Methods for semantic communication in a communication network

The method aligns sender and receiver nodes in mobile communication networks by using alignment nodes to semantically encode and decode content samples, addressing the challenge of establishing a common understanding without shared knowledge, thereby enabling efficient semantic communication.

WO2026235576A1PCT designated stage Publication Date: 2026-11-12TELEFONAKTIEBOLAGET LM ERICSSON (PUBL)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/SE2025/050904
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-07
Filing Date
2025-10-07
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Achieving alignment between sender and receiver nodes in mobile communication networks for semantic communication is challenging due to the vast range of information to be encoded or decoded at higher layers and the reluctance of chipset manufacturers to share private information, making it difficult to establish a common knowledge base.

Method used

A method involving alignment nodes that use a feedback mechanism to align sender and receiver nodes by semantically encoding and decoding content samples, generating a measure of difference, and providing alignment instructions to improve decoding accuracy without relying on predefined shared information.

Benefits of technology

Enables alignment of sender and receiver nodes for semantic communication by building a common understanding of the content to be transmitted, reducing data overhead and facilitating efficient communication sessions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure SE2025050904_12112026_PF_FP_ABST
    Figure SE2025050904_12112026_PF_FP_ABST
Patent Text Reader

Abstract

Methods (300, 400, 500, 600) are disclosed for facilitating semantic communication between a sender node and a receiver node in a communication network. The methods involve initiating a semantic encoder at the sender node to semantically encode a content sample (320), and obtaining a version of the content sample that has been reconstructed by a semantic decoder at the receiver node (340). The methods further comprise using a measure of difference between the original and reconstructed content sample to generate an alignment instruction for the semantic decoder at the receiver node (360), and implementing the alignment instruction at the receiver node (650).
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Methods for Semantic Communication in a Communication Network

[0002] Technical Field

[0003] The present disclosure relates to methods for facilitating semantic communication between a sender node and a receiver node in a communication network. The methods may be performed by first and second alignment nodes, by a sender node, and by a receiver node respectively. The present disclosure also relates to first and second alignment nodes, a sender node, and a receiver node, and to a computer program product configured, when run on a computer, to carry out methods for facilitating semantic communication between a sender node and a receiver node in a communication network.

[0004] Background

[0005] Semantic communication (also referred to as semantic compression), represents a paradigm shift from traditional communication by focusing not just on the transmission of raw data but on the accurate exchange of meaning between two endpoints. Unlike traditional communication systems, which prioritize fidelity in signal reconstruction, semantic communication systems aim to understand and convey the intent or meaning of the transmitted information. This approach has the potential to significantly reduce the data overhead by transmitting only the essential semantic elements relevant to the communication goal, rather than all the data.

[0006] Figure 1 illustrates comparison of semantic communication (lower part of the Figure) with respect to traditional compression (upper part of the Figure) by way of an example. For a given piece of content (an image of a flower in the illustrated example), traditional compression techniques would involve lossless or lossy compression followed by transmission of the compressed content and decompression on the receiving end. Semantic communication by contrast describes the content instead of compressing it, transmitting the description of the content, which enables the receiving end to reproduce the content or a semantically equivalent version of the content.

[0007] In the literature state of art, encoder-decoder types of Artificial Neural Networks (ANNs), such as transformers, have provided an interesting solution to the problem of semantic communication across multiple modalities that include text, video and audio. Adiscussion of this is included in Yating Liu, Xiaojie Wang, Zhaolong Ning, MengChu Zhou, Lei Guo, Behrouz Jedari, A survey on semantic communications: Technologies, solutions, applications and challenges, Digital Communications and Networks, Volume 10, Issue 3, 2024, Pages 528-545, ISSN 2352-8648.

[0008] This technology is also quite promising for the wireless interface (also known as “air interface”) of mobile networks, as it can enable support for use cases that involve transmission of data-intensive content, such as high-resolution extended reality (XR), and allow them to scale to multiple sessions. Such scaling would be impossible with traditional compression algorithms.

[0009] A common denominator of all the approaches to semantic communication is that they require alignment of sender and receiver such that there is a common understanding between sender and receiver regarding the content for transmission. Typical means of achieving this alignment is sharing of a common knowledge base. This knowledge base can for example be a common dataset that both semantic encoder (sender) and semantic decoder (receiver) are trained on. In another example, the semantic encoder and semantic decoder share a (predefined) set of ontologies, which describe a common understanding of content that is to be semantically encoded or decoded.

[0010] Achieving alignment of sender and receiver based on predefined information, whether symbolic (for example, in the form of ontologies) or in datasets, presents certain difficulties in the context of mobile communication networks. Firstly, when semantic communication occurs at higher layers, such as the application layer, the range of information to be encoded or decoded becomes vast, as different applications may require encoding and decoding of diverse types of content. The quantity of shared information required to align sender and receiver is consequently excessively large. Secondly, when semantic communication occurs at lower layers, it may require support from chipset manufacturers, who may not be willing to share private information (such as model architecture, common training datasets or large ontologies) with competitors or with mobile network vendors. Obtaining a common knowledge base in a multi-vendor setting can therefore be extremely difficult, and alignment of sender and receiver to facilitate semantic communication in communication networks thus remains an open challenge.

[0011] SummaryIt is an aim of the present disclosure to provide methods, first and second alignment nodes, a sender node, and a receiver node, and a computer program product which at least partially address one or more of the challenges mentioned above. It is a further aim of the present disclosure to provide methods, first and second alignment nodes, a sender node, and a receiver node, and a computer program product which cooperate to achieve alignment of a sender node and receiver node for a semantic communication session between the sender node and the receiver node. Features of the methods disclosed herein are provided in the independent claims. Additional features are set out in the dependent claims.

[0012] According to a first aspect of the present disclosure, there is provided a method for facilitating semantic communication between a sender node and a receiver node in a communication network. The method, performed by a first alignment node in the communication network, comprises obtaining a sample of content, wherein the content is to be transmitted from the sender node to the receiver node within a semantic communication session. The method further comprises initiating a semantic encoder at the sender node to semantically encode the content sample, and transmitting the semantically encoded content sample to a second alignment node in the communication network. The method further comprises obtaining, from the second alignment node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample, and generating a measure of difference between the content sample and the reconstructed version of the content sample. The method further comprises using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node, and transmitting the alignment instruction to the second alignment node for implementation in the receiver node.

[0013] According to another aspect of the present disclosure, there is provided another method for facilitating semantic communication between a sender node and a receiver node in a communication network. The method, performed by the sender node, comprises extracting a sample of content, wherein the content is to be transmitted from the sender node to the receiver node within a semantic communication session, and providing the sample of content to a first alignment node in the communication network. The method further comprises, on request from the first alignment node, semantically encoding thecontent sample using a semantic encoder at the sender node, and providing to the first alignment node the semantically encoded content sample.

[0014] According to another aspect of the present disclosure, there is provided another method for facilitating semantic communication between a sender node and a receiver node in a communication network. The method, performed by a second alignment node in the communication network, comprises receiving, from a first alignment node in the communication network, a semantically encoded content sample, and obtaining, from the receiver node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample. The method further comprises transmitting the reconstructed version of the content sample to the first alignment node, receiving, from the first alignment node, an alignment instruction for implementation in the receiver node, and providing the alignment instruction to the receiver node for implementation.

[0015] According to another aspect of the present disclosure, there is provided another method for facilitating semantic communication between a sender node and a receiver node in a communication network. The method, performed by the receiver node, comprises receiving, from a second alignment node in the communication network, a semantically encoded content sample, and generating, using a semantic decoder at the receiver node, a reconstructed version of the content sample. The method further comprises providing the reconstructed version of the content sample to the second alignment node, receiving, from the second alignment node, an alignment instruction for implementation in the receiver node, and implementing the alignment instruction with respect to the semantic decoder at the receiver node.

[0016] According to another aspect of the present disclosure, there is provided a computer program product comprising a computer readable non-transitory medium, the computer readable medium having computer readable code embodied therein, the computer readable code being configured such that, on execution by a suitable computer or processor, the computer or processor is caused to perform a method according to any one of the aspects or examples of the present disclosure.

[0017] According to another aspect of the present disclosure, there is provided a first alignment node for facilitating semantic communication between a sender node and a receiver node in a communication network. The first alignment node comprises processingcircuitry configured to cause the first alignment node to obtain a sample of content to be transmitted from the sender node to the receiver node over a semantic communication session, and to initiate a semantic encoder at the sender node to semantically encode the content sample. The processing circuitry is further configured to cause the first alignment node to transmit the semantically encoded content sample to a second alignment node in the communication network, and to obtain, from the second alignment node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample. The processing circuitry is further configured to cause the first alignment node to generate a measure of difference between the content sample and the reconstructed version of the content sample, to use the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node, and to transmit the alignment instruction to the second alignment node for implementation in the receiver node.

[0018] According to another aspect of the present disclosure, there is provided a second alignment node for facilitating semantic communication between a sender node and a receiver node in a communication network. The second alignment node comprises processing circuitry configured to cause the second alignment node to receive, from a first alignment node in the communication network, a semantically encoded content sample, and to obtain, from the receiver node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample. The processing circuitry is further configured to cause the second alignment node to transmit the reconstructed version of the content sample to the first alignment node, to receive, from the first alignment node, an alignment instruction for implementation in the receiver node, and to provide the alignment instruction to the receiver node for implementation.

[0019] According to another aspect of the present disclosure, there is provided a sender node for facilitating semantic communication between the sender node and a receiver node in a communication network. The sender node comprises processing circuitry configured to cause the sender node to extract a sample of content to be transmitted from the sender node to the receiver node over a semantic communication session, and to provide the sample of content to a first alignment node in the communication network. The processing circuitry is further configured to cause the sender node, on request from the first alignment node, to semantically encode the content sample using a semanticencoder at the sender node, and to provide to the first alignment node the semantically encoded content sample.

[0020] According to another aspect of the present disclosure, there is provided a receiver node for facilitating semantic communication between a sender node and the receiver node in a communication network. The receiver node comprises processing circuitry configured to cause the receiver node to receive, from a second alignment node in the communication network, a semantically encoded content sample, and to generate, using a semantic decoder at the receiver node, a reconstructed version of the content sample. The processing circuitry is further configured to cause the receiver node to provide the reconstructed version of the content sample to the second alignment node, to receive, from the second alignment node, an alignment instruction for implementation in the receiver node, and to implement the alignment instruction with respect to the semantic decoder at the receiver node.

[0021] Aspects of the present disclosure thus provide methods and nodes that cooperate to align a sender and receiver for a semantic communication session based on the type of content to be transmitted from the former to the latter. Example methods according to the present disclosure present signaling which, when carried out before the semantic communication session begins, allows sender and receiver to build a common understanding of the concepts to be transmitted through a feedback mechanism, and without requiring knowledge of previously agreed shared information in the form of ontologies or shared datasets.

[0022] Brief Description of the Drawings

[0023] For a better understanding of the present disclosure, and to show more clearly how it may be carried into effect, reference will now be made, by way of example, to the following drawings in which:

[0024] Figure 1 illustrates comparison of semantic communication with respect to traditional compression by way of an example;

[0025] Figure 2 is a block diagram illustrating components of an example architecture that may implement methods disclosed herein;Figures 3A and 3B show a flow chart illustrating process steps in a computer implemented method for facilitating semantic communication between a sender node and a receiver node in a communication network;

[0026] Figure 4 is a flow chart illustrating process steps in another computer implemented method for facilitating semantic communication between a sender node and a receiver node in a communication network;

[0027] Figure 5 is a flow chart illustrating process steps in another computer implemented method for facilitating semantic communication between a sender node and a receiver node in a communication network;

[0028] Figure 6 is a flow chart illustrating process steps in another computer implemented method for facilitating semantic communication between a sender node and a receiver node in a communication network;

[0029] Figure 7 is a block diagram illustrating functional modules in an example first alignment node;

[0030] Figure 8 is a block diagram illustrating functional modules in an example sender node;

[0031] Figure 9 is a block diagram illustrating functional modules in an example second alignment node;

[0032] Figure 10 is a block diagram illustrating functional modules in an example receiver node;

[0033] Figures 11A and 11 B show a sequence diagram illustrating an example implementation of the methods disclosed herein;

[0034] Figure 12A illustrates pairs of semantic encoders and decoders;

[0035] Figure 12B illustrates an architecture for an alignment ML model;

[0036] Figure 13 shows a network node in accordance with some examples; and

[0037] Figure 14 shows a virtualization environment in accordance with some examples.Detailed

[0038]

[0039] Examples of the present disclosure propose methods according to which, prior to a semantic communication session between a sender node and a receiver node, a signalling exchange is used to enable sender and receiver to build a common understanding of the content to be transmitted. In this manner alignment of semantic encoder at the sender, and semantic decoder at the receiver, is achieved without relying on predefined shared information.

[0040] Example methods disclosed herein thus use incoming content to be transmitted over a semantic communication session to achieve alignment. This content may be actual or predicted. Prediction of content for transmission allows for preparation of the semantic communication session even before the actual content for transmission is available. Given incoming content to be semantically encoded, the sender may initialize a “probing period” during which alignment of sender semantic encoder and receiver semantic decoder is achieved via execution of example methods disclosed herein. The length of the probing period may be determined by an indication of the complexity and variety of the content. Content that is substantially homogeneous and / or of low complexity may require less time for alignment than highly heterogeneous or complex content, for which a longer signaling exchange may be required to achieve suitable shared understanding of the content between sender and receiver. During the probing period, as part of execution of example methods according to the present disclosure, a sample of the content for transmission is prepared by the sender, semantically encoded, and transmitted to the receiver. A version of the content sample is then reconstructed by the receiver, and returned to the sender side. The reconstructed content is then assessed at the sender side, and suggestions prepared as to corrections that can be applied to improve the accuracy of the reconstructed content. These suggestions are provided to the receiver. This loop may be repeated as long as is required to achieve an acceptable level of accuracy in the reconstruction of the content sample by the receiver. Different metrics may be used to asses the degree of accuracy, as discussed below with reference to Figure 3. A record of the alignment process may then be stored in an experience buffer, and can be reused in future semantic communication sessions.

[0041] Figure 2 is a block diagram illustrating components of an example architecture that may implement methods disclosed herein in a communication network. Referring to Figure2, the sender 210 and receiver 220 each comprise network nodes capable of some means of communication. Examples of such nodes include, but are not limited to, a User Equipment (UE), and radio base stations such as 5G Node-B (gNB), each of which may assume the roles of sender and receiver interchangeably, for uplink and downlink traffic. Each of the sender 210 and receiver 220 is associated with an alignment node, illustrated in the figures as an alignment function 212, 222. This alignment function is a logical node and can be physically collocated with the associated sender 210 or receiver 220 (as illustrated in the Figure) or can be cloud services.

[0042] Figure 2 also illustrates a process flow between the architecture elements demonstrating an example implementation of methods disclosed herein. The process flow of Figure 2 provides an overview of how methods performed at respective alignment nodes, sender node and receiver node may cooperate to achieve alignment of sender and receiver for a semantic communication session. Referring to Figure 2, the example process flow proceeds as follows:

[0043] 1. The sender 210 receives some incoming content.

[0044] 2. Based on the incoming content a “content analysis” entity 214 performs an analysis and sends samples of the content to the alignment function 212 (step 2.b) and also sends all content that is to be transmitted for temporary storage in a content buffer 216 (step 2. a).

[0045] The alignment function 212 requests a semantic encoder 218 at the sender 210 to encode the content sample (step 2.c).

[0046] 3. A capabilities exchange may be carried out between the sender and receiver and or their respective alignment functions.

[0047] 4. The alignment function 212 then sends the semantically encoded content sample to the alignment function 222 on the receiver side.

[0048] The receiver 220 causes its semantic decoder 228 to generate a reconstructed version of the content sample based on the received description (semantically encoded content sample) for sending back to the sender side.

[0049] 5. The receiver alignment function 222 then sends the reconstructed version of the content sample back to the sender side alignment function 212.

[0050] The sender side alignment function 212 performs a comparison between the reconstructed version of the content sample received from the receiver 220 and the original content sample and outputs a similarity indicator.6. Based on the value of the similarity indicator the sender side alignment function 212 may provide instructions to the receiver, in the form of a prompt, or using symbols or other notation, to adjust the receiver decoding capabilities.

[0051] The sample transmission, comparison, suggestion feedback loop may repeat as long as necessary, i.e., until the sender’s similarity indicator is above (or consistently above) a specific value.

[0052] 7. Once alignment has been achieved, the sender side alignment function 212 initiates the semantic communication session.

[0053] 8. The alignment function 212 may also store the sequence of corrections made to the receiver, together with a description of the content, and optionally a profile of the sender and receiver to an experience buffer 230. The experience buffer may be used by the alignment function in the future to accelerate the alignment process between the sender and receiver.

[0054] 9. With the semantic communication session initiated, the sender semantic encoder 218 starts pulling and encoding content to be transmitted from the content buffer 216.

[0055] 10. The semantic communication session can then continue as long as is required.

[0056] It will be appreciated that the process flow described above involves methods carried out at each of the sender and receiver alignment functions 212, 222 as well as the sender node 210 and the receiver node 220. In some examples, the alignment functions, which are themselves logical nodes, may comprise component parts of the sender and receiver nodes, but in other examples, the sender and receiver alignment functions may be instantiated in the cloud or some other distributed architecture. The following discussion therefore addresses methods carried out in respective sender and receiver alignment nodes separately from the methods conducted in the corresponding sender node and receiver node. It will be appreciated that the communication between sender node and associated sender alignment node, and receiver node and associated receiver alignment node, may therefore vary according to the deployment scenario for a given use case (collocated or separate alignment functions).

[0057] Figures 3A and 3B show a flow chart illustrating process steps in a computer implemented method 300 for facilitating semantic communication between a sender node and a receiver node in a communication network. The facilitation may involve supporting a process for alignment between a semantic encoder at the sender node, and a semantic decoder at the receiver node. The sender and receiver nodes may beincluded in a plurality of nodes forming a distributed system, and the method may be conducted with reference to semantic communication between any pair of nodes in a network. Individual nodes may act as sender and / or receiver nodes for a semantic communication session, according to different use cases.

[0058] The method 300 may be performed by a first alignment node, which may comprise a physical or virtual node, and may be implemented in a computer system, computing device or server apparatus and / or in a virtualized environment, for example in a cloud, edge cloud, or fog deployment. Examples of a virtual node may include a piece of software or computer program, a code fragment operable to implement a computer program, a virtualised function, or any other logical entity. The first alignment node may comprise a component entity in a communications network, which may comprise a 4th, 5thor other generation 3GPP network, a local network such as WiFi, an enterprise network, a data centre network, an Internet of Things network, or any other existing or future communications network. The first alignment node may in some examples be implemented within a sender node, as discussed below, and / or within functionality for communications network management, such as in the Network Management Domain. The first alignment node may encompass multiple logical entities, as discussed in greater detail below, and may for example comprise a Virtualised Network Function (VNF). In some examples, the first alignment node may be implemented as an rApp within an O-RAN architecture. The first alignment node cooperates with a sender node performing an example of the method 400 discussed below. The first alignment node may be collocated with the sender node, or may be separate from the sender node, for example operating as a virtualised function in a cloud deployment.

[0059] Referring initially to Figure 3A, the method 300 comprises, in a first step 310, obtaining a sample of content. As illustrated in step 310, the content is to be transmitted from the sender node to the receiver node within a semantic communication session. Step 310 may comprise obtaining a single sample of content, or a plurality of content samples. In some examples, step 310 may comprise obtaining a stream of content samples over a period of time. The content samples may in some examples be understood as a control flow of information for supporting the alignment of sender semantic encoder and receiver semantic decoder. By using, during the subsequent method steps, samples of the content that is to be transmitted within the semantic communication session, examples of the method 300 ensure that on completion of the method, the sender semantic encoder and receiver semantic decoder are aligned for the content that is to betransmitted. That is, the sender and receiver have a shared understanding of the content that is to be transmitted. The alignment is therefore specific to this content.

[0060] In step 320, the method 300 comprises initiating a semantic encoder at the sender node to semantically encode the content sample. The method then comprises, in step 330, transmitting the semantically encoded content sample to a second alignment node in the communication network. The second alignment node may be associated with the receiver node for the sematic communication session. In some examples the second alignment node may be collocated with the receiver node, or the second alignment node may be instantiated in a distributed deployment, and be in communication with the receiver node. The semantically encoded content sample may be transmitted to the second alignment node over any suitable communication channel.

[0061] Referring now to Figure 3B, in step 340, the method 300 comprises obtaining, from the second alignment node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample. The method then comprises generating a measure of difference between the content sample and the reconstructed version of the content sample in step 350, and using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node in step 360. In step 370, the method comprises transmitting the alignment instruction to the second alignment node for implementation in the receiver node. In some examples of the method 300, transmitting the alignment instruction to the second alignment node for implementation in the receiver node may comprise causing the sender node to transmit the alignment instruction.

[0062] In some examples of the method 300, the alignment instruction may comprise data operable to enable the semantic decoder at the receiver node to adjust the output generated by the semantic decoder for a given input. The alignment instruction may consequently comprise any one or more of training data for training the semantic decoder, instructions for generating training data for training the semantic decoder, original and semantically encoded versions of content (the semantically encoded version of the content having been encoded by the semantic encoder of the sender node), and / or contextual information for use by the semantic decoder at the receiver node during generation of its output. The nature of the alignment instruction may vary according to the type of semantic decoder present at the receiver node. The different examples of alignment instruction are discussed in greater detail below with reference to Figures 11Aand 11 B, however it will be appreciated that the alignment instruction effectively serves to enable the receiver node to update the functioning of its semantic decoder to ensure that the output that it generates from semantically encoded content more closely resembles the original content that it is intended to reproduce.

[0063] In some examples, the step 350 of generating a measure of difference between the content sample and the reconstructed version of the content sample may comprise performing a comparison of the content sample and the reconstructed version of the content sample, and generating, from the comparison, the difference measure as a representation of a difference between the content sample and the reconstructed version of the content sample. The comparison may be a semantic, symbol level comparison, and / or may be statistical comparison, according to the nature of the content, and the degree of accuracy required for the reconstruction of the content during the semantic communication session. Similarly, the difference measure may comprise a semantic description of the difference, a numerical representation of the difference such as a percentage difference, etc. Further detail of the difference measure is discussed below with reference to Figures 11A and 11 B.

[0064] Following step 350, the method 300 may further comprise checking whether the measure of difference fulfils an alignment criterion, and, if the measure of difference does fulfil the alignment criterion, initiating a semantic communication session between the sender node and the receiver node. The alignment criterion may be specified in any manner suitable for the measure of difference, thus the alignment criterion may be a threshold, such as THRESHOLD2 illustrated in Figure 11 A, or may be a criterion expressed in natural language, symbolically, or in any other appropriate manner. Different metrics can be considered to assess the alignment criterion, including for example mean squared error, measuring the distance between original and reconstructed material, but also other types such as sentence similarity if the content comprises natural language. If the content is in the vision domain, metrics such as Peak Signal-to-Noise Ratio (PSNR) and structural similarity index measure (SSIM) can be considered.

[0065] In some examples of the method 300, it may be that performing the method steps 360 (using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node), and 370 (transmitting the alignment instruction to the second alignment node for implementation in the receiver node) is conditional upon the measure of difference failing to fulfil the alignment criterion. That is, if the measure ofdifference fulfils the alignment criterion, then the sender node semantic encoder and receiver node semantic decoder are sufficiently aligned for the content to be transmitted (as represented by the obtained samples), and consequently further alignment instruction is not required, and the semantic communication session can be initiated.

[0066] In some examples, step 360 of using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node may comprise mapping the measure of difference to an alignment instruction using at least one of a manual mapping, a heuristic process, and / or a Machine Learning, ML, model. The nature of the mapping may in some examples be selected according to the nature of the comparison performed in to generate the measure of difference. For example, if a semantic comparison is performed then a symbolic reasoner may be used to generate a description of the difference between the reconstructed and original content, and consequently what should be changed in the reconstructed content in order to improve the reconstruction. Different examples of generation of the alignment instruction are discussed below with reference to Figure 11 A.

[0067] In further examples of the method 300, the method may also comprise using the measure of difference to generate an alignment instruction for the semantic encoder at the sender node, and providing the alignment instruction to the sender node for implementation in the sender node. Thus it may be envisaged that the sender node semantic encoder is also updated based on the reconstructed content from the receiver node, supporting the process developing a shared understanding between encoder and decoder of the content to be transmitted. It will be appreciated that the above discussed detail with respect to the alignment instruction for the receiver node semantic decoder, and its generation, applies equally to the alignment instruction for the sender node semantic encoder, and its generation.

[0068] In some examples of the method 300, the method may further comprise repeating iterations of the steps 340 (obtaining a version of the content sample that has been reconstructed by a semantic decoder at the receiver node), 350 (generating a measure of difference between the content sample and the reconstructed version of the content sample), checking whether the measure of difference fulfils an alignment criterion, and if not, 360 (using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node), and 370 (transmitting the alignment instruction to the second alignment node for implementation in the receiver node). The iterations ofthese steps may be repeated until the measure of difference fulfils the alignment criterion. Each iteration may thus be understood as a check to see whether the previous alignment instruction was sufficient to bring the reconstructed version close enough to the original content sample to fulfil the alignment criterion. Once the alignment criterion is fulfilled, the first alignment node may initiate the semantic communication session, as discussed above.

[0069] In some examples of the present disclosure, it may be that the receiver node cannot, or for some reason will not, implement the alignment instruction. This may be the case for example if the receiver node perceives a risk that its decoder model will be contaminated with data that could negatively impact future communication sessions. In such examples, the method 300 may further comprise receiving, from at least one of the second alignment node or the receiver node, at least one of a refusal by the receiver node to implement the alignment instruction and / or a request from the receiver node for an alternative alignment instruction. The first alignment node may then generate and provide an alternative instruction, or may instruct or initiate the sender and receiver nodes to fall back to conventional (non-semantic) communication for transmission of the content.

[0070] In some examples of the method 300, the method may further comprise storing an experience record for semantic alignment of the receiver node in a repository such as an Experience Buffer. The experience record may comprise a profile of the content, a profile of the receiver node semantic decoder, and an adaptation strategy for the semantic decoder at the receiver node, the adaptation strategy comprising at least one alignment instruction for the semantic decoder at the receiver node. In the case of multiple iterations of the steps of 340 to 370 as discussed above, the adaptation strategy may comprise all of the alignment instructions generated and transmitted for the receiver node semantic decoder. It will be appreciated that the profile of the content, and of the receiver node semantic decoder, may be obtained from the sender node.

[0071] In some examples of the present disclosure, and as discussed in further detail with reference to the method 400 below, it may be that the sender node has been able to obtain a suitable adaptation strategy from the repository, based on a model profile of the receiver node and the content to be transmitted. In such examples, it may be that the method 300 further comprises receiving, from the sender node, an adaptation strategy of an experience record stored in the repository, the adaptation strategy comprising atleast one alignment instruction for a semantic decoder at a receiver node. The method 300 may then further comprise transmitting the adaptation strategy to a second alignment node for implementation in the receiver node, and initiating a semantic communication session between the sender node and the receiver node. It will be appreciated that receipt of the adaptation strategy and subsequent steps may in some examples take place before method step 310, in which case it may be that the remaining method steps are omitted. Thus in one alternative, the first alignment node either receives content samples and performs the steps of examples of the method 300 as set out above, and in another alternative, the first alignment node simply receives an adaptation strategy from the sender node, and provides this strategy to the second alignment node for implementation in the receiver node before initiating the semantic communication session.

[0072] In some examples of the present disclosure, it may be envisaged that a Machine Learning (ML) model is trained using supervised learning approaches to approximate the functionality of the first alignment node. This training may be implemented via additional method steps in the method 300, as discussed below. In such examples, the method 300 may further comprise collecting a training data set to train an alignment ML model to estimate the measure of difference between the content sample and the reconstructed version of the content sample, and using the collected training data set to train the alignment ML model. The method 300 may further comprise, while the output of the alignment ML model fulfils an accuracy criterion, and during additional iterations of the method 300, using the alignment ML model to replace the steps of:

[0073] transmitting the semantically encoded content sample to a second alignment node in the communication network (330);

[0074] obtaining, from the second alignment node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample (340);

[0075] generating a measure of difference between the content sample and the reconstructed version of the content sample (350).

[0076] The accuracy criterion may be defined according to the needs and circumstances of a given use case. For example the accuracy criterion may detect when data drift in the content to be encoded means that the alignment ML model is no longer providing a good estimation of the delta between the desired and obtained decoded content. In some examples of the present disclosure the first alignment node may revert to performing themethod steps as set out above when the accuracy criterion is not fulfilled, so as to generate more training data and update the training of the alignment ML model.

[0077] In some examples, the alignment ML model may be operable to receive as input a first version of the content sample that has been semantically encoded by a semantic encoder at the sender node, and a second version of the content sample that has been semantically encoded by a semantic encoder at the receiver node. The alignment ML model may be further operable to process the input according to current values of its trainable parameters, and generate an output comprising an estimation of the measure of difference between the content sample and the reconstructed version of the content sample. The alignment ML model may be further operable to generate an alignment instruction corresponding to the estimated measure of difference.

[0078] In some examples, the alignment ML model may comprise a first classifier model operable to classify an output of the semantic encoder at the sender node, a second classifier model operable to classify an output of the semantic encoder at the receiver node, and a conversion model operable to convert the output of the first and second classifiers into the estimated measure of difference.

[0079] In some examples, the training dataset may comprise a plurality of training pairs of input and output data, each training pair comprising input data comprising a first version of the content sample that has been semantically encoded by a semantic encoder at the sender node and a second version of the content sample that has been semantically encoded by a semantic encoder at the receiver node, and output data comprising a measure of difference between the content sample and a reconstructed version of the content sample obtained by the alignment node during execution of the method 300. The output data may in some examples further comprise an alignment instruction generated by the alignment node for the semantic decoder at the receiver node during execution of the method 300.

[0080] For the purposes of the present disclosure, the term “ML model” encompasses within its scope the following concepts:

[0081] machine learning algorithms, comprising processes or instructions through which data may be used in a training process to generate a model artefact for performing a given task, or for representing a real-world process or system; andthe model artefact that is created by such a training process, and which comprises the computational architecture that performs the task.

[0082] The method 300 may be complemented by a method 400 performed by a sender node. Figure 4 is a flow chart illustrating process steps in a method 400 for facilitating semantic communication between a sender node and a receiver node in a communication network. The method 400 may be performed by a sender node, which may comprise a physical or virtual node, and may be implemented in a computer system, computing device or server apparatus and / or in a virtualized environment, for example in a cloud, edge cloud, or fog deployment. Examples of a virtual node may include a piece of software or computer program, a code fragment operable to implement a computer program, a virtualised function, or any other logical entity. The sender node may comprise a component entity in a communications network, which may comprise a 4th, 5thor other generation 3GPP network, a local network such as WiFi, an enterprise network, a data centre network, an Internet of Things network, or any other existing or future communications network. The sender node may be implemented in a Radio Access node, which itself may comprise a physical node and / or a virtualized network function that is operable to exchange wireless signals. In some examples, a Radio Access node may comprise a base station node such as a NodeB, eNodeB, gNodeB, or any future implementation of this functionality. The sender node may encompass multiple logical entities, as discussed in greater detail below, and may for example comprise a Virtualised Network Function (VNF). In other examples, the sender node may be implemented in a computing device, which may be a wireless device, a User Equipment, a wired device, a constrained device, etc. The sender node cooperates with a first alignment node performing an example of the method 300. The first alignment node may be collocated with the sender node, or may be separate from the sender node, for example operating as a virtualised function in a cloud deployment.

[0083] Referring to Figure 4, the method 400 comprises, in step 410, extracting a sample of content. As illustrated in step 410, the content is to be transmitted from the sender node to the receiver node over a semantic communication session. Step 410 may comprise extracting a single sample of content, or a plurality of content samples. In some examples, step 410 may comprise extracting a stream of content samples over a period of time. The content samples may in some examples be understood as a control flow of information for supporting the alignment of sender semantic encoder and receiver semantic decoder. Extracting content samples in step 410 may for example compriseextracting samples of the content that are unique, and / or that are representative of the variety and breadth of the content to be transmitted. In further examples, extracting one or more content samples at step 410 may comprise extracting parts of the content that do not violate privacy, or that can be decoded and later compared in due time, so as not to delay the alignment process. Content samples could for example include parts of an image, frames of a video, a summary of text, etc.

[0084] In step 420, the method comprises providing the sample or samples of content to a first alignment node in the communication network. In step 430, the method 400 comprises, on request from the first alignment node, semantically encoding the content sample using a semantic encoder at the sender node, and providing to the first alignment node the semantically encoded content sample.

[0085] In some examples the method 400 may further comprise receiving, from the first alignment node, an alignment instruction for implementation in the sender node, and implementing the alignment instruction with respect to the semantic encoder at the sender node. The alignment instruction may be as described above with reference to the method 300, and may enable the semantic encoder at the sender node to be updated such that its semantic encoding results in a more accurate reconstruction of the content by the receiver node semantic decoder. Thus it may be envisaged that the sender node semantic encoder is also updated using an instruction generated by the first alignment node based on the reconstructed content from the receiver node. This may support the process of developing a shared understanding between encoder and decoder of the content to be transmitted.

[0086] In some examples the method 400 may further comprise obtaining a profile of the receiver node semantic decoder, and providing the profile of the receiver node semantic decoder, as well as a profile of the content to be transmitted from the sender node to the receiver node, to a repository. The content profile may be generated by the sender node as an additional step in the method 400. The content profile may for example include content metadata, timestamps, subject matter, resolution, etc. as well as predictions of how the content may change over time. The method 400 may further comprise receiving, from the repository (i) an adaptation strategy of an experience record stored in the repository, the adaptation strategy comprising at least one alignment instruction for a semantic decoder at a receiver node, and (ii) a measure of similarity between the content profile and semantic decoder profile provided to the repository, and the content profileand semantic decoder profile of the experience record stored in the repository. In some examples the sender node may additionally provide the model profile of the receiver node semantic decoder to the first alignment node (for example in the event that the retrieved adaptation strategy is not a good match - as discussed in further detail below - so as to enable the model profile to be stored with the content profile and alignment instruction(s) as a new experience record).

[0087] In some examples of the method 400, if the measure of similarity fulfils a matching criterion, the method 400 may further comprise providing the adaptation strategy to the first alignment node to be transmitted to a second alignment node for implementation in the receiver node. In some examples, performing the method steps 410 (extracting a sample of content to be transmitted from the sender node to the receiver node over a semantic communication session), 420 (providing the sample of content to a first alignment node in the communication network), and 430 (on request from the first alignment node, semantically encoding the content sample using a semantic encoder at the sender node, and providing to the first alignment node the semantically encoded content sample), may be conditional upon the measure of similarity failing to fulfil the matching criterion. Thus the alignment process of steps 410 to 430 may only be carried out if the experience fetched from the repository is not a good enough match.

[0088] The method 400 may further comprise receiving an instruction from the first alignment node to commence a semantic communication session with the receiver node, and commencing semantic encoding and transmission of the content to be transmitted from the sender node to the receiver node. In some examples the content for transmission may be stored by the sender node in a content buffer while waiting to receive the instruction.

[0089] It will be appreciated that in some examples of the present disclosure, the semantic encoding may be performed by an ML model, which may understood as set out above with reference to the alignment ML model discussed earlier.

[0090] The methods 300 and 400 may be complemented by a method 500 performed by a second alignment node. Figure 5 is a flow chart illustrating process steps in a computer implemented method 500 for facilitating semantic communication between a sender node and a receiver node in a communication network. The method 500 may be performed by a second alignment node, which may comprise a physical or virtual node,and may be implemented in a computer system, computing device or server apparatus and / or in a virtualized environment, for example in a cloud, edge cloud, or fog deployment. Examples of a virtual node may include a piece of software or computer program, a code fragment operable to implement a computer program, a virtualised function, or any other logical entity. The second alignment node may comprise a component entity in a communications network, which may comprise a 4th, 5thor other generation 3GPP network, a local network such as WiFi, an enterprise network, a data centre network, an Internet of Things network, or any other existing or future communications network. The second alignment node may in some examples be implemented within a receiver node, as discussed below, and / or within functionality for communications network management, such as in the Network Management Domain. The second alignment node may encompass multiple logical entities, as discussed in greater detail below, and may for example comprise a Virtualised Network Function (VNF). In some examples, the second alignment node may be implemented as an rApp within an O-RAN architecture. The second alignment node cooperates with a receiver node performing an example of the method 600 discussed below. The second alignment node may be collocated with the receiver node, or may be separate from the receiver node, for example operating as a virtualised function in a cloud deployment.

[0091] Referring to Figure 5, the method 500 comprises, in step 510, receiving, from a first alignment node in the communication network, a semantically encoded content sample. In step 520, the method 500 comprises obtaining, from the receiver node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample. The method 500 then comprises transmitting the reconstructed version of the content sample to the first alignment node in step 530, and receiving, from the first alignment node, an alignment instruction for implementation in the receiver node in step 540. In some examples, receiving the alignment instruction from the first alignment node may comprise receiving the alignment instruction from the first alignment node via the sender node. In step 550, the method 500 comprises providing the alignment instruction to the receiver node for implementation.

[0092] As discussed above with reference to the method 300, an alignment instruction comprises data operable to enable the semantic decoder at the receiver node to adjust the output generated by the semantic decoder for a given input. Examples of an alignment instruction may therefore include training data for training the semanticdecoder, instructions for generating training data for training the semantic decoder, original and semantically encoded versions of content, the semantically encoded version of content having been encoded by the semantic encoder of the sender node, and / or contextual information for use by the semantic decoder at the receiver node during generation of its output.

[0093] In some examples, the method 500 may further comprise receiving, from the first alignment node, an adaptation strategy of an experience record stored in a repository, the adaptation strategy comprising at least one alignment instruction for a semantic decoder at a receiver node, and providing the adaptation strategy to the receiver node for implementation. In some examples the second alignment node may receive and provide the adaptation strategy before steps 510 to 550 of the method 500 are performed. In such examples, it may be that steps 510 to 550 of the method 500 are omitted, such that in a first alternative, the second alignment node receives semantically encoded content samples and performs steps 520 to 530 accordingly, and in another alternative, the second alignment node receives an adaptation strategy, and provides this strategy to the receiver node.

[0094] In some examples of the method 500, the method may further comprise repeating iterations of the steps 520 (obtaining a reconstructed version of the content sample from a semantic decoder at the receiver node), 530 (transmitting the reconstructed version to the first alignment node), 540 (receiving an alignment instruction from the first alignment node), and 550 (providing the alignment instruction to the receiver node for implementation).

[0095] In further examples of the present disclosure, the method 500 may comprise receiving, from the receiver node, at least one of a refusal by the receiver node to implement the alignment instruction and / or a request from the receiver node for an alternative alignment instruction, and transmitting the refusal and / or request to the first alignment node.

[0096] The methods 300, 400 and 500 may be complemented by a method 600 performed by a receiver node. Figure 6 is a flow chart illustrating process steps in a method 600 for facilitating semantic communication between a sender node and a receiver node in a communication network. The method 600 may be performed by a receiver node, which may comprise a physical or virtual node, and may be implemented in a computer system,computing device or server apparatus and / or in a virtualized environment, for example in a cloud, edge cloud, or fog deployment. Examples of a virtual node may include a piece of software or computer program, a code fragment operable to implement a computer program, a virtualised function, or any other logical entity. The receiver node may comprise a component entity in a communications network, which may comprise a 4th, 5thor other generation 3GPP network, a local network such as WiFi, an enterprise network, a data centre network, an Internet of Things network, or any other existing or future communications network. The receiver node may be implemented in a Radio Access node, which itself may comprise a physical node and / or a virtualized network function that is operable to exchange wireless signals. In some examples, a Radio Access node may comprise a base station node such as a NodeB, eNodeB, gNodeB, or any future implementation of this functionality. The receiver node may encompass multiple logical entities, as discussed in greater detail below, and may for example comprise a Virtualised Network Function (VNF). In other examples, the receiver node may be implemented in a computing device, which may be a wireless device, a User Equipment, a wired device, a constrained device, etc. The receiver node cooperates with a second alignment node performing an example of the method 500. The second alignment node may be collocated with the receiver node, or may be separate from the receiver node, for example operating as a virtualised function in a cloud deployment.

[0097] Referring to Figure 6, the method 600 comprises, in step 610, receiving, from a second alignment node in the communication network, a semantically encoded content sample. The method 600 then comprises generating a reconstructed version of the content sample using a semantic decoder at the receiver node in in step 620, and providing the reconstructed version of the content sample to the second alignment node in step 630. The method 600 then comprises receiving, from the second alignment node, an alignment instruction for implementation in the receiver node in step 640, and implementing the alignment instruction with respect to the semantic decoder at the receiver node in step 650.

[0098] As discussed above with reference to the method 300, an alignment instruction comprises data operable to enable the semantic decoder at the receiver node to adjust the output generated by the semantic decoder for a given input. Examples of an alignment instruction may therefore include training data for training the semantic decoder, instructions for generating training data for training the semantic decoder, original and semantically encoded versions of content, the semantically encoded versionof content having been encoded by the semantic encoder of the sender node, and / or contextual information for use by the semantic decoder at the receiver node during generation of its output. It will be appreciated that the exact nature of the step 650 (implementing the alignment instruction with respect to the semantic decoder at the receiver node) will depend on the nature of the alignment instruction. Implementing the alignment instruction may for example comprise updating training of the semantic decoder, generating a dataset and using the dataset to update training of the semantic decoder, using the instruction to modulate the output of the semantic decoder, etc.

[0099] In some examples of the present disclosure, the method 600 may further comprise repeating iterations of the steps 620 (generating, using a semantic decoder at the receiver node, a reconstructed version of the content sample), 630 (providing the reconstructed version of the content sample to the second alignment node), 640 (receiving, from the second alignment node, an alignment instruction for implementation in the receiver node), and 650 (implementing the alignment instruction with respect to the semantic decoder at the receiver node).

[0100] In further examples of the present disclosure, the method 600 may further comprise determining that implementation of the alignment instruction is not compatible with a functionality criterion for the receiver node. This may for example arise if the receiver node perceives a risk that its decoder model will be contaminated with data that could negatively impact future communication sessions. In such examples, the method may further comprise generating at least one of a refusal by the receiver node to implement the alignment instruction and / or a request from the receiver node for an alternative alignment instruction, and providing the refusal and / or request to the second alignment node. The receiver node may then receive an alternative instruction, or may be instructed or initiated to fall back to conventional (non-semantic) communication for transmission of the content.

[0101] It will be appreciated that in some examples of the present disclosure, the semantic decoding may be performed by an ML model, which may understood as set out above with reference to the alignment ML model discussed earlier.

[0102] As discussed above, the method 300 may be performed by a first alignment node, and the present disclosure provides a first alignment node that is adapted to perform any or all of the steps of the above discussed method. The first alignment node may comprisea physical node such as a computing device, server etc., or may comprise a virtual node. A virtual node may comprise any logical entity, such as a Virtualized Network Function (VNF) which may itself be running in a cloud, edge cloud or fog deployment. The first alignment node may be operable to be instantiated in a physical or virtual server in a centralised or cloud based deployment, in a sender node as discussed below and / or in any other logical entity in a communication network.

[0103] Figure 7 is a block diagram illustrating an example first alignment node 700 which may implement the method 300, as illustrated in Figures 3A and 3B, according to examples of the present disclosure, for example on receipt of suitable instructions from a computer program 750. Referring to Figure 7 the first alignment node 700 comprises a processor or processing circuitry 702, and may comprise a memory 704 and interfaces 706. The processing circuitry 702 is operable to perform some or all of the steps of the method 300 as discussed above with reference to Figures 3A and 3B. The memory 704 may contain instructions executable by the processing circuitry 702 such that the first alignment node 700 is operable to perform some or all of the steps of the method 300, as illustrated in Figures 3A and 3B. The instructions may also include instructions for executing one or more telecommunications and / or data communications protocols. The instructions may be stored in the form of the computer program 750. In some examples, the processor or processing circuitry 702 may include one or more microprocessors or microcontrollers, as well as other digital hardware, which may include digital signal processors (DSPs), special-purpose digital logic, etc. The processor or processing circuitry 702 may be implemented by any type of integrated circuit, such as an Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA) etc. The memory 704 may include one or several types of memory suitable for the processor, such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, solid state disk, hard disk drive, etc.

[0104] As discussed above, the method 400 may be performed by a sender node, and the present disclosure provides a sender node that is adapted to perform any or all of the steps of the above discussed method. The sender node may comprise a physical node such as a computing device, server etc., or may comprise a virtual node. A virtual node may comprise any logical entity, such as a Virtualized Network Function (VNF) which may itself be running in a cloud, edge cloud or fog deployment. The sender node may be operable to be instantiated in a physical or virtual server in a centralised or cloudbased deployment, and / or in any other logical entity in a communication network such as a base station, computing device, wireless device, etc.

[0105] Figure 8 is a block diagram illustrating an example sender node 800 which may implement the method 400, as illustrated in Figure 4, according to examples of the present disclosure, for example on receipt of suitable instructions from a computer program 850. Referring to Figure 8, the sender node 800 comprises a processor or processing circuitry 802, and may comprise a memory 804 and interfaces 806. The processing circuitry 802 is operable to perform some or all of the steps of the method 400 as discussed above with reference to Figure 4. The memory 804 may contain instructions executable by the processing circuitry 802 such that the sender node 800 is operable to perform some or all of the steps of the method 400, as illustrated in Figure 4. The instructions may also include instructions for executing one or more telecommunications and / or data communications protocols. The instructions may be stored in the form of the computer program 850. In some examples, the processor or processing circuitry 802 may include one or more microprocessors or microcontrollers, as well as other digital hardware, which may include digital signal processors (DSPs), special-purpose digital logic, etc. The processor or processing circuitry 802 may be implemented by any type of integrated circuit, such as an Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA) etc. The memory 804 may include one or several types of memory suitable for the processor, such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, solid state disk, hard disk drive, etc.

[0106] As discussed above, the method 500 may be performed by a second alignment node, and the present disclosure provides a second alignment node that is adapted to perform any or all of the steps of the above discussed method. The second alignment node may comprise a physical node such as a computing device, server etc., or may comprise a virtual node. A virtual node may comprise any logical entity, such as a Virtualized Network Function (VNF) which may itself be running in a cloud, edge cloud or fog deployment. The second alignment node may be operable to be instantiated in a physical or virtual server in a centralised or cloud based deployment, in a receiver node as discussed below and / or in any other logical entity in a communication network.

[0107] Figure 9 is a block diagram illustrating an example second alignment node 900 which may implement the method 500, as illustrated in Figure 5, according to examples of thepresent disclosure, for example on receipt of suitable instructions from a computer program 950. Referring to Figure 9 the second alignment node 900 comprises a processor or processing circuitry 902, and may comprise a memory 904 and interfaces 906. The processing circuitry 902 is operable to perform some or all of the steps of the method 500 as discussed above with reference to Figure 5. The memory 904 may contain instructions executable by the processing circuitry 902 such that the second alignment node 900 is operable to perform some or all of the steps of the method 500, as illustrated in Figure 5. The instructions may also include instructions for executing one or more telecommunications and / or data communications protocols. The instructions may be stored in the form of the computer program 950. In some examples, the processor or processing circuitry 902 may include one or more microprocessors or microcontrollers, as well as other digital hardware, which may include digital signal processors (DSPs), special-purpose digital logic, etc. The processor or processing circuitry 902 may be implemented by any type of integrated circuit, such as an Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA) etc. The memory 904 may include one or several types of memory suitable for the processor, such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, solid state disk, hard disk drive, etc.

[0108] As discussed above, the method 600 may be performed by a receiver node, and the present disclosure provides a receiver node that is adapted to perform any or all of the steps of the above discussed method. The receiver node may comprise a physical node such as a computing device, server etc., or may comprise a virtual node. A virtual node may comprise any logical entity, such as a Virtualized Network Function (VNF) which may itself be running in a cloud, edge cloud or fog deployment. The receiver node may be operable to be instantiated in a physical or virtual server in a centralised or cloud based deployment, and / or in any other logical entity in a communication network such as a base station, computing device, wireless device, etc.

[0109] Figure 10 is a block diagram illustrating an example receiver node 1000 which may implement the method 600, as illustrated in Figure 6, according to examples of the present disclosure, for example on receipt of suitable instructions from a computer program 1050. Referring to Figure 10, the receiver node 1000 comprises a processor or processing circuitry 1002, and may comprise a memory 1004 and interfaces 1006. The processing circuitry 1002 is operable to perform some or all of the steps of the method 600 as discussed above with reference to Figure 6. The memory 1004 maycontain instructions executable by the processing circuitry 1002 such that the receiver node 1000 is operable to perform some or all of the steps of the method 600, as illustrated in Figure 6. The instructions may also include instructions for executing one or more telecommunications and / or data communications protocols. The instructions may be stored in the form of the computer program 1050. In some examples, the processor or processing circuitry 1002 may include one or more microprocessors or microcontrollers, as well as other digital hardware, which may include digital signal processors (DSPs), special-purpose digital logic, etc. The processor or processing circuitry 1002 may be implemented by any type of integrated circuit, such as an Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA) etc. The memory 1004 may include one or several types of memory suitable for the processor, such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, solid state disk, hard disk drive, etc.

[0110] Figures 3A to 6 discussed above provide an overview of methods which may be performed according to different examples of the present disclosure. These methods may be performed by a first alignment node, sender node, second alignment node, and receiver node respectively, as illustrated in Figures 7 to 10. The methods enable alignment of a sender semantic encoder and receiver semantic encoder in preparation for the semantic communication session between sender node and receiver node. There now follows a detailed discussion of how different process steps illustrated in Figures 3A to 6 and discussed above may be implemented. The functionality and implementation detail described below is discussed with reference to the modules of Figures 7 to 10 performing examples of the methods 300, 400, 500 and / or 600, substantially as described above.

[0111] Figures 11A and 11 B show a sequence diagram illustrating an example implementation of the methods discussed above. The sequence diagram of Figures 11A and 11 B illustrates message exchange between logical elements of a sender node, a first alignment function, a Receiver (including a receiver node and its associated second alignment node), and an experience buffer. The logical elements of the sender node illustrated in Figures 11 A and 11 B are a content analysis function CA, semantic encoder SE, and content buffer CB. The first alignment node is illustrated as sender alignment function AF. In the implementation illustrated in Figures 11 A and 11 B, the first alignment node is collocated with the sender node, but it will be appreciated that this is merely an example, and the first alignment node may alternatively be instantiated in a distributeddeployment. In Figure 11 A and 11 B, the “Receiver” entity encompasses actions at both the logical elements of the receiver node (including a semantic decoder), and at the second alignment node, which may be collocated with the receiver node or in a distributed deployment.

[0112] Referring initially to Figure 11 A:

[0113] In step 1, incoming content is sent to the Content Analysis (CA) entity of the sender node “Sender”, which is a logical entity that can be embedded in a device (UE), base station (NodeB, eNodeB, gNodeB, etc.), other network node, or its entities can be located in different network nodes.

[0114] In step 2, the CA performs an analysis of the content and extracts a content profile. The profile contains metadata about the content including its format (e.g., one or more of audio, image, video, text), its subject matter (a rock song, the sound of waves if the format is audio, or a description of the renaissance period if it’s a piece of text, etc.), its resolution (e.g. 320 Kbps, MP3 for audio or 1080p MPEG-4 for video, etc.). The analysis may also involve multiple timestamps of format, subject matter, resolution as the content may change, and may also involve predictions, e.g., how the content profile will evolve in the future.

[0115] In step 3, the Receiver shares its own profile with CA. The Receiver profile typically includes a description of the receiver node semantic decoder, detailing its architecture and specifying the model version. This version can be indicated either directly (e.g., through an identifier like version number 2.3.1) or indirectly (e.g., by describing the training dataset in terms of the multivariate probability distributions of input and output features).

[0116] In step 4, CA aggregates the content profile and model profile from previous steps in a request transmitted to the ExpBuffer (experience buffer) entity, which serves as a repository for storing experiences, and accelerates the initialization of the semantic communication session. ExpBuffer can also function as a standalone node, capable of collecting experiences from multiple semantic sessions, whether from the same senderreceiver pair or across different pairs. In this broader context, as the buffer accumulates more experiences over time, the likelihood of having a relevant content profile for a specific receiver node increases progressively. In this step ExpBuffer would perform amatching of the profile (content / model) provided by the Sender’s CA with the profiles stored in its buffer and return an Adaptation Strategy, and a confidence match of the profile stored in the ExpBuffer that is the closest match to that provided by CA. The confidence match can be an indication of how close the content / model profiles sent by CA are to the closest content / model profile stored in ExpBuffer. In an example, assuming that the profiles are vectors of values, the match can be performed using vector comparison measurements such as cosine similarity. The Adaptation Strategy comprises the steps the Receiver needs to take to align the output of its content generator (semantic decoder) to the expectations of the sender. The format of the adaptation strategy may differ depending on the architecture of the decoder model being used and the use case.

[0117] In a first example, the Adaptation Strategy can be training data needed for the Receiver to train its generative model to bias it towards the content that the sender expects to be generated.

[0118] In another example, the Adaptation Strategy can be instructions for the receiver node to generate its own training data. The instructions can be expressed with symbols such as a graph or natural language (e.g., you should generate houses with blue roofs instead of houses with red roofs).

[0119] In another example the Alignment Strategy may comprise alignment instructions sent as a mixed stream of semantic encoding performed by the sender semantic encoder alongside the original content, thus providing for on-line alignment. The receiver node uses the given encoding to align itself and signal the sender to switch to fully semantic mode once it is confident that an accuracy threshold is reached. The sender can sample the original content during encoding and may at any time determine that the context has significantly changed (e.g. there was a major change in the video stream) requiring new alignment. In this circumstance, the sender node can switch back to the mixed mode allowing the receiver to align itself again.

[0120] In another example the Adaptation Strategy can be contextual information that the receiver node can use during the generation process (semantic decoding) to constrain its output. For example, in case of a decoder / transformer being used (such as a GPT-type of model), it could be a prompt sent in embedded format or in natural language. If a conditional generative adversarial network (C-GAN) or a decoder from a conditionalvariational autoencoder (C-VAE) is used then auxiliary information can be provided to the decoder model, for example in the form of scalar or vector.

[0121] Referring still to Figure 11A, in step 5, the incoming content that was analyzed is temporarily stored in a content buffer (CB) for the time required to align the sender and the receiver.

[0122] Depending on how well the ExpBuffer was able to match the profile of the receiver node model and incoming content as analyzed from CA, i.e. how high the value of confidence match is, it may be possible to follow the provided Adaptation Strategy directly, as retrieved from ExpBuffer. Alternatively, it may be appropriate to generate adaptation instructions, if there is no suitable precedence recorded in ExpBuffer. The threshold for accepting use of a retrieved Adaptation Strategy from the ExpBuffer, illustrated as THRESHOLD1 in Figure 11 A, can be set by design according to a particular use case or deployment scenario. This is an example of a matching criterion, as discussed above with reference to the method 400.

[0123] In a first alternative (illustrated as alt in Figure 11A), if the confidence match is lower than THRESHOLD1, the Adaptation Strategy is to be identified as there is no suitable experience in ExpBuffer. Steps 6 to 16 of Figures 11 A and 11 B illustrate how this is done.

[0124] In step 6, CA extracts content samples from the content for transmission (step 410 of method 400). The extraction process may be based any one or any combination of: selecting parts of the content that are unique (to illustrate the breadth of content) selecting parts of the content that do not violate privacy (in case for example some content cannot be shared beyond the demarcation point of the Sender).

[0125] selecting parts of the content that can be decoded (and later compared, in due time) in order not to delay the alignment process. Avoiding delay can be important, particularly for time-sensitive content.

[0126] In step 7, CA sends the content samples to its AF (first alignment node) (step 420 of method 400, step 310 of method 300).

[0127] In step 8, AF will request the sender semantic encoder (SE) entity to encode the content samples, and receive the encoded content from SE (step 320 of method 300, step 430 of method 400).In step 9, AF sends the semantically encoded content samples to the Receiver entity (step 330 of method 300, step 510 of method 500, step 610 of method 600).

[0128] Steps 10 to 13 are then performed in a loop until a comparison of reconstructed content sample and original content sample meets a similarity threshold (illustrated as THRESHOLD2 in Figure 11A). The similarity threshold (THRESHOLD2) is an example of an alignment criterion, as discussed above with reference to the method 300.

[0129] In step 10, Receiver’s SE decodes the content (generating a reconstructed version of the content samples) (step 620 and 630 of method 600).

[0130] In step 11, the Receiver transmits the generated reconstructed version back to the Sender’s AF (step 630 of method 300, step 520 and 530 of method 500, step 340 of method 300).

[0131] In step 12, Sender’s AF performs a comparison between the generated reconstructed version (decoded content) and the original sampled content, and generates a measure of difference between the reconstructed and original content (STEP 350 of method 300). Based on this comparison, sender may either send a request to Receiver’s AF to modify the decoder model (step 360 and 370 of method 300), or, if the comparison is favorable, conclude the alignment process. The comparison can be performed in different ways, examples of which are set out below.

[0132] In a first example, the content can be compared semantically, on a symbol level. For example, assuming a picture, a capable visual object detector can be used to extract symbols. In one example, on the decoded content a house_with_red_roof class could be extracted, and on the sampled content it could be a house_with_blue_roof class. Then the AF can instruct the Receiver to change the color of the roof from red to blue, either by supplying a dataset of houses with blue-colored roofs or providing instructions for the Receiver to generate such dataset or prepare its existing dataset by changing roof color.

[0133] In another example, the content can also or alternatively be compared statistically on a value level, e.g. on a pixel-by-pixel basis in images, which may be better suited in use cases where 100% identical content needs to be generated on the Receiver side.In step 13, the sender AF sends alignment instructions to the Receiver based on the comparison (step 370 of method 300, step 540, 550 of method 500, step 640, 650 of method 600).

[0134] As discussed above, this process of describing modifications and comparing generated content may repeat several times, until the sender AF is satisfied with the result (i.e. , the comparison yields a value higher than or equal to THRESHOLD2).

[0135] Referring now to Figure 11B:

[0136] Once the Sender AF is satisfied with the accuracy of the decoded content (meaning the alignment criterion is fulfilled, and the sender semantic encoder and receiver semantic decoder are now aligned for the content to be transmitted), then in step 14, the sender AF obtains the model profile of the receiver from the sender CA.

[0137] In step 15, the sender AF then aggregates the alignment instructions sent suing the loop of steps 10 to 13 into an Adaptation_Strategy message, and embeds this message, with the content profile and receiver model profile, to a request to store this as an experience in ExpBuffer for future sessions.

[0138] In step 16, the sender AF sends the request to the ExpBuffer.

[0139] Step 17 illustrates actions in a second alternative, if the confidence match with the experience retried from the ExpBuffer in step 4 is at or above THRESHOLD1. In this alternative, the Adaptation_Strategy from the ExpBuffer can be used, and in step 17 the sender AF transmits the retrieved Adaptation Strategy to the Receiver, with an instruction to execute the transmitted Adaptation Strategy.

[0140] Following completion either of steps 6 to 16, or step 17, the Sender and Receiver are aligned, and in step 18, the sender AF initiates the semantic communication session by signaling SE to start semantic encoding of incoming content.

[0141] In step 19, SE retrieves the content from CB that was stored there temporarily until alignment process was complete.In step 20, SE encodes this content, and in step 21, SE transmits the encoded content to the Receiver.

[0142] Once all buffered content has been encoded and transmitted, SE receives new incoming content to sender directly, and subsequently encodes it in a normal semantic communication session towards the receiver.

[0143] As discussed above with reference to the methods 300 to 600, there may exist in some scenarios a risk of contaminating the receiver node decoder model with data that could negatively impact future communication sessions. For instance, in the example involving red and blue roofs, one sender node might guide the receiver's generator (semantic decoder) to produce houses with blue roofs. However, a different sender node in a subsequent session might expect red roofs, creating a scenario in which the receiver's generator (semantic decoder) must be continuously realigned with the sender node's expectations. This process adds additional costs in terms of time and computational resources.

[0144] In one example, this issue may be addressed by having separate instances of sender and receiver models per session. This could be feasible for some use cases where simple VAE-based models can be used for contents of lower resolution and size. However, it will be appreciated that for other use cases such as those using transformers, this approach would not scale. For these cases, and to shelter the receiver model from contamination, some examples methods according to the present disclosure propose additional interface for the Receiver. According to these examples, the receiver can deny a request from a sender to adapt the content if it deems that such adaptation will poison / bias its generative capability. Alternatively or additionally, the receiver can propose counter offers to the original request of the sender, by indicating for example that it would prefer alternative content to that selected and sent from the Sender.

[0145] In some examples of the methods discussed above, it may be envisaged that an ML model is trained using supervised learning approaches to approximate the functionality of a first alignment node. This possibility is described further below.

[0146] As shown in Figure 12A, in semantic communications there are usually pairs of encoders and decoders. For simplicity, Figure 12A illustrates two pairs, one for the sender and the other for the receiver. If the Sender Semantic Encoder (SSE) receives content X itwill generate a semantic encoding ai of the content X. On receiving the semantic encoding ai , the Sender Semantic Decoder (SSD) is operable to reconstruct a version Xi’ of the content X. If the SSE and SSD are aligned, as is reasonable for a matched encoder / decoder pair, then X is a reasonably accurate reconstruction of X. Similarly, if the Receiver Semantic Encoder (RSE) receives content X it will generate a semantic encoding a2 of the content. On receiving the semantic encoding a2, the Receiver Semantic Decoder (RSD) is operable to reconstruct a version X2’ of the content X. Again, if the RSE and RSD are aligned, as is reasonable for a matched encoder / decoder pair, then X2’ is also a reasonably accurate reconstruction of X.

[0147] If the sender and receiver are to participate in a semantic communication session, then initially it may be assumed that the SSE is not aligned with the RSD. Consequently, it is expected that when the RSD receives ai , the reconstructed version X” of the content X that is produced by the RSD is incorrect, and very different from X , which is the expected output from the SSD. The methods proposed herein provide a signaling exchange that enables alignment of the RSD, such that X” is brought to more closely resemble X . Once an implementation of these methods has been running for enough time to generate a training dataset, then it may be envisaged to train an alignment ML model to estimate either a measure of difference between the content sample and the reconstructed version of the content sample (i.e. between X and X”), and / or an alignment instruction corresponding to the estimated measure of difference.

[0148] In order to train an alignment ML model, it is envisaged to use a supervised learning approach, as illustrated in Figure 12B, which shows an example alignment ML model. Referring to Figure 12B, the alignment ML model, which may be understood as an estimation of an alignment node or AF, consists of two classifiers, one for the sender and the other for the receiver. The sender classifier receives as input the semantically encoded content ai from the SSE, and the receiver classifier receives as input the semantically encoded content a2 from the RSE. The output of both classifiers may be concatenated or otherwise combined, and is provided to an output Neural Network (NN), which is tasked to estimate the expected reconstructed version of the content X, that is the Xi’ that is the expected output of the SSD. The architecture illustrated in Figure 12B is proposed owing to its capability to combines both activations (from SSE and RSE), and to teach a model to associate these activations with the expected output. In order to train the model, a dataset is used which contains the activations (ai) from the SSE (which can be communicated by the Sender to the Receiver), the estimated X produced by theSSD (which can be communicated by the Sender to the Receiver), and the activations a2 from the RSD (which can be computed locally at the Receiver using the local encoder). The combination of activation from both encoders is beneficial as it allows the use of a greater amount of information when training the model, thus increasing knowledge in the input samples which leads to better classification results.

[0149] It will be appreciated that while the methods disclosed herein have primarily been described in the content of semantic communication between a device such as a UE and a base station, the methods disclosed herein may be used for the preparation of a semantic communication session between any logical entities within a communication network. The sender node and receiver node disclosed herein may consequently be instantiated within any logical entity a communication network. It will also be appreciated that the vast majority of logical entities operable to communicate within a communication network may act both as a sender and as a receiver of information, according to a particular use case or scenario. It may thus be envisaged for example that both a sender node and a receiver node as described herein may be instantiated within a logical entity in a communication network. Similarly, both a first alignment node and a second alignment node may be instantiated in a single logical entity in a communication network. Alternatively, first and second alignment nodes may be instantiated in a distributed deployment such as a cloud, and may be operable to communicate with sender and receiver nodes instantiated within any one or more of a wide range of logical entities within a communication network. As discussed in the background section, semantic communication between logical entities in a communication network may be advantageous for a wide range of use cases, and particularly for communication over the wireless, or air, interface, where it can support use cases such as high resolution XR.

[0150] Example methods according to the present disclosure enable semantic communication without requiring a priori alignment between sender and receiver, which is an expensive, data intense, and potentially insecure and privacy-invasive process. Rather than requiring shared data in the form of common training sets or ontologies, examples of the present disclosure achieve alignment between sender and receiver through a signaling exchange prior to initiation of a semantic communication session. In addition, previous alignment experience may be leveraged to accelerate the process.

[0151] Figure 13 shows a network node 1300 in accordance with some examples. As used herein, network node refers to equipment capable, configured, arranged and / or operableto communicate directly or indirectly with a UE and / or with other network nodes or equipment, in a telecommunications network. In accordance with respective examples, network node 1300 may be configured to operate in a communication system Examples of network nodes include, but are not limited to, access points (APs) (e.g., radio access points), base stations (BSs) (e.g., radio base stations, Node Bs, evolved Node Bs (eNBs) and NR NodeBs (gNBs)), O-RAN nodes or components of an O-RAN node (e.g., 0-Rll, O-DU, O-CU).

[0152] Network nodes 1300 may be categorized based on the amount of coverage they provide (or, stated differently, their transmit power level) and so, depending on the provided amount of coverage, may be referred to as femto base stations, pico base stations, micro base stations, or macro base stations. Network node 1300 may be a relay node or a relay donor node controlling a relay. Network nodes 1300 may also include one or more (or all) parts of a distributed radio base station such as centralized digital units, distributed units (e.g., in an O-RAN access node) and / or remote radio units (RRUs), sometimes referred to as Remote Radio Heads (RRHs). Such remote radio units may or may not be integrated with an antenna as an antenna integrated radio. Parts of a distributed radio base station may also be referred to as nodes in a distributed antenna system (DAS).

[0153] Other examples of network nodes 1300 include multiple transmission point (multi-TRP) 5G access nodes, multi-standard radio (MSR) equipment such as MSR BSs, network controllers such as radio network controllers (RNCs) or base station controllers (BSCs), base transceiver stations (BTSs), transmission points, transmission nodes, multi-cell / multicast coordination entities (MCEs), Operation and Maintenance (O&M) nodes, Operations Support System (OSS) nodes, Self-Organizing Network (SON) nodes, positioning nodes (e.g., Evolved Serving Mobile Location Centers (E-SMLCs)), and / or Minimization of Drive Tests (MDTs).

[0154] In particular examples, network node 1300 includes a processing circuitry 1302, a memory 1304, a communication interface 1306, and a power source 1308. In general, in a particular example of network node 1300, processing circuitry 1302, memory 1304, communication interface 1306, and power source 1308 may, in whole or in part, represent or include physical components common to or shared by one or more of the other elements of network node 1300.The network node 1300 may be composed of multiple distinct network entities (e.g., a NodeB entity and a RNC entity, or a BTS entity and a BSC entity, etc.), which may each have or utilize their own respective physical components. In certain scenarios in which the network node 1300 comprises multiple such entities (e.g., BTS and BSC), one or more of the separate entities may be shared among several network nodes. For example, a single RNC may control multiple NodeBs. In such a scenario, each unique NodeB and RNC pair, may in some instances be considered a single separate network node. In some examples, the network node 1300 may be configured to support multiple radio access technologies (RATs). In such examples, some components may be duplicated (e.g., separate memories 1304 or portions of memory 1304 for different RATs) and some components may be reused (e.g., a same antenna 1310 may be shared by different RATs). The network node 1300 may also include multiple sets of the various illustrated components for different wireless technologies integrated into network node 1300, for example GSM, WCDMA, LTE, NR, Wi-Fi (e.g., according to an IEEE 802.11 family standard), Zigbee, Z-wave, LoRaWAN, Radio Frequency Identification (RFID) or Bluetooth wireless technologies. These wireless technologies may be integrated into the same or different chip or set of chips and other components within network node 1300.

[0155] The processing circuitry 1302 may comprise a combination of one or more of a microprocessor, controller, microcontroller, central processing unit, digital signal processor, application-specific integrated circuit, field programmable gate array, or any other suitable computing device, resource, or combination of hardware, software and / or encoded logic operable to provide, either alone or in conjunction with other components, such as the memory 1304, to provide network node 1300 functionality. For example, the processing circuitry 1302 may be configured to cause the network node 1300 to perform the methods as described with reference to Figures 3A to 12B.

[0156] In some examples, the processing circuitry 1302 includes a system on a chip (SOC). In some examples, the processing circuitry 1302 includes one or more of radio frequency (RF) transceiver circuitry 1312 and baseband processing circuitry 1314. In some examples, the RF transceiver circuitry 1312 and the baseband processing circuitry 1314 may be on separate chips (or sets of chips), boards, or units, such as radio units and digital units. In alternative examples, part or all of RF transceiver circuitry 1312 and baseband processing circuitry 1314 may be on the same chip or set of chips, boards, or units.The memory 1304 may comprise any form of volatile or non-volatile computer-readable memory including, without limitation, persistent storage, solid-state memory, remotely mounted memory, magnetic media, optical media, random access memory (RAM), readonly memory (ROM), mass storage media (for example, a hard disk), removable storage media (for example, a flash drive, a Compact Disk (CD) or a Digital Video Disk (DVD)), and / or any other volatile or non-volatile, non-transitory device-readable and / or computerexecutable memory devices that store information, data, and / or instructions that may be used by the processing circuitry 1302. The memory 1304 may store any suitable instructions, data, or information, including a computer program, software, an application including one or more of logic, rules, code, tables, and / or other instructions capable of being executed by the processing circuitry 1302 and utilized by the network node 1300. The memory 1304 may be used to store any calculations made by the processing circuitry 1302 and / or any data received via the communication interface 1306. In some examples, the processing circuitry 1302 and memory 1304 is integrated.

[0157] The communication interface 1306 is used in wired or wireless communication of signalling and / or data with UEs, other network nodes, and / or any other network equipment. In the illustrated example, communication interface 1306 comprises port(s) / terminal(s) 1316 to send and receive data, for example to and from a network over a wired connection. In particular examples, network node 1300 may be capable of wireless communication and communication interface 1306 may also include radio frontend circuitry 1318 that may be coupled to, or in certain examples a part of, an antenna 1310. Particular examples of radio front-end circuitry 1318 include filter(s) 1320 and amplifier(s) 1322. The radio front-end circuitry 1318 may be connected to an antenna 1310 and processing circuitry 1302. The radio front-end circuitry may be configured to condition signals communicated between antenna 1310 and processing circuitry 1302. The radio front-end circuitry 1318 may receive digital data that is to be sent out to other network nodes or UEs via a wireless connection. The radio front-end circuitry 1318 may convert the digital data into a radio signal(s) having the appropriate channel and bandwidth parameters using a combination of filters 1320 and / or amplifiers 1322. The radio signal(s) may then be transmitted via the antenna 1310. Similarly, when receiving data, the antenna 1310 may collect radio signals which are then converted into digital data by the radio front-end circuitry 1318. The digital data may be passed to the processing circuitry 1302. In other examples, the communication interface may comprise different components and / or different combinations of components.In certain alternative examples, network node 1300 may be capable of wireless communication but does not include separate radio front-end circuitry 1318, instead, the processing circuitry 1302 includes radio front-end circuitry and is connected to the antenna 1310. Similarly, in some examples, all or some of the RF transceiver circuitry 1312 is part of the communication interface 1306. In still other examples, the communication interface 1306 includes one or more ports or terminals 1316, the radio front-end circuitry 1318, and the RF transceiver circuitry 1312, as part of a radio unit (not shown), and the communication interface 1306 communicates with the baseband processing circuitry 1314, which is part of a digital unit (not shown).

[0158] The antenna 1310 may include one or more antennas, or antenna arrays, configured to send and / or receive wireless signals. The antenna 1310 may be coupled to the radio front-end circuitry 1318 and may be any type of antenna capable of transmitting and receiving data and / or signals wirelessly. In certain examples, the antenna 1310 is separate from the network node 1300 and connectable to the network node 1300 through one or more interfaces or ports.

[0159] The antenna 1310, communication interface 1306, and / or the processing circuitry 1302 may be configured to perform some or all of the receiving operations and / or obtaining operations described herein as being performed by the network node 1300. Any information, data and / or signals may be received from a UE, another network node and / or any other network equipment. Similarly, the antenna 1310, the communication interface 1306, and / or the processing circuitry 1302 may be configured to perform some or all of the transmitting or sending operations described herein as being performed by the network node 1300. Any information, data and / or signals may be transmitted to a UE, another network node and / or any other network equipment.

[0160] The power source 1308 provides power to the various components of network node 1300 in a form suitable for the respective components (e.g., at a voltage and current level needed for each respective component). The power source 1308 may further comprise, or be coupled to, power management circuitry to supply the components of the network node 1300 with power for performing the functionality described herein. For example, the network node 1300 may be connectable to an external power source (e.g., the power grid, an electricity outlet) via an input circuitry or interface such as an electrical cable, whereby the external power source supplies power to power circuitry of the power source 1308. As a further example, the power source 1308 may comprise a source of power inthe form of a battery or battery pack which is connected to, or integrated in, power circuitry. The battery may provide backup power should the external power source fail.

[0161] Examples of the network node 1300 may include additional components beyond those shown in Figure 13 for providing certain aspects of the network node’s functionality, including any of the functionality described herein and / or any functionality necessary to support the subject matter described herein. For example, the network node 1300 may include user interface equipment to allow input of information into the network node 1300 and to allow output of information from the network node 1300. This may allow a user to perform diagnostic, maintenance, repair, and other administrative functions for the network node 1300.

[0162] Figure 14 is a block diagram illustrating a virtualization environment 1400 in which functions implemented by some examples may be virtualized. In the present context, virtualizing means creating virtual versions of apparatuses or devices which may include virtualizing hardware platforms, storage devices and networking resources. As used herein, virtualization can be applied to any device described herein, or components thereof, and relates to an implementation in which at least a portion of the functionality is implemented as one or more virtual components. Some or all of the functions described herein may be implemented as virtual components executed by one or more virtual machines (VMs) implemented in one or more virtual environments 1400 hosted by one or more of hardware nodes, such as a hardware computing device that operates as an access network node, UE, core network node, or host. Further, in examples in which a virtual node does not require radio connectivity (e.g., a core network node or host), then the node may be entirely virtualized. In some examples, the virtualization environment 1400 includes components defined by the O-RAN Alliance, such as an O-Cloud environment orchestrated by a Service Management and Orchestration Framework via an O-2 interface.

[0163] Applications 1402 (which may alternatively be called software instances, virtual appliances, network functions, virtual nodes, virtual network functions, etc.) are run in the virtualization environment Q400 to implement some of the features, functions, and / or benefits of some of the examples disclosed herein.

[0164] Hardware 1404 includes processing circuitry, memory that stores software and / or instructions executable by hardware processing circuitry, and / or other hardware devicesas described herein, such as a network interface, input / output interface, and so forth. Software may be executed by the processing circuitry to instantiate one or more virtualization layers 1406 (also referred to as hypervisors or virtual machine monitors (VMMs)), provide VM 1408A and VM 1408B (which may be collectively referred to as VMs 1408), and / or perform any of the functions, features and / or benefits described in relation with some examples described herein. The virtualization layer 1406 may present a virtual operating platform that appears like networking hardware to one or more of the VMs 1408.

[0165] The VMs 1408 comprise virtual processing, virtual memory, virtual networking or interface and virtual storage, and may be run by virtualization layer 1406. Different examples of the instance of a virtual appliance 1402 may be implemented on one or more of VMs 1408, and the implementations may be made in different ways. Virtualization of the hardware is in some contexts referred to as network function virtualization (NFV). NFV may be used to consolidate many network equipment types onto industry standard high volume server hardware, physical switches, and physical storage, which can be located in data centers, and customer premise equipment.

[0166] In the context of NFV, each of the VMs 1408 may be a software implementation of a physical machine that runs programs as if they were executing on a physical, nonvirtualized machine. Each of the VMs 1408, and that part of hardware 1404 that executes that VM, be it hardware dedicated to that VM and / or hardware shared by that VM with others of the VMs, forms separate virtual network elements. Still in the context of NFV, a virtual network function is responsible for handling specific network functions that run in one or more of the VMs 1408 on top of the hardware 1404 and corresponds to an application 1402.

[0167] Hardware 1404 may be implemented in a standalone network node with generic or specific components. Hardware 1404 may implement some functions via virtualization. Alternatively, hardware 1404 may be part of a larger cluster of hardware (e.g. such as in a data center or CPE) where many hardware nodes work together and are managed via management and orchestration 1410, which, among others, oversees lifecycle management of applications 1402. In some examples, hardware 1404 is coupled to one or more radio units that each include one or more transmitters and one or more receivers that may be coupled to one or more antennas. Radio units may communicate directly with other hardware nodes via one or more appropriate network interfaces and may beused in combination with the virtual components to provide a virtual node with radio capabilities, such as a radio access node or a base station. In some examples, some signalling can be provided with the use of a control system 1412 which may alternatively be used for communication between hardware nodes and radio units.

[0168] It should be noted that the above-mentioned examples illustrate rather than limit the disclosure, and that those skilled in the art will be able to design many alternative examples without departing from the scope of the appended claims or numbered examples. The word “comprising” does not exclude the presence of elements or steps other than those listed in a claim or example, “a” or “an” does not exclude a plurality, and a single processor or other unit may fulfil the functions of several units recited in the claims or numbered examples. Any reference signs in the claims or numbered examples shall not be construed so as to limit their scope.

Claims

CLAIMS1. A method (300) for facilitating semantic communication between a sender node and a receiver node in a communication network, the method, performed by a first alignment node in the communication network, comprising:obtaining a sample of content, wherein the content is to be transmitted from the sender node to the receiver node within a semantic communication session (310); initiating a semantic encoder at the sender node to semantically encode the content sample (320);transmitting the semantically encoded content sample to a second alignment node in the communication network (330);obtaining, from the second alignment node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample (340);generating a measure of difference between the content sample and the reconstructed version of the content sample (350);using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node (360); andtransmitting the alignment instruction to the second alignment node for implementation in the receiver node (370).

2. A method as claimed in claim 1, wherein the alignment instruction comprises data operable to enable the semantic decoder at the receiver node to adjust the output generated by the semantic decoder for a given input.

3. A method as claimed in claim 1 or 2, wherein the alignment instruction comprises at least one of:training data for training the semantic decoder;instructions for generating training data for training the semantic decoder; original and semantically encoded versions of content, wherein the semantically encoded version of content was encoded by the semantic encoder of the sender node;contextual information for use by the semantic decoder at the receiver node during generation of its output.

4. A method as claimed in any one of the preceding claims, wherein transmitting the alignment instruction to the second alignment node for implementation in the receiver node (370) comprises causing the sender node to transmit the alignment instruction.

5. A method as claimed in any one of the preceding claims, wherein generating a measure of difference between the content sample and the reconstructed version of the content sample (350) comprises:performing a comparison of the content sample and the reconstructed version of the content sample;generating, from the comparison, the difference measure as a representation of a difference between the content sample and the reconstructed version of the content sample.

6. A method as claimed in any one of the preceding claims, wherein using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node (360) comprises mapping the measure of difference to an alignment instruction using at least one of:a manual mapping;a heuristic process;a Machine Learning, ML, model.

7. A method as claimed in any one of the preceding claims, further comprising: using the measure of difference to generate an alignment instruction for the semantic encoder at the sender node; andproviding the alignment instruction to the sender node for implementation in the sender node.

8. A method as claimed in any one of the preceding claims, further comprising: checking whether the measure of difference fulfils an alignment criterion; and if the measure of difference does fulfil the alignment criterion, initiating the semantic communication session between the sender node and the receiver node.

9. A method as claimed in claim 8, wherein performing the method steps of:using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node (360); andtransmitting the alignment instruction to the second alignment node for implementation in the receiver node (370)is conditional upon the measure of difference failing to fulfil the alignment criterion.

10. A method as claimed in claim 8 or 9, further comprising, repeating iterations of the steps:obtaining, from the second alignment node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample (340);generating a measure of difference between the content sample and the reconstructed version of the content sample (350);checking whether the measure of difference fulfils an alignment criterion; and, if not,using the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node (360); andtransmitting the alignment instruction to the second alignment node for implementation in the receiver node (370);until the measure of difference fulfils the alignment criterion.

11. A method as claimed in any one of the preceding claims, further comprising: storing an experience record for semantic alignment of the receiver node in a repository, wherein the experience record comprises:a profile of the content;a profile of the receiver node semantic decoder;an adaptation strategy for the semantic decoder at the receiver node, the adaptation strategy comprising at least one alignment instruction for the semantic decoder at the receiver node.

12. A method as claimed in any one of the preceding claims, further comprising: receiving, from the sender node, an adaptation strategy of an experience record stored in a repository, the adaptation strategy comprising at least one alignment instruction for a semantic decoder at a receiver node; andtransmitting the adaptation strategy to a second alignment node for implementation in the receiver node; andinitiating the semantic communication session between the sender node and the receiver node.

13. A method as claimed in any one of the preceding claims, further comprising: collecting a training data set to train an alignment ML model to estimate the measure of difference between the content sample and the reconstructed version of the content sample;using the collected training data set to train the alignment ML model; and while the output of the alignment ML model fulfils an accuracy criterion, and during additional iterations of the method steps of claim 1, using the alignment ML model to replace the steps of:transmitting the semantically encoded content sample to a second alignment node in the communication network;obtaining, from the second alignment node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample;generating a measure of difference between the content sample and the reconstructed version of the content sample.

14. A method as claimed in claim 13, wherein the alignment ML model is operable to:receive as input a first version of the content sample that has been semantically encoded by a semantic encoder at the sender node, and a second version of the content sample that has been semantically encoded by a semantic encoder at the receiver node;process the input according to current values of its trainable parameters; and generate an output comprising an estimation of the measure of difference between the content sample and the reconstructed version of the content sample.

15. A method as claimed in claim 13 or 14, wherein the alignment ML model is further operable to generate an alignment instruction corresponding to the estimated measure of difference.

16. A method as claimed in any one of claims 13 to 15, wherein the alignment ML model comprises a first classifier model operable to classify an output of the semantic encoder at the sender node, a second classifier model operable to classify an output of the semantic encoder at the receiver node, and a conversion model operable toconvert the output of the first and second classifiers into the estimated measure of difference.

17. A method as claimed in any one of claims 13 to 16, wherein the training dataset comprises a plurality of training pairs of input and output data, each training pair comprising:input data comprising a first version of the content sample that has been semantically encoded by a semantic encoder at the sender node and a second version of the content sample that has been semantically encoded by a semantic encoder at the receiver node;output data comprising a measure of difference between the content sample and a reconstructed version of the content sample obtained by the alignment node during execution of the method of claim 1.

18. A method as claimed in claim 17, when dependent on claim 15, wherein the output data further comprises an alignment instruction generated by the alignment node for the semantic decoder at the receiver node during execution of the method of claim 1.

19. A method as claimed in any one of the preceding claims, further comprising receiving, from at least one of the second alignment node or the receiver node, at least one of:a refusal by the receiver node to implement the alignment instruction;a request from the receiver node for an alternative alignment instruction.

20. A method (400) for facilitating semantic communication between a sender node and a receiver node in a communication network, the method, performed by the sender node, comprising:extracting a sample of content, wherein the content is to be transmitted from the sender node to the receiver node within a semantic communication session (410); providing the sample of content to a first alignment node in the communication network (420); andon request from the first alignment node, semantically encoding the content sample using a semantic encoder at the sender node, and providing to the first alignment node the semantically encoded content sample (430).

21. A method as claimed in claim 20, further comprising:receiving an instruction from the first alignment node to commence a semantic communication session with the receiver node; andcommencing semantic encoding and transmission of the content to be transmitted from the sender node to the receiver node.

22. A method as claimed in claim 20 or 21 , further comprising:receiving, from the first alignment node, an alignment instruction for implementation in the sender node; andimplementing the alignment instruction with respect to the semantic encoder at the sender node.

23. A method as claimed in any one of claims 20 to 22, further comprising:obtaining a profile of the receiver node semantic decoder;providing, to a repository, a profile of the content to be transmitted from the sender node to the receiver node and the profile of the receiver node semantic decoder;receiving, from the repository:an adaptation strategy of an experience record stored in the repository, the adaptation strategy comprising at least one alignment instruction for a semantic decoder at a receiver node; anda measure of similarity between the content profile and semantic decoder profile provided to the repository, and the content profile and semantic decoder profile of the experience record stored in the repository.

24. A method as claimed in claim 23, further comprising:if the measure of similarity fulfils a matching criterion, providing the adaptation strategy to the first alignment node to be transmitted to a second alignment node for implementation in the receiver node.

25. A method as claimed 24, wherein performing the method steps of:extracting a sample of content to be transmitted from the sender node to the receiver node over a semantic communication session (410);providing the sample of content to a first alignment node in the communication network (420); andon request from the first alignment node, semantically encoding the content sample using a semantic encoder at the sender node, and providing to the first alignment node the semantically encoded content sample (430)is conditional upon the measure of similarity failing to fulfil the matching criterion.

26. A method (500) for facilitating semantic communication between a sender node and a receiver node in a communication network, the method, performed by a second alignment node in the communication network, comprising:receiving, from a first alignment node in the communication network, a semantically encoded content sample (510);obtaining, from the receiver node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample (520);transmitting the reconstructed version of the content sample to the first alignment node (530);receiving, from the first alignment node, an alignment instruction for implementation in the receiver node (540); andproviding the alignment instruction to the receiver node for implementation (550).

27. A method as claimed in claim 26, wherein the alignment instruction comprises data operable to enable the semantic decoder at the receiver node to adjust the output generated by the semantic decoder for a given input.

28. A method as claimed in claim 26 or 27, wherein the alignment instruction comprises at least one of:training data for training the semantic decoder;instructions for generating training data for training the semantic decoder; semantically encoded and unencoded versions of content, wherein the semantically encoded version of content was encoded by the semantic encoder of the sender node;contextual information for use by the semantic decoder at the receiver node during generation of its output.

29. A method as claimed in any one of claims 26 to 28, wherein receiving, from the first alignment node, the alignment instruction for implementation in the receiver node(540) comprises receiving the alignment instruction from the first alignment node via the sender node30. A method as claimed in any one of claims 26 to 29, further comprising, repeating iterations of the steps:obtaining, from the receiver node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample (520);transmitting the reconstructed version of the content sample to the first alignment node (530);receiving, from the first alignment node, an alignment instruction for implementation in the receiver node (540); andproviding the alignment instruction to the receiver node for implementation (550).

31. A method as claimed in any one of claims 26 to 30, further comprising:receiving, from the first alignment node, an adaptation strategy of an experience record stored in a repository, the adaptation strategy comprising at least one alignment instruction for a semantic decoder at a receiver node; andproviding the adaptation strategy to the receiver node for implementation.

32. A method as claimed in any one of claims 26 to 31 , further comprising:receiving, from the receiver node, at least one of:a refusal by the receiver node to implement the alignment instruction; a request from the receiver node for an alternative alignment instruction; and transmitting the refusal and / or request to the first alignment node.

33. A method (600) for facilitating semantic communication between a sender node and a receiver node in a communication network, the method, performed by the receiver node, comprising:receiving, from a second alignment node in the communication network, a semantically encoded content sample (610);generating, using a semantic decoder at the receiver node, a reconstructed version of the content sample (620);providing the reconstructed version of the content sample to the second alignment node (630);receiving, from the second alignment node, an alignment instruction for implementation in the receiver node (640); andimplementing the alignment instruction with respect to the semantic decoder at the receiver node (650).

34. A method as claimed in claim 33, wherein the alignment instruction comprises data operable to enable the semantic decoder at the receiver node to adjust the output generated by the semantic decoder for a given input.

35. A method as claimed in claim 33 or 34, wherein the alignment instruction comprises at least one of:training data for training the semantic decoder;instructions for generating training data for training the semantic decoder; semantically encoded and unencoded versions of content, wherein the semantically encoded version of content was encoded by the semantic encoder of the sender node;contextual information for use by the semantic decoder at the receiver node during generation of its output.

36. A method as claimed in any one of claims 33 to 35, further comprising, repeating iterations of the steps:generating, using a semantic decoder at the receiver node, a reconstructed version of the content sample (620);providing the reconstructed version of the content sample to the second alignment node (630);receiving, from the second alignment node, an alignment instruction for implementation in the receiver node (640); andimplementing the alignment instruction with respect to the semantic decoder at the receiver node (650).

37. A method as claimed in any one of claims 33 to 36, further comprising:determining that implementation of the alignment instruction is not compatible with a functionality criterion for the receiver node;generating at least one of:a refusal by the receiver node to implement the alignment instruction; a request from the receiver node for an alternative alignment instruction;and providing the refusal and / or request to the second alignment node.

38. A computer program product comprising a computer readable medium, the computer readable medium having computer readable code embodied therein, the computer readable code being configured such that, on execution by a suitable computer or processor, the computer or processor is caused to perform a method as claimed in any one of the preceding claims.

39. A first alignment node (700) for facilitating semantic communication between a sender node and a receiver node in a communication network, the first alignment node comprising processing circuitry (702) configured to cause the first alignment node to:obtain a sample of content, wherein the content is to be transmitted from the sender node to the receiver node within a semantic communication session;initiate a semantic encoder at the sender node to semantically encode the content sample;transmit the semantically encoded content sample to a second alignment node in the communication network;obtain, from the second alignment node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample;generate a measure of difference between the content sample and the reconstructed version of the content sample;use the measure of difference to generate an alignment instruction for the semantic decoder at the receiver node; andtransmit the alignment instruction to the second alignment node for implementation in the receiver node.

40. A first alignment node as claimed in claim 37, wherein the processing circuitry is further configured to cause the first alignment node to carry out a method according to any one of claims 2 to 19.

41. A second alignment node (900) for facilitating semantic communication between a sender node and a receiver node in a communication network, the second alignment node comprising processing circuitry (902) configured to cause the second alignment node to:receive, from a first alignment node in the communication network, a semantically encoded content sample;obtain, from the receiver node, a version of the content sample that has been reconstructed by a semantic decoder at the receiver node from the transmitted semantically encoded content sample;transmit the reconstructed version of the content sample to the first alignment node;receive, from the first alignment node, an alignment instruction for implementation in the receiver node; andprovide the alignment instruction to the receiver node for implementation.

42. A second alignment node as claimed in claim 39, wherein the processing circuitry is further configured to cause the second alignment node to carry out a method according to any one of claims 27 to 32.

43. A sender node (800) for facilitating semantic communication between the sender node and a receiver node in a communication network, the sender node comprising processing circuitry (802) configured to cause the sender node to:extract a sample of content to be transmitted from the sender node to the receiver node over a semantic communication session;provide the sample of content to a first alignment node in the communication network; andon request from the first alignment node, semantically encode the content sample using a semantic encoder at the sender node, and providing to the first alignment node the semantically encoded content sample.

44. A sender node as claimed in claim 41, wherein the processing circuitry is further configured to cause the sender node to carry out a method according to any one of claims 21 to 25.

45. A receiver node (1000) for facilitating semantic communication between a sender node and the receiver node in a communication network, the receiver node comprising processing circuitry (1002) configured to cause the receiver node to:receive, from a second alignment node in the communication network, a semantically encoded content sample;generate, using a semantic decoder at the receiver node, a reconstructed version of the content sample;provide the reconstructed version of the content sample to the second alignment node;receive, from the second alignment node, an alignment instruction for implementation in the receiver node; andimplement the alignment instruction with respect to the semantic decoder at the receiver node.

46. A receiver node as claimed in claim 43, wherein the processing circuitry is further configured to cause the receiver node to carry out a method according to any one of claims 33 to 37.

47. A computing device comprising at least one of:A first alignment node as claimed in claimsA second alignment node as claimed in claimsA sender node as claimed in claimsA receiver node as claimed in claims