Obtaining samples for learning-based resource management by adjusting flow characteristics

Abstract

A system (1, 21) for obtaining samples for a dynamic resource management system in a wireless network is configured to analyze (101) data for a dynamic resource management system to determine one or more combinations of input values that are not present or underrepresented in the data. The dynamic resource management system uses supervised learning to estimate resource requirements. The system is further configured to adjust (103) one or more flow characteristics of a data flow between a user equipment and a base station to obtain one of the combinations of input values, determine a number of resources used for the data flow and actual conditions and / or cell characteristics related to the data flow, create a new sample based on the number of resources used, the adjusted flow characteristics and the actual conditions and / or cell characteristics, and store the new sample in the data.

Description

Technical Field

[0001] This invention relates to a system for obtaining samples for a dynamic resource management system in a wireless network, the dynamic resource management system using supervised learning to estimate resource requirements.

[0002] The present invention further relates to a method for obtaining samples in a wireless network for a dynamic resource management system that uses supervised learning to estimate resource requirements.

[0003] The present invention also relates to a computer program product that enables a computer system to perform such a method. Background Technology

[0004] In mobile communication networks, resources are dynamically allocated to user equipment. For example, in 5G, base stations (gNodeBs) operate schedulers that assign blocks of resources to different user equipment (UEs) every millisecond, for example. Resources can also be dynamically assigned to slices and / or cells. For example, 5G network slicing is a network architecture feature that enables virtualized and independent logical networks to be reused on the same physical network infrastructure. A slice can involve one or more cells. Resource utilization can be optimized by dynamically assigning resources to different slices.

[0005] Typically, resource allocation decisions are based on predictions of resource demands imposed by (a class of) users or applications. Machine learning can be used to predict such resource demands. For example, the paper “RAN Resource Usage Prediction for a 5G Slice Broker” by Craig Gutterman et al. in Mobihoc'19: Proceedings of the Twentieth ACM International Symposium on Mobile Ad Hoc Networking and Computing (July 2019) describes the use of machine learning for time-series forecasting, where historical information is used to predict resource demands at future time steps.

[0006] While time series forecasting works well in scenarios where time dependencies are observable, it is not effective for predicting rare occurrences. In such cases, supervised machine learning models, such as artificial neural networks, can be used to predict resource demands based on observable network factors such as service requirements, wireless channel conditions, interference levels, and cell load.

[0007] Supervised machine learning models learn to make predictions based on a set of labeled observations. This set, also known as the data or training data, largely determines the accuracy of the model's predictions. Generally, the data must be large, diverse, and unbiased for the model to make good predictions. Furthermore, the data must adequately represent the scenario in which the model is expected to make predictions. In these cases, good generalization is indeed helpful, but it is a well-known fact that finding a model that does not overfit the data is contradictory to finding a model that captures all the essential characteristics of the data. Therefore, adequate generalization is a known and difficult problem in deriving machine learning models. Moreover, high bias in the data or training data can cause underfitting, potentially causing the resulting machine learning model to miss the correlation between the observed variables and the target output. Summary of the Invention

[0008] The first objective of this invention is to provide a system that can be used to improve data for resource demand estimation based on supervised learning.

[0009] A second objective of this invention is to provide a method for improving data used in supervised learning-based resource demand estimation.

[0010] In a first aspect of the invention, a system for obtaining samples for a dynamic resource management system in a wireless network (the dynamic resource management system uses supervised learning to estimate resource requirements) includes at least one processor configured to analyze data from the dynamic resource management system to determine one or more combinations of input values ​​that are absent or underrepresented in the data, the data including a plurality of samples, each of which includes flow characteristics and state and / or cell characteristics as input values, and the quantity of resources as an output value.

[0011] The at least one processor is further configured to: adjust one or more flow characteristics of the data stream between the user equipment and the base station to obtain one or more combinations of input values ​​that are not present or are underrepresented in the data; determine the amount of resources used for the data stream, and the actual state and / or cell characteristics associated with the data stream; create new samples based on the amount of resources used, the adjusted flow characteristics, and the actual state and / or cell characteristics; and store the new samples in the data.

[0012] This system allows processing a portion of the traffic corresponding to a given, sufficiently existing second application scenario (corresponding to a combination of input values) as if it were corresponding to a less-than-sufficient first application scenario (i.e., an application scenario that requires data samples for training purposes). For example, the first application scenario could include applications that account for a small portion of the total processed traffic (e.g., applications that require delayed critical guaranteed bit rate streams), unused applications planned for future use, or more frequently used applications that now have associated uncommon quality of service (QoS) requirements or other streaming characteristics.

[0013] To obtain better data, traffic in the second application scenario can be processed differently than what the scheduler at the base station would expect, in order to mimic the processing of the flow corresponding to the first application scenario. As a specific example, the adjusted traffic processing might require a QoS-aware scheduler to temporarily override the actual QoS requirements associated with the current flow in the second application scenario with the QoS requirements associated with the first application scenario of interest. Using this approach, samples can be obtained during the operational use of the wireless network without generating any additional artificial traffic, potentially having a greater impact on subscribers. The data can be training data, and the samples can be training samples.

[0014] Data flows can be initiated by, for example, user equipment or core network functions (e.g., SMF, UPF, or AMF). For example, a data flow can be a 5G Quality of Service (QoS) flow. Preferably, the adjusted flow characteristics are similar to the unadjusted flow characteristics and include QoS requirements that are more stringent or equally stringent than those included in the unadjusted flow characteristics.

[0015] The number of resources included as output values ​​in the samples can be specified as, for example, the number of Physical Resource Blocks (PRBs) for each data stream. In this case, supervised learning can be used to estimate the resource requirements for each data stream, and, for example, an aggregation method can then be used to estimate the resource requirements for each slice or cell using this resource requirement for each data stream. If many data streams with low latency requirements exist within the same time interval, the sum of the resource requirements of the simultaneous streams can be multiplied by a multiplication factor, which may depend on the state and / or cell characteristics, or a margin of a specific amount can be added to the sum to obtain the resource requirements for each slice or cell. The estimated resource requirements for each slice or cell can then be assigned to those slices or cells, for example, using a reinforcement learning method.

[0016] Current and / or future resource requirements can be estimated. Estimates of future resource requirements are also referred to as predicted resource requirements. Resource requirement forecasting and resource allocation can be performed periodically and / or triggered by certain events. For example, resource requirement forecasting and resource allocation can be performed periodically on timescales of milliseconds, minutes, or even higher. For example, a supervised learning process can involve training a neural network. The PRBs allocated to a data stream by the scheduler can be assumed to be the PRBs used for the data stream, but not all PRBs assigned to uplink traffic can always be used in practice, for example, due to device transmit power limitations.

[0017] The data stream can be an existing data stream or a new data stream. Stream characteristics can include a Quality of Service (QoS) level selected from multiple QoS levels. For example, these multiple QoS levels can include one or more of the following: non-guaranteed bit rate, non-latency-critical guaranteed bit rate, and latency-critical guaranteed bit rate. For example, these multiple QoS levels could be 5G 5QI level.

[0018] The at least one processor can be configured to select multiple characteristics from flow characteristics and state and / or cell characteristics based on the corresponding quality of service (QoS) level for each of the plurality of QoS levels, and to determine one or more combinations of input values ​​by determining whether the combination of selected characteristics is absent or underrepresented in the data according to the QoS level. The plurality of selected characteristics includes the corresponding QoS level. For example, RSRP, average neighboring cell load, and experienced throughput can be selected for non-guaranteed bit rate and non-delay-critical guaranteed bit rate data streams, and RSRP, average neighboring cell load, experienced throughput, and experienced latency can be selected for delay-critical guaranteed bit rate data streams.

[0019] The at least one processor can be configured to determine whether a combination of selected characteristics is absent or underrepresented in the data according to the Quality of Service (QoS) level by determining how many of these samples have combinations of selected characteristics that deviate from each other by less than a specific amount (e.g., a specific percentage like 25% or 33%). The range of values ​​for the characteristics can be divided (manually or automatically) into a specific number of groups, for example, of equal size, and values ​​within the same group can be considered to deviate from each other by less than a specific amount. As a first example, if none of the samples for the QoS level “Delay-Critical Guaranteed Bit Rate” have an RSRP of -60dBm, an average neighboring cell load of 80%, an experienced throughput of 20Mb / s, and an experienced latency of 75ms, then the QoS level of a new or existing data stream can be adjusted from “Non-Guaranteed Bit Rate” to “Delay-Critical Guaranteed Bit Rate” if the actual RSRP is -60dBm or optionally close to -60dBm and the actual average neighboring cell load is 80% or optionally close to 80%.

[0020] As a second example, if none of the samples of QoS level "Delay-Critical Guaranteed Bit Rate" have an RSRP between -50dBm and -70dBm, an average neighboring cell load between 65% and 85%, an experienced throughput between 15Mb / s and 25Mb / s, and an experienced latency between 60ms and 90ms, then in the case where the actual RSRP is between -50dBm and -70dBm and the average neighboring cell load is between 65% and 85%, the QoS level of a new or existing data stream can be adjusted from "Non-Guaranteed Bit Rate" to "Delay-Critical Guaranteed Bit Rate". As a third example, if none of the samples for the QoS level "Delay-Critical Guaranteed Bit Rate" have an RSRP between -50dBm and -70dBm, an average neighboring cell load between 65% and 85%, an experienced throughput between 15Mb / s and 25Mb / s, and an experienced latency between 60ms and 90ms, then the QoS level of a new or existing data stream can be adjusted from "Non-Guaranteed Bit Rate" to "Delay-Critical Guaranteed Bit Rate" without considering the actual RSRP and actual average neighboring cell load. Compared to the first and second examples, the flow characteristics of the data stream will need to be adjusted more frequently in the third example.

[0021] The number of resources included in the sample as output values ​​can be specified by time interval and by slice and / or cell, and the flow characteristics can include one or more values ​​indicating at least the number of data flows for each quality of service level within that time interval. Instead of using supervised learning to estimate the resource requirements for each data flow and then estimating the resource requirements for each slice or cell in a different way (e.g., using an aggregation method) based on that resource requirement for each data flow, supervised learning can be used to directly estimate the resource requirements for each slice and / or cell or even to directly determine the resource allocation for each slice and / or cell.

[0022] In this context, performance metrics can be additionally included as output values ​​in the samples to allow for specifying loss functions that include those metrics, thus enabling performance improvements. Slice utilization and the number of flows rejected per slice are examples of such performance metrics. If a particular combination of data flows for each QoS level is not represented or adequately represented in the samples within the same time interval, flow characteristics can be adjusted to obtain those samples.

[0023] State and / or cell characteristics may include one or more values ​​indicating one or more of radio channel conditions, interference level, throughput, delay, antenna orientation, scheduler parameters, and duplex mode. Radio channel conditions, interference level, throughput, and delay are examples of state characteristics. Antenna orientation, scheduler parameters, and duplex mode (e.g., FDD / TDD) are examples of cell characteristics. Average neighboring cell load is an example of a value indicating interference level. RSRP is an example of a value indicating radio channel conditions.

[0024] The at least one processor can be configured to induce adjustments to at least one of these flow characteristics by sending a request to the core network functions to use at least one adjusted flow characteristic. This can be used to ensure the quality of service of the core network functions and, therefore, for example, to ensure that the corresponding applications are aware of the actual application's service.

[0025] The system may include a controller and a base station, the controller including a first processor and the base station including a second processor, the first processor being configured to determine one or more combinations of input values ​​that are not present or are underrepresented in the data and to notify the base station of the one or more combinations, and the second processor being configured to adjust one or more streaming characteristics of the data stream to obtain one of the one or more combinations of input values ​​that are not present or are underrepresented in the data.

[0026] As a first example, the controller specifies which adjustments to make, and the base station only complies with this request for all subsequent streams that match these adjustments. As a second example, the controller only notifies the base station of one or more combinations and adjustments, and the base station decides in which cases (and which) the adjustments are performed. For example, if adjusting from a "non-guaranteed bit rate" to a "delay-critical guaranteed bit rate," the base station might decide to perform the adjustment only on 10% of all data streams that require the "non-guaranteed bit rate." The reason for this might be to avoid creating too many data samples based on the characteristics of the adjusted streams, which could potentially introduce the opposite bias.

[0027] The second processor can be configured to send a message to the user equipment notifying it of a temporarily adjusted streaming characteristic of the data stream, the adjusted streaming characteristic being one or more of the adjusted streaming characteristics. The message may include a time interval associated with the adjusted streaming characteristic that the user equipment is to use. Temporary adjustments can help reduce the impact of sample collection on the resources used and / or reduce the impact of service quality on the user experience.

[0028] The second processor can be configured to send a request to the user equipment, receive a response to the request, and, depending on whether the response is positive, adjust one or more streaming characteristics, the request requesting permission to use the one or more adjusted streaming characteristics. This can be beneficial, for example, in situations where it is necessary to relax quality of service requirements to obtain a desired sample.

[0029] In a second aspect of the invention, a user equipment for use or interaction with the system described above includes at least one processor configured to receive a message from a base station requesting permission from the user equipment to use adjusted streaming characteristics of a data stream, or notifying the user equipment to temporarily use adjusted streaming characteristics of a data stream, and using rules associated with the adjusted streaming characteristics.

[0030] In a third aspect of the invention, a method for obtaining samples for a dynamic resource management system (which uses supervised learning to estimate resource requirements) in a wireless network includes analyzing data from the dynamic resource management system to determine one or more combinations of input values ​​that are absent or underrepresented in the data, the data including a plurality of samples, each of which includes flow characteristics and state and / or cell characteristics as input values, and the quantity of resources as output values.

[0031] The method further includes: adjusting one or more flow characteristics of the data stream between the user equipment and the base station to obtain one or more combinations of input values ​​that are absent or underrepresented in the data; determining the amount of resources used for the data stream, and the actual state and / or cell characteristics associated with the data stream; creating new samples based on the amount of resources used, the adjusted flow characteristics, and the actual state and / or cell characteristics; and storing the new samples in the data. The data can be training data and the samples can be training samples. The method can be executed by software running on a programmable device. This software can be provided as a computer program product.

[0032] Furthermore, a computer program for performing the methods described herein is provided, as well as a non-transitory computer-readable storage medium for storing the computer program. For example, the computer program may be downloaded or uploaded to an existing device, or stored during the manufacture of these systems.

[0033] A non-transitory computer-readable storage medium stores at least a first software code portion that, when executed or processed by a computer, is configured to perform executable operations for obtaining samples for a dynamic resource management system in a wireless network, the dynamic resource management system using supervised learning to estimate resource requirements.

[0034] The executable operations include obtaining samples for a dynamic resource management system in a wireless network that uses supervised learning to estimate resource requirements. These executable operations include analyzing data from the dynamic resource management system to determine one or more combinations of input values ​​that are absent or underrepresented in the data, which includes multiple samples, each of which includes flow characteristics and state and / or cell characteristics as input values ​​and the quantity of resources as output values.

[0035] The executable operations further include: adjusting one or more flow characteristics of the data stream between the user equipment and the base station to obtain one or more combinations of input values ​​that are not present or are underrepresented in the data; determining the amount of resources used for the data stream, and the actual state and / or cell characteristics associated with the data stream; creating new samples based on the amount of resources used, the adjusted flow characteristics, and the actual state and / or cell characteristics; and storing the new samples in the data.

[0036] As will be understood by those skilled in the art, aspects of the present invention can be implemented as devices, methods, or computer program products. Therefore, aspects of the present invention can take the form of entirely hardware embodiments, entirely software embodiments (including firmware, resident software, microcode, etc.), or combined software and hardware embodiments, which are generally referred to herein as “circuit,” “module,” or “system.” The functionality described in this disclosure can be implemented as algorithms executed by a computer’s processor / microprocessor. Furthermore, aspects of the present invention can take the form of computer program products implemented on one or more computer-readable media having, for example, computer-readable program code implemented thereon, stored thereon.

[0037] Any combination of one or more computer-readable media can be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, for example, but not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the context of this invention, a computer-readable storage medium can be any tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0038] Computer-readable signal media may include propagated data signals having computer-readable program code implemented therein (e.g., in baseband or as part of a carrier wave). Such propagated signals may take any of a variety of forms, including, but not limited to, electromagnetic, optical, or any suitable combination thereof. Computer-readable signal media may be any computer-readable medium that is not a computer-readable storage medium and can communicate, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0039] The program code implemented on a computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, fiber optic, cable, RF, or any suitable combination thereof. The computer program code for performing the operations of various aspects of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java™, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" programming language or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0040] The aspects of the invention will now be described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, specifically a microprocessor or central processing unit (CPU), to produce a machine that, when executed via the processor of the computer or other programmable data processing apparatus or other device, creates means for implementing the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams.

[0041] These computer program instructions may also be stored in a computer-readable medium that can instruct a computer, other programmable data processing apparatus, or other device to function in a particular manner, such that the instructions stored in the computer-readable medium produce manufactured articles comprising the instructions, which implement the functions / actions specified in one or more blocks of a flowchart and / or block diagram.

[0042] Computer program instructions may also be loaded onto a computer, other programmable data processing apparatus or other equipment to cause a series of operational steps to be performed on the computer, other programmable apparatus or other equipment to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide for implementing the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0043] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code comprising one or more executable instructions for implementing a specified logical function(s). It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a non-linear order. For example, depending on the function involved, two consecutively shown blocks may actually be executed substantially simultaneously, or these blocks may sometimes be executed in reverse order. It will also be noted that each block of the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented by a system based on dedicated hardware or a combination of dedicated hardware and computer instructions that performs the specified function or action. Attached Figure Description

[0044] These and other aspects of the invention will become apparent by way of example with reference to the accompanying drawings, in which:

[0045] Figure 1 This is a flowchart of the first embodiment of the method;

[0046] Figure 2 This is a flowchart of the second embodiment of the method;

[0047] Figure 3 This is a block diagram of an embodiment of a system for obtaining samples;

[0048] Figure 4 This illustrates a first example of a dynamic resource management system;

[0049] Figure 5 A second example of a dynamic resource management system is shown;

[0050] Figure 6 A third example of a dynamic resource management system is shown;

[0051] Figure 7 This illustrates a first example of adjusting the flow characteristics of a network-initiated data stream;

[0052] Figure 8 This illustrates a first example of adjusting the streaming characteristics of a data stream initiated by a user equipment;

[0053] Figure 9 A second example of adjusting the flow characteristics of a network-initiated data stream is shown;

[0054] Figure 10 A second example of adjusting the streaming characteristics of a data stream initiated by a user equipment is shown; and

[0055] Figure 11 This is a block diagram of an exemplary data processing system for performing the methods of the present invention.

[0056] Corresponding elements in the accompanying drawings are indicated by the same reference numerals. Detailed Implementation

[0057] Collecting data from actual wireless / mobile network traffic processing is beneficial for resource management systems using supervised learning. While it is often possible to collect large amounts of data, the majority of wireless / mobile traffic is associated with a limited set of applications, including: video streaming (approximately 63%), web applications (social networks, browsing, navigation; approximately 19%), audio streaming, messaging, and file downloads.

[0058] Poor datasets for supervised learning are characterized by being insufficiently broad (containing too few data points to adequately represent the given problem scenario) and / or including biased or unrepresentative data. The lack of sufficient data points leads to high variability in model predictions for a given input; even if the model perfectly fits (or may even overfit) the data, it may then fail to provide adequate predictions for previously unseen data.

[0059] Since most wireless / mobile traffic is associated with a limited set of applications, the data collected in 5G will be biased for applications of URLLC or mMTC types that are little to no representative (or unrepresentative). Poor representativeness can further include rare or even common applications with unusual Quality of Service (QoS) requirements in terms of throughput, latency, or reliability, or other application / stream characteristics that are unusual in terms of duration or size.

[0060] The method described below makes it possible to create data or training data that includes a sufficient sample of data (wireless / mobile network statistics) for use in situations where the data is not representative or is underrepresented in total wireless / mobile traffic.

[0061] exist Figure 1The first embodiment of a method for obtaining samples for a dynamic resource management system in a wireless network is illustrated. This dynamic resource management system uses supervised learning to estimate resource requirements. Step 101 includes analyzing the data of the dynamic resource management system to determine one or more combinations of input values ​​that are absent or underrepresented in the data.

[0062] The data or training data includes multiple samples. Each of these samples includes flow characteristics and state and / or cell characteristics as input values, and the amount of resources as an output value. For example, the state and / or these cell characteristics may include one or more values ​​indicating one or more of the following: radio channel conditions, interference level, throughput, latency, antenna orientation, scheduler parameters, and duplex mode.

[0063] Step 103 includes adjusting one or more flow characteristics (e.g., QoS level) of the data stream between the user equipment and the base station to obtain one or more combinations of input values ​​that are absent or underrepresented in the data. Step 105 includes determining the amount of resources available for the data stream, as well as the actual state and / or cell characteristics associated with the data stream.

[0064] Step 107 includes creating new samples or training samples based on the amount of resources used, adjusted flow characteristics, and actual state and / or cell characteristics. Step 109 includes storing the new samples in the data.

[0065] exist Figure 2 A second embodiment of a method for obtaining samples for a dynamic resource management system in a wireless network is illustrated. This second embodiment is... Figure 1 An extension of the first embodiment. In Figure 2 In the embodiments, Figure 1 Step 101 is implemented by steps 121 and 123. Figure 2 In this embodiment, the flow characteristics include a Quality of Service (QoS) level selected from multiple QoS levels. For example, the multiple QoS levels may include one or more of a non-guaranteed bit rate, a non-delay-critical guaranteed bit rate, and a delay-critical guaranteed bit rate.

[0066] Step 121 involves selecting multiple characteristics from flow characteristics and state and / or cell characteristics for each of a plurality of quality of service (QoS) levels, based on the corresponding QoS level. The selected characteristics include the corresponding QoS level. This will be relative to... Figure 4 An example describing this selection is provided. Step 123 includes determining whether one or more combinations of selected features are absent or underrepresented in the data or training data according to the quality of service level, thereby determining one or more combinations of input values.

[0067] exist Figure 2In this embodiment, step 123 is implemented via step 131. Step 131 includes determining how many combinations of values ​​in these samples have selected characteristics that deviate from each other by less than a specific amount, according to the service quality level. This will be relative to... Figure 4 Describe this specific example.

[0068] Figure 3 An embodiment of a system for obtaining samples for a dynamic resource management system in a wireless network is illustrated. In this embodiment, the system includes a base station 1 and a controller 21. The dynamic resource management system uses supervised learning to estimate resource requirements. The wireless network 31 includes the base station 1, the controller 21, and a core network 33. For example, the wireless network 31 could be a 5G network. For example, the base station 1 could be a gNodeB. Figure 3 In one embodiment, base station 1 comprises a single unit. In an alternative embodiment, for example, the base station may include a central unit and one or more distributed units.

[0069] Base station 1 includes receiver 3, transmitter 4, processor 5, and memory 7. Controller 21 includes receiver 23, transmitter 24, processor 25, and memory 27. Figure 3 In this embodiment, processor 25 is configured to analyze data from the dynamic resource management system to determine one or more combinations of input values ​​that are absent or underrepresented in the data. For example, the data may include multiple samples and may be stored in memory 27. Each of these samples includes flow characteristics and state and / or cell characteristics as input values, and the quantity of resources as an output value.

[0070] exist Figure 3 In this embodiment, processor 5 is configured to adjust one or more flow characteristics of the data flow between user equipment and base station to obtain one or more combinations of input values ​​that are absent or underrepresented in the data. Preferably, the adjusted flow characteristics are similar to the unadjusted flow characteristics and include QoS requirements that are more stringent or equally stringent than those included in the unadjusted flow characteristics. Base station 1 is equipped with a QoS-aware scheduler and communicates with core network 33 and user equipment 11 to exchange information. For example, the information to be adjusted can be stored in memory 7.

[0071] Adjusted traffic processing may require a QoS-aware scheduler to temporarily override the actual QoS requirements associated with the current flow of a second application by using the QoS requirements associated with the first application. This approach allows for sampling during operational use of the wireless network without generating any additional artificial traffic, potentially having a greater impact on subscribers. For example, the first application could be an application representing a small portion of the total processed traffic (e.g., an application requiring delayed critical guaranteed bit rate flows), an unused application planned for future use, or an application that is more frequently used but now has associated uncommon quality of service (QoS) requirements or other flow characteristics.

[0072] At least one of processors 5 and 25 is configured to determine the amount of resources used for the data flow and the actual state and / or cell characteristics associated with the data flow, create new samples based on the amount of resources used, adjusted flow characteristics, and actual state and / or cell characteristics, and store the new samples in data, such as in memory 27 and / or memory 7. For example, the amount of resources used for the data flow can be determined by a QoS-aware scheduler running on base station 1.

[0073] User equipment 11 includes a receiver 13, a transmitter 14, a processor 15, and a memory 17. Figure 3 In some embodiments or alternative embodiments, processor 15 is configured to receive a message from base station 1 requesting user equipment 11 permission to use adjusted streaming characteristics of a data stream or notifying user equipment 11 to temporarily use adjusted streaming characteristics of a data stream. Processor 15 is further configured to use rules associated with the adjusted streaming characteristics.

[0074] exist Figure 3 In the illustrated embodiment, base station 1 includes a processor 5. In alternative embodiments, base station 1 includes multiple processors. For example, the processor 5 of base station 1 may be a general-purpose processor (e.g., an Intel or AMD processor) or a dedicated processor. For example, processor 5 may include multiple cores. For example, processor 5 may run a Unix-based operating system or a Windows operating system. For example, memory 7 may include solid-state storage (e.g., one or more solid-state drives (SSDs) made of flash memory) or one or more hard disks.

[0075] For example, receiver 3 and transmitter 4 can communicate with user equipment 11 using one or more cellular communication technologies such as GPRS, CDMA, UMTS, LTE, and / or 5G new radio. For example, receiver 3 and transmitter 4 can communicate with controller 21 and core network 33 using one or more wired communication technologies. Receiver 3 and transmitter 4 can be combined into a transceiver. Base station 1 may include other components typical of a base station in a mobile communication network, such as a power supply.

[0076] exist Figure 3 In the illustrated embodiment, controller 21 includes a processor 25. In alternative embodiments, controller 21 includes multiple processors. For example, processor 25 of controller 21 may be a general-purpose processor (e.g., an Intel or AMD processor) or a dedicated processor. For example, processor 25 may include multiple cores. For example, processor 25 may run a Unix-based operating system or a Windows operating system. For example, memory 27 may include solid-state storage (e.g., one or more solid-state drives (SSDs) made of flash memory) or one or more hard disks.

[0077] For example, receiver 23 and transmitter 24 can communicate with base station 1 using one or more wired communication technologies. Receiver 23 and transmitter 24 can be combined into a transceiver. Controller 21 may include other components typical for functions in a mobile communication network, such as a power supply.

[0078] exist Figure 3 In one embodiment, controller 21 is a separate network node and is located outside the core network 33. In an alternative embodiment, controller 21 may be located at base station 1 or within the core network 33, depending on its function. Figure 3 In the illustrated embodiment, controller 21 is a single, independent device. In another embodiment, controller 21 may include multiple devices and / or may be combined with another function in the mobile communication network (e.g., a base station).

[0079] Those skilled in the art may also refer to user equipment as mobile station (MS), user station, mobile unit, user unit, radio unit, radio terminal, radio device, wireless communication device, remote device, mobile user station, access terminal (AT), mobile terminal, user equipment (UE), remote terminal, mobile phone, terminal, user agent, mobile client, client, or any other suitable term.

[0080] Examples of wireless terminals include cellular phones, smartphones, Session Initiation Protocol (SIP) phones, laptop computers, notebook computers, netbooks, smart notebooks, personal digital assistants (PDAs), tablet computers, satellite radios, Global Positioning System (GPS) devices, multimedia devices, video devices, digital audio players, cameras, game consoles, or any other similar functional devices. For example, user equipment may have a slot for a UICC (also known as a SIM card), or be provisioned with an embedded version or an enhanced version for storing credentials.

[0081] exist Figure 3 In the illustrated embodiment, user equipment 11 includes a processor 15. In alternative embodiments, user equipment 11 includes multiple processors. Processor 15 may be a general-purpose processor (e.g., an ARM processor or a Qualcomm processor) or a dedicated processor. For example, processor 15 may run Google Android or Apple iOS as an operating system.

[0082] For example, the receiver 13 and transmitter 14 of user equipment 11 can communicate with base station 1 using one or more cellular communication technologies such as GPRS, CDMA, UMTS, LTE, and / or 5G new radio. The receiver 13 and transmitter 14 can be combined into a transceiver. User equipment 11 may include other components typical of user equipment, such as a display and a battery. For example, user equipment 11 may be a mobile phone.

[0083] Figure 4 A first example of a dynamic resource management system is shown. In this first example, the dynamic resource management system includes a neural network 51. The neural network 51 is trained using samples of each data stream, each sample corresponding to a data stream. Each sample includes the characteristics (input value) and the number of resources (output value) of each data stream. Figure 4 Data 61 is depicted, which includes data streams DF1 to DF... n The characteristics C and the number of resources #R for each data stream.

[0084] Table 1 shows an exemplary sample of data 61. The sample shown in Table 1 represents six data streams established at six different times (i.e., not simultaneously) for six different user equipment. The stream characteristics shown in Table 1 include a Quality of Service (QoS) level selected from multiple QoS levels. The QoS requirements for the data streams are latency-sensitive (DS) or latency-tolerant (DT) with or without a minimum guaranteed throughput requirement (GBR / non-GBR). The assigned radio resources are specified as the number of Physical Resource Blocks (PRBs).

[0085] Table 1

[0086]

[0087] Such as about Figure 2 As described in step 121, multiple characteristics can be selected from flow characteristics and state and / or cell characteristics for each of the multiple quality of service (QoS) levels, based on the corresponding QoS level. For example, RSRP, average neighboring cell load, and experienced throughput can be selected for non-GBR and DT / GBR data flows, and RSRP, average neighboring cell load, experienced throughput, and experienced latency can be selected for DS / GBR data flows.

[0088] Such as about Figure 2 As described in step 123, this step can then determine how many combinations of selected characteristics in these samples deviate from each other by a specific amount, according to the service quality level. For example, experienced throughput rounded down to 45 Mb / s can be considered as experienced throughput deviating from each other by a specific amount, rounded down to 44 Mb / s. Alternatively, the range of characteristic values ​​can be assumed (manually or automatically) divided into a specific number of groups, for example, of equal size, and values ​​within the same group can be considered as deviating from each other by a specific amount.

[0089] As an example of the latter, if Table 1 above includes all samples, then none of the samples for QoS class DS / GBR have an RSRP between -90dBm and -110dBm, an average neighboring cell load between 85% and 100%, an experienced throughput between 0Mb / s and 10Mb / s, and an experienced latency between 60ms and 90ms. Therefore, if the actual RSRP is between -90dBm and -110dBm and the average neighboring cell load is between 85% and 100%, the QoS class of a new or existing data stream can be adjusted from DT / non-GBR to DS / GBR.

[0090] After training the neural network 51, it can be used to estimate resource requirements in an inference mode. The neural network 51 estimates resource requirements according to the data flow. For slice S... p Multiple data streams DF x …DF y Each of them provides features (C x …C y () as input value. In Figure 4 The text describes DF. x Input values ​​71 and DF y The input value is 72. Then, the neural network 51 outputs the resource quantity (#R) for each of the multiple data streams. x …#R y ).exist Figure 4 The text describes DF. xOutput value 74 and DF y The output value is 75.

[0091] Then slice S p Data Stream DF x …DF y The estimated resource requirements are provided to aggregation function 53, which aggregates the resource requirements determined for the data stream, i.e., the quantity of resources. The sum of resource requirements from simultaneous streams can be multiplied by a multiplication factor, which may depend on state and / or cell characteristics, or a margin of a specific amount can be added to the sum to obtain the resource requirements for each slice or cell. In this way, to guarantee performance, additional resources can be allocated to slices to account for possible estimation errors in the demand estimation step. For example, possible estimation errors may arise from insufficient training or inaccurate modeling, such as suboptimal choices of input parameters and / or parameters of the machine learning algorithm (e.g., the number of hidden layers and / or the number of neurons per layer).

[0092] The output of aggregation function 53 is slice S. p Resource requirement 81 is determined. This resource requirement 81, along with the resource requirements of other slices (determined in a similar manner), is provided to resource assignment function 55. Resource assignment function 55 assigns resources to slices based on resource requirement 81 and the resource requirements of other slices. For example, resource assignment function 55 can use reinforcement learning. The output of resource assignment function 55 is slice S. p Resource assignment 83 and resource assignments for other slices.

[0093] Figure 5 A second example of a dynamic resource management system is shown. In this second example, dynamic resource management includes a neural network 57. The neural network 57 is trained using samples from each slice. Each sample includes a characteristic (input value) and the number of resources (output value). Figure 5 Slices S1 to S2 were depicted. n The sample, in which Figure 5 The image shows sample 63 of slice S1 and slice S in more detail. n Sample 65.

[0094] Each of these samples, used as input values, includes each Quality of Service (QoS) level (#QoS1, ..., #QoS). k The amount of data flow within a specific time interval is considered as part of the flow characteristics, and includes other flow, state, and / or cell characteristics (OC). For example, other characteristics may include average RSRP and / or average neighboring cell load. Each of these samples, as an output value, further includes the amount of resources (#R) used for the corresponding slice within that time interval.

[0095] Table 2 shows an example of a sample of neural network 57. The QoS requirements for the data stream are either latency tolerance (DT) with or without a guaranteed minimum throughput requirement (GBR / non-GBR) or latency sensitivity (DS) with a guaranteed minimum throughput requirement. The assigned radio resources are specified as the number of physical resource blocks (PRBs).

[0096] Table 2

[0097]

[0098] If Table 2 above includes all samples, the samples with a large number of DS / GBR data streams would be underrepresented. Therefore, the QoS level of several data streams could be adjusted from DT / non-GBR or DT / GBR to DS / GBR to obtain those samples.

[0099] After training the neural network 57, it can be used to estimate resource requirements in an inference mode. The neural network 57 estimates the resource requirements for each slice. For slice S... p Each Quality of Service (QoS) level (#QoS1, ..., #QoS) k The number of data streams, as well as other streams, states, and / or cell characteristics (OC), are provided as input values ​​77 to the neural network 57. In the example in Table 2, these other characteristics include the average RSRP of all UEs in the cell and the average neighboring cell load. The neural network 57 then outputs slice S. p Resource requirements 81.

[0100] This resource requirement is 81. Figure 4 The same method shown in the diagram is used to provide the resource requirements of other slices together with the resource assignment function 55. The output of resource assignment function 55 is slice S. p Resource assignment 83 and resource assignments for other slices.

[0101] Figure 6 A third example of a dynamic resource management system is shown. In this third example, dynamic resource management includes a neural network 59. Figure 5 Similar to neural network 57, neural network 59 is trained using samples from each slice. However, neural network 59 outputs slice S. p Resource assignment for 83 and other slices is performed, and therefore a separate resource assignment function is not required. In this case, in addition to the output of radio resource assignment mentioned in Table 2, outputs related to performance metrics need to be added so that a loss function that includes those performance metrics can be specified to improve performance. Slice utilization and the number of rejected streams per slice are examples of such performance metrics. If the function is based on reinforcement learning, these performance metrics can also be used. Figure 4 and Figure 5 The resource assignment function and the reward function of 55.

[0102] exist Figures 4 to 6 In the example, resource requirements are estimated by cell according to slices, and resources are assigned to slices for that cell. Alternatively, resource requirements can be estimated by cell, and resources can be assigned to cells even without using slices.

[0103] Figure 7 A first example of adjusting the flow characteristics of a network-initiated data stream is shown. First, in step 101, controller 21 determines one or more combinations of input values ​​that are absent or underrepresented in the data. In a first embodiment of the method, controller 21 specifies which adjustments to make, and base station 1 complies with this request only for all subsequent streams that match these adjustments.

[0104] In a second embodiment of the method, controller 21 notifies base station 1 of one or more combinations and adjustments only, and base station 1 decides in which cases to perform the adjustment. For example, if adjusting from a "non-guaranteed bit rate" to a "delay-critical guaranteed bit rate," base station 1 may decide to perform the adjustment only on 10% of all data streams requiring the "non-guaranteed bit rate." The reason for this is likely to avoid creating too many data samples based on the characteristics of the adjusted stream, which could lead to the introduction of the opposite bias.

[0105] exist Figure 7 In this example, controller 21 determines that the data stream with QoS1 level associated with the first application is underrepresented in the data and decides that QoS1 level, rather than QoS2 level associated with the second application, should be used for the second application. In this example, a combination of input values ​​including QoS1 level is required. Controller 21 uses message 201 to notify base station 1 of these one or more combinations.

[0106] In step 103, when the core network 33 (e.g., the session management function of the core network 33) transmits message 203 to establish a QoS flow with QoS2 for the second application, the base station 1 adjusts the QoS level of this QoS flow from QoS2 to QoS1. Then, the base station 1 transmits message 205 to the user equipment 11 to establish a QoS flow between the base station 1 and the user equipment 11. For example, message 205 could be a 5G NR RRC reconfiguration message.

[0107] Message 205 could simply specify that the rules associated with QoS level 1 should be used, and then user equipment 11 might not even be aware that this QoS flow will typically require QoS level 2. Alternatively, message 205 could inform user equipment 11 to temporarily use QoS level 1 for the QoS flow and subsequently use the normal QoS level 2 for that QoS flow. In this case, message 205 could include the time interval for which the rules associated with the adjusted flow characteristics (i.e., QoS level 1) should be used by the user equipment. The latter might require RRC reconfiguration of unstandardized fields in the message.

[0108] Next, user equipment 11 sends message 207 to base station 1 to confirm the establishment of a QoS flow with QoS level 1. Then, base station 1 sends message 209 to core network 33 to confirm the establishment of a QoS flow with QoS level 2. Therefore, core network 33 is unaware of the connection between the user equipment 11 and base station 1. Figure 7 Adjustments to the flow properties in the example.

[0109] User equipment 11 receives and processes data from the second application according to the rules associated with QoS1. Therefore, user equipment 11 applies resource mapping according to QoS1 rather than QoS2, and accordingly sends packets belonging to the flow uplink to base station 1. For example, different queues, different radio bearers, and / or different frequencies can be used for QoS1 traffic instead of QoS2 traffic.

[0110] Figure 8 This illustrates a first example of adjusting the streaming characteristics of a data stream initiated by a user device. Step 101 is performed, and in this example, it is compared with... Figure 7 The example transmits message 201 in the same manner. However, in this example of a data stream initiated by the user equipment, user equipment 11 transmits message 221 to establish a QoS stream with QoS level 2 for the second application.

[0111] Then, base station 1 sends message 223 to core network 33 to establish a QoS flow with QoS level 2 between base station 1 and core network 33. Next, core network 33 sends message 225 to base station 1 to confirm the establishment of the QoS flow with QoS level 2.

[0112] Next, in step 103, base station 1 adjusts the QoS level of the QoS flow between base station 1 and user equipment 11 from QoS2 to QoS1, and base station 1 sends message 205 to user equipment 11 to establish a QoS flow between base station 1 and user equipment 11 at the adjusted QoS level (i.e., QoS1 instead of QoS2). Then, user equipment 11 sends message 207 to base station 1 to confirm the establishment of a QoS flow with QoS1 level.

[0113] Figure 9 This illustrates a second example of adjusting the flow characteristics of a network-initiated data stream. Step 101 is performed, and in this example, it is compared with... Figure 7 The same method is used to transmit message 201. However, in this example, instead of base station 1 adjusting the QoS level itself, base station 1 causes the QoS level to be adjusted by transmitting request 241 to core network 33 (e.g., Session Management Function (SMF) of the second application) in step 151 to use QoS level 1 in certain circumstances.

[0114] Then, in step 103, core network 33 complies with the request and subsequently transmits message 243 to establish a QoS flow with QoS1 for the second application, thereby overriding the original QoS (QoS2) of the second application with the desired QoS (QoS1) of the first application. After the QoS flow (with QoS1 level) between base station 1 and user equipment 11 has been established using messages 205 and 207, base station 1 transmits message 225 to core network 33 to confirm the establishment of the QoS flow with QoS1 level. Therefore, in this example, and in this embodiment of the method, core network 33 is aware of the adjustment to the flow characteristics.

[0115] exist Figure 9 Following the message exchange described herein, core network 33 (e.g., UPF as indicated by SMF) sends downlink packets belonging to the second application now associated with QoS1 to base station 1, and optionally to other base stations if the data is sent via multiple base stations.

[0116] Figure 10 A second example of adjusting the streaming characteristics of a data stream initiated by a user device is shown. This example is similar to... Figure 8 For example, before base station 1 adjusts the QoS level in step 103, base station 1 sends request 261 to user equipment 11. In this example, request 261 requests permission to use one or more adjusted flow characteristics, i.e., to use QoS level 1 instead of QoS level 2. User equipment 11 then sends response 263 to request 261 and base station 1 performs step 103, i.e., adjusts the QoS level depending on whether response 263 is affirmative.

[0117] Next, base station 1, depending on response 263, sends message 205 to user equipment 11 to establish a QoS flow between base station 1 and user equipment 11 at an unadjusted QoS level (i.e., QoS2 level) or an adjusted QoS level (i.e., QoS1 level). Next, user equipment 11 sends message 207 to base station 1 to confirm the establishment of a QoS flow with the QoS level specified in message 205.

[0118] exist Figure 8 and Figure 10 In the example, the user equipment utilizes an application in the core network to establish a QoS flow. Alternatively, the user equipment can utilize another user equipment to establish a QoS flow.

[0119] Figure 11 The description shows that the demonstration can be performed as in the reference. Figure 1 and Figure 2 A block diagram of an exemplary data processing system describing the method.

[0120] like Figure 11 As shown, the data processing system 300 may include at least one processor 302 coupled to a memory element 304 via a system bus 306. Thus, the data processing system can store program code within the memory element 304. Furthermore, the processor 302 can execute program code accessed from the memory element 304 via the system bus 306. In one aspect, the data processing system may be implemented as a computer suitable for storing and / or executing program code. However, it should be understood that the data processing system 300 may be implemented in the form of any system including a processor and memory capable of performing the functions described herein.

[0121] Memory element 304 may include one or more physical memory devices, such as local memory 308 and one or more mass storage devices 310. Local memory may refer to random access memory or other non-persistent memory(s) typically used during the actual execution of the program code. Mass storage devices may be implemented as hard disk drives or other persistent data storage devices. Processing system 300 may also include one or more cache memories (not shown) that provide temporary storage for at least some program code to reduce the number of times program code must be retrieved from mass storage device 310 during execution.

[0122] The input / output (I / O) devices, depicted as input device 312 and output device 314, may be optionally coupled to the data processing system. Examples of input devices may include, but are not limited to, keyboards, pointing devices such as mice, etc. Examples of output devices may include, but are not limited to, monitors or displays, speakers, etc. The input devices and / or output devices may be coupled to the data processing system directly or through an intermediate I / O controller.

[0123] In embodiments, the input device and the output device can be implemented as a combined input / output device (in... Figure 11(The dashed lines surrounding input device 312 and output device 314 are shown in the image). An example of such a combined device is a touch-sensitive display, sometimes also called a "touchscreen display" or simply a "touchscreen". In such embodiments, input to the device can be provided by the movement of a physical object (such as a stylus or a user's finger) on or near the touchscreen display.

[0124] Network adapter 316 can also be coupled to the data processing system to enable it to couple to other systems, computer systems, remote network devices, and / or remote storage devices via an intermediate private or public network. The network adapter may include a data receiver for receiving data transmitted from the systems, devices, and / or networks to the data processing system 300, and a data transmitter for transmitting data from the data processing system 300 to the systems, devices, and / or networks. Modems, cable modems, and Ethernet cards are examples of different types of network adapters that can be used with the data processing system 300.

[0125] As in Figure 11 As depicted, memory element 304 can store application program 318. In various embodiments, application program 318 may be stored in local memory 308, one or more mass storage devices 310, or separately from local memory and mass storage devices. It should be understood that data processing system 300 may further execute an operating system that facilitates the execution of application program 318. Figure 11 (Not shown in the image). The application program 318, implemented in the form of executable program code, can be executed by the data processing system 300 (e.g., by the processor 302). In response to executing the application program, the data processing system 300 can be configured to perform one or more operational or method steps described herein.

[0126] Various embodiments of the present invention can be implemented as a program product for use with a computer system, wherein the program(s) of the program product define the functionality of the embodiments (including the methods described herein). In one embodiment, the program(s) may be contained on a variety of non-transitory computer-readable storage media, wherein, as used herein, the expression “non-transitory computer-readable storage media” includes all computer-readable media, with the sole exception of temporarily propagated signals. In another embodiment, the program(s) may be contained on a variety of transient computer-readable storage media. Illustrative computer-readable storage media include, but are not limited to: (i) non-writable storage media on which information is permanently stored (e.g., read-only memory devices within a computer, such as CD-ROM discs readable by a CD-ROM drive, ROM chips, or any type of solid-state non-volatile semiconductor memory); and (ii) writable storage media on which variable information is stored (e.g., flash memory, floppy disks within a floppy disk drive or hard disk drive, or any type of solid-state random access semiconductor memory). The computer program may run on the processor 302 described herein.

[0127] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that when the terms “comprises” and / or “comprising” are used in this specification, they specify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0128] All means or steps in the claims, plus corresponding structures, materials, actions, and equivalents of functional elements, are intended to include any structure, material, or action for performing a function in conjunction with other claimed elements as specifically claimed. Descriptions of embodiments of the invention have been presented for illustrative purposes, but such description is not intended to be exhaustive or to limit implementation to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the invention. The embodiments were chosen and described to best explain the principles of the invention and some practical applications, and to enable others skilled in the art, when suited to a particular intended use, to understand the invention with respect to various embodiments having various modifications.

Claims

1. A system (1, 21) for obtaining samples for a dynamic resource management system (51, 53, 55, 57, 59) in a wireless network, the dynamic resource management system (51, 53, 55, 57, 59) using supervised learning to estimate resource requirements, the system (1, 21) comprising at least one processor (5, 25), the at least one processor being configured to: - Analyze the data used in the dynamic resource management system (51, 53, 55, 57, 59) to determine one or more combinations of input values ​​that are absent or underrepresented in the data, said data comprising multiple samples, each of said samples including flow characteristics and state and / or cell characteristics as input values, and the quantity of resources as an output value. - Adjust one or more streaming characteristics of the data stream between the user equipment (11) and the base station (1) to obtain one or more combinations of input values ​​that are absent or underrepresented in the data. - Determine the amount of resources available for the data stream, as well as the actual state and / or cell characteristics associated with the data stream. - Create new samples based on the amount of resources used, adjusted flow characteristics, and the actual state and / or cell characteristics, and - Store the new sample in the data.

2. The system (1, 21) as described in claim 1, wherein, The number of resources included in the sample as output values ​​is specified according to the data stream.

3. The system (1, 21) as described in claim 1 or 2, wherein, The flow characteristics include a Quality of Service (QoS) level selected from multiple QoS levels.

4. The system (1, 21) as described in claim 3, wherein, The at least one processor (5, 25) is configured to select multiple features from the flow characteristics and the state and / or cell characteristics based on the corresponding service quality level for each of the plurality of service quality levels, and to determine the one or more combinations of input values ​​by determining whether the combination of the multiple selected features is absent or insufficiently representative in the data according to the service quality level, the multiple selected features including the corresponding service quality level.

5. The system (1, 21) as described in claim 4, wherein, The at least one processor (5, 25) is configured to determine, according to the quality of service level, whether a combination of the multiple selected characteristics is absent or underrepresented in the data by determining how many combinations of the multiple selected characteristics in the sample have a deviation from each other by less than a specific amount.

6. The system (1, 21) as claimed in claim 3, wherein, The multiple quality of service levels include one or more of the following: non-guaranteed bit rate, non-delay-critical guaranteed bit rate, and delay-critical guaranteed bit rate.

7. The system (1, 21) as claimed in claim 3, wherein, The number of resources included in the sample as output values ​​is specified by time interval and by slice or cell, and the flow characteristics include one or more values ​​that indicate at least the number of data flows per quality of service level within the time interval.

8. The system (1, 21) as claimed in any one of claims 1 to 3, wherein, The state and / or cell characteristics include one or more values ​​indicating one or more of the following: radio channel conditions, interference level, throughput, delay, antenna orientation, scheduler parameters, and duplex mode.

9. The system (1, 21) as claimed in any one of claims 1 to 3, wherein, The system (1, 21) includes a controller (21) and a base station (1), the controller (21) including a first processor (25) and the base station (1) including a second processor (5). The first processor (25) is configured to determine one or more combinations of input values ​​that are absent or underrepresented in the data, and to notify the base station of the one or more combinations. The second processor (5) is configured to adjust one or more streaming characteristics of the data stream to obtain one or more combinations of input values ​​that are not present or are underrepresented in the data.

10. The system (1, 21) as claimed in claim 9, wherein, The second processor (5) is configured to send a message to the user equipment (11) notifying the user equipment (11) to temporarily use the adjusted streaming characteristics of the data stream, the adjusted streaming characteristics being included in one or more of the adjusted streaming characteristics.

11. The system (1, 21) as claimed in claim 10, wherein, The message includes the time interval for which the rules associated with the adjusted streaming characteristics will be used by the user equipment.

12. The system (1, 21) as claimed in claim 9, wherein, The second processor (5) is configured to send a request to the user equipment (11), receive a response to the request, and adjust one or more streaming characteristics depending on whether the response is affirmative, the request requesting permission to use one or more adjusted streaming characteristics.

13. The system (1, 21) as claimed in any one of claims 1 to 3, wherein, The at least one processor (5, 25) is configured to cause an adjustment of at least one of the flow characteristics by sending a request to the core network function (33) to use the at least one adjusted flow characteristic.

14. The system (1, 21) as claimed in any one of claims 1 to 3, wherein, The data is training data and the samples are training samples.

15. A user equipment (11) that interacts with a system (1, 21) to obtain samples for a dynamic resource management system (51, 53, 55, 57, 59) in a wireless network, the dynamic resource management system (51, 53, 55, 57, 59) using supervised learning to estimate resource requirements, the system (1, 21) comprising at least one processor (5, 25) configured to: - Analyze the data used in the dynamic resource management system (51, 53, 55, 57, 59) to determine one or more combinations of input values ​​that are absent or underrepresented in the data, said data comprising multiple samples, each of said samples including flow characteristics and state and / or cell characteristics as input values, and the quantity of resources as an output value. - Adjust one or more streaming characteristics of the data stream between the user equipment (11) and the base station (1) to obtain one or more combinations of input values ​​that are absent or underrepresented in the data. - Determine the amount of resources available for the data stream, as well as the actual state and / or cell characteristics associated with the data stream. - Create new samples based on the amount of resources used, the adjusted flow characteristics, and the actual state and / or cell characteristics, and - Store the new sample in the data. And among them, The system (1, 21) further includes a controller (21) and the base station (1). The controller (21) includes a first processor (25) configured to determine one or more combinations of input values ​​that are absent or underrepresented in the data, and to notify the base station of the one or more combinations. The base station (1) includes a second processor (5) configured to adjust one or more streaming characteristics of the data stream to obtain one or more combinations of input values ​​that are absent or underrepresented in the data. The user equipment (11) includes at least one processor (15), which is configured to: - Receive a message from the base station (1) requesting the user equipment (11) permission to use the adjusted flow characteristics of the data stream or notifying the user equipment (11) to temporarily use the adjusted flow characteristics of the data stream, and - Use the rules associated with the adjusted flow characteristics.

16. A method for obtaining samples for a dynamic resource management system in a wireless network, the dynamic resource management system using supervised learning to estimate resource requirements, the method comprising: - Analysis (101) of the data used in the dynamic resource management system to determine one or more combinations of input values ​​that are absent or underrepresented in the data, the data comprising multiple samples, each of the samples comprising flow characteristics and state and / or cell characteristics as input values, and the quantity of resources as output values; - Adjust one or more flow characteristics of the data stream between the user equipment and the base station to obtain one or more combinations of input values ​​that are not present or are not representative in the data; - Determine (105) the amount of resources used for the data stream, and the actual state and / or cell characteristics associated with the data stream; - Create (107) new samples based on the amount of resources used, the adjusted flow characteristics, and the actual state and / or cell characteristics; as well as - Store the new sample (109) in the data.

17. A computer program product storing at least one software code portion, said software code portion being configured to perform the method of claim 16 when running on a computer system.

Images