Method, communication system and computer program for clustering data flows established through a wireless bridge - Patents.com

Abstract

The invention relates to a method for clustering data flows established through a wireless bridge of a communication system in a time sensitive network, the method comprising obtaining, in a functional entity of a core network component of the wireless bridge, for a plurality of data flows, indicators of sequential and simultaneous communication performance of each said data flow. The data flows are then clustered based on the obtained indicators such that if the difference between the sequential and simultaneous communication performance is lower than a predefined threshold, two given data flows are independent, otherwise they are dependent, and any two data flows belonging to two different clusters are independent and any two data flows of the same cluster are dependent. The invention further relates to a corresponding communication system and a corresponding computer program.

Description

[Technical field]

[0001] The present disclosure relates to the field of telecommunications. [Background technology]

[0002] The present disclosure relates more particularly to a method for clustering data flows established through wireless bridges of a communication system in a time sensitive network, and also to a corresponding communication system and a corresponding computer program.

[0003] This disclosure addresses the problem of using 5G networks as a communication vehicle for time sensitive networks (TSN), which were initially developed to provide guaranteed communications over Ethernet wired networks for industrial applications.

[0004] According to the conventions in current TSN standards, when flowing through a TSN bridge, the end-to-end delay is calculated by taking into account the aggregated delays of the transport links between the preceding bridge (external link) on the path and the internal links of the bridge between the input (ingress) port and the output (egress) port. In a typical TSN network, these aggregated delays are taken into account in a central scheduler, or central network configuration node (CNC), which ensures that no packet collisions occur on the transport links in the time domain.

[0005] In the case of one-to-many bridges, the bottleneck is in the external transport links where a TDMA (Time Division Multiple Access) strategy is applied as a result of CNC scheduling. The internal links of the bridge can be used simultaneously, but they are in fact used consecutively as a result of TDMA on the incoming transport links. The impact on performance is minor when the capacity of the internal links is much larger than that of the transport links. However, the convention of considering aggregated delays induces a large performance degradation when the capacity of the internal links is comparable to or smaller than that of the external links. Indeed, the bridge capacity is limited by not exploiting the possibility of using the internal links simultaneously.

[0006] Such a situation arises when considering a 5G TSN bridge, where the internal link capacity is associated with a wireless transmission capacity that is generally inferior to the input link capacity that relies on wired Ethernet. Using the current state-of-the-art model and representing 5GS as a one-to-many topology, a TDMA strategy is applied between all packets flowing through the 5G TSN bridge. Unfortunately, TDMA is often not the best multiplexing scheme for wireless networks, since other multiplexing dimensions can be utilized with improved performance. Examples of multiplexing dimensions are frequency (using several sub-bands in the frequency domain), space (using multiple antennas), and site (frequency reuse between several non-interfering sites). Thus, the current state-of-the-art definition of the model of the 5G TSN bridge does not utilize the multiplexing capabilities of 5GS, resulting in less than optimal performance.

[0007] The time multiplexing capabilities of 5GS are often considerably below those of Ethernet fixed networks. For example, the smallest time unit of 5GS is a slot lasting at least 100 microseconds, while the maximum frame duration on a Gigabit Ethernet link is about 12 microseconds. The low time granularity of 5GS is therefore taken into account in the declaration of independent delays at the cost of any frame transmission. This is necessary since dependent and independent delays are used in the CNC to calculate the guaranteed delay. However, 5GS necessarily relies on a radio interface that allows broadcasting. This means that several data flows can be transmitted simultaneously with a variable capacity depending on the configuration, terminal position and channel conditions, the number of data flows sharing the radio channel at the same time, etc. This multiplexing capability of 5GS is not essentially taken into account when considering the current state-of-the-art 5GS TSN bridge representation. Indeed, according to this bridge model, the CNC provides packets at the bridge input in TDMA format according to the low internal time granularity of 5GS. Summary of the Invention [Problem to be solved by the invention]

[0008] In order to improve the system performance of wireless bridges such as 5GS, take advantage of the multiplexing capability of the system, and compensate for its low time granularity, there is a need to design a better TSN bridge model for such wireless bridges.

[0009] The present disclosure improves this situation. [Means for solving the problem]

[0010] A method for clustering data flows established through wireless bridges integrated in a time-sensitive network formed by a wireless communication system, comprising: A wireless bridge handles multiple user terminals. The time-sensitive network includes a core network side end station, The time sensitive network further includes a plurality of device side end stations, the device side end stations being attached to the user terminals; At least two data flows are established in the time-sensitive network and forwarded by the wireless bridge, where one data flow is established between a core network side end station and a single device side end station; The method comprises: obtaining, in a functional entity of a core network component of the wireless bridge, for each data flow of the at least two data flows, an indicator of continuous communication performance of said data flow corresponding to when the at least two data flows are established consecutively; obtaining, for each data flow of the at least two data flows, an indicator of concurrency capabilities of said data flows corresponding to when the at least two data flows are established simultaneously; Based on the obtained indicators, Two given data flows are independent if the difference between their sequential and simultaneous communication performance is less than a predefined threshold, and dependent otherwise; Any two data flows that belong to two different clusters are independent and Any two data flows in the same cluster are dependent, clustering at least two data flows into clusters of dependent data flows such that A method is proposed, which includes:

[0011] The particular clustering defined above allows to automatically discover from the data flow measurements a virtual topology of wireless bridges, the so-called TSN model, as a set of independent channels, where each channel corresponds to a cluster and is associated with a pair of end-to-end guaranteed delays, i.e. dependent and independent delays.

[0012] In addition, the end-to-end guaranteed delay of each independent channel described by the TSN model topology can be specifically measured by independently activating the data flows in each cluster and, together with the TSN model, can be provided to a central network configuration entity (CNC) that handles data flow scheduling in TSN.

[0013] In a particular clustering as defined above, the communication performance of a given channel may be defined by one or more delays associated with one or more communications through said given channel.

[0014] In some examples, these delays may be, for example, end-to-end delays between two user terminals connected through the channel, as described above, where these delays may vary depending on the channel activation scheme. Alternatively, in some examples, these delays may include over-the-air transmission delays associated with wireless links between the user terminals and their serving end stations, where activating a given wireless link between a given user terminal and its serving end station is equivalent to activating a given channel between the given user terminal and another user terminal in the communication network.

[0015] The TSN model is accurate and takes into account any type of multiplexing capability of wireless bridges, and the corresponding end-to-end guaranteed delays allow the CNC to calculate routing and scheduling of packets in TSN networks using wireless bridges with improved data flow performance over known methods.

[0016] In another aspect, proposed is a computer software or program comprising one or more instructions for performing at least a part of a method as defined herein when the software is executed by a processor. In another aspect, proposed is a computer readable non-transitory storage medium, in which software is registered to perform a method as defined herein when the software is executed by a processor.

[0017] In another aspect, a communication system in a time sensitive network, the communication system comprises a wireless bridge handling a plurality of user terminals, the time sensitive network comprises a core network side end station, the time sensitive network further comprises a plurality of device side end stations, the device side end stations are attached to the user terminals, at least two data flows are established in the time sensitive network, where one data flow is established between the core network side end station and a single device side end station, the communication system comprises: In a functional entity of a core network component of a wireless bridge, for each data flow of the at least two data flows, an indicator of sequential communication capability of the data flow corresponding to when the at least two data flows are established sequentially, and for each data flow of the at least two data flows, an indicator of simultaneous communication capability of the data flow corresponding to when the at least two data flows are established simultaneously; Based on the obtained indicators, Two given data flows are independent if the difference between their sequential and simultaneous communication performance is less than a predefined threshold, and dependent otherwise; Any two data flows that belong to two different clusters are independent and Any two data flows in the same cluster are dependent, clustering at least two data flows into clusters of dependent data flows such that A communication system is proposed that is configured for:

[0018] The following features can optionally be implemented separately or in combination with each other.

[0019] In one example, the communication performance of a data flow established between a pair of end stations is: a first value of end-to-end latency for uplink communications between a pair of end stations; a second value of end-to-end latency for downlink communications between the pair of end stations; and a maximum value between the first value and the second value; and It may relate to at least one element of the list containing

[0020] The first value and / or the second value may be associated with best effort communication between the end stations.

[0021] In this example, latency, according to several alternative definitions, is selected as the metric for determining communication performance due to its relevance for determining end-to-end guaranteed delay, however, any known QoS metric may be selected instead or in combination with latency.

[0022] When the communication performance of the data flows all relate to values ​​of uplink communication only, the clustered data flows all have the same direction, which is uplink. Conversely, when the communication performance of the data flows all relate to values ​​of uplink communication only, the clustered data flows all have the same direction, which is downlink. This means that a separate set of clustered data flows can be determined and provided to serve as two separate TSN models of the wireless bridge for the uplink and downlink.

[0023] In one example, a given user terminal attached to a device-side end station is active when it receives or transmits a given data flow and is inactive otherwise; The indicator of continuous communication performance of a first data flow of the at least two data flows comprises: a first measurement of communication performance for the first data flow; and a first activity indicator indicating that at least at the time of the first measurement, a first user terminal attached to a device side end station involved in receiving or transmitting a first data flow is active while a second user terminal attached to a device side end station involved in receiving or transmitting a second data flow is inactive; may include The indicator of the concurrency of the first data flow is a second measurement of communication performance of the first data flow; and a second activity indicator indicating that at least the first user terminal and the second user terminal are both active at the time of the second measurement; may include.

[0024] The activity indicator defined above makes it possible to characterize a measurement of communication performance by determining whether such a measurement is to be interpreted as continuous or simultaneous communication performance.

[0025] The activity indicator can be determined from a binary matrix, the binary matrix including a list of binary variables, one binary variable corresponding to one device side end station, the binary variable having a first value when a user terminal attached to the corresponding device side end station is active, and the binary variable having a second value when the user terminal attached to the corresponding device side end station is inactive.

[0026] Such a binary matrix is ​​an effective scheme for storing activity indicators with minimal data storage size requirements, without any restrictions on the number of UEs handled by a wireless bridge.

[0027] The method comprises: and activating the at least two flows according to a sequence of a plurality of predefined activation topologies before obtaining the indicators of sequential and simultaneous communication performance for the at least two flows; For each data flow of the at least two data flows, an indicator of sequential communication performance of said data flow and an indicator of simultaneous communication performance of said data flow are respectively obtained by measurements made under different activation topologies of a sequence of activation topologies.

[0028] Activating the flows according to the above sequence of activation topologies corresponds to injecting traffic with the specific purpose of quickly performing robust clustering by quickly identifying the relative dependencies of a set of data flows in order to enable quickly providing the CNC with an accurate virtual topology of wireless bridges.

[0029] Various schemes are proposed in the following examples to optimize the accuracy of the virtual topology while minimizing the amount of injected traffic.

[0030] In one example, a sequence of activation topologies can be optimized to maximize the variability of communication performance of at least two flows between two successive activation topologies. For example, a given activation topology of a sequence can be iteratively inferred through machine learning based on previous activation topologies of the sequence and associated measured communication performance.

[0031] In another example, the sequence of activation topologies can be designed such that for any two consecutive activation topologies in the sequence of activation topologies, a fixed number of user terminals are active in one of the two consecutive activation topologies and inactive in the other, which is equivalent to setting a fixed distance between the two consecutive activation topologies, where the distance is the number of user terminals that are active in a single one of the two consecutive activation topologies.

[0032] In one example, clustering at least two data flows into a cluster of dependent data flows includes: obtaining an autoencoder comprising an encoder for mapping an input to a code according to a compression matrix and a decoder for mapping the code to a reconstruction of the input according to a restoration matrix, the code being a representation of the input as a set of deep feature variables in a reduced feature space; training an autoencoder using the obtained indicators as input; After training the autoencoder, a compression matrix, a restoration matrix, and a set of deep feature variables are obtained; determining a clustering scheme based at least on the obtained compression matrix; Clustering at least two data flows according to a clustering scheme by using the obtained set of deep feature variables as centroids of clusters; may include.

[0033] Autoencoders are powerful tools that automatically identify patterns in input data through unsupervised learning, making them preferable over more traditional methods such as principal component analysis.

[0034] This particular use of an autoencoder is particularly effective in determining whether groups of data flows are dependent or independent, and consequently in determining clusters of dependent data flows.

[0035] Other features, details and advantages are set forth in the following detailed description and figures. [Brief description of the drawings]

[0036] [Figure 1] FIG. 1 is a diagram of a current state-of-the-art representation of a 5G system as a TSN bridge. [Diagram 2] FIG. 1 is a diagram of a current state-of-the-art representation of a 5G system as a TSN bridge in a TSN network. [Diagram 3] FIG. 1 illustrates a TSN network according to one embodiment. [Figure 4] FIG. 1 illustrates the current state of the art integration of wireless bridges in a TSN network. [Diagram 5] 1 is a flowchart of a software implemented method of clustering input data and associated data transformations according to one embodiment. [Figure 6] FIG. 1 illustrates principal component analysis for data pre-processing, according to one embodiment. [Figure 7] FIG. 1 illustrates an autoencoder for data processing according to one embodiment. [Figure 8] FIG. 1 illustrates a TSN model of a wireless bridge according to one embodiment. [Figure 9] 4 is a flow chart of a software-implemented method for determining the topology of a wireless bridge, according to one embodiment. [Figure 10] 4 is a flow chart of a software-implemented method for determining the topology of a wireless bridge, according to one embodiment. [Figure 11] FIG. 11 is a diagram of an example processing circuit suitable for executing any of the software represented in FIGS. 5, 9, and 10, according to one embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0037] This disclosure addresses the problem of black-box identification of an equivalent TSN bridge model of a wireless system, such as a 5G system, suitable for integrating the wireless system as a logical TSN bridge into a TSN network.

[0038] The identification of this equivalent TSN model can be performed based on application layer measurements from a wireless system (e.g., 5GS), and a deep learning framework can be used.

[0039] Measurements in the application layer (the seventh layer of the OSI model) are considered to be readily available in the core network components of the wireless bridge to determine an equivalent TSN bridge model for the wireless system.

[0040] Application layer measurements can be correlated with various communication resources of the wireless system that are not directly available in a black-box approach. Such communication resources that are not known in advance can include, for example, spatial multiplexing capabilities, broadcast capabilities, and / or frequency reuse factors.

[0041] This disclosure proposes techniques and frameworks for black-box identification of an equivalent TSN bridge model from these application layer measurements.

[0042] <Integration of 5G Systems in TSN Networks: State of the Art at Present> This disclosure focuses on the transmission of packets with delay constraints. To achieve guaranteed end-to-end latency for the transmission of packets between a talker and a listener, it is necessary to rely on a given data path or route and the scheduling of the operation of network elements along the path.

[0043] In the current state of the art approach to integrating 5G systems (5GS) in TSN networks, 5GS is represented using a generic TSN bridge model 100 with a one-to-many topology including ports / gates as shown in Figure 1. Here, the bridge has one NW (network side) port 102 and two DS (device side) ports 104, 106. The internal structure of the 5GS involves the network side translator (NW-TT), which is an element of the 5GS that creates the interface between the TSN network and the 5G user plane (UP) in the core network, and the device side translator (DS-TT), which creates the interface between the 5G UP and the device.

[0044] <TSNネットワーク> The general structure of a TSN network is shown in Figure 3.

[0045] The talker 200 and the listener 202 are both represented as end stations (ES) of the TSN network. A series of bridges 204 are further represented.

[0046] Bridge 204 is a packet switching device of the TSN network. Packets arriving at an ingress port of the bridge can be routed to an egress port with bounded delay according to some parameters specific to each port pair represented by an internal link. The parameters are typically a dependent delay factor (i.e., payload dependent and can be considered as the inverse of the instantaneous throughput of the link) and an independent delay (i.e., a fixed delay per communication).

[0047] The TSN network further comprises a centralized user configuration node (CUC) 206 configured to communicate with the end stations 200, 202 to receive flow requirements.

[0048] The TSN network further comprises a centralized network configuration node (CNC) 208. The CNC can receive QoS indicators from the different TSN bridges 204, such as egress and ingress port identification information, traffic classes, and minimum and maximum delays per port pair.

[0049] The CNC 208 may further receive data related to a user configuration from the CUC 206 through the user / network interface. The data related to the user configuration may include flow requirements of an end station. The data related to the user configuration may further include a TSN network topology related to the flow requirements of the end station. The data related to the user configuration may further include a capacity of a network link according to the topology.

[0050] Having collected the topology and the capacities of the network links, as well as the requirements of the data streams as explained above, the CNC 208 makes a routing decision, i.e., bridge selection, i.e., defines the data path for transmitting packets from one end station 200, through the bridge 204 to another end station 202. This data path is calculated by the CNC in a known manner to meet the stream requirements of the TSN flow.

[0051] The CNC 208 further calculates the scheduling of gate openings for packets to flow from one ES to another ES while guaranteeing transmission delays, with the consistent objective of meeting the stream requirements of the TSN flow. To do this, the CNC calculates the cumulative delay from egress port to egress port of two consecutive bridges in the calculated data path, where the egress port is defined at the output port of the bridge in the flow direction of the packet along the path (while the ingress port is the bridge input port).

[0052] The scheduling calculated in the CNC is a time-gating window table according to which the gates are opened and closed successively. In fact, for each bridge port, packets are buffered and transmitted successively when the corresponding gate is opened. Thus, while the time multiplexing is specific to each bridge port, simultaneous transmissions between two different ports of a TSN bridge are possible.

[0053] <5G system> Although this disclosure focuses primarily on the example of a 5G system, its general principles can also be applied to other wireless systems that have multiplexing capabilities in different dimensions such as time, frequency and space.

[0054] The deployment of 5G systems in factories is related to the integration of 5G systems in time-sensitive networks. In the context of this integration, the 5G system (5GS) acts as at least one Ethernet bridge belonging to the OSI data link layer, i.e. layer 2, which is integrated within the IEEE TSN network.

[0055] According to the fully centralized model of IEEE 802.1Qcc, an overall view of the internal structure of 5GS, together with some elements of the TSN network, is given in Figure 2. The main components of 5GC (5G Core Network) and 5G RAN (5G Radio Access Network) are represented in Figure 2 in the form of nodes or functional entities.

[0056] In particular, a functional division between the control plane 210 and the user plane 214 is observed.

[0057] The control plane involves the PCF (Policy Charging Function) and the TSN-AF (Time Sensitive Network Application Function), both of which are 5GC nodes 212.

[0058] The user plane involves a single UPF (User Plane Function) for routing packets from the 5G-RAN to network (NW) ports. The user plane further involves a gNB (Next Generation Node B) configured to communicate with the 5G-RAN, and more specifically, the UPF, and handle UEs (User Equipment).

[0059] The CUC206 (centralized user configuration node) and CNC208 (centralized network configuration node) shown in Figures 2 and 3 are control elements of the TSN network connected to the 5GS control plane via the TSN-AF in Figure 2.

[0060] Figure 4 shows a known integration of 5GS acting as an Ethernet bridge in a TSN network according to 3GPP technical report TR23.734. In this framework, the 5GS includes one or more ports in a single User Plane Function (UPF) 306 on the network side (NW-TT) 310, which serve as a user plane tunnel between the UPF and the user equipment (UE) 302 through the radio access network 304. The 5GS acting as an Ethernet bridge further includes one port per user equipment on the device side (DS-TT) 300. For each 5GS bridge in the TSN network, the ports on the NW-TT side support connectivity to the TSN network, while the ports on the DS-TT side are each associated with a corresponding protocol data unit (PDU) session and provide connectivity to the TSN network according to the 3GPP technical specification TS23.501.

[0061] In Fig. 4 it is shown that logical bridge configuration information is sent to the TSN network via a trusted TSN Application Function (AF) 308. This configuration information relates to network status and performance parameters, which are received from the Access Management Function (AMF) by the Session Management Function (SMF) in the Unified Data Management (UDM) and sent to the AF via the Network Exposure Function (NEF) using the N33 interface according to 3GPP technical specification TS 23.502, or directly from the Policy and Charging Function (PCF) via the N5 interface if the AF is considered trusted.

[0062] The measurements considered in this disclosure are end-to-end (E2E) measurements related to QoS KPIs at the application level, i.e., at the application layer of the OSI model, i.e., layer 7. The QoS KPIs are E2E latency measured in uplink or downlink, E2E throughput measured in uplink or downlink, or E2E link reliability measured in uplink / downlink, and the reference points for this E2E measurements can be, for example, DS-TT (device-side TSN converter) and NW-TT (network-side TSN converter).

[0063] For multiple deployments of TSN-capable UEs in a network, a UE is activated if it is sending / receiving a TSN-PDU session corresponding to a TSN data flow from a TSN talker. A UE is not activated if it is not sending / receiving a PDU session corresponding to a TSN flow from a TSN talker to a TSN listener.

[0064] QoS KPIs are measured in the 3GPP domain over the N6 interface linking the User Plane Function (UPF) to the NW-TT in the uplink, and over interface N60 for downlink transmission.

[0065] The QoS KPI measurements necessarily relate to data flows associated with an active (or activated) UE, hereafter referred to as the "reference UE" and denoted as UE(0), which is considered as part of a deployment of N UEs handled by the 5GS, where a UE is denoted as UE(i), where i is a natural number ranging from 0 to N-1.

[0066] It is proposed herein to record, for each QoS KPI measurement associated with UE(0), the active UEs that are activated together with UE(0) during the measurement. The other UEs in the deployment are assumed to be inactive. Then, for each QoS KPI measurement associated with UE(0), the record is stored as, for example, a binary matrix [x1, x2, ..., x N-1 ], where, when a UE(i) of the deployment is activated, a binary variable x i = 1, otherwise the binary variable x i =0.

[0067] The binary matrix defined above is hereinafter referred to as measurement label or activation topology label, or simply label. The QoS KPI measurements are organized as a labeled performance table, where the measurements of activated UEs are labeled by activation topology labels.

[0068] <Clustering> Clustering is a key step that aims to learn groups of dependent TSN data flows from their QoS KPI measurements and from their associated activation topology labels.

[0069] A virtual TSN bridge topology is then determined from the group of dependent TSN data flows.

[0070] Consider a pair of TSN data flows. The pair are said to be "dependent" if the measured end-to-end QoS KPIs of one TSN data flow in the pair change depending on whether the other TSN data flow in the pair is simultaneously active or not.

[0071] More specifically, it is possible to define sequential and simultaneous communication performance. Simultaneous communication performance is reflected by the measured end-to-end QoS KPIs of a given data flow of a pair for an activation topology in which both data flows of the pair are activated together. Sequential communication performance, in contrast, is reflected by the measured end-to-end QoS KPIs of a given data flow of a pair for an activation topology in which both data flows of the pair are activated one at a time, rather than together.

[0072] For example, a pair of UEs formed by UE(0) and UE(1) are dependent if the QoS KPI measurements associated with the UEs differ depending on whether x1=1 (concurrent communication performance) or x1=0 (sequential communication performance).

[0073] This principle applies more generally to any group of two or more data flows, where the sequential communication performance specifies the measured end-to-end QoS KPIs of a given data flow for an activation topology where none of the other data flows of the group are activated at the same time, and the concurrent communication performance of a given data flow specifies the measured end-to-end QoS KPIs of a given data flow for an activation topology where at least one other data flow of the group is activated at the same time.

[0074] The absolute value of the difference between the sequential and concurrent performance of a given data flow can be calculated, which is referred to in the following as "QoS KPI change". A group of data flows is said to be dependent when the data flows of the group exhibit a QoS KPI change that exceeds a predefined threshold. Otherwise, the group of data flows is said to be independent. The threshold can be preset to a non-zero value, which can be positive or negative. The sign and absolute value of the threshold are selected depending on the application.

[0075] <Clustering metric> The following section presents possible options for QoS KPIs that can be used in the clustering step. The QoS KPIs are described in the Service Level Specifications (SLS) that are sent from the CUC to the TSN talkers and listeners and used by the CNC to calculate the schedule across the TSN network.

[0076] By way of example, a QoS KPI contemplated herein is E2E latency or delay. An example of a key QoS KPI clustering metric corresponding to E2E latency is: - E2E latency of communication of TSN data flows between one DS-TT and one NW-TT, i.e., uplink TSN data flow transmission across a 5GS network; - E2E latency of communication of TSN data flows between one NW-TT and one DS-TT, i.e., uplink-downlink data flow transfer across a 5GS network; - the maximum E2E latency between uplink and downlink data flow transmission in a 5GS network; - communication of TSN data flows between DS-TT and NW-TT, i.e. best-effort E2E latency of best-effort uplink TSN data flow transport over 5GS; - communication of TSN data flows between NW-TT and DS-TT, i.e. best-effort E2E latency of best-effort downlink TSN data flow transfer over 5GS; - the maximum best effort E2E latency between uplink and downlink data flow transmissions in a 5GS network; Includes.

[0077] <Clustering algorithm> A clustering algorithm is used to learn to cluster UEs into groups of dependent QoS KPIs. The input of the clustering algorithm is the labeled performance table defined above, which contains the recorded QoS KPI measurements organized and labeled by the activation topology labels.

[0078] Discovering the performance table can be done in a variety of ways.

[0079] For example, data flows can be ordered (labeled) and UEs associated with the data flows can be activated pairwise in succession. The performance of the data flows is measured for each activation topology and stored in a data table. This corresponds to the successive discovery of pairs of data flows.

[0080] Alternatively, data flows may be labeled and activated pairwise in succession in the domain of the expansion, and then the performance of the data flows is measured and stored in a data table, which corresponds to a lexicographical discovery of the data flows.

[0081] When a UE is activated for the purpose of measuring the performance of a data flow, said data flow may be associated with an active PDU session or a dummy PDU session.

[0082] Clustering algorithms belong to a general class of algorithms that are suitable for finding features in a performance table in a similar way that image segmentation algorithms find regions in an image: these features represent sets of data flows that have a certain performance across multiple activation topologies, i.e., sets of independent data flows.

[0083] From this obtained set of independent data flows, clusters of dependent data flows are constructed and output by a clustering algorithm.

[0084] Reference is now made to FIG. 5, which is a flow chart illustrating the general steps of the clustering algorithm in an exemplary embodiment.

[0085] These general steps are: - data pre-processing 400 of the input data 406 (labeled performance tables); - then combining feature detection in the pre-processed data 408 with segmentation 402 of the pre-processed data based on the detected features; - Finally, post-processing 404 of the segmented data 410 to obtain and output clustered data 414 (clusters formed from a set of dependent data flows); Includes.

[0086] <Data pre-processing> The goal of data preprocessing is to optimize deep clustering performance by permuting the input data, i.e., the performance tables, to a particular activation topology order that allows for simplified discovery of clusters of dependent data flows.

[0087] The activation topology order can be determined by various criteria.

[0088] An example of a criterion is to maximize the variability of the flow performance between two successive activation topologies. For example, a principal component analysis of the flow performance data can be performed. This makes it possible to obtain the activation topology that maximizes the variance of the flow performance as an eigenvector of the covariance matrix of the performance data.

[0089] Another example of a criterion is to combine the distance between two consecutive activation topologies. For example, a lexicographical order can be selected in which a selected subset of flows are activated such that every two consecutively activated flow sets are distinct and have a distance of k flows.

number

number

number

[0090] Another example is a combination or aggregation of at least two of the orderings described above.

[0091] FIG. 6 shows a principal components analysis on a typical set of input data 406, here represented as a scatter plot, where the principal components determine the directions of maximum variance in the flow data set.

[0092] <Feature detection> Feature detection allows us to perform an actual clustering of the pre-processed dataset (the sorted performance table) in order to learn some features shared by different flows for different activation topologies, i.e. similar levels of performance variability over activation topology changes.

[0093] One common technique for performing this clustering is to perform K-means clustering on the pre-processed dataset.

[0094] First, a subset of flows is selected arbitrarily or randomly, which means that each flow in the subset is selected arbitrarily or randomly. Furthermore, the number of flows forming the subset can also be selected arbitrarily or randomly.

[0095] Each selected flow is assigned to a different cluster as its cluster center.

[0096] Then, iteratively, - for an existing cluster center, the flow exhibiting the closest change in its QoS performance measure to the activation topology change is added to said cluster; - The cluster center is updated as the average of the QoS performance measures within the cluster.

[0097] Once all flows have been assigned to clusters, K-means clustering automatically converges.

[0098] However, this standard clustering does not converge well in practice for large input data sets, and in particular, due to the initialization of the clusters, this standard clustering may not provide a satisfactory cluster representation of the input data set.

[0099] For this reason, we propose to use an autoencoder for the purpose of deep clustering. An autoencoder is a part of a regenerative neural network that is trained to reconstruct the dataset as shown in Figure 7.

[0100] As shown in FIG. 7, an autoencoder is a pair of an encoder 500 and a decoder 502 that learns a deep representation of a dataset by a backpropagation algorithm that attempts to reconstruct the input data at the output of the encoder / decoder pair.

[0101] The encoder portion of the autoencoder compresses the preprocessed data 408 (x1, x2, x3, x4) as shown in FIG. 7 into a reduced feature space 504 represented by the cluster centers (z1, z2) through the following compression relationship:

number

[0102] The decoder portion reconstructs the original dataset 506 from the feature space through the reconstruction equation:

number

[0103] Function f α and f β is a non-linear function that can be, for example, a sigmoid function defined as follows:

number

[0104] The autoencoder learns the parameters W for clustering the data and the parameters R for reconstructing the data jointly with the sigmoid slope α by performing the following procedure, known as the backpropagation technique: First, the entries of matrices W and R are assigned random values, and the slope of the sigmoid is assigned loose values, such as α=β=0.1. - Using the compression and reconstruction equations, we obtain the deep feature variables (z1, z2) - Input data x and reconstructed data

number

[0105] Once the autoencoder parameters W, R, α, and β are learned from the backpropagation technique, the encoder part of the autoencoder is used as a clustering scheme for the input data as follows: - Compression matrix W and nonlinear function f α is used to define the clustering scheme, -The deep feature variables (z1, z2) are used as the centroids of the clusters of dependent data flows to be constructed.

[0106] <Post-processing> To further improve the clustering scheme, post-processing is performed.

[0107] The post-processing can rely on the K-means clustering method as described above, initialized by using the deep feature variables (z1, z2) output by the autoencoder as the centroids of the clusters of dependent data flows. Other known clustering methods can also be applied, as long as the deep feature variables (z1, z2) output by the autoencoder are used as descriptive data for the clusters of dependent data flows in the initialization phase.

[0108] For example, another option is to further improve the separation of the clusters by calculating the probability of association of the data to the clusters and adjusting the clusters by inter-cluster Kullback-Leibler divergence maximization.

[0109] <Proposed TSN model for 5GS> Now referring to FIG.

[0110] Once a group of UEs has been formed and identified, it is possible to determine a main bridge model 600 that indicates whether different UEs have contention access to time resources or, conversely, whether their simultaneous activation does not affect data flow performance.

[0111] The topology of the main bridge model 600 is as follows:

[0112] The main bridge model includes as many NW ports 602 as the number of identified groups of UEs. The main bridge model further includes as many DS ports 604, 606 as the total number of UEs across all identified groups of UEs. The main bridge model further includes internal links between each NW port and all UEs that belong to the associated group.

[0113] Alternatively, the main bridge model can be determined as multiple model parts, one for each group of UEs, with each model part having only one NW port and as many internal links and DS ports as there are UEs in the associated group.

[0114] Any configuration that integrates these model parts is possible, as long as the configuration link between the NW port associated with the group and the DS port associated with the UE is preserved.

[0115] The capacity of the internal links of the main bridge model can be learned through measurements, i.e., a data packet is transmitted to any UE and the capacity or delay is measured. As a note, UEs of the same group should be learned independently, which means that the obtained capacity for a given internal link is the capacity when the corresponding UE is the only activated UE among the group of UEs to which it belongs.

[0116] External links arriving at any ingress port of the main bridge model can be assumed to have zero dependent and independent delays.

[0117] Current state-of-the-art models of TSN bridges often include a single NW port, and therefore, for compatibility reasons, it is expected that any model of a TSN bridge will also be assumed to include a single NW port.

[0118] To meet this assumption, a Demux bridge model 608 can be optionally introduced with a single NW port 610 and as many DS ports 612 as the number of specified groups of UEs. The Demux bridge further includes, for each DS port, an internal link with the NW port. These internal links have zero independent and dependent delays, which corresponds to infinite capacity. The external link has the characteristics of a physical interface that links the 5GS CN to the TSN network, which is usually an Ethernet interface.

[0119] As a result, the end-to-end delay for transmitting two identical packets from upstream bridge B0 to an end station associated with an egress port of B is related as 1 / C'1+1 / C'2 when considering the current state-of-the-art model, whereas in the new model of this disclosure, it is related as 1 / min(C'1,C'2), thus fully utilizing the multiplexing capability of 5GS and reducing the end-to-end latency.

[0120] <Application to uplink> Although the above examples focus on the downlink, one skilled in the art can easily apply the same principles for the uplink, where the ingress port is a DS port and the egress port is a NW port.

[0121] Then, two different models are constructed for the downlink and uplink.

[0122] In practice, the uplink radio access technique may differ from that of the downlink.

[0123] This may result in different schemes for grouping UEs and may result in different capacity measurements for the internal links of the bridge.

[0124] Considering the uplink in particular, the main bridge model has as many ingress ports as the number of UEs handled by the gNB of the wireless bridge, and as many egress ports as the number of groups of UEs.

[0125] Optionally, the egress ports of the main bridge models are each connected to an ingress port of a multiplexing (Mux) model having a single egress port.

[0126] <Algorithms and processing circuits> FIG. 9 is a flowchart representing an overall algorithm of a computer program that can be stored on a storage medium and / or conveyed as a signal for executing a method for determining the topology of a time-sensitive network model of a wireless bridge of a communication system in a time-sensitive network for downlink applications.

[0127] The algorithm includes obtaining 700 a partitioning of a set of P user equipments handled by a wireless bridge into N groups of user equipments, where N and P are both natural numbers strictly greater than 1. Such partitioning is the clustered data 414 obtained as output of post-processing 404 when the performance table 406 pertains to downlink applications, i.e., downlink data flows from talker 200 to listener 202.

[0128] As already mentioned, a user equipment is activated when it is involved in the transmission of at least one data flow through the wireless bridge, and data flow performance is associated with a given data flow of the at least one data flow, such that for any first subset of user equipment belonging to the same group of user equipment, simultaneous activation of all user equipment of the first subset degrades the data flow performance in the wireless bridge relative to activation of a single user equipment among the first subset, and for any second subset of user equipment, all belonging to a different group, simultaneous activation of all user equipment of the second subset does not degrade the data flow performance in the wireless bridge relative to activation of a single user equipment among the second subset.

[0129] Next, the algorithm includes constructing (702) a topology of a time-sensitive network model of the wireless bridge, where the time-sensitive network model includes at least a main bridge model, where the main bridge model includes N network side ports, where the network ports are associated with corresponding groups of user equipment handled by the wireless bridge, P device side ports, where the device side ports are associated with corresponding user equipment, and P internal links, where the internal links connect the device side ports associated with a given group of user equipment with the network side ports associated with the given group.

[0130] The algorithm finally includes providing (704) the topology of the time-sensitive network model to a centralized network configuration node within the communication system.

[0131] 10 is a flow chart representing an overall algorithm of a computer program that can be stored on a storage medium and / or conveyed as a signal for executing the method for determining the topology of a time-sensitive network model of a wireless bridge of a communication system in a time-sensitive network for an uplink application. Such a division is the clustered data 414 obtained as the output of the post-processing 404 when the performance table 406 relates to an uplink application, i.e., an uplink data flow from the listener 202 to the talker 200.

[0132] The algorithm includes dividing the set of P user equipment into M groups of user equipment (800), such that uplink data flow performance is associated with an uplink data flow defined as a data transfer from each activated user equipment to the wireless bridge, and such that for any first subset of user equipment belonging to the same group of user equipment, simultaneous activation of all user equipment of the first subset does not degrade uplink data flow performance at the wireless bridge relative to activation of a single user equipment among the first subset, and such that for any second subset of user equipment, all belonging to a different group, simultaneous activation of all user equipment of the second subset does not degrade uplink data flow performance at the wireless bridge relative to activation of a single user equipment among the second subset.

[0133] Next, the algorithm includes constructing (802) a topology of an uplink primary bridge model, the uplink primary bridge model including M network side ports, where the network ports are associated with corresponding groups of user equipment handled by the wireless bridge, P device side ports, where the device side ports are associated with corresponding user equipment, and P internal links, where the internal links connect the device side ports associated with a given group of user equipment with the network side ports associated with the given group.

[0134] The algorithm finally includes providing (804) the topology of the uplink main bridge model, or (more generally) the topology of a time-sensitive network model of wireless bridges including such an uplink main bridge model, to a centralized network configuration node in the communication system.

[0135] Figure 11 shows a schematic representation of a processing circuit 900 suitable for executing any one or more of the algorithms of figures 5, 9 and 10. The processing circuit comprises a memory 904 for storing a computer program including said algorithm or algorithms. The processing circuit further comprises a processor 902 for accessing the memory and executing the computer program. The processing circuit further comprises a communication interface 906 controllable by the processor for transmitting at least said topology, or multiple topologies, of the time sensitive network model to a centralized network configuration node in the communication system.

[0136] Of course, it does not matter whether the processor 902 uses a single core or multiple cores to execute a computer program. Cloud computing technologies can also be used in which a computer program is executed across processors of multiple processing circuits.

[0137] List of citations For the purposes, the following non-patent literature is cited: - nplcit1: Lu Si et Li Ruisi. DAC: Deep Autoencoder-based Clustering, a General Deep Learning Framework of Representation Learning. arXiv preprint arXiv:2102.07472, 2021; - nplcit2: Tom M. Mitchell, “Machine learning”, McGraw Hill education; and - nplcit3: Asperti, A., & Trentin, M. (2020). Balancing reconstruction error and Kullback-Leibler divergence in Variational Autoencoders. IEEE Access, 8, 199440-199448.

Claims

1. A method for clustering data flows established through wireless bridges integrated in a time-sensitive network formed by a wireless communication system, comprising: the wireless bridge handles a plurality of user terminals; the time-sensitive network includes a core network side end station; The time-sensitive network further includes a plurality of device-side end stations, the device-side end stations being attached to user terminals; At least two data flows are established in the time-sensitive network and forwarded by the wireless bridge, where one data flow is established between the core network side end station and a single device side end station; The method comprises: obtaining, in a functional entity of a core network element of the wireless bridge, for each data flow of the at least two data flows, an indicator of continuous communication performance of the data flow corresponding to when the at least two data flows are established successively; obtaining, for each data flow of the at least two data flows, an indicator of concurrency capabilities of the data flows corresponding to when the at least two data flows are established simultaneously; Based on the obtained indicator, Two given data flows are independent if the difference between the sequential communication performance and the simultaneous communication performance is less than a predetermined threshold, and are dependent otherwise; Any two data flows that belong to two different clusters are independent, Any two data flows in the same cluster are dependent, clustering the at least two data flows into clusters of dependent data flows such that A method comprising:

2. The communication performance of a data flow established between a pair of end stations is a first value of end-to-end latency for uplink communications between the pair of end stations; a second value of the end-to-end latency of downlink communications between the pair of end stations; and a maximum value between the first value and the second value; and The method of claim 1 , wherein the at least one element of the list comprises:

3. The method of claim 2 , wherein the first value and / or the second value are associated with best effort communication between the end stations.

4. A given user terminal attached to a device-side end station is active when it receives or transmits a given data flow and is inactive otherwise; The indicator of the continuous communication performance of a first data flow of the at least two data flows comprises: a first measure of communication performance for the first data flow; and a first activity indicator indicating that at least a first user terminal attached to the device side end station involved in receiving or transmitting the first data flow is active while a second user terminal attached to the device side end station involved in receiving or transmitting the second data flow is inactive at the time of the first measurement; Including, The indicator of the concurrency of the first data flow is: a second measure of communication performance for the first data flow; and a second activity indicator indicating that both the first user terminal and the second user terminal are active at least at the time of the second measurement; and The method according to any one of claims 1 to 3, comprising:

5. 5. The method of claim 4, wherein the activity indicator is determined from a binary matrix, the binary matrix including a list of binary variables, one binary variable corresponding to one device side end station, the binary variable having a first value when the user terminal attached to the corresponding device side end station is active, and the binary variable having a second value when the user terminal attached to the corresponding device side end station is inactive.

6. and activating the at least two flows according to a sequence of a plurality of predefined activation topologies prior to obtaining the indicators of sequential and simultaneous communication capabilities for the at least two flows; 6. The method of claim 4 or 5, wherein for each data flow of the at least two data flows, the indicator of the sequential communication performance of the data flow and the indicator of the simultaneous communication performance of the data flow are respectively obtained by measurements made under different activation topologies of the sequence of activation topologies.

7. The method of claim 6 , wherein the sequence of activation topologies is optimized to maximize variability of the communication performance of the at least two flows between two successive activation topologies.

8. The method according to claim 6 or 7, wherein for any two successive activation topologies of the sequence, a fixed number of user terminals are active in one of the two successive activation topologies and inactive in the other one.

9. The clustering of the at least two data flows into a cluster of dependent data flows comprises: obtaining an autoencoder comprising an encoder for mapping an input to a code according to a compression matrix and a decoder for mapping said code to a reconstruction of said input according to a restoration matrix, said code being a representation of said input as a set of deep feature variables in a reduced feature space; training the autoencoder using the obtained indicators of sequential and simultaneous communication performance as the input; After training the autoencoder, obtaining the compression matrix, the decompression matrix, and the set of deep feature variables; determining a clustering scheme based at least on the obtained compression matrix; clustering the at least two data flows according to the clustering scheme by using the obtained set of deep feature variables as centroids of the clusters; The method according to any one of claims 1 to 8, comprising:

10. The method according to any one of claims 1 to 9, wherein the clustered data flows all have the same direction, either uplink or downlink.

11. A communication system in a time-sensitive network, the communication system comprising a wireless bridge handling a plurality of user terminals, the time-sensitive network comprising a core-network side end station, the time-sensitive network further comprising a plurality of device side end stations, the device side end stations being attached to user terminals, at least two data flows being established in the time-sensitive network, where one data flow is established between the core-network side end station and a single device side end station, the communication system comprising: obtaining, in a functional entity of a core network element of the wireless bridge, for each of the at least two data flows, an indicator of sequential communication capability of the data flow corresponding to when the at least two data flows are established sequentially, and obtaining, for each of the at least two data flows, an indicator of simultaneous communication capability of the data flow corresponding to when the at least two data flows are established simultaneously; Based on the obtained indicator, Two given data flows are independent if the difference between the sequential communication performance and the simultaneous communication performance is less than a predetermined threshold, and are dependent otherwise; Any two data flows that belong to two different clusters are independent, Any two data flows in the same cluster are dependent, clustering the at least two data flows into clusters of dependent data flows such that A communication system configured for:

12. A computer program comprising instructions which, when said program is executed by a computer, cause said computer to carry out the method according to any one of claims 1 to 10.

Images