a fixed KPI parameter and current network characteristics in the deployment phase.

Further, according to an embodiment, in the deployment phase, the reward agent 503 learns the policy πθ which is a mapping function as expressed in the equation 5.

πθ: S → A --- (5)

where πθ(st) denotes the action to be performed by the agent in state st. The actions are RoHC on/off and the state is the KPI list of the RoHC dataset.

The policy πθ is the learning that the agent has done over time, with each KPI list that the reward agent 503 determines, it takes an action on the basis of the policy learnt by it. On the basis of the action taken and subsequent reward received the model parameter θ is updated.

Thus, the KPI parameters are updated while training, however, the KPI parameters are fixed in the deployment mode. Further, the EVAL engine 301 also updates the KPI list. It will eliminate the KPIs which are not proving effective for RoHC prediction and will simultaneously explore new KPIs.

Accordingly, the sub plurality of KPI parameters that obtained by shortlisting, are the relevant KPI parameters for the deployment of the AI model for enablement/disablement of the RoCH. Further, the determination of list of KPI parameters and a validation of the monitored plurality of the network characteristics performed by the reinforcement learning (RL) engine 301. RL engine 301 is able to predict when to enable RoHC with 98.9% accuracy, reducing packet retransmissions and power consumption.

Continuing with the method 400, at step 407, the system 101 performs by the AI model the enablement or the disablement the Robust header compression (RoHC), based on a result of the analysis. And when the current network characteristics are below a threshold value as a result of the analysis.

According to an embodiment, when the system 101 decides to enable the Robust header compression(RoHC) based on the result of the analysis, the RoHC compressor 201-1 starts in Initialization and Refresh (IR) state of Unidirectional (U) mode, wherein in the U mode, the RoHC compressor 201-1 sends IR packets to decompressor 203-1 of the receiver 203.

According to another embodiment, when the system 101 decides to enable the Robust header compression(RoHC) based on the result of the analysis, the system 101 determine optimal mode of state of the Robust header compression and the RoHC compressor 201-1 starts in the determined optimal mode.

According to an embodiment, when the system 101 decides to disable the Robust header compression(RoHC) based on the result of the analysis, the system 101 repeatly performs step 401 to 407 until the system 101 decides to enable the Robust header compression(RoHC).

According to another embodiment, when the system 101 decides to disable the Robust header compression(RoHC) based on the result of the analysis, the system 101 starts to perform step 401 to 407 in case specific event occurs.

According to an embodiment, a supervised learning methods like neural network (DNN, CNN, RNN,GAN), decision trees, etc can be used to identify the suitability for RoHC deployment. These supervised methods may work on static KPI list. However, the EVAL engine 301 works with dynamic KPI list. Further, the reward agent 503 may remove or include KPIs according to the KPI's effectiveness in the learning process. Further, the supervised ML mechanism may work only with smaller set of KPIs. However, for a larger set of KPIs, an automated process of selecting the most relevant KPIs affecting the deployment of RoHC is required. This issue was addressed by the is done by the reward agent 503. Furthermore, the EVAL engine 301 capable to operate in online learning, thus it is capable to learn in deployment phase. This enables the model to correct itself on the basis of past performance. Various simulation results will be explained in the forthcoming paragraphs.

Figure 10 illustrates an example architecture of the neural network used in the RL agent, according to an embodiment of the present disclosure. A simulation is carried using the following parameters:

A set of 16 KPIs are used. These KPIs were fed to the RL engine 301 which had to choose the most appropriate set of KPIs for RoHC prediction.

RL model using deep Q learning was trained for 200 iterations.

In each iteration of RL, a neural network model was trained for 200 iterations. This neural network takes the KPI list as input and decides on the action to be initiated.

For the neural network, an adam optimizer with a binary cross entropy loss function may be used.

For Q value updates the discount factor as 0.5 was kept.

The KPI list was randomly initialized to have a starting state. This, additional information is available for the KPI values, it can also feed provided along with the start state.

According to the parameters as set in the figure 10, Figure 11 illustrates a curve depicting the variation of reward with training iterations. According to the figure 11 the RL engine calculates the reward according to the neural network prediction accuracy. Final reward is the summation of immediate reward and future reward. The future reward is added to the cumulative record in proportion to the discount factor. By doing this for 200 iterations, the EVAL engine 301 selects 13 KPIs out of 16 KPIs. With every action of adding or removing the KPI from the KPI list, the Reward agent 503 affects it rewards. It tends to increase it reward over time and hence learn an effective policy. We can see from the graph at figure 11 that the rears are increasing with iterations. The rate of increase of reward decreases as the model training progresses. This indicated the model has converged to an optimal solution.

Figure 12 depicts a curve showing a variation of KPI list size with training iterations, according to an embodiment of the present disclosure. The graph shows the variation of KPI size with respect to training. Depending on the actions taken (KPI is added or removed), the variation in the KPI size occurs. The mean KPI size was 9. It can be seen that the KPI list size was showing more fluctuation in the beginning of training, in the later half of training the list's size variation has considerably decreased. This indicates that RL engine has achieved the optimal solution.

Accuracy of prediction using the best KPI list= 98.8%

Inference time for each KPI list - 0.000105 seconds

Now referring back to figure 4, the network node 101 may further include transmitter (Tx)/ receiver (Rx) 105 coupled with one or more processors 107. The processor(s) 107 may be further operatively coupled with the vCU)/ o-RAN 301 for performing various operations. The processor 107 may further coupled with a memory (not shown here). The method as disclosed above may be implemented in network node 101.

Figure 13 illustrates another exemplary diagram of a network node. The network node 1200 may include a communication unit 1205 (e.g., communicator or communication interface), a memory unit 1203 (e.g., storage), and at least one processor 12001. Further, the network node 1200 may also include the Cloud -RAN (C-RAN), a Central Unit (CU), a core Network (NW), a Distributed unit (DU) or a TRP controller or any other possible network (NW) entity. The communication unit 1205 may perform functions for transmitting and receiving signals via a wireless channel.

In an example, the processor 1005 may be a single processing unit or a number of units, all of which could include multiple computing units. The processor 203 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. Among other capabilities, the processor 1005 is configured to fetch and execute computer-readable instructions and data stored in the memory. The processor may include one or a plurality of processors. At this time, one or a plurality of processors may be a general purpose processor, such as a central processing unit (CPU), an application processor (AP), or the like, a graphics-only processing unit such as a graphics processing unit (GPU), a visual processing unit (VPU), and/or an AI-dedicated processor such as a neural processing unit (NPU). The one or a plurality of processors control the processing of the input data in accordance with a predefined operating rule or artificial intelligence (AI) model stored in the non-volatile memory and the volatile memory. The predefined operating rule or artificial intelligence model is provided through training or learning.

The memory may include any non-transitory computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read-only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes.

Figure 14 is a diagram illustrating the configuration of a terminal 1300 in a wireless communication system according to an embodiment of the present disclosure. The configuration of Fig. 14 may be understood as a part of the configuration of the terminal 1300. Hereinafter, it is understood that terms including "unit" or "er" at the end may refer to the unit for processing at least one function or operation and may be implemented in hardware, software, or a combination of hardware and software.

Referring to Fig. 14, the terminal 1300 may include a communication unit 1303 (e.g., communicator or communication interface), a storage unit 1305 (e.g., storage), and at least one processor 1301. By way of example, the terminal 1300 may be a User Equipment, such as a cellular phone or other devices that communicate over a plurality of cellular networks (such as a 4G, a 5G or pre-5G network or any future wireless communication network).

The communication unit 1303 may perform functions for transmitting and receiving signals via a wireless channel.

In an example, the module(s)/ engine may include a program, a subroutine, a portion of a program, a software component or a hardware component capable of performing a stated task or function. As used herein, a module(s)/ engine may be implemented on a hardware component such as a server independently of other modules, or a module can exist with other modules on the same server, or within the same program. The module(s)/ engine may be implemented on a hardware component such as processor one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions. The module(s)/ engine when executed by the processor may be configured to perform any of the described functionalities.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skilled in the art to which this invention belongs. The system, methods, and examples provided herein are illustrative only and not intended to be limiting.

Embodiments of the disclosure will be described below in detail with reference to the accompanying drawings. Moreover, the actions of any flow diagram need not be implemented in the order shown; nor do all of the acts necessarily need to be performed. Also, those acts that are not dependent on other acts may be performed in parallel with the other acts. The scope of embodiments is by no means limited by these specific examples. Numerous variations, whether explicitly given in the specification or not, such as differences in structure, dimension, and use of material, are possible. The scope of embodiments is at least as broad as given by the following claims.

Benefits, other advantages, and solutions to problems have been described above with regard to specific embodiments. However, the benefits, advantages, solutions to problems, and any component(s) that may cause any benefit, advantage, or solution to occur or become more pronounced are not to be construed as a critical, required, or essential feature or component of any or all the claims.

While specific language has been used to describe the present subject matter, any limitations arising on account thereto, are not intended. As would be apparent to a person in the art, various working modifications may be made to the method in order to implement the inventive concept as taught herein. The drawings and the foregoing description give examples of embodiments. Those skilled in the art will appreciate that one or more of the described elements may well be combined into a single functional element. Alternatively, certain elements may be split into multiple functional elements. Elements from one embodiment may be added to another embodiment.

Claims

A method for managing Robust Header Compression (RoHC), comprising:

monitoring over a time period a plurality of network characteristics and a plurality of key performance indicator (KPI) parameters;

generating an AI model having co-relation between the KPI parameters and the network characteristics;

analysing a current network characteristic with reference to the co-relation between the KPI parameters and the network characteristics in the AI model; and

performing by the AI model the enablement or the disablement of the Robust header compression (RoHC), based on a result of the analysis.
The method as claimed in claim 1, wherein the AI model perform the enablement or the disablement of the Robust header compression (RoHC) when the current network characteristics are below a threshold value as a result of the analysis.
The method as claimed in claim 1, further comprising:

shortlisting a sub plurality of the KPI from the plurality of the KPI parameters for a performance of an action defined by the enablement or disablement of the RoHC based on the correlation.
The method as claimed in claim 3, wherein the AI model is generated based on the correlation between the shortlisted sub-plurality of KPI parameters and the network characteristics, and the method further comprising:

deploying the generated AI model for analysing the current network characteristics thereby performing enablement or disablement of the RoHC.
The method as claimed in the claim 1, wherein generating the AI model further comprises:

feeding, to a reinforcement learning (RL) engine, at least one of a pre-classified set of KPI parameters, the monitored plurality of network characteristics, and a result associated with the pre-classified set of KPI parameters; and

predicting the enablement or the disablement of the RoHC by the RL engine based on a feedback for maximum reward as the output of the RL engine.
The method as claimed in the claim 5, further comprising:

learning a policy through the RL engine over a time period for deciding RoHC to be enabled or disabled based on the prediction of the enablement or the disablement of the RoHC and the reward received; and

generating the AI model for deployment based on the learned policy through the RL engine.
The method as claimed in the claim 6, further comprising:

calculating the reward based on an accuracy of the prediction of the enablement or the disablement of the RoHC, wherein the calculated reward includes at least one of a positive reward and the negative reward based on an effect of the enablement or the disablement of the RoCH, wherein the calculated reward assigned as a negative reward or a positive reward enabling ROHC had a positive or negative effect on the system.
The method as claimed in the claim 7, further comprising updating at least one of the policy, the reward, and the list of KPI parameters according to the accuracy of the prediction of the enablement or the disablement of the RoHC.
The method as claimed in the claim 6, wherein the correlating the plurality of network characteristics with the plurality of KPI parameters comprising:

performing a sequence of actions while assigning a reward to each of the actions to generate a total reward for each of the corresponding correlated plurality of network characteristics with the plurality of KPI parameters;

assigning the reward to each of the corresponding correlated plurality of network characteristics with the plurality of KPI parameters;

generating a log comprising the correlated plurality of network characteristics with the plurality of KPI parameters and the assigned reward thereof, wherein the policy over a time period is based on the generated log.
The method as claimed in the claims 1 and 5, wherein the sub plurality of KPI parameters are the relevant KPI parameters for the deployment of the AI model for enablement/disablement of the RoCH, and

wherein the determination of list of KPI parameters and a validation of the monitored plurality of the network characteristics are based on the reinforcement learning (RL) engine.
A system for managing Robust Header Compression (RoHC), comprising one or more processors coupled with a memory, the one or more processors is configured to:

monitor over a time period a plurality of network characteristics and a plurality of key performance indicator (KPI) parameters;

generate an AI model having co-relation between the KPI parameters and the network characteristics;

analyse a current network characteristic with reference to the co-relation between the KPI's parameters and the network characteristics in the AI model; and

perform by the AI model the enablement or the disablement of the Robust header compression (RoHC), based on a result of the analysis.
The system as claimed in claim 11, wherein the AI model perform the enablement or the disablement of the Robust header compression (RoHC) when the current network characteristics are below a threshold value as a result of the analysis.
The system as claimed in claim 11,wherein the one or more processors is configured to :

shortlist a sub plurality of the KPIs from the plurality of the KP's parameters for a performance of an action defined by the enablement or disablement of the RoHC based on the correlation.
The system as claimed in claim 13, wherein the AI model is generated based on the correlation between the shortlisted sub-plurality of KPI parameters and and the network characteristics, and the one or more processors is configured to:

deploy the generated AI model for analysing the current network characteristics thereby performing enablement or disablement of the RoHC.
The system as claimed in the claim 11, wherein for generating the AI model, the one or more processors is configured to: feed, to a reinforcement learning (RL) engine, at least one of a pre-classified set of KPI parameters, the monitored plurality of network characteristics, and a result associated with the pre-classified set of KPI parameters; and

predict the enablement or the disablement of the RoHC by the RL engine based on a feedback for maximum reward as the output of the RL engine.