Devices and methods for ai-assisted urllc
Patent Information
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-06-15
- Publication Date
- 2026-03-11
Smart Images
Figure EP2023066157_19122024_PF_FP_ABST
Abstract
Description
[0001] DEVICES AND METHODS FOR AI-ASSISTED URLLC
[0002] TECHNICAL FIELD
[0003] The present disclosure relates to wireless communications. More specifically, the present disclosure relates to devices and methods for Al-assisted ultra-reliable low-latency communication, URLLC.
[0004] BACKGROUND
[0005] Ultra-reliable low-latency communication (URLLC) is a key service for future vertical applications like smart factory, agriculture, healthcare, intelligent transportation, and many more. Centralized grant-based and semi-persistent scheduling of communication resources, while promoting reliable communication, may fail to achieve ultra-low latency for the following reasons. For grant-based scheduling resource allocation requires prior exchange of scheduling grants, which introduces delays. Semi-persistent scheduling may reduce if not eliminate the burden for sending and receiving grants, but it requires pre-defined and known traffic patterns, limiting its flexibility.
[0006] Grant-free random allocation, as envisioned for 5G NR, stands out as a good candidate for decentralized systems, but can still be improved in realistic scenarios with high traffic load of sporadic or aperiodic nature while exhibiting some correlations. User-centric distributed resource allocation allows machines to make autonomous decisions, “disaggregated” from the network. This approach may create collisions during communication, though, which is not compatible with URLLC applications.
[0007] SUMMARY
[0008] It is an objective of the present disclosure to provide improved devices and methods for ultrareliable low-latency communication, URLLC, in particular for Al-assisted URLLC.
[0009] The foregoing and other objectives are achieved by the subject matter of the independent claims. Further implementation forms are apparent from the dependent claims, the description and the figures. According to a first aspect a user equipment, UE, is provided, wherein the UE is one of a plurality of UEs for communication, in particular URLLC in a wireless communication network with a base station over a respective communication channel using one or more resource blocks, RBs, of a plurality of RBs during one or more time slots of a plurality of time slots (herein also referred to as scheduling units, SUs). An RB may span a frequency sub-band (several subcarriers) over multiple subsequent transmission symbols (e.g., OFDM symbols).
[0010] The UE is configured to receive from the base station a multicast feedback signal, wherein the multicast feedback signal comprises information representative of the usage of one or more RBs of the plurality of RBs by the plurality of UEs for one or more past time slots of the plurality of time slots and / or information representative of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots. Moreover, the UE is configured to allocate for a communication with the base station for one or more future time slots one or more RBs of the plurality of RBs, based on the multicast feedback signal, i.e. based on the information representative of the usage of one or more RBs of the plurality of RBs by the plurality of UEs for one or more past time slots of the plurality of time slots and / or information representative of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots. Thus, an improved UE for URLLC with a base station is provided with a resource allocation based on a recorded past and / or predicted future communication behavior of a plurality of UEs with the base station.
[0011] In a further possible implementation form, the UE is configured to allocate the one or more RBs of the plurality of RBs for the communication with the base station for the one or more future time slots, based on the information contained in the multicast feedback signal and based on one or more of RBs of the plurality of RBs allocated by the UE for one or more past time slots of the plurality of time slots.
[0012] In a further possible implementation form, the multicast feedback signal comprises information representative of the usage of one or more RBs of the plurality of RBs by the plurality of UEs for one or more past time slots of the plurality of time slots and wherein the UE is configured to implement a machine learning, ML, model, in particular a neural network configured to estimate the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots based on the information representative of the usage of the one or more RBs of the plurality of RBs by the plurality of UEs for the one or more past time slots of the plurality of time slots.
[0013] In a further possible implementation form, the UE is configured to allocate for the communication with the base station for the one or more future time slots the one or more RBs of the plurality of RBs, based on the estimate of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots.
[0014] In a further possible implementation form, the UE is configured to allocate for the communication with the base station for the one or more future time slots the one or more RBs of the plurality of RBs, based on the estimate of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots, using a Linear Thompson Sampling, LTS, scheme.
[0015] In a further possible implementation form, the information representative of the usage of one or more RBs of the plurality of RBs by the plurality of UEs for one or more past time slots of the plurality of time slots comprises at least one of the following: information representative of whether a transmission using the one or more RBs of the plurality of RBs was successful; information representative of, e.g. a respective identifier of one or more UEs of the plurality of UEs using the one or more RBs of the plurality of RBs; information representative of a collision and / or an outage using the one or more RBs of the plurality of RBs by the plurality of UEs; information indicating that the one or RBs of the plurality of RBs have not been used for a transmission.
[0016] In a further possible implementation form, the multicast feedback signal comprises information representative of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots and wherein the UE is configured to allocate for the communication with the base station for the one or more future time slots the one or more RBs of the plurality of RBs, based on the information representative of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots, using a Linear Thompson Sampling, LTS, scheme.
[0017] According to a second aspect a base station is provided for communication, in particular URLLC in a wireless communication network with a plurality of user equipments, UEs, over a respective communication channel using one or more resource blocks, RBs, of a plurality of RBs during one or more time slots of a plurality of time slots. The base station is configured to transmit a multicast feedback signal to the plurality of UEs, wherein the multicast feedback signal comprises information representative of the usage of one or more RBs of the plurality of RBs by the plurality of UEs for one or more past time slots of the plurality of time slots and / or information representative of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots. Thus, an improved base station for URLLC with a plurality of UEs is provided allowing each UE to allocate resources based on a recorded past and / or predicted future communication behavior of the plurality of UEs with the base station.
[0018] In a further possible implementation form, the base station is configured to implement a machine learning, ML, model, in particular a neural network configured to estimate the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots based on the information representative of the usage of the one or more RBs of the plurality of RBs by the plurality of UEs for the one or more past time slots of the plurality of time slots and wherein the information representative of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots comprises the estimate of the availability of the one or more RBs of the plurality of RBs for the one or more future time slots of the plurality of time slots.
[0019] According to a third aspect a network entity is provided for a wireless communication network, including a base station and a plurality of user equipments, UEs, wherein each UE is configured to communicate with the base station over a respective communication channel using one or more resource blocks, RBs, of a plurality of RBs during one or more time slots of a plurality of time slots. The network entity is configured to transmit to a further UE of the wireless communication network information representative of a machine learning, ML, model, in particular a neural network configured to estimate the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots based on information representative of the usage of one or more RBs of the plurality of RBs by the plurality of UEs for one or more past time slots of the plurality of time slots.
[0020] In a further possible implementation form, the network entity is configured to implement the ML model configured to estimate the availability of the one or more RBs of the plurality of RBs for the one or more future time slots of the plurality of time slots based on the information representative of the usage of one or more RBs of the plurality of RBs by the plurality of UEs for the one or more past time slots of the plurality of time slots.
[0021] In a further possible implementation form, the network entity is configured to receive from one or more of the plurality of UEs the information representative of the ML model configured to estimate the availability of the one or more RBs of the plurality of RBs for the one or more future time slots of the plurality of time slots based on the information representative of the usage of the one or more RBs of the plurality of RBs by the plurality of UEs for the one or more past time slots of the plurality of time slots.
[0022] In a further possible implementation form, the network entity is configured to transmit to the further UE of the wireless communication the network information representative of the machine learning, ML, model, in particular the neural network, in response to receiving a request from the further UE.
[0023] In a further possible implementation form, the request comprises information representative of the ML capabilities of the further UE.
[0024] In a further possible implementation form, the network entity is the base station or a part thereof.
[0025] According to a fourth aspect a method is provided for operating a user equipment, UE, of a plurality of UEs in a wireless communication network for communication, in particular URLLC with a base station over a respective communication channel using one or more resource blocks, RBs, of a plurality of RBs during one or more time slots of a plurality of time slots. The method comprises the steps of: receiving from the base station a multicast feedback signal, wherein the multicast feedback signal comprises information representative of the usage of one or more RBs of the plurality of RBs by the plurality of UEs for one or more past time slots of the plurality of time slots and / or information representative of the availability of one or more RBs of the plurality of RBs for one or more future time slots of the plurality of time slots; and allocating for a communication with the base station for one or more future time slots one or more RBs of the plurality of RBs, based on the multicast feedback signal, i.e. based on the information contained in the multicast feedback signal. Thus, an improved method for operating a UE for URLLC with a base station is provided with a resource allocation based on a recorded past and / or predicted future communication behavior of a plurality of UEs with the base station.
[0026] The method according to the fourth aspect of the present disclosure can be performed by the UE according to the first aspect of the present disclosure. Thus, further features of the method according to the fourth aspect of the present disclosure result directly from the functionality of the UE according to the first aspect of the present disclosure as well as its different implementation forms described above and below.
[0027] According to a fifth aspect, a computer program product is provided, comprising a computer- readable storage medium for storing a program code which causes a computer or a processor to perform the method according to the fourth aspect, when the program code is executed by the computer or the processor.
[0028] Details of one or more embodiments are set forth in the accompanying drawings and the description below. Other features, objects, and advantages will be apparent from the description, drawings, and claims.
[0029] BRIEF DESCRIPTION OF THE DRAWINGS
[0030] In the following, embodiments of the present disclosure are described in more detail with reference to the attached figures and drawings, in which:
[0031] Fig. 1 shows a schematic diagram illustrating a wireless network comprising a plurality of UEs according to an embodiment, a base station according to an embodiment and a network entity according to an embodiment;
[0032] Fig. 2 is a schematic diagram illustrating the interaction between a plurality of UEs according to an embodiment and a base station according to an embodiment for providing Al-assisted URLLC;
[0033] Fig. 3 is a schematic diagram illustrating the operation of a neural network implemented by a UE according to an embodiment; Fig. 4 is a diagram illustrating exemplary feedback about a plurality of resource blocks to be used for allocating resource blocks for a future communication by a UE according to an embodiment;
[0034] Fig. 5 is a schematic diagram illustrating the interaction between a network entity according to an embodiment, a base station according to an embodiment and a further UE for enabling the further UE to participate in the Al-assisted URLLC;
[0035] Fig. 6 is a schematic diagram illustrating the operation of a neural network implemented by a base station according to an embodiment;
[0036] Figs. 7a and 7b show graphs illustrating a signalling performance of a plurality of UEs according to an embodiment and a base station according to an embodiment; and
[0037] Fig. 8 is a flow diagram illustrating a method for operating a UE according to an embodiment.
[0038] In the following, identical reference signs refer to identical or at least functionally equivalent features.
[0039] DETAILED DESCRIPTION OF THE EMBODIMENTS
[0040] In the following description, reference is made to the accompanying figures, which form part of the disclosure, and which show, by way of illustration, specific aspects of embodiments of the present disclosure or specific aspects in which embodiments of the present disclosure may be used. It is understood that embodiments of the present disclosure may be used in other aspects and comprise structural or logical changes not depicted in the figures. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present disclosure is defined by the appended claims.
[0041] For instance, it is to be understood that a disclosure in connection with a described method may also hold true for a corresponding device or system configured to perform the method and vice versa. For example, if one or a plurality of specific method steps are described, a corresponding device may include one or a plurality of units, e.g. functional units, to perform the described one or plurality of method steps (e.g. one unit performing the one or plurality of steps, or a plurality of units each performing one or more of the plurality of steps), even if such one or more units are not explicitly described or illustrated in the figures. Moreover, if a specific apparatus is described based on one or a plurality of units, e.g. functional units, a corresponding method may include one step to perform the functionality of the one or plurality of units (e.g. one step performing the functionality of the one or plurality of units, or a plurality of steps each performing the functionality of one or more of the plurality of units), even if such one or plurality of steps are not explicitly described or illustrated in the figures. Further, it is understood that the features of the various exemplary embodiments and / or aspects described herein may be combined with each other, unless specifically noted otherwise.
[0042] Figure 1 shows a schematic diagram illustrating a wireless communication network 100, in particular a 5G network 100, comprising a plurality of user equipments 110 according to an embodiment, a base station 120 according to an embodiment and a core network entity 130 according to an embodiment. Each UE 110 is configured to communicate with the base station 120 over a respective communication channel (such as the exemplary communication channel 116 shown in figure 2) using one or more resource blocks, RBs, of a plurality of time-frequency RBs (such as the exemplary RBs shown in figure 4) during one or more time slots of a plurality of time slots (such as the exemplary time slots 401 shown in figure 4). A RB may span a frequency sub-band (several subcarriers) over multiple subsequent transmission symbols (e.g., OFDM symbols).
[0043] The wireless communication network 100 may be implemented, for instance, in a smart factory, where the plurality of UEs 110 are connected wirelessly with the base station 120. In such a scenario, smart machines are organized in production lines, whose setup may be subject to frequent changes (due to customization), and each smart machine is associated with a UE 110 creating periodic traffic with jitter. Not all machines may be in service simultaneously, but there are interdependencies, resulting in spatial-temporal correlations of the communications. The UEs 110 may become active / inactive during operation, which may create dynamicity in the activity patterns. As will be described in more detail in the following, embodiments disclosed herein implement a distributed resource allocation scheme for the plurality of UEs 110 to independently (i.e., without centralized control) transmit data in a periodic fashion, while taking into account jitter and spatial-temporal correlations, with the final objective of fulfilling URLLC constraints. As illustrated in figure 1, each UE 110 may comprise a processing circuitry 111 and a communication interface, in particular transceiver 113. The processing circuitry 111 may be implemented in hardware and / or software. The hardware may comprise digital circuitry, or both analog and digital circuitry. Digital circuitry may comprise components such as applicationspecific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), or one or more general-purpose processors. Moreover, the UE 110 may comprise a memory 115 configured to store executable program code which, when executed by the processing circuitry 111, causes the UE 110 to perform the functions and operations described herein.
[0044] Likewise, the base station 120 may comprise a processing circuitry 121 and a communication interface, in particular transceiver 123. The processing circuitry 121 may be implemented in hardware and / or software. The hardware may comprise digital circuitry, or both analog and digital circuitry. Digital circuitry may comprise components such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), or one or more general-purpose processors. Moreover, the base station 120 may comprise a memory 124 configured to store executable program code which, when executed by the processing circuitry 121, causes the base station 120 to perform the functions and operations described herein.
[0045] As further illustrated in figure 1, the core network entity 130 may comprise a processing circuitry 131 and a communication interface 133. The processing circuitry 131 may be implemented in hardware and / or software. The hardware may comprise digital circuitry, or both analog and digital circuitry. Digital circuitry may comprise components such as applicationspecific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), or one or more general-purpose processors. Moreover, the core network entity 130 may comprise a memory 135 configured to store executable program code which, when executed by the processing circuitry 131, causes the core network entity 130 to perform the functions and operations described herein.
[0046] As will be described in more detail below, each UE 110 is configured to receive from the base station 120 a multicast feedback signal (such as the multicast feedback signal 125 shown in figure 2), wherein the multicast feedback signal 125 comprises information representative of the usage of one or more RBs of the plurality of RBs (such as the plurality of RBs 400 illustrated in figure 4) by the plurality of UEs 110 for one or more past time slots of the plurality of time slots 401 and / or information representative of the availability of one or more RBs of the plurality of RBs 400 for one or more future time slots of the plurality of time slots 401. Moreover, each UE 110 is configured to allocate for a communication with the base station 120 for one or more future time slots of the plurality of time slots 401 one or more RBs of the plurality of RBs 400, based on the multicast feedback signal 125, i.e. based on the information provided by the multicast feedback signal 125.
[0047] As will be described in more detail below, in an embodiment, each UE 110 may be equipped with Al, i.e. machine learning, ML, capabilities to learn the spatial-temporal traffic correlations among the plurality of UEs 110 for enabling them to perform informed resource allocation strategies based on a distributed user-centric grant-free access. This may reduce latency, since there is no need for prior exchange of scheduling grants or resource reservation requests, as well as the number of potential (re)transmission attempts (at the cost of an initial configuration phase).
[0048] Figure 2 shows the interaction according to an embodiment between the plurality of UEs 110 and the base station 120 for providing Al-assisted resource selection for URLLC. In the embodiment of figure 2, each UE 110, in particular the processing circuitry 111 of each UE 110, is configured to implement a neural network I l la. In an embodiment, the neural network 11 la of each UE is configured to receive as input data an input state Xt. As will be described in more detail below, the processing circuitry 111 of each UE 110 may be configured to generate a state representation ^Xt) based on the feedback from the base station 120 (referred to as gNB 120 in figure 2), i.e. the multicast feedback signal 125 containing the information representative of the usage of one or more RBs of the plurality of RBs 400 by the plurality of UEs 110 for one or more past time slots of the plurality of time slots 401 and / or information representative of the availability of one or more RBs of the plurality of RBs 400 for one or more future time slots of the plurality of time slots 401. In the embodiment of figure 2, the multicast feedback signal 125 from the base station contains the information representative of the usage of one or more RBs of the plurality of RBs 400 by the plurality of UEs 110 for one or more past time slots of the plurality of time slots 401, i.e. information about the usage of communication resources, in particular RBs used for previous transmissions. Based on the input state Xt the neural network I l la of the UE 110 generates as output the state representation t) providing information about the availability of the RBs for the current time slot, i.e. the scheduling unit t. In the embodiment shown in figure 2, the processing circuitry 111 of each UE 110 is further configured to implement a Thompson Sampling, TS, module 11 lb, which is configured to make, based on the state representation ^Xt) provided by the neural network I l la, a resource allocation decision for the data transmission for the current time slot, i.e. the scheduling unit t, i.e. to select one or more RBs for the current time slot t. Thus, whenever the UE 110 has to transmit data, it may draw the resource allocation for the time slot t from the TS module 111b and then transmit during time slot t using this allocation of RBs. Based on the resource allocation decision of each UE 110 and the conditions of the channel 115 experienced during the time slot, i.e. scheduling unit t, the base station 120 generates the feedback 125, i.e. the multicast feedback signal 125 comprising information representative of the usage of RBs by the plurality of UEs 110 during time slot t.
[0049] Figure 3 is a schematic diagram illustrating in more detail the operation of the neural network I l la and the TS module 111b implemented by the processing circuitry 111 of each UE 110 according to an embodiment. As already described above in the context of figure 2, the neural network I l la illustrated in figure 3 is configured to predict the resource availability for the future time slot t. More specifically, based on the input state Xtgenerated from the feedback signal 125 received from the base station 120, the neural network I l la outputs the state representation t), which may be extracted by the last layer of the neural network I l la and which specifies the resource availability for the future time slot t.
[0050] The TS module 111b illustrated in figure 3 is configured to allocate, i.e. select the communication resources, in particular RBs to be used for data transmission in the future time slot t. More specifically, based on the input in form of the state representation t), the TS module 111b is configured to output a resource allocation decision, i.e., the resource blocks (RBs) to be used for the data transmission.
[0051] As will be appreciated from figure 3, the neural network I l la may be trained with the data obtained from the feedback signals 125 received from the base station 120, and, thus, the training data set may consist of the entire history of states Xt that have been used as the input for the neural network I l la at the earlier time instances {t-1, ..., t-T}. Also, the TS module 111b may exhibit a learning behavior and will adapt its decisions according to the evolution of the state representation ^Xt) over time used as its input.
[0052] In an embodiment, the TS module 111b implemented by the processing circuitry 111 of each UE 110 may be configured to approximate the average number of bits successfully sent in each RB, given the specific state representation ^Xt) by using a linear regression algorithm. The outputs of this algorithm may be used by the TS module 111b to choose the most promising RBs for uplink communications in the future time slot t. By analyzing the feedback 125, the TS module 111b may assess the quality of its allocation decisions and can be optimized accordingly. For example, when collisions happen due to its allocation decision, the TS module 111b may learn to modify its choice when being in that specific state. In this way, the linear regressor may be optimized to properly adapt to the traffic statistics, i.e. the predicted communication pattern of the other UEs 110 in the network 100 for the time slot t. After a certain number of transmitted time slots (and thus TS updates after processing the feedback 125), the neural network I l la is trained to correctly predict the RBs occupancy. In an embodiment, whenever the neural network I l la is trained in this way, the TS module 111b may be re-calibrated to follow the output statistics of the freshly trained neural network I l la. More specifically, a batch of historical state-feedback data may be sampled from a local dataset 115a (which may be stored in the memory 115 of each UE 110) and fed into the neural network I l la for obtaining the new state representations t) for re-calibrating the TS module 111b. In this way, the TS module 111b is tuned to follow the learned statistics reflected by the new state representations. This two-stage learning process may be repeated until convergence.
[0053] Figure 4 is a diagram illustrating exemplary feedback about a plurality of resource blocks 400 for allocating resource blocks (which may be allocated for a plurality of time slots 401 and a plurality of frequency channels 403) for a communication by a UE 110 according to an embodiment and how each UE 110 may generate the state Xtbased on the latest H feedback signals rt125 received from the base station 120 after each completed transmission during the previous time slots t-1, . . ., t-H (also referred to as scheduling units, SU). As already described above, the feedback signal rt125 is multicast by the base station 120 to all the UEs 110 and may indicate the usage of each RB of a plurality of RBs 400 for time slot t (or resource block groups, RBGs, if several RBs are grouped into one unit for reducing signaling). In an embodiment, the base station 120 may be configured to include one or more of the following indicators in the multicast feedback signal 125:
[0054] • rt = i -> successful transmission by UE i,
[0055] • rt= 0 - RB not used,
[0056] • rt= -2 -> collision (non-successful transmission),
[0057] • rt= -1 outage (non-successful transmission).
[0058] Using those indicators, log2(N+3)bits per RB may be employed for usage indication, where N is the total number of UEs 110 within the network 100. In an embodiment, the information provided in the multicast feedback signal 125 distinguishes between outages and collisions, where those two cases both do not allow to decode the signal at the base station 120, but can be discriminated by a threshold on the received power; in particular, for the case of collision, the received power will be above the decoding threshold. This distinction may be beneficial for the Al-based coordinated communication strategy implemented by the embodiments disclosed herein. In the further scope of this document, the information contained in the feedback signal rt125 according to the description given here will be denoted as Feedback Control Information, FCI.
[0059] Figure 5 is a schematic diagram illustrating the interaction between a network entity 130 according to an embodiment, a base station 120 according to an embodiment and a further UE for enabling the further UE to participate in the Al-assisted URLLC. In an embodiment, the network entity 130, in particular the processing circuitry 131 thereof is configured to implement an Artificial Intelligence Program Manager, AIPM, as a support function in the network 100 enabling and coordinating the embodiments disclosed above. The network entity 130, e.g. AIPM 130 may be implemented as part of a mobile core network 140, which may include further entities and functions, such as a policy control function, PCF, 150.
[0060] As will be described in more detail below, the network entity 130, e.g. AIPM 130 is configured to store trained modules (not data) and coordinate the initialization phase for new UEs entering the network 100. In an embodiment, a new UE entering the network 100 may communicate its capability class to the network entity 130, e.g. AIPM 130 indicating whether it is equipped with Al capabilities or not (and thus whether it is capable of using the Al-assisted resource allocation disclosed herein). In an embodiment, the network entity 130, e.g. AIPM 130 may manage one or more of the following tasks: receiving the information on the UEs capability class; managing new UEs joining the Al-assisted resource allocation provided by the network 100; and / or communicating the initial data to a new UE for setting up the Al-assisted resource allocation disclosed herein.
[0061] For incorporating a new UE into the Al-assisted resource allocation provided by the network 100, the network entity 130, e.g. AIPM 130 may be configured to implement the following procedure. At the end of the training process of the plurality of UEs 110, the network entity 130, e.g. AIPM 130 may request a randomly selected UE of the plurality of UEs 110 to upload its trained neural network I l la (using, for instance, a standard data communication protocol). As will be appreciated, all the UEs 110 may be trained with the same data (history of feedback signals 125 multicast to all UEs 110), but may end up with different neural networks I l la. Whenever a new UE wants to participate in the Al-assisted resource allocation implemented by the network 100, it may communicate to the network entity 130, e.g. AIPM 130, its Al capabilities. If the new UE has sufficient Al capabilities, such as computational and / or storage resources, it may download the already trained neural network I l la from the network entity 130, e.g. AIPM 130 (where the trained neural network I l la may be stored in the memory 135 of the network entity 130).
[0062] Based on the already trained neural network I l la, the new UE may optimize just its TS module 11 lb, yielding fast convergence. This TS optimization can be yield by letting the UE perform a “dry-run”. That is, the new UE may select RBs for an upcoming data transmission, but actually does not transmit any data using the allocated RBs. Rather, the new UE collects the feedback signals 125 from the base station 120 for ongoing transmissions, modifies them based on its resource selection, and stores the modified feedback signals 125). The new UE uses the stored feedback 125 to train its neural network I l la and to update its TS module 111b. After sufficient training, the TS module 111b may provide an RB selection that does not result in potential collisions. Once this state is reached, the new UE can be set to an active state and participate in the Al-assisted communication between the plurality of UEs 110 and the base station 120. As will be appreciated, the process described above works best, if only one new UE at a time is optimizing its decisions (since information on other new UEs being trained before participating in the communication may not be present in the feedback signals 125, as they do not actually transmit any data). Thus, in an embodiment, the network entity 130, e.g. AIPM 130 is configured to coordinate the order of new UEs wanting to participate in the Al- assisted resource allocation implemented by the network 100 and running the local training. Figure 6 is a schematic diagram illustrating a further main embodiment, which primarily differs from the embodiments described above in that the neural networks 11 la are not operated by the individual UEs 110, but the base station 120 (in particular its processing circuitry 121) is configured to implement a centralized neural network 121a for providing the functionality described above. In other words, the embodiment illustrated in figure 6 employs a split architecture, where the state is first processed at the base station 120 to provide global information representative of the plurality of UEs 110 and then is multicast as feedback 125 to the plurality of UEs 110 for local optimization of their respective TS modules 111b. Thus, for this embodiment, the feedback signal 125 contains information representative of the availability of one or more RBs of the plurality of RBs 400 for one or more future time slots. In an embodiment, application-specific information or other network-specific details may also be included in the feedback signal 125.
[0063] As will be appreciated, in the embodiment shown in figure 6 the burden of data processing (specifically, training and inference of the neural networks I l la) is shifted from the plurality of UEs 110 (which may have limited computational and / or energy resources) to the base station 120. More specifically, for each time slot the state Xtis processed at the base station 120 (rather than at the UEs 110) by the centralized neural network 121a, producing the shared state representation %). The state representation t) is then multicast as feedback signal 125 to the plurality of UEs 110 and used by each UE 110 to make resource allocation decisions using its respective local TS module 11 lb. However, in an embodiment, only during optimization of the TS module 1 lb, also the FCI (as defined above in Figure 4) for the current time slot may be provided by the base station as additional feedback 125, as it may be used for optimizing the TS module (as described above in the context of figure 3). After a certain number of time slots, the centralized neural network 121a implemented by the processing circuitry 121 of the base station 120 is sufficiently trained. At this stage, the base station may multicast a batch of data (processed with the sufficiently trained neural network 121a) as feedback 125 to the plurality of UEs 110, so that each UE 110 may adapt, i.e. re-calibrate its local TS module 111b to the new state representations t), as already described above.
[0064] In the embodiment shown in figure 6, an additional communication overhead may be introduced, but only during the optimization phase, and not after convergence. This is because, in an embodiment, the base station 120 may communicate the FCI for optimizing the local TS modules 111b together with the state representation ^Xt) to the UEs 110 as feedback 125 every time the centralized neural network 121a is updated, to further re-calibrate the local TS modules 111b of the plurality of UEs 110. However, as already mentioned above, one main advantage of the embodiment shown in figure 6, is to shift the computational burden generated by the neural networks I l la from the UEs 110 to the base station 120, while the TS modules 111b implemented at the UEs 110 are not computationally heavy, since they mainly represent linear regressors.
[0065] For the embodiment shown in figure 6, the network entity, e.g. AIPM 130 may have the same functionalities, as described above, but does not need to store the trained neural network 121a, and does not need to further send it to the UEs. For the embodiment shown in figure 6, the procedure for integrating a new UE into the Al-assisted resource allocation implemented by the network 100 may comprise the following stages. Whenever a new UE wants to participate in the Al-assisted resource allocation implemented by the network 100, it may communicate to the network entity, e.g. AIPM 130 its own Al capabilities. The new UE may then start its “dryrun”. More specifically, the new UE selects RBs for data, but does not transmit. Rather, it informs the base station 120 of its selection by sending, for instance a resource reservation information (RRI) message (as used in 5G V2X sidelink) to the base station 120. The base station 120 uses the RRI information to multicast to the UEs two versions of the state representation t) and the FCI as feedback 125. One version of the feedback 125 is for the new UE, wherein the state representation t) and the FCI contain information regarding the notional transmission of the new UE. The other version of the feedback 125 is for the plurality of UEs already participating in the Al-assisted resource allocation implemented by the network 100, wherein the state representation t) and the FCI only reflect the real transmissions of the actively communicating UEs 110. The feedback 125 from the base station 120 is used by the new UE to complete the dry-run and to optimize (and re-calibrate) the local TS module 111b (in the way already described above). After sufficient training, the TS module 111b provides an RB selection that does not result in potential collisions. Once this state is reached, the new UE can be set active and participate in the communication. Also in this embodiment, the network entity, e.g. AIPM 130 may be configured to coordinate the order of new UEs wanting to join the Al-assisted resource allocation implemented by the network 100, so that only one new UE at a time is performing the initial dry-run. This is done to avoid collisions during the RRI communications. In the following some further embodiments for implementing the signaling between the base station 120 and the plurality of UEs 110 will be described in the context of the PHY channels according to 3GPP. In an embodiment, the multicast feedback signal 125 is readable by all UEs 110. In an embodiment, the base station 120 may be configured to send the multicast feedback signal 125 as Feedback Control Information, FCI, 125 in a form similar to the Downlink Control Information, DCI, defined by the 3GPP standard. The standardized DCI format is characterized by its size and type, which may be a priori unknown by the device that blindly decodes a PDCCH candidate. In an embodiment, each UE 110 may obtain during an initial configuration phase the size and the type of the FCI 125. As for the DCI, for the FCI 125 the identity of the receiving device may be implicitly encoded in the CRC (through the RNTI) and not explicitly transmitted. Thus, in an embodiment, each UE 110 may use the same group-level RNTI, as the information should be decoded by all UEs. In an embodiment, the FCI 125 may be located at the end of an SU, so that the feedback 125 can be sent after each data transmission.
[0066] In an embodiment, the base station 120 may be configured to modify compared with the standard the Master Information Block, MIB, structure for inserting framework-specific information, such as: the size, structure and / or periodicity of the FCI 125; a maximum number of UEs 110 manageable by the system; a maximum number of orthogonal channels that can be used by one UE 110; the size of the state to be used as input to the neural network I l la (specifically, the value H specifying the history size to train the neural network I l la, as described above in the context of figure 4); the size of the neural network 11 la for optimizing the decision strategy, i.e., number of layers and neurons; and / or information on pre-trained models.
[0067] After receiving the MIB, each UE 110 may use the Physical Random Access Channel (PRACH) for communication with the base station 120 via the AIPM 130, if it is equipped with the Al capabilities described herein.
[0068] In an embodiment, the FCI 125 may contain specific information regarding the transmission outcomes at the RB-level, but no indication of the uplink resource allocation (as opposed to typical DCI formats 0 0, 0 1, 0 2) and / or no additional Hybrid- ARQ & Transport fields. Figures 7a and 7b show graphs illustrating a signalling performance of the plurality of UEs 110 and the base station 120 according to an embodiment. More specifically, figure 7a shows the number of bits needed for the FCI and the two DCI versions, as a function of the number of RBs in the system. As will be appreciated, the FCI-based embodiments disclosed herein achieve a comparable / lower overhead than a DCI-based solution in many reasonable configurations. This is because, the FCI-based embodiments disclosed herein scale linearly with the number of RBs manageable by the UEs 110 and logarithmically with the number of UEs 110 in the network 100, while for the DCI it is almost the opposite. Figure 7a shows the two configurations of the DCI, as supported by the 3 GPP standard. As will be appreciated, the results shown in figure 7a do not take into account the uplink overhead needed by the DCI-based protocol to request the resources (which is not required by the FCI-based embodiments disclosed herein).
[0069] Figure 7b shows the number of bits needed for the FCI and two DCI versions, as a function of the number of transmitting UEs 110 in the network 100. As will be appreciated, the FCI-based embodiments disclosed herein achieve a comparable / lower overhead than a DCI-based solution in many reasonable configurations. Figure 7b shows the two configurations of the DCI, as supported by the 3 GPP standard. As will be appreciated, the results shown in figure 7b do not take into account the uplink overhead needed by the DCI-based protocol to request the resources (which is not required by the FCI-based embodiments disclosed herein).
[0070] Figure 8 is a flow diagram illustrating a method 800 for operating a UE of the plurality of UEs 110 in the wireless communication network 100 for communication with the base station 120 over a respective communication channel 115 using one or more RBs of a plurality of RBs 400 during one or more time slots of a plurality of time slots 401. The method 800 comprises a step 801 of receiving from the base station 120 a multicast feedback signal 125, wherein the multicast feedback signal 125 comprises information representative of the usage of one or more RBs of the plurality of RBs 400 by the plurality of UEs 110 for one or more past time slots of the plurality of time slots 401 and / or information representative of the availability of one or more RBs of the plurality of RBs 400 for one or more future time slots of the plurality of time slots 401. Moreover, the method 800 comprises a step 803 of allocating for a communication with the base station 120 for one or more future time slots of the plurality of time slots 401 one or more RBs of the plurality of RBs 400, based on the multicast feedback signal 125. The method 800 can be performed by the UE 110 according to an embodiment. Thus, further features of the method 800 result directly from the functionality of the UE 110 as well as the different embodiments thereof described above and below.
[0071] 5G V2X communication via sidelink promotes an autonomous mode of operation (mode 2), which allows UEs to select their resources for transmission autonomously after observing the ongoing communication for a while (“listen before talk” - LBT) and determining the communication patterns of the other users. Comparing embodiments described above with the autonomous resource allocation scheme in this mode 2, the following distinguishing aspects may be highlighted. In mode 2, resources used by the UEs are indicated by resource reservation information (RRI). Each UE can only receive other UEs’ RRI messages if it does not transmit itself its RRI (unless it has full-duplex capabilities, implying higher hardware costs). The ACK / NACK feedback in mode 2 will be given by the base station, e.g. gNB in DL, in the form of signals dedicated to individual users (i.e., unicast, may even be beamformed). Letting the UEs decode all the ACK / NACK feedback signals requires significant effort for decoding. Moreover, beamformed signals may not be heard by other UEs being located outside of the beam. No distinction between collision and outage is supported in mode 2. Additional uplink resources need to be used to transmit the RRI in mode 2.
[0072] Embodiments disclosed herein provide, for instance, the following advantages with respect to centralized approaches. The feedback signal (FCI, as described in the context of Figure 4) may replace standard ACK / NACK feedback, while according to the embodiments disclosed herein, it is multicast to a (large) group of UEs 110. By reusing fields in the DCI for the FCI 125, the signaling overhead is comparable to that of centralized solutions. Conventional semi-persistent schedulers cannot optimize non-periodic unpredictable traffic or consider jitter. The autonomous decisions made by the UEs 110 according to the embodiments described above allow cutting down the time they would have to wait to access the physical resources in a centralized scheme and, thereby, significantly reducing latency.
[0073] The person skilled in the art will understand that the "blocks" ("units") of the various figures (method and apparatus) represent or describe functionalities of embodiments of the present disclosure (rather than necessarily individual "units" in hardware or software) and thus describe equally functions or features of apparatus embodiments as well as method embodiments (unit = step). In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus, and method may be implemented in other manners. For example, the described embodiment of an apparatus is merely exemplary. For example, the unit division is merely a logical function division and may be another division in an actual implementation. For example, a plurality of units or components may be combined or integrated into another system, or some features may be ignored or not performed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections may be implemented by using some interfaces. The indirect couplings or communication connections between the apparatuses or units may be implemented in electronic, mechanical, or other forms.
[0074] The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
[0075] In addition, functional units in the embodiments of the disclosure may be integrated into one processing unit, or each of the units may exist alone physically, or two or more units may be integrated into one unit.
Claims
CLAIMS1. A user equipment, UE, (110) of a plurality of UEs (110) for communication in a wireless communication network (100) with a base station (120) over a respective communication channel (115) using one or more resource blocks, RBs, of a plurality of RBs (400) during one or more time slots of a plurality of time slots (401), wherein the UE (110) is configured to: receive from the base station (120) a multicast feedback signal (125), wherein the multicast feedback signal (125) comprises information representative of the usage of one or more RBs of the plurality of RBs (400) by the plurality of UEs (110) for one or more past time slots of the plurality of time slots (401) and / or information representative of the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401); and allocate for a communication with the base station (120) for one or more future time slots of the plurality of time slots (401) one or more RBs of the plurality of RBs (400), based on the multicast feedback signal (125).
2. The UE (110) of claim 1, wherein the UE (110) is configured to allocate the one or more RBs of the plurality of RBs (400) for the communication with the base station (120) for the one or more future time slots of the plurality of time slots (401), based on the multicast feedback signal (125) and based on one or more of RBs of the plurality of RBs (400) allocated by the UE (110) for one or more past time slots of the plurality of time slots (401).
3. The UE (110) of claim 1 or 2, wherein the multicast feedback signal (125) comprises information representative of the usage of one or more RBs of the plurality of RBs (400) by the plurality of UEs (110) for one or more past time slots of the plurality of time slots (401) and wherein the UE (110) is configured to implement a machine learning, ML, model (I l la) configured to estimate the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401) based on the information representative of the usage of the one or more RBs of the plurality of RBs (400) by the plurality of UEs (110) for the one or more past time slots of the plurality of time slots (401).
4. The UE (110) of claim 3, wherein the UE (110) is configured to allocate for the communication with the base station (120) for the one or more future time slots of the plurality of time slots (401) the one or more RBs of the plurality of RBs (400), based on the estimate ofthe availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401).
5. The UE (110) of claim 4, wherein the UE (110) is configured to allocate for the communication with the base station (120) for the one or more future time slots of the plurality of time slots (401) the one or more RBs of the plurality of RBs (400), based on the estimate of the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401), using a Linear Thompson Sampling, LTS, scheme (111b).
6. The LE (110) of any one of the preceding claims, wherein the information representative of the usage of one or more RBs of the plurality of RBs (400) by the plurality of LEs (110) for one or more past time slots of the plurality of time slots (401) comprises at least one of the following: information representative of whether a transmission using the one or more RBs of the plurality of RBs (400) was successful; information representative of one or more LEs of the plurality of LEs (110) using the one or more RBs of the plurality of RBs (400); information representative of a collision and / or an outage using the one or more RBs of the plurality of RBs (400) by the plurality of LEs (110); information indicating that the one or more RBs of the plurality of RBs (400) have not been used for a transmission.
7. The LE (110) of claim 1, wherein the multicast feedback signal (125) comprises information representative of the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401) and wherein the LE (110) is configured to allocate for the communication with the base station (120) for the one or more future time slots of the plurality of time slots (401) the one or more RBs of the plurality of RBs (400), based on the information representative of the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401), using a Linear Thompson Sampling, LTS, scheme (11 lb).
8. A base station (120) for communication in a wireless communication network (110) with a plurality of user equipments, LEs, (110) over a respective communication channel (115) using one or more resource blocks, RBs, of a plurality of RBs (400) during one or more time slots of a plurality of time slots (401), wherein the base station (120) is configured to:transmit a multicast feedback signal (125) to the plurality of UEs (110), wherein the multicast feedback signal (125) comprises information representative of the usage of one or more RBs of the plurality of RBs (400) by the plurality of UEs (110) for one or more past time slots of the plurality of time slots (401) and / or information representative of the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401).
9. The base station (120) of claim 8, wherein the base station (120) is configured to implement a machine learning, ML, model (121a) configured to estimate the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401) based on the information representative of the usage of the one or more RBs of the plurality of RBs (400) by the plurality of UEs (110) for the one or more past time slots of the plurality of time slots (401) and wherein the information representative of the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401) comprises the estimate of the availability of the one or more RBs of the plurality of RBs (400) for the one or more future time slots of the plurality of time slots (401).
10. A network entity (120, 130) for a wireless communication network (100), including a base station (120) and a plurality of user equipments, UEs, (110), wherein each UE (110) is configured to communicate with the base station (120) over a respective communication channel (115) using one or more resource blocks, RBs, of a plurality of RBs (400) during one or more time slots of a plurality of time slots (401), wherein the network entity (120, 130) is configured to: transmit to a further UE of the wireless communication network (100) information representative of a machine learning, ML, model (I l la, 121a) configured to estimate the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401) based on information representative of the usage of one or more RBs of the plurality of RBs (400) by the plurality of UEs (110) for one or more past time slots of the plurality of time slots (401).
11. The network entity (120, 130) of claim 10, wherein the network entity (120, 130) is configured to implement the ML model (121a) configured to estimate the availability of the one or more RBs of the plurality of RBs (400) for the one or more future time slots of the plurality of time slots (401) based on the information representative of the usage of one or more RBs ofthe plurality of RBs (400) by the plurality of UEs (110) for the one or more past time slots of the plurality of time slots (401).
12. The network entity (120, 130) of claim 11, wherein the network entity (120, 130) is configured to receive from one or more of the plurality of UEs (110) the information representative of the ML model (I l la) configured to estimate the availability of the one or more RBs of the plurality of RBs (400) for the one or more future time slots of the plurality of time slots (401) based on the information representative of the usage of the one or more RBs of the plurality of RBs (400) by the plurality of UEs (110) for the one or more past time slots of the plurality of time slots (401).
13. The network entity (120, 130) of any one of claims 10 to 12, wherein the network entity (120, 130) is configured to transmit to the further UE of the wireless communication network (100) the information representative of the machine learning, ML, model (I l la, 121a), in response to receiving a request from the further UE.
14. The network entity (120, 130) of claim 13, wherein the request comprises information representative of the ML model capabilities of the further UE.
15. The network entity (120, 130) of any one of claims 10 to 14, wherein the network entity (130) is a part of the base station (120).
16. A method (800) for operating a user equipment, UE, of a plurality of UEs (110) in a wireless communication network (100) for communication with a base station (120) over a respective communication channel (115) using one or more resource blocks, RBs, of a plurality of RBs (400) during one or more time slots of a plurality of time slots (401), wherein the method (800) comprises: receiving (801) from the base station (120) a multicast feedback signal (125), wherein the multicast feedback signal (125) comprises information representative of the usage of one or more RBs of the plurality of RBs (400) by the plurality of UEs (110) for one or more past time slots of the plurality of time slots (401) and / or information representative of the availability of one or more RBs of the plurality of RBs (400) for one or more future time slots of the plurality of time slots (401); andallocating (803) for a communication with the base station (120) for one or more future time slots of the plurality of time slots (401) one or more RBs of the plurality of RBs (400), based on the multicast feedback signal (125).
17. A computer program product comprising a computer-readable storage medium for storing a program code which causes a computer or a processor to perform the method (800) of claim 16, when the program code is executed by the computer or the processor.