WO2023143698A1

WO2023143698A1 - Network resource control based on neighborhood measurements

Info

Publication number: WO2023143698A1
Application number: PCT/EP2022/051624
Authority: WO
Inventors: Zied Ben Houidi; Jonatan KROLIKOWSKI; Dario Rossi
Original assignee: Huawei Technologies Co., Ltd.
Priority date: 2022-01-25
Filing date: 2022-01-25
Publication date: 2023-08-03

Abstract

The present disclosure relates to resource allocation in a network, for example, a wireless local area network (WLAN). The disclosure provides a control device that is configured to allocate the resources to a plurality of network access devices of the network. The control device collects measurement data over a period of time, wherein the measurement data indicates signal qualities of signals between the network access devices and the terminal devices. Then, the control device determines a historical density distribution of the terminal devices in the network based on the collected measurement data, and finally allocates resources to the network access devices based on the determined historical density distribution.

Description

NETWORK RESOURCE CONTROL BASED ON NEIGHBORHOOD MEASUREMENTS

TECHNICAL FIELD

The present disclosure relates to resource allocation in a communications network, for example, in a wireless local area network (WLAN). The disclosure provides a control device that is configured to manage the resources by allocating or reallocating the resources to a plurality of network access devices of the network. The control device is able to base the resource allocation on neighborhood measurements, which allow the control device to derive a historical density distribution of a plurality of terminal devices in the network.

BACKGROUND

As an example of a commonly known communications network, a WLAN is defined in the IEEE 802.11 standard and its amendments. The most popular setup is infrastructure-based, wherein terminal devices, which may be called stations (STAs), such as smartphones or laptops, connect to fixed network access point devices, which may be called Access Points (APs). In a similar manner, in a mobile communications network, like in a 5^th generation (5G) or a 6^th generation (6G) mobile network, the terminal devices, which may be called user equipment (UEs), connect to the network access devices, which may be called base stations.

In a specific example of a WLAN, namely a campus WLAN that is often established in office buildings or university campuses, a fleet of APs is deployed and may connect to numerous STAs. For example, FIG. 1 shows a schematic representation of such a campus WLAN.

In a campus WLAN, the APs are conventionally governed by a central Access Controller (AC) that coordinates their resource configurations. On the AC, a number of actions may be taken to manage the network resources. As an example, two important AP configurations are: (1) primary channel selection and bonding; and (2) transmission power management. The significance of these two aspects is briefly introduced in the following, in order to clarify the interplay between the different aspects

(1) Each AP in the WLAN may be configured to use a specific primary channel (e.g., within a 2.4 GHZ or 5 GHz band), for performing downlink (DL) and uplink (UL) transmissions in a half-duplex manner. In FIG. 1, these channels are represented by the circles around the respective coverage areas of the respective APs. Ideally, only one device within a receiver’s vicinity transmits on one channel at the same time, avoiding collisions and data loss. This may be achieved through the listen-before-talk mechanism of Carrier-sense multiple access with collision avoidance (CSMA/CA). As a consequence, the APs and STAs on the same channel share airtime. The time that a transmitter waits while the channel is busy is called interference (time).

Unfortunately, interference cannot be avoided in highly dense areas, since channels are scarce: depending on the regulatory region, only 3-4 channels are non-overlapping on the 2.4 GHz band, and around 20 channels on the 5 GHz band. It is thus not always possible to set neighboring APs on different channels. In the example of FIG. 1, for instance, the two lower APs share a channel, whose usage time is split between UL and DL communications of STAs A and B, idle time and interference. Interference is particularly detrimental in busy parts of the network, such as in conference rooms, which regularly experience flash crowds of STAs. As the network conditions evolve over time, it is beneficial to re-adjust the channel allocation accordingly.

Additionally, in modem WLANs, an AP may be configured to allow the aggregation (also known as ‘bonding’) of several channels to increase bandwidth and consequently throughput. In the example of FIG. 1, the top-left AP uses two channels (dark and light circle). While this brings obvious benefits, bonding also increases the number of neighbors competing for the same channels, which complicates the channel allocation.

(2) STAs attach to a serving AP based on criteria that differ by vendor and device generation. The most important aspect, however, is the DL received signal strength indicator (RSSI), which needs at least to exceed a device-specific threshold. A higher DL RSSI is generally beneficial for the user (STA), since it allows for a higher data rate and thus better throughput. While the DL RSSI is influenced by factors that are not directly controlled through network management, such as the distance between the AP and the STA and environmental factors such as shadowing through walls and other obstacles, it is partially determined though the configurable AP transmit power. While increasing the AP transmit power would benefit the data rates of closer STAs, it would also increase the number of attached STAs, and hence competitors on the radio resource. Additionally, surrounding low-power APs may be underused, and more distant STAs and APs may see the high-power AP as an interferer such that network performance may degrade. Hence, an optimal power allocation needs to provide good coverage quality in the entire network area, while limiting interference and balancing load among APs.

Objectives when configuring the APs in a campus network are to provide the best service demands during the deployment period of the configuration. The objectives for a configuration may, for example, comprise the following aspects: achieve good coverage quality across the network, e.g., as many as possible of the STAs present in the network during the deployment period receive a high DL RSSI; minimize co-channel interference during the deployment period; balance the user load across APs during the deployment period.

One conventional solution consists of two distinct algorithms that adjust transmit power with respect to each of the following two objectives: coverage quality (i.e., signal strength) maximization (first algorithm), and interference minimization (second algorithm). Both algorithms use as input the AP topology, which is a measurement, or a series of measurements, that determines whether two APs are expected to interfere with each other or not, i.e. if their transmissions share airtime or not. The first algorithm iteratively calculates an ideal power based on the AP topology, and makes adjustments until the top- 3 neighbors meet some RSSI threshold for each AP. The second algorithm reduces coverage overlaps. The overlap is estimated based on the AP topology. The precise adjustment criteria remain unknown. Notably, neither the first nor the second algorithm takes user traffic into account.

SUMMARY

The present disclosure and its solutions are based further on the following considerations. Seemingly, all conventional configuration mechanisms do not take the points of presence or the distribution of users into account. Thus, densely populated areas and empty areas in the network are equally allocated the scare radio resources. Further, conventional solutions typically require human interaction, and can thus only be applied locally.

In view of the above, this disclosure has the objective to provide a solution for a resource configuration (allocation) of a plurality of network access devices in a network, wherein the resource configuration fits the traffic pattern of terminal devices in the deployment period (the period of time for which the resource configuration is valid). In particular, the resource configuration should be applicable to a fleet of APs in a campus WLAN. The solution of this disclosure should rely solely on widely available network telemetry, and should not require human intervention.

These and other objectives are achieved by the solutions of this disclosure as described in the independent claims. Advantageous implementation forms are further defined in the dependent claims.

A first aspect of this disclosure provides a control device for managing resources in a network, wherein the network comprises a plurality of network access devices and a plurality of terminal devices, and wherein the control device is configured to: collect measurement data over a period of time, wherein the measurement data indicates signal qualities of signals between the network access devices and the terminal devices; determine a historical density distribution of the terminal devices in the network based on the collected measurement data; and allocate resources to the network access devices based on the determined historical density distribution.

The network may be a WLAN, for example, a campus WLAN. However, the network may be any other network, like a 5G mobile network or a 6G mobile network. Resources may comprise a transmit power and/or a channel or channel bandwidth. That is, allocating resources may comprise providing the network with a resource configuration for the network access devices, which determines transmit powers and/or channels or channel bandwidth on which the network access devices communicate. However, resources may also (alternatively or additionally) comprise time and/or frequency resources, for example, a time slot, a frequency band, a frequency subcarrier, a resource block, a subchannel, or the like.

The control device of the first aspect is able to leverage the measurement data of the signal qualities, e.g. 802.11k neighbor measurement data, which is collected over the period of time. The measurement data is used to infer historical densities of the terminal devices (users) in the network, and to optimize the (radio) resources, i.e. their allocation, based on these historical densities. For instance, the control device may give more weight to those areas in the network that are usually dense with terminal devices, while giving less resources to areas where terminal devices are rarely present.

Thus, a resource configuration can be provided for the plurality of network access devices in the network, which fits the traffic pattern in the deployment period. In particular, the resource configuration is applicable to a fleet of APs in a campus WLAN. The solution does not require human interaction.

In an implementation form of the first aspect, the measurement data indicates, for each terminal device, the respective signal quality between the terminal device and each network device.

The control device can also use raw measurements to indicate the signal qualities in the measurement data.

In an implementation form of the first aspect, the measurement data does not include positional information of the terminal devices and the network access devices in the network.

Thus, the solution of this disclosure is fast and efficient, as a localization is not required.

In an implementation form of the first aspect, the network is a wireless local area network, and the measurement data comprises one or more 802.1 Ik neighbor reports.

In an implementation form of the first aspect, the measurement data is a measurement data matrix, wherein each row of the measurement data matrix corresponds to one terminal device at one time point or time interval in the period of time, and wherein each column entry in the row of the terminal device comprises the signal quality between the terminal device and one respective network access device of the plurality of network access devices.

Further, each column of the measurement data matrix may correspond to one network access device (it may be assumed that a network access device is fixed over the entire period of time), and each column entry in the column of the network access device may comprise the signal quality between a respective terminal device of the plurality of network access devices at one time point or time interval in the period of time.

In an implementation form of the first aspect, the control device is configured to allocate more resources to a network access device located in an area of the network corresponding to a higher historical density of terminal devices, and/or allocate less resources to a network access device located in an area of the network corresponding to a lower historical density of terminal devices.

Accordingly, the resource allocation may better fit the traffic pattern of the terminal devices in the network.

In an implementation form of the first aspect, the control device is configured to select only a part of the measurement data as reference data, and to determine the historical density distribution based on the selected reference data.

Thus, the amount of data can be reduced, which makes the solution faster and less computationally complex.

In an implementation form of the first aspect, for determining the historical density distribution of the terminal devices in the network based on the collected measurement data, the control device is configured to: convert the measurement data to points in a latent space; estimate a density distribution of the points in the latent space; and determine the historical density distribution of the terminal devices in the network based on the estimated density distribution of the points in the latent space.

The conversion into the latent space may enable a more efficient determination of the historical density distribution

In an implementation form of the first aspect, each row entry of the measurement data matrix corresponds to one point in the latent space, and wherein a distance between two points in the latent space is proportional to a difference between two corresponding row entries of the measurement data matrix.

In an implementation form of the first aspect, the dimensionality of the latent space is smaller than the dimensionality of the measurement data matrix.

This makes the solution more convenient and less computationally complex. In an implementation form of the first aspect, the control device is configured to select a subset of the points in the latent space as reference points.

Thus, fewer points have to be processed by the control device.

In an implementation form of the first aspect, the control device is configured to randomly select the reference points from the points in the latent space.

In an implementation form of the first aspect, the control device is configured to determine a plurality of clusters of points in the latent space, wherein the points of a cluster are distanced from each other by less than a threshold difference or are not distanced from each other; and select one or more reference points from each cluster of points.

This may reduce the amount of data, without sacrificing representativeness of the points.

In an implementation form of the first aspect, the control device is further configured to: compute a utility function for each selected reference point and/or compute a global utility function based on utility functions of all selected reference points, wherein a utility function is representative of at least one of: a coverage quality of the terminal device associated with the reference point, an interference at the network access point associated with the reference point, and a load of the network access point associated with the reference point; and allocate the resources to the network access devices such that one or more utility functions and/or the global utility function are optimized.

A second aspect of this disclosure provides a method for managing resources in a network, wherein the network comprises a plurality of network access devices and a plurality of terminal devices, and wherein the method comprises: collecting measurement data over a period of time, wherein the measurement data indicates signal qualities of signals between the network access devices and the terminal devices; determining a historical density distribution of the terminal devices in the network based on the collected measurement data; and allocating resources to the network access devices based on the determined historical density distribution.

In an implementation form of the second aspect, the measurement data indicates, for each terminal device, the respective signal quality between the terminal device and each network device. In an implementation form of the second aspect, the measurement data does not include positional information of the terminal devices and the network access devices in the network.

In an implementation form of the second aspect, the network is a wireless local area network, and the measurement data comprises one or more 802.1 Ik neighbor reports.

In an implementation form of the second aspect, the measurement data is a measurement data matrix, wherein each row of the measurement data matrix corresponds to one terminal device at one time point or time interval in the period of time, and wherein each column entry in the row of the terminal device comprises the signal quality between the terminal device and one respective network access device of the plurality of network access devices.

In an implementation form of the second aspect, the method comprises allocating more resources to a network access device located in an area of the network corresponding to a higher historical density of terminal devices, and/or allocating less resources to a network access device located in an area of the network corresponding to a lower historical density of terminal devices.

In an implementation form of the second aspect, the method comprises selecting only a part of the measurement data as reference data, and to determine the historical density distribution based on the selected reference data.

In an implementation form of the second aspect, for determining the historical density distribution of the terminal devices in the network based on the collected measurement data, the method comprises: converting the measurement data to points in a latent space; estimating a density distribution of the points in the latent space; and determining the historical density distribution of the terminal devices in the network based on the estimated density distribution of the points in the latent space.

In an implementation form of the second aspect, each row entry of the measurement data matrix corresponds to one point in the latent space, and wherein a distance between two points in the latent space is proportional to a difference between two corresponding row entries of the measurement data matrix.

In an implementation form of the second aspect, the dimensionality of the latent space is smaller than the dimensionality of the measurement data matrix. In an implementation form of the second aspect, the method comprises selecting a subset of the points in the latent space as reference points; and determine the density distribution of the points in the latent space based only on the selected reference points.

In an implementation form of the first aspect, the method comprises randomly selecting the reference points from the points in the latent space.

In an implementation form of the first aspect, the method comprises determining a plurality of clusters of points in the latent space, wherein the points of a cluster are distanced from each other by less than a threshold difference or are not distanced from each other; and selecting one or more reference points from each cluster of points.

In an implementation form of the second aspect, the method further comprises: computing a utility function for each selected reference point and/or compute a global utility function based on utility functions of all selected reference points, wherein a utility function is representative of at least one of: a coverage quality of the terminal device associated with the reference point, an interference at the network access point associated with the reference point, and a load of the network access point associated with the reference point; and allocating the resources to the network access devices such that one or more utility functions and/or the global utility function are optimized.

The method of the second aspect and its implementation forms achieve the same advantages and effects

A third aspect of this disclosure provides a computer program comprising instructions which, when the program is executed by a computer, cause the computer to perform the method according to the second aspect or any implementation form thereof.

A fourth aspect of this disclosure provides a non-transitory storage medium storing executable program code which, when executed by a processor, causes the method according to the second aspect or any of its implementation forms to be performed.

It has to be noted that all devices, elements, units and means described in the present application could be implemented in the software or hardware elements or any kind of combination thereof. All steps which are performed by the various entities described in the present application as well as the functionalities described to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be performed by external entities is not reflected in the description of a specific detailed element of that entity which performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be implemented in respective software or hardware elements, or any kind of combination thereof.

BRIEF DESCRIPTION OF DRAWINGS

The above described aspects and implementation forms will be explained in the following description of specific embodiments in relation to the enclosed drawings, in which

FIG. 1 shows an exemplary WLAN schema.

FIG. 2 shows a control device according to a solution of this disclosure.

FIG. 3 illustrates a solution of this disclosure that leverages STA-to-AP measurements as measurement data.

FIG. 4 illustrates a solution of this disclosure that leverages 802.11k neighbor measurement data.

FIG. 5 illustrates experimental results of a solution of this disclosure on a real network

FIG. 6 shows an example of an end-to-end pipeline from 802.1 Ik neighbor measurement data to a network resource configuration for a control device according to a solution of this disclosure.

FIG. 7 shows an example of a pipeline for reference data (reference point) selection from the measurement data for a control device according to a solution of this disclosure.

FIG. 8a shows an exemplary projection of 802.11k measurement data points using umap (dimension 5), which is visualized using T-SNE of dimension 2.

FIG. 8b shows the projection of FIG. 8a with only selected reference points. FIG. 9 shows a solution of this disclosure for selecting reference points, for instance, from a cluster of points.

FIG. 10 illustrates a utility function that may be computed by the control device according to a solution of this disclosure.

FIG. 11 shows a method according to a solution of this disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS

FIG. 2 shows a control device 200 according to this disclosure. The control device 200 is configured to manage resources in a network 201. Managing resources may comprise allocating or reallocating resources for determined time periods (deployment periods). The network 201 comprises a plurality of network access devices 202 and a plurality of terminal devices 203. The control device 200 may be a central AC of, for example, a WLAN. The control device 200 may also be a management network function of a mobile communications network, for example, a 5G network. The control device may be a controller or control function. The control device 200 may also be part of a network access device 202, or may be instantiated by two or more of the plurality of network access devices 200 in a distributed manner.

The control device 200 may comprise a processor or processing circuitry (not shown) configured to perform, conduct or initiate the various operations of the control device 200 described herein. The processing circuitry may comprise hardware and/or the processing circuitry may be controlled by software. The hardware may comprise analog circuitry or digital circuitry, or both analog and digital circuitry. The digital circuitry may comprise components such as application- specific integrated circuits (ASICs), field-programmable arrays (FPGAs), digital signal processors (DSPs), or multi-purpose processors. The control device 200 may further comprise memory circuitry, which stores one or more instruction(s) that can be executed by the processor or by the processing circuitry, in particular under control of the software. For instance, the memory circuitry may comprise a non-transitory storage medium storing executable software code which, when executed by the processor or the processing circuitry, causes the various operations of the control device 200 to be performed. In one embodiment, the processing circuitry comprises one or more processors and a non-transitory memory connected to the one or more processors. The non-transitory memory may carry executable program code which, when executed by the one or more processors, causes the control device 200 to perform, conduct or initiate the operations or methods described herein.

The control device 200 is configured to collect measurement data 204 over a period of time. The measurement data 204 indicates signal qualities of signals between the network access devices 202 and the terminal devices 203. In particular, the measurement data 204 may indicate, for each of the terminal devices 203, the respective signal quality between the terminal device 203 and each of the network access devices 202. The signal quality may change over the period in time, and the measurement data 204 may indicate this change in the signal quality for each of the terminal devices 203. For example, the measurement data

204 may come in the form of a measurement data matrix. Each row of this measurement data matrix may correspond to one terminal device 203 at one time point or time interval in the period of time. That is, multiple rows may correspond to the same terminal device 203, and may reflect the temporal evolution of the signal qualities related to this terminal device. Each row entry in the row of the terminal device 203 comprises the signal quality between the terminal device 203 (at the respective time or time interval) and one respective network access device 202 of the plurality of network access devices 202. The measurement data 204 may, for example, comprise pathlosses as the indications of the signal qualities. The measurement data may, for example, be or comprise one or more 802.1 Ik neighbor reports.

Further, the control device 200 is configured to determine a historical density distribution

205 of the terminal devices 203 in the network 201 based on the collected measurement data 204. The control device 200 may, for instance, be configured to determine one or more areas of the network 201, which correspond to a higher historical density of terminal devices 203, e.g., where over the period of time there were more terminal devices 203 present and/or terminal devices 203 where present longer. Additionally, or alternatively, the control device 200 may be configured to determine one or more areas of the network 201, which correspond to a lower historical density of terminal devices 203, e.g., where over the period of time there were fewer terminal devices 203 present and/or the terminal devices 203 were present shorter. The control device 200 may accordingly be configured to infer historical densities of the presence of the terminal devices 203 in the network 201. The control device 200 may determine a traffic pattern of the terminal devices 203 in the network during the period of time. Further, the control device 200 is configured to allocate resources 206, particularly radio resources, to the network access devices 202 based on the determined historical density distribution 205. Thus, the network device 200 may provide the plurality of network access devices 202 with a resource configuration. For example, the control device 200 may allocate more resources 206 to a network access device 202 located in an area of the network 201, which corresponds to a higher historical density of terminal devices 203, and/or may allocate fewer resources 206 to a network access device 202 located in an area of the network 201, which corresponds to a lower historical density of terminal devices 203. For instance, a network access device 202 in a denser area may be provided a higher transmit power and/or more channel bandwidth. The control device 200 may thus be configured to provide a resource configuration, by allocating the resources 206, which fits the density distribution and/or traffic pattern of the terminal devices 203 in the network 201 over the period of time. No human interaction is needed. Also, the positions of the network access devices 202 in the network 200 are not needed.

FIG. 3 and FIG. 4 show a solution of this disclosure that builds on the solution shown in FIG. 2. Same elements in FIG. 2 and FIG. 3 are labelled with the same reference signs, and are likewise implemented. The solution shown in FIG. 3 leverages STA-to-AP measurements comprising an 802.11k neighbor measurement report as the measurement data 204. The network 201 is a WLAN, and the terminal devices 203 and the network access points 202 are, respectively, STAs 203 and APs 202. The control device 200 may collect and store this measurement data 204, and may use it to infer the historical densities of the presence of the terminal devices 203 in the network 201. The control device 200 may then allocate and/or optimize the radio resources 206 in the network 201, for example, by giving more weights to those areas of the network 201 that are usually dense with terminal devices 203, and giving less weights to those areas in the network 201 that are usually less dense with terminal devices 203.

The 802.11k WLAN standard standardizes the way signal quality measurements are done and exchanged between the STAs and the APs. Its role originally is to help in the handoff process allowing STAs in motion to roam to the neighbor with the best signal (since the list of measurements of signals to the best APs is available). Thanks to 802.11k, it thus becomes possible for the network operator to collect, if desired, the signal quality between each STA present in the network and a set of neighboring APs. As shown in FIG. 3, each determined time interval (e.g., each second), for each STA 203, it is possible to obtain the signal quality/strength (particularly, indicated by pathloss in this case) towards neighboring APs 202. The 802.1 Ik neighbor report may be in the form of a data matrix having a certain dimensionality (rows x columns). Each row corresponds to one STA 203 and one time interval in the period of time. Each column corresponds to one AP 202. Each row entry in the row of each STA 203 comprises the pathloss between the STA 203 and one respective AP 202 of the plurality of APs.

The 802.11k neighbor report does not carry the physical positions of STAs 203 and the APs 202 in the network 201 (e.g. in the physical plane of a building), but still expresses a distance in the pathloss space from each STA 203 to each AP 202 in the network 201. The control device 200 of this disclosure can leverage this information to extract a historical density distribution 205 of the presence of the STAs 203, and then assign more resources 206 to denser areas in the network 201. That is, the control device 200 can determine the historical density distribution 205 of the STAs 203 in the network 201 based on the 802.1 Ik neighbor report and can allocate radio resources 206 to the APs 202 based on the determined historical density distribution 205 of the STAs 203.

The advantage is an improved coverage quality, while having less or similar interference. Experiments on a real network 201 show this improvement. In particular, FIG. 5 shows results on the real network 201, wherein “as-is” is a conventional solution and “data-driven” is the solution of this disclosure.

According to the above-described, an idea of this disclosure consists in leveraging terminal device 203 to network access device 202 measurement data 204 (e.g., STA-to-AP measurements in a WLAN) guide the network resource configuration to fit a historical terminal device density in the network 201. This is unlike conventional solutions, which do not account for users (terminal devices). The solution of this disclosure is adapted to operate directly in a high-dimensional signal quality (e.g., pathloss) space, without requiring intermediate conversion, e.g. from signal quality vectors to positional information like (x, y) coordinates of the terminal devices 203 and network access devices 202 in the network 201. Unlike some conventional work, which performed only theoretical studies under unrealistic assumptions such as full and continuous knowledge of user positions, the present solution is able to operate in a way that is both computationally feasible and able to preserve information value. FIG. 6 shows an exemplary implementation of a solution of this disclosure, which may be implemented by the control device 200. In particular, FIG. 6 shows an end-to-end pipeline of the solution, which may be carried out by the control device 200.

In a first step, the measurement data 204 is collected in the form of a neighborhood measurement data matrix as an input. Each row of the matrix refers to one terminal device 203, at a given point or during a given time interval in a period of time, in the network 201. Each row entry indicates a signal quality (e.g., a pathloss) with respect to all the network access devices 202 in the network 201.

In a second step 601, the measurement data 204 is projected to a low dimensional space (also referred to as latent space). The points 604 in this latent space preserve the distance that they have in the measurement data 204. For instance, each row entry of the measurement data matrix corresponds to one point 604 in the latent space, and a distance between two points 604 in the latent space is relatively representative of a distance between two corresponding row entries. Notably, the measurement data matrix has a certain dimension, and this dimension may be higher than the dimension of the low dimensional space. A purpose of this projection is that the measurements (row entries) of one and the same terminal device 203, but for different points in time or time intervals - i.e. different rows in the measurement data matrix for the same terminal device 203 - will fall on identical or at least closely clustered points in the low dimensional space, if the terminal device 203 remains in the same physical location over these different points in time or during the time intervals.

In a third step 602, a representative set of points 605 is selected from the low dimensional space. In particular, the control device 200 may be configured to select the representative set of points 605 to optimize the allocation of the resources 206 to the network access devices 202.

In a fourth step 603, once the reference points 605 are selected, a resource allocation algorithm may leverage an objective function that takes into account all reference points 605, so as to find a particular resource allocation configuration (configuration search) that for example, maximizes a given reward or a utility function for all reference points 605. The step 603 may take as input the selected reference points 605 and topology and/or telemetry data of the network access devices 202 in the network 201. The resource configuration (resources 206) may be allocated to the network access devices 202. Notably, as a simple example, the measurement data 204, e.g. in the form of the 802.11k neighbor report, could be used as is without further processing. However, processing may reduce the amount of data. FIG. 7 shows, as an example, a pipeline for reference point selection, as it may be performed by the control device 200. The following data reductions are possible by the processing.

Firstly, the dimensionality of each reference point 605 (e.g., from the number of network access devices 202, which can be 30 to 50, or even more for a large network) can be reduced to a 2D or 3D (two-dimensional or three-dimensional) latent space. For this, one can use techniques like Umap (e.g., ‘Mclnnes, L., Healy, J., & Melville, J., “Umap: Uniform manifold approximation and projection for dimension reduction' j arXiv preprint arXiv: 1802.03426, 2018’) or like t-SNE (e.g., ‘Van der Maaten, Laurens, and Geoffrey Hinton. "Visualizing data using t-SNE" Journal of machine learning research 9.11, 2008’).

Secondly, additionally or alternatively, the number of reference points 605, for which to optimize the resource allocation, can be reduced. Many points 604 in the network 201 are redundant (e.g., terminal devices 203 that spend the whole period of time in the same location). For example, in one network 201, for a duration of e.g. one month, more than 2 million points 604 can be observed. It may be convenient to optimize for a smaller number of still representative points 605 (e.g., 50k points may be a more convenient number). In this respect, FIG. 8a and 8b may be compared.

One trivial reference point selection strategy is to select the reference points 605 randomly from all points 604. However, this may favor the most popular points 604, and may underserve rare points 604, which may still be beneficial to cover, yet will get less resources 206 during the resource allocation. Thus, another beneficial approach is to adopt a density-based selection, as is shown in FIG. 9. For example, only one point 604 from each cluster 901 of points may be sampled. The control device 200 may determine a plurality of clusters 901 of points 604 in the latent space - wherein points 604 of a cluster 901 are distanced from each other by less than a threshold difference or are not distanced from each other - and may select one reference point 605 from each cluster 901. To do so, the control device 200 may go over all the points 604, may select one point 604 of each cluster, and may ‘invalidate” all the other points within the cluster 901. It is also possible to select reference points 605 based on their radius from a center of the latent space, for instance, one or more points 604 having the same radius. FIG. 9 shows resulting reference points 605 depending on the radius and density. Notably, the number of reference points 605 should not be too low, to avoid information bottlenecks. However, the number of reference points 605 should also not be too high, to avoid a computational bottleneck.

Based on the selected reference points 605, the resource allocation may be optimized by the control device 200. In one example, a user centric objective may be used by the control device 200, for which the network 201 may be resource optimized. For example, the control device 200 may compute a utility function 1000 (see FIG. 10) for each selected reference point 605 and/or may compute a global utility function based on the utility functions of all selected reference points 605. The global utility function may be the sum of per reference point utility functions. As an example, it is possible to compute/define a utility function 1000 for each of the reference points 605 directly in the latent space (e.g., a pathloss space). Each utility function 1000 of a certain reference point 605 may be representative of at least one of: a coverage quality of the terminal device 203 that corresponds to the certain reference point 605; an interference at the network access point 202 associated with the certain reference point 605; a load of the network access point 202 that corresponds to the certain reference point 605. An example is shown in FIG. 10a. The control device 200 may allocate the resources 206 to the network access devices 202 such that one or more utility functions 1000 and/or the global utility function are optimized.

First, a user-centric objective may be defined. A goal may be to adjust power and to reduce channel overlap (resource configuration) to enhance signal quality in the network 201. The objective may, in particular, take into account the coverage quality, the interference, and/or load balancing. Example objectives are: minimize a number of terminal devices 203 attached to the closest network access device 202 (the serving network access device 203 should not be very loaded); maximize a RSSI from the serving network access device 202 (good RSSI is desired); minimize a number of terminal devices 203 for each interfering network access device 202 (the lower the number of interferers to a certain point (x, y), the better).

The objective may be mapped from points in the real space (x, y) (see FIG. 10b, left side) to reference points 605 built from readily available measurement data 202 (see FIG 10b, right side). A terminal device 203 may be represented on the left hand side objective by its spatial (x, y) coordinates, which may be not available in the measurement data 204. A by a reference point 605 may be represented by a vector of measurements between a terminal device 203 and network access device 202 on the right hand side objective, which may be available in the measurement data 204.

Furthermore, given a power allocation and a reference point 605 defined e.g. by a pathloss vector, it is possible for the control device 200 to estimate the following: the coverage quality, e.g., the signal quality of signals that the terminal device 202 associated with a reference point 605 sees from each network access device 203 in the network 201; interference, e.g., the amount of interference that each network access device 203 in the network 201 sees; and the load of each network access device 202 (given assumptions, to which network access device 203 the terminal device 202 associated with the reference point 605 is connected, e.g. best network access device signal). This gives the possibility to measure the utility function for each reference point 605 in the latent space (defined only by e.g. its pathloss vector).

To find the best power and channel combination (an example for resource allocation), algorithmically, the control device 200 may be configured to determine, from within a space of power configurations, an optimal one according to a set of reference points 605. Finding the best power and channel combination that yields the best U(ref) is a combinatorial optimization problem, for which complete enumeration is a solution for small instances. As this is practically not possible starting from a number of network access devices 202, alternative heuristics (e.g. local search based ones) can be leveraged.

Fig. 11 shows a method 1100 according to a solution of this disclosure. The method 1100 can be used for managing resources in a network 201, wherein the network comprises a plurality of network access devices 202 and a plurality of terminal devices 203. The method 1100 may be performed by the control device 200.

The method 1100 comprises a step 1101 of collecting measurement data 204 over a period of time, wherein the measurement data 204 indicates signal qualities of signals between the network access devices 202 and the terminal devices 203. The method 1100 further comprises a step 1102 of determining a historical density distribution 205 of the terminal devices 202 in the network 201 based on the collected measurement data 204. Then, the method 1100 comprises a step 1103 of allocating resources 206 to the network access devices 202 based on the determined historical density distribution 205. The present disclosure has been described in conjunction with various examples as well as aspects and implementations. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed matter, from the studies of the drawings, this disclosure and the independent claims. In the claims as well as in the description the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.

Claims

1. A control device (200) for managing resources in a network (201), wherein the network (201) comprises a plurality of network access devices (202) and a plurality of terminal devices (203), and wherein the control device (200) is configured to: collect measurement data (204) over a period of time, wherein the measurement data (204) indicates signal qualities of signals between the network access devices (202) and the terminal devices (203); determine a historical density distribution (205) of the terminal devices (203) in the network (201) based on the collected measurement data (204); and allocate resources (206) to the network access devices (202) based on the determined historical density distribution (205).

2. The control device (200) according to claim 1, wherein the measurement data (204) indicates, for each terminal device (203), the respective signal quality between the terminal device (203) and each network access device (202).

3. The control device (200) according to claim 1 or 2, wherein the measurement data (204) does not include positional information of the terminal devices (203) and the network access devices (202) in the network (201).

4. The control device (200) according to one of the claims 1 to 3, wherein the network (201) is a wireless local area network, and the measurement data (204) comprises one or more 802.1 Ik neighbor reports (404).

5. The control device (200) according to one of the claims 1 to 4, wherein the measurement data (204) is a measurement data matrix, wherein each row of the measurement data matrix corresponds to one terminal device (203) at one time point or time interval in the period of time, and wherein each row entry in the row of the terminal device (203) comprises the signal quality between the terminal device (203) and one respective network access device (202) of the plurality of network access devices (202).

6. The control device (200) according to one of the claims 1 to 5, configured to allocate more resources (206) to a network access device (202) located in an area of the network (201) corresponding to a higher historical density of terminal devices (203), and/or allocate less resources (206) to a network access device (202) located in an area of the network (201) corresponding to a lower historical density of terminal devices (203).

7. The control device (200) according to one of the claims 1 to 6, configured to select only a part of the measurement data (204) as reference data (605), and to determine the historical density distribution (205) based on the selected reference data (605).

8. The control device (200) according to one of the claims 1 to 7, wherein for determining the historical density distribution (205) of the terminal devices (203) in the network (201) based on the collected measurement data (204), the control device (200) is configured to: convert the measurement data (204) to points (604) in a latent space; estimate a density distribution of the points (604) in the latent space; and determine the historical density distribution (205) of the terminal devices (203) in the network (201) based on the estimated density distribution of the points (604) in the latent space.

9. The control device (200) according to the claims 8 and 5, wherein each row entry of the measurement data matrix corresponds to one point (604) in the latent space, and wherein a distance between two points (604) in the latent space is proportional to a difference between two corresponding row entries of the measurement data matrix.

10. The control device (200) according to claim 8 or 9, wherein the dimensionality of the latent space is smaller than the dimensionality of the measurement data matrix.

11. The control device (200) according to one of the claims 8 to 10, configured to: select a subset of the points (604) in the latent space as reference points (605).

12. The control device (200) according to claim 11, configured to randomly select the reference points (605) from the points (604) in the latent space.

13. The control device (200) according to claim 11, configured to: determine a plurality of clusters (901) of points (604) in the latent space, wherein the points (604) of a cluster are distanced from each other by less than a threshold difference or are not distanced from each other; and select one or more reference points (605) from each cluster (901) of points (604).

14. The control device (200) according to one of the claims 11 to 13, further configured to: compute a utility function (1000) for each selected reference point (605) and/or compute a global utility function based on utility functions of all selected reference points (605), wherein a utility function is representative of at least one of: a coverage quality of the terminal device (203) associated with the reference point (605), an interference at the network access point (202) associated with the reference point (605), and a load of the network access point (202) associated with the reference point (605); and allocate the resources (206) to the network access devices (202) such that one or more utility functions (2000) and/or the global utility function are optimized.

15. A method (1100) for managing resources in a network (201), wherein the network comprises a plurality of network access devices (202) and a plurality of terminal devices (203), and wherein the method (1100) comprises: collecting (1101) measurement data (204) over a period of time, wherein the measurement data (204) indicates signal qualities of signals between the network access devices (202) and the terminal devices (203); determining (1102) a historical density distribution (205) of the terminal devices (202) in the network (201) based on the collected measurement data (204); and allocating (1103) resources (206) to the network access devices (202) based on the determined historical density distribution (205).

16. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to perform the method (1100) according to claim 15.