Air interface apparatus and method for utilizing artificial intelligence to perform air interface algorithms

Abstract

An air interface apparatus (210, 314) configured to utilize artificial intelligence (212) to perform air interface algorithms (316) is provided. The air interface apparatus is further configured to (i) obtain channel state information, CSI, including information on channel properties, at a global time (t) corresponding with internal sampling index (ℓ), (ii) obtain environment state information, ESI, including information on environmental properties, at the sampling time (ℓ), (iii) tokenize the CSI and ESI together into a state information token (x(t)), (iv) predict a future sequence of state information tokens (formula(I)) for a future internal sampling index (ℓ + Ρ) based on the sequence of past state information tokens until the global time t ({x(t)}), (v) generate a set of air interface representation inputs (formula (II)) for the future internal sampling index (ℓ + Ρ) by performing representation learning on the channel and environmental properties and the predicted future sequence of state information tokens and then to perform the air interface algorithm(s) utilizing artificial intelligence air interface algorithms (316) based on the air interface representation inputs (formula (II)).

Description

[0001] AIR INTERFACE APPARATUS AND METHOD FOR UTILIZING ARTIFICIAL INTELLIGENCE TO PERFORM AIR INTERFACE ALGORITHMS TECHNICAL FIELD

[0002] The disclosure relates generally to an Artificial Intelligence, Al, based air Interface and more particularly, the disclosure relates to an air interface apparatus configured to utilize Artificial Intelligence, Al, to perform air interface algorithms. The disclosure also relates to a method for the air interface apparatus configured to utilize Al to perform the air interface algorithms.

[0003] BACKGROUND

[0004] Rapid evolution of wireless communication systems demands new approaches to manage the increasing complexity of radio networks. One such approach is the design of an Artificial Intelligence, Al, based air interface, which uses Al models to manage key communication tasks, including specific wireless functionalities, which usually rely on conventional algorithms. As nextgeneration wireless communication systems are expected to extend beyond traditional communication roles, Al may be anticipated to play a pivotal role in shaping the air interface to support these advanced capabilities. By replacing the conventional algorithms with Al models, the Al-based air interface has the potential to optimize performance by leveraging data such as Channel State Information, CSI. The Al models may adapt their decision-making processes to current wireless propagation conditions through training.

[0005] In both industrial and academic research, visions for an Al-based air interface are highly diverse. These range from using Al to replace specific wireless functions or algorithms, as explored by the 3GPP, to employ the Al for the full design of the air interface through end-to-end protocol learning. In the latter approach, there is no need for explicit CSI acquisition or representation (e.g., pilotless transmission as discussed in Hoydis 2021), with air interface algorithms directly processing the received signal.

[0006] FIG. 1B illustrates a conventional air interface 100 and FIG. 1C illustrates a process / procedure of the conventional air interface 100 in a wireless communication system. The conventional air interface 100, as shown in FIG. 1B, includes a channel state information acquisition module 102, a channel state information prediction module 104, and an air interface computation and communication module 106. The channel state information acquisition module 102 obtains raw channel state information, CSI, at internal sampling index T. The raw CSI data contains the input data about the frequency-domain propagation channels between every transmit and receive antenna pair over either the whole resource bandwidth or a part resource bandwidth. The channel state information prediction module 104 predicts the raw CSI data for a transmission block starting at internal sampling index T + P. The transmission block is as a subset comprising L ■ F radio resource elements within a time-frequency grid as illustrated in FIG. 1 A. The time-frequency grid is constituted as a matrix of resource elements containing L consecutive resource elements in the time domain and F consecutive resource elements in the frequency domain centered at some carrier frequency fc. The channel state information provided by the prediction module 104 is used by the air interface computation and communication module 106 to decide on the transmission parameters to be used in the transmission block starting at internal sampling index T + P as illustrated in FIG. 1C. The air interface computation and communication module 106 includes an air interface algorithm. The air interface algorithms may also take input data from restricted or quantized CSI data, such as channel quality indicators (CQI), precoding matrix indicators (PMI), and rank indicators (RI).

[0007] As shown in FIG. 1C, prediction techniques can be applied to predict the CSI for the transmission block beginning at internal sampling index T + P. However, the high dimensionality of CSI, due to the large bandwidth and numerous antennas at both the transmitter and receiver, makes it expensive to obtain and difficult to predict. This predicted CSI is then used by wireless functionalities to determine the transmission parameters for the upcoming transmission block starting at internal sampling index P + P.

[0008] Even when combined with Al by incorporating Al models to the constituent modules, the conventional air interface 100 faces several constraints and limitations such as (i) the air interface algorithms may only have access to multiple channel state information observations until internal sampling index P, (ii) the air interface algorithms make decisions for a transmission block covering sampling indices between the P + P until P + P + L, where the raw CSI data is outdated with respect to the index in which actual communication takes place, and (iii) the air interface algorithms do not exploit additional information on the environment such as video information, and weather measurements, which are captured by sensors at the different nodes and could be used to improve the decisions.

[0009] In more detail, the designs of the conventional air interface 100 face several challenges that limit their efficiency. First, the Channel State Information, CSI, is often collected at internal sampling index P, but the decisions have to be made for future transmission blocks, creating a gap between the CSI acquisition and usage. This can lead to incomplete or outdated CSI, especially when it only covers part of the transmission bandwidth. Additionally, the CSI data is highly dimensional, involving a large number of antennas and wide bandwidth, making the design of algorithms (or training in the case of Al models) difficult to process such complex inputs. Finally, the conventional air interface often overlooks environmental prior information, which may be leveraged to better characterize propagation conditions. The inability to integrate this extra information further limits the adaptability and performance of the air interface algorithms. Thus, existing solutions / conventional air interface has drawbacks such as prediction models for the raw CSI data are hard to design due to hardware impairments, estimation noise, and other sources of distortions. Additionally, Al models that requires a large raw CSI dataset as input for efficient training due to the high dimensionality of the raw CSI data. Furthermore, Al-based algorithms within the air interface often learn redundant CSI representations, adding unnecessary complexity to the models and hindering overall efficiency.

[0010] Therefore, there arises a need to address the aforementioned technical problem / drawbacks in the conventional / existing AI-based air interface.

[0011] SUMMARY

[0012] It is an object of the disclosure to provide an air interface apparatus configured to utilize Artificial Intelligence, Al, to perform air interface algorithms, and a method for the air interface apparatus configured to utilize Al to perform air interface algorithms while avoiding one or more disadvantages of prior art approaches.

[0013] This object is achieved by the features of the independent claims. Further, implementation forms are apparent from the dependent claims, the description, and the figures.

[0014] According to a first aspect, an air interface apparatus configured to utilize Artificial Intelligence, Al, to perform air interface algorithms is provided. The air interface apparatus is further configured to obtain channel state information, CSI, at a global time (t) corresponding with internal sampling index (P). The CSI includes information on channel properties. The air interface apparatus is further configured to obtain environment state information, ESI, at the global time (t). The ESI includes information on environmental properties. The air interface apparatus is further configured to tokenize the CSI and the ESI together into a state information token (x(t)). The air interface apparatus is further configured to predict a future sequence of state information tokens (y [P + P ] ) for a future internal sampling index (P + P ) based on the sequence of past state information tokens until the global time t ({x(t)}). The air interface apparatus is further configured to generate a set of air interface representation inputs (Z[P + P]) for the future internal sampling index (P + P) by performing representation learning on at least some of the channel properties and environmental properties and the predicted future sequence of state information tokens. Each property on which representation learning is performed on provides one air interface representation input (property i -> Z<LI[P + P ]). The air interface apparatus is then further configured to perform the air interface algorithm(s) utilizing Artificial I

[0015]

[0016] ntelligence air interface algorithm(s) based on the air interface representation inputs (Z\P + P]). The air interface apparatus employs wireless environment state information, WESI, which contains CSI enriched by time-synchronized ESI. The ESI is a representation of the state of the environment obtained by fusing the information captured by different sensors such as cameras, lidars, weather-related sensors, GPS sensors and other prior information such as geographical maps, site-specific hardware information at a network node, a receiver node, and a transmitter node. The air interface apparatus employs representation learning techniques to determine a WESI representation model which is easier to predict than raw CSI. The air interface apparatus decouples the Al models related to wireless environment state information, WESI, representation and the Al models implementing some wireless functionality, i.e., Al-based air interface algorithms. This allows for independently management of the life cycle of both type of Al models. Further, the air interface apparatus employ s / deploys less complex and less scenario or wireless-propagation dependent Al-based algorithms, which requires less online training and fine-tuning and less frequent Al model updates.

[0017] Optionally, the predicted future sequence of state information tokens (y\P + P]) for the future internal sampling index (T + P) comprises L predicted tokens (y) and the set of air interface representation inputs (Z\P + P]) for the future internal sampling index (T + P) includes a set of representation input for property i (Z^ [T + P]) for i = 1,..., D, wherein each representation input for property i (Z^ [P + P]) includes a set of vectors (z®. [P + P]) corresponding to an individual feature of the state information (CSI, ESI) for a transmission block containing F ■ L resource elements (REs) starting at an internal sampling index (P + P), wherein each vector is for a time index ( / ') and frequency index ( / ) at a specified time and frequency granularity. Optionally, the time granularity is r;= L / L,- and the frequency granularity is A,- = F / F;, wherein the set of interface representations includes all vectors for all features (i = 1,...,£)), all time indices (j = 1, and all frequency indices ( / = 1. Ft). Optionally, the air interface apparatus is further configured to predict the future sequence of state information tokens (y\P + P]) for a future internal sampling index (P + P) up to an end transmission sampling index (P + P + L) and to generate the set of air interface representation inputs (Z\P + P]). Optionally, the air interface apparatus is further configured to predict the token sequences (y [P + P] ) as sampled for the whole resource frame starting at internal sampling index f + P. in which transmission is taking place. Optionally, the predicted token sequence (y\P + P]) is sampled at an internal sampling rate at minimum time

[0018]

[0019] granularity. Optionally, a property (channel and / or environmental) is defined by a representation dimensionality, a representation size, and a time-frequency span. Optionally, the ESI is a representation of the state of the environment obtained by fusing information captured by sensor(s) environment information at a node.

[0020] Optionally, the air interface apparatus is further configured to perform representation learning on the at least some of the channel properties and environmental properties related to the air interface algorithm(s) to be performed. Optionally, the channel state information, CSI, and the environment state information, ESI, are time-synchronized for the global time (t). Optionally, the CSI and ESI are measured at a transmitter node. Optionally, the air interface apparatus is further configured to perform an air interface algorithm utilizing artificial intelligence air interface algorithm(s) based on one or more elements (Z(i)) of the set of air interface representation inputs (Z).

[0021] Optionally, the combination of the channel state information, CSI, and the environment state information, ESI, is grouped as a Wireless Environment State Information, WESI, wherein the channel and environmental properties are WESI properties. Optionally, the air interface apparatus is a controller configured to be used in a wireless communications node. Optionally, the air interface apparatus is a wireless communications node.

[0022] According to a second aspect, a method for an air interface apparatus configured to utilize Artificial Intelligence to perform air interface algorithms is provided. The method includes the air interface apparatus to obtain / receive channel state information, CSI, at a global time (t) corresponding with internal sampling index ( ), wherein the CSI includes information on channel properties. The method includes obtaining environment state information, ESI, at the global time (t), wherein the ESI includes information on environmental properties. The method includes tokenizing the CSI and the ESI together into a state information token (x(t)). The method includes predicting a future sequence of state information tokens (y[-t + P]) for a future internal sampling index (T + P) based on the sequence of past state information tokens until the global time t ({x(t)}). The method includes generating a set of air interface representation inputs (Z[ℓ + P]) for the future internal sampling index (T + P) by performing representation learning on at least some of the channel properties and environmental properties and the predicted future sequence of state information tokens. Each property on which representation learning is performed on provides one air interface representation input (property i -> Z^[f + P]). The method includes performing the air interface algorithm(s) utilizing Artificial Intelligence air interface algorithm(s) based on the air interface representation inputs (Z[ℓ + P]).

[0023] The method employs wireless environment state information, WESI, which contains CSI enriched by time-synchronized ESI. The ESI is a representation of the state of the environment obtained by fusing the information captured by different sensors such as cameras, lidars, weather-related sensors, GPS sensors and other prior information such as geographical maps, sitespecific hardware information at a network node, a receiver node, and a transmitter node. The method employs representation learning techniques to determine a WESI representation model which is easier predicted than raw CSI. The method decouples the Al models related to wireless environment state information, WESI, representation and the Al models implementing some wireless functionality, i.e., Al-based air interface algorithms. This allows for independently management of the life cycle of both type of Al models. Further, the air interface apparatus employs / deploys less complex and less scenario or wirelesspropagation dependent Al-based algorithms, which requires less online training and fine-tuning and less frequent Al model updates.

[0024] According to a third aspect, a computer program product includes program instructions for performing the method as described above, when executed by one or more processors in an air interface apparatus.

[0025] Therefore, in contradistinction to the existing solutions, the air interface apparatus allows for deploying less complex and less scenario or wireless-propagation dependent Al-based algorithms, which require less online training and fine-tuning and less frequent Al model updates as described above.

[0026] These and other aspects of the disclosure will be apparent from the implementation s) described below.

[0027] BRIEF DESCRIPTION OF DRAWINGS

[0028] Implementations of the disclosure will now be described, by way of example only, with reference to the accompanying drawings, in which:

[0029] FIG. 1 A (PRIOR ART) illustrates a typical transmission block;

[0030] FIG. 1B (PRIOR ART) illustrates a block diagram of a conventional (Al-based or not) air interface;

[0031] FIG. 1C (PRIOR ART) illustrates a process / procedure of the conventional (Al-based or not) air interface; FIG. 2A illustrates a block diagram of a wireless communication network in accordance with an implementation of the disclosure;

[0032] FIG. 2B illustrates a block diagram of an air interface apparatus in accordance with an implementation of the disclosure; FIG. 3 A illustrates an air interface apparatus that includes a Wireless Environment State Information, WESI, architecture, implementing prediction and representation in a wireless communication network in accordance with an implementation of the disclosure;

[0033] FIG. 3B illustrates a block diagram of a wireless communication network for acquiring Wireless Environment State Information, WESI when Environment State Information tokenizers are positioned at a Transmitter Node (TX) in accordance with an implementation of the disclosure;

[0034] FIG. 3C illustrates a Wireless Environment State Information, WESI, tokenization block implemented at a Transmitter Node, TX, configured to be used in a wireless communication network in accordance with an implementation of the disclosure; FIG. 3D illustrates a Wireless Environment State Information, WESI, sequence predictor block implemented at a Transmitter Node, TX, configured to be used in a wireless communication network in accordance with an implementation of the disclosure; FIG. 4 illustrates an exemplary Artificial intelligence, Al, based air interface algorithm implemented in an air interface apparatus in accordance with an implementation of the disclosure;

[0035] FIG. 5 illustrates an architecture of a Structured Wireless Environment State Information, WESI, structured representation in accordance with an implementation of the disclosure;

[0036] FIG. 6 illustrates how a Wireless Environment State Information, WESI, structured representation / air interface representation input features are mapped into the time-frequency grid associated with a transmission block in accordance with an implementation of the disclosure;

[0037] FIG. 7 illustrates how ithfeature of a Wireless Environment State Information, WESI, structured representation / air interface representation input are mapped into the time-frequency grid associated with a transmission block in accordance with an implementation of the disclosure;

[0038] FIG. 8 illustrates an exemplary view of Air interface algorithms’ decisions granularities in accordance with an implementation of the disclosure;

[0039] FIG. 9 is a flow diagram that illustrates a method for an air interface apparatus configured to utilize artificial intelligence to perform air interface algorithms in a wireless communication network in accordance with an implementation of the disclosure; and

[0040] FIG. 10 is an illustration of a computer system in which the various architectures and functionalities of the various previous implementations may be implemented.

[0041] DETAILED DESCRIPTION OF THE DRAWINGS

[0042] Implementations of the disclosure provide an air interface apparatus configured to utilize Artificial Intelligence, Al, to perform to perform wireless environment state representation and air interface algorithms. Moreover, the disclosure relates to a method for the air interface apparatus configured to utilize Artificial Intelligence to perform wireless environment state representation and air interface algorithms in a wireless communication network.

[0043] To make solutions of the disclosure more comprehensible for a person skilled in the art, the following implementations of the disclosure are described with reference to the accompanying drawings.

[0044] Terms such as “a first”, “a second”, “a third”, and “a fourth” (if any) in the summary, claims, and foregoing accompanying drawings of the disclosure are used to distinguish between similar objects and are not necessarily used to describe a specific sequence or order. It should be understood that the terms so used are interchangeable under appropriate circumstances, so that the implementations of the disclosure described herein are, for example, capable of being implemented in sequences other than the sequences illustrated or described herein. Furthermore, the terms “include” and “have” and any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, a method, a system, a product, or a device that includes a series of steps or units, is not necessarily limited to expressly listed steps or units but may include other steps or units that are not expressly listed or that are inherent to such process, method, product, or device.

[0045] Definitions:

[0046] Air Interface: An Air Interface is defined as the specification of technologies and protocols (usually involving algorithms from the physical layer (LI) and data link layer (L2) of the OSI model) enabling the wireless communication between a transmitter node and a receiver node in a wireless communication network using a set of radio resources elements in a time-frequency grid, denoted as a transmission block. Examples of Air Interface algorithms are as follows: precoder design, scheduler design, user localization, beamforming, etc.

[0047] Artificial Intelligence, Al, based air interface: An Al-based air interface refers to an air interface, in which the set of protocols or algorithms implementing the specific wireless functionalities is Al-based. That is, they use pre-trained models based on data to take the corresponding actions / decisions.

[0048] Resource Element: A resource element, RE, is a communication resource considered in the standard specification of specified duration TREseconds and is specified frequency bandwidth BREHz. This is the highest / elementary considered granularity (e.g., in time and frequency) in the 3GPP standard. The following disclosure describes the time and frequency instants using this granularity.

[0049] Transmission block is defined as a set of L • F radio resource elements in the time-frequency grid. A transmission block contains F resource frames, each one including L consecutive resource elements in the time domain. A transmission block contains L resource blocks, each one including F consecutive resource elements in the frequency domain.

[0050] Air interface algorithm: An air interface algorithm or wireless functionality is defined as a process or a set of rules taking decisions / actions enabling the transmission between a transmitter node and a receiver node using a Transmission Block. Note that each wireless functionality may choose a Transmission Block of different size (i.e., different L and F) depending on the needed properties. Indeed, in modem communications systems, the actions performed by a wireless functionality usually depend on the propagation channel over which the communication takes place and the dynamics of this channel is dependent on the considered Resource Block.

[0051] Feature representation learning: In machine learning, feature learning, representation learning, or feature representation learning, is a set of techniques that allows a system to automatically discover the representations needed for feature detection or classification from raw data. This replaces manual feature engineering and allows a machine to both learn the features and use them to perform a specific task. The feature learning is motivated by the fact that machine learning tasks such as prediction or inference often require input that is mathematically and computationally convenient to process.

[0052] FIG. 2A illustrates a block diagram of a wireless communication network 202 in accordance with an implementation of the disclosure. The wireless communication network 202 includes a Network Node, NW, 204, a Receiver Node, RX, 206, and the Transmitter Node, TX, 208. The network node 204 serves as a control center within the wireless communication network 202.

[0053] The network node 204 manages network resources, coordinates the communication between the transmitter node 208 and the receiver node 206, and ensures seamless data exchange across the network 202. The network node 204 may also coordinate the collection of Channel State Information, CSI, and Environment State Information, ESI, to optimize the communication process and improve network performance. The transmitter node 208 is responsible for generating and transmitting data signals over the air to other nodes within the network 202. The receiver node 206 is responsible for receiving and decoding the signals transmitted by the transmitter node 208.

[0054] FIG. 2B illustrates a block diagram of an air interface apparatus 210 in accordance with an implementation of the disclosure. The air interface apparatus 210 includes an artificial intelligence, Al for wireless environment representation, 212, and an Al for wireless functionalities computation 214. The air interface apparatus 210 is configured to utilize the Al 212 to obtain and predict the wireless environment and to utilize the Al 214 to perform air interface algorithms. The air interface apparatus 210 obtains channel state information, CSI, at a global time (t) corresponding with internal sampling index ( ). The channel state information comprises information on channel properties. The air interface apparatus 210 is configured to obtain environment state information, ESI, at the global time (t). The environment state information comprises information on environmental properties. The air interface apparatus 210 tokenizes the CSI and the ESI together into a state information token ( (t)). The air interface apparatus 210 predicts a future sequence of state information tokens (ŷ[ℓ + P]) for a future internal sampling index (T + P) based on the sequence of past state information tokens until the global time t ({ (£)}). The air interface apparatus 210 generates a set of air interface representation inputs (Z[ℓ + P]) for the future internal sampling index (T + P) by performing representation learning on at least some of the channel properties and environmental properties and the predicted future sequence of state information tokens. Each property on which representation learning is performed on provides the air interface representation input (property i -> Z^[f + P]). The air interface apparatus 210 then performs the air interface algorithms utilizing artificial intelligence air interface algorithms based on the air interface representation inputs (Z[ℓ + P]).

[0055] The air interface apparatus 210 employs wireless environment state information, WESI, which contains CSI enriched by time-synchronized ESI. The ESI is a representation of the state of the environment obtained by fusing the information captured by different sensors such as cameras, lidars, weather-related sensors, GPS sensors, and other prior information such as geographical maps, site-specific hardware information at a network node, a receiver node, and a transmitter node. The air interface apparatus 210 employs representation learning techniques to determine a WESI representation model which is easier predicted than raw CSI. The air interface apparatus 210 allows for decoupling the Al models related to wireless environment state information, WESI, representation 212 and the Al models implementing some wireless functionality, i.e., Al-based air interface algorithms 214. This allows for independently management of the life cycle of both type of Al models. Further, the air interface apparatus 210 employs / deploys less complex and less scenario or wireless-propagation dependent Al-based algorithms 214, which requires less online training and fine-tuning and less frequent Al model updates.

[0056] Optionally, the predicted future sequence of state information tokens (ŷ[ℓ + P]) for the future internal sampling index (T + P) comprises L’ predicted tokens (y) and the set of air interface representation inputs (Z[ℓ + P]) for the future internal sampling index (T + P) includes a set of representation inputs for property i (Z^), wherein each representation input for property i (Z^ [T + P]) includes a set of vectors (z®. [T + P]) corresponding to an individual feature of the state information (CSI, ESI) for a transmission block containing F ■ L resource elements (REs) starting at an internal sampling index (T + P), wherein each vector is for a time index (J) and frequency index ( ) at a specified time and frequency granularity. Optionally, the time granularity is r;= L / L,- and the frequency granularity is A,- = F / Ft, wherein the set of interface representations includes all vectors for all features (i = 1,...,£)), all time indices (j = 1, and all frequency indices ( = 1,,;).

[0057] Optionally, the air interface apparatus 210 is further configured to predict the future sequence of state information tokens (1 / [ + P]) f°r afuture internal sampling index (ℓ + P) up to an end transmission sampling index (£ + P + L) and to generate the set of air interface representation inputs (Z[£ + P ]). Optionally, the air interface apparatus 210 is further configured to predict the token sequence (ŷ[ℓ + P]) as sampled for the whole resource frame starting at internal sampling index £ + P, in which transmission is taking place. Optionally, the predicted token sequence (y[£ + P]) is sampled at an internal sampling rate at minimum time granularity.

[0058] Optionally, a property (channel and / or environmental) is defined by a representation dimensionality, a representation size, and a time-frequency span. Optionally, the ESI is a representation of the state of the environment obtained by fusing information captured by sensor(s) environment information at a node.

[0059] Optionally, the air interface apparatus 210 is further configured to perform representation learning on the at least some of the channel properties and environmental properties related to the air interface algorithm(s) to be performed. Optionally, the channel state information, CSI, and the environment state information, ESI, are time-synchronized for the global time (t). Optionally, the CSI and ESI are measured at a transmitter node. Optionally, the air interface apparatus 210 is further configured to perform an air interface algorithm utilizing artificial intelligence air interface algorithm(s) based on one or more elements (Z^F) of the set of air interface representation inputs (Z).

[0060] Optionally, the combination of the channel state information, CSI, and the environment state information, ESI, is grouped as a Wireless Environment State Information, WESI, wherein the channel and environmental properties are WESI properties. Optionally, the air interface apparatus 210 is a controller configured to be used in a wireless communications node. Optionally, the air interface apparatus 210 is a wireless communications node.

[0061] FIG. 3A illustrates an air interface apparatus 314 that includes a Wireless Environment State Information, WESI, architecture 300, implementing prediction and representation in a wireless communication network in accordance with an implementation of the disclosure. The WESI architecture 300 includes a WESI acquisition block including a channel state information, CSI, acquisition block 302, and an environment state information, ESI, acquisition block 304, a WESI tokenizer block 306, a WESI sequence predictor block 308, a WESI property list 310, a WESI structured representation block 312. The output of the WESI representation block 312 represent the input of the artificial intelligence, Al, based air interface algorithms 316.

[0062] The CSI acquisition block 302 obtains the CSI at a global time (t) corresponding with internal sampling index ( ). The CSI includes information on channel properties. The ESI acquisition block 304 obtains the ESI at the global time (t). The ESI includes information on environmental properties. The environment state information may contain at least any one information captured by sensors such as cameras, lidars, weather-related sensors, GPS sensors at a network node, a receiver node, and a transmitter node. Optionally, the environment state information may contain prior information available at a transmitter node, TX, a receiver node, RX, and a network node, NW. The prior information may be at least any of a geographical map and a sitespecific hardware information, at the network node, the receiver node, and the transmitter node. Optionally, the ESI is a time-synchronized ESI. The ESI acquisition block 304 may collect, clean, timestamp, tokenize, and store the ESI from the environmental property and the prior information.

[0063] CSI (t) is a time-stamped CSI vector of size F • N obtained either at the transmitter node or the receiver node and containing the frequency-domain propagation channels in h(t), connecting IV(t) of the N = NTX•

[0064]

[0065] antenna pairs for a subset of F(t) out of the F resources contained in a resource block. The remaining (F — F(t)) • (IV — IV(t)) entries are filled with the special symbol 0csi to indicate missing elements / information.

[0066] ESI(t) is a time-stamped environment state information vector with S = SNW+ SRX+ STXstate information variables, which fuses the environment state information acquired at the NW node, at the RX node and at the TX node, denoted by ESINW(t), ESIRX(t), and ESITX( t), respectively. ESINW(t) is obtained from the local NW node state information acquired at global time tNW, SINW(tNW), and received by the TX node at global time rNW, which is a time-stamped vector gathering SNW(tNW) out of SNWpossible local state information variables (i.e., with 0 < SNW(tNW) < SNW), by applying data-modality specific tokenizers to the SNW(tNW) state variables and by filling the remaining (SNW— SNW(tNW)) state variables with special symbol 0NWto indicate missing information. ESIRX(t) is obtained from the local RX node state information acquired at global time tRX, SIRX(tRX), and received by the TX node at global time TRX, which is a time-stamped vector gathering SRX(tRX) out of SRXpossible local state information variables (i.e., with 0 < SRX(tRX) < SRX), by applying data-modality specific tokenizers to the SRX(tRX) state variables and by filling the remaining (SRX— S'RX(tRX)) state variables with special symbol 0RXto indicate missing information. ESITX(t) is obtained from the local TX node state information acquired at global time tTX, SITX(tTX), which is a time-stamped vector gathering STX(tTX) out ofSTXpossible local state information variables (i.e., with 0 < STX(tTX) < STX), by applying data-modality specific tokenizers to the STX(tTX) state variables and by filling the remaining (STX—^TXC^TX)) state variables with special symbol 0TXto indicate missing information.

[0067] The WESI tokenizer block 306 includes a combination of pre-processing and embedding techniques that are configured to (i) handle hardware impairments (e.g., phase offset) in the CSI data and address special symbols in the ESI data, (ii) combine the ESI and CSI data, and (iii) provide an embedding block to map WESI into a lower-dimensional space. The WESI tokenizer block 306 may tokenize the CSI and ESI data into a state information token (x(t)) using a pre-trained model. The WESI tokenizer block 306 applies an embedding process to map the WESI data into a lower-dimensional space. Optionally, x(t) is a token vector of size d < F • N, including a token joint representation of ESI and CSI for all N antenna pairs and all F resource elements within the resource block.

[0068] The WESI sequence predictor block 308 predicts a future sequence of state information tokens (ŷ[ℓ + P]) for a future internal sampling index (T + P) based on the sequence of past state information tokens until the global time t ({x(t)}). The WESI sequence predictor block 308 obtains a length L sequence of state information tokens (ŷ[ℓ + P]) by predicting the state information tokens for the next transmission block (of length L) starting at an internal sampling index T + P.

[0069] The predicted future sequence of state information tokens (ŷ[ℓ + P]) is used to obtain the structured WESI representation (Z[F + P] ) at internal sampling index T + P, which gathers the D individual features (Z1>[T + P] for i = 1,..., D) defined in the WESI property list 310 at a specific time and frequency granularity covering the transmission block starting at an internal sampling index T + P. The WESI property list 310 defines each feature characterizing a physical quantity of a wireless environment. The feature may be a signal strength, an interference level, a channel conditions, and other environmental metrics. The predicted future sequence of state information tokens (ŷ[ℓ + P]) is represented in minimum time granularity for the whole transmission block starting at an internal sampling index T + P. Each wireless environment property in the WESI property list specifies a representation dimensionality such as an output dimension, a representation size, and a time-frequency span of property validity.

[0070] The WESI structured representation block 312 generates a set of air interface representation inputs (Z[F + P] ) for the future internal sampling index (T + P) by performing representation learning on at least some of the channel properties and environmental properties and the predicted future sequence of state information tokens (ŷ[ℓ + P]). Each property on which representation learning is performed on provides one air interface representation input (property i -> Z^ [T + P]). The air interface apparatus 314 takes decisions on the transmission parameters to enable communication at internal sampling index T + P between the transmitter node (e.g., a wireless transmitter node) and the receiver node (e.g., a wireless receiver node) within the transmission block using the Al- based air interface algorithm 316. The transmission block is a set of resource elements in a time-frequency grid. Each of the Al-based algorithms adapt its decision to the wireless channel propagation conditions based on a subset of individual features in the WESI property list 310. The Al-based air interface algorithms 316 may include a precoder design, a scheduler design, a user localization, and a beamforming.

[0071] FIG. 3B illustrates a block diagram of a wireless communication network 318 for acquiring Wireless Environment State Information, WESI, when Environment State Information tokenizers are positioned at a Transmitter Node (TX) 324 in accordance with an implementation of the disclosure. The wireless communication system 318 includes a Network Node (NW) 320, a Receiver Node (RX) 322, and the TX Node 324. The network node 320 includes a local network state information, SI, acquisition block 326. The receiver node 322 includes a local receiver SI acquisition block 328. The transmitter node 324 includes a channel state information, CSI, acquisition block 330, a transmitter node SI acquisition block 332, a transmitter node SI tokenizer 334, a receiver node SI tokenizer 336, a network node SI tokenizer 338 and a ESI update and data fusion module 340. The wireless communication system 318 is configured to receive environment state Information / RAW state information from a source device over air. The RAW state information may be local network node state information, SINW, local receiver node state information, SIRX, or local transmitter node state information, SITX.

[0072] At the network node 320, the local network SI acquisition block 326 obtains the local network node state information at a global time tNW, SINW(tNW). At the receiver node 322, the local receiver SI acquisition module 328 obtains the local receiver node state information at a global time

[0073]

[0074] SIRX(tRX). At the transmitter node 324, the transmitter SI acquisition module 332 obtains the local transmitter node state information at the a global time (tTX), SITX(tTX) ■ The network node 320 transmits the SINW(tNW) to the transmitter node 324 over air at the global time rNW. The receiver node 322 transmits the SIRX(tRX) to the transmitter node 324 over the air at the global time TRX.

[0075] The wireless communication network 318 is configured to store the local state information in a table where each property of the state information is associated with a variable identifier, ID, an over-the-air (OTA) update timestamp, r, an acquisition timestamp, t, a validity time, T, property data, and property format. The OTA update timestamp (r) indicates the timestamp of the last over-the-air update for any change or modification to a variable associated with the variable ID. The acquisition timestamp (t) indicates a timestamp at which state information (SI) or data associated with the variable related to the variable ID is acquired. The validity time (T) indicates the duration for which the data, associated with the variable related to the variable ID, is considered valid after it is acquired, i.e., how long the acquired data remains usable before it needs to be updated or refreshed. The property data indicates that the data corresponding to the variable associated with the variable ID is stored. The property format indicates the data representation method used. This means how the data is structured, formatted, or transformed for storage, transmission, or processing. The table may be a network SI table, a receiver SI table, or a transmitter SI table. The local state information may be the local transmitter node state information, SITX, the local receiver node state information, SIRX, or the local network node state information, SINW.

[0076] The network SI tokenizer 338 tokenizes the local network node state information SINW(tNW) and obtains ESINW(tNW). The receiver SI tokenizer 336 tokenizes the local receiver node state information SIRX(tRX) and obtains ESIRX(tRX). The transmitter SI tokenizer 334 tokenizes the local transmitter node state information SITX(tTX) and obtains ESITX(tTX).

[0077] At the transmitter node 324, at the global time t, with t > tNW> tRX> tTX, the CSI acquisition module 330 obtains CSI (t) and the ESI update and data fusion module 340 obtains ESI(t). The ESI update and data fusion module 340 checks the network environment state information ESINW(tNW), obtained from the local NW node state information acquired at the global time tNW, for missing or outdated local state information variables and substitutes them by special symbol 0NWand obtains ESINW(t). The ESI update and data fusion module 340 checks the receiver environment state information ESIRX(tRX), obtained from the local RX node state information acquired at the global time for missing or outdated local state information variables, substitutes them by special symbol 0!Xand obtains ESIRX(t). The ESI update and data fusion module 340 checks the transmitter environment state information ESITX(tTX), obtained from the local TX node state information acquired at the global time tTX, for missing or outdated local state information variables, substitutes them by special symbol 0TXand obtains ESITX(t). For each node (i.e. the network node 320, the receiver node 322, or the transmitter node 324), SINW, SIRX, SITXare sets of variables or fields that describe specific aspects of state information relevant to that node. The state information may be rain sensor data, temperature data, or receiver position data. The ESI update and data fusion module 340 fuses the resulting network environment state information ESINW(t), receiver environment state information ESIRX(t), and transmitter environment state information ESITX( t) at the global time t into ESI ( t). The wireless communication system 318 acquires the WESI(t) at the global time (t) by concatenating the ESI (t) and CSI(t).

[0078] FIG. 3C illustrates a Wireless Environment State Information, WESI, tokenization block 344 implemented at a Transmitter Node, TX, configured to be used in a wireless communication network in accordance with an implementation of the disclosure. The wireless communication network includes a Network Node, NW, a Receiver Node, RX, and the Transmitter Node, TX. The wireless communication network is configured to utilize Artificial Intelligence, Al, to perform air interface algorithms. The Wireless Environment State Information, WESI, tokenization block 344 is implemented at the transmitter node of the wireless communication network. The WESI is derived from sensors and prior data available at the network node, the receiver node and the transmitter node of the wireless communication network. The WESI includes two main components: (i) channel state information, CSI, and (ii) time-synchronized environment state information, ESI.

[0079] The WESI tokenization block 344 includes an impairment removal block 346, a CSI embedding block 348, an ESI embedding block 350, a WESI embedding block 352, and a WESI feature scaling block 354. The impairment removal block 346 performs preprocessing to mitigate hardware and CSI acquisition protocol impairments (e.g., phase offset due to miscalibration or imperfect synchronization) affecting the input CSI data. The CSI embedding block 348 manages the missing CSI information (i.e., entries with special symbol 0CSI) and maps the CSI input into a CSI representation xCSI(t) in a dCSI-dimensional domain, with dCSI< F • N.

[0080] The ESI embedding block 350 manages the missing ESI information (i.e., entries with special symbols 0NW, 0RX, and 0TX) and maps the ESI input into an ESI representation xESI(t) in a dESI-dimensional domain, with dESIbeing smaller than the size of the ESI. The WESI embedding block 352 maps the CSI representation xCSI(t) into a WESI representation xWESI(t) in a d-dimensional domain, with d < dCSI(t). The ESI embedding xESI(t) is used as side information to enrich the CSI representation. The WESI feature scaling block 354 transforms the input features in vector xWESI(t) into a fixed range (e.g., [0,1], [-1,1]). The scaled output is the final token vector x(t) ∈

[0081]

[0082] .

[0083] The tokenization of the WESI is applied at a specific global time instant (denoted as t). The WESI tokens can be timestamped in line with their respective WESI inputs (i.e., the CSI input and the ESI input). The time granularity that is used may correspond to the resource element index, a subframe ID, a frame ID, or other relevant metrics.

[0084] If the WESI input is incomplete, the tokenization of the WESI may output a special failure symbol 0token. The "insufficient completeness" flag may be triggered by the following conditions: the CSI input is unavailable, the CSI input is available, but its bandwidth is too limited, and the ESI input is inconsistent.

[0085] The CSI embedding block 348, the ESI embedding block 350, and the WESI embedding block 352 may be designed as cascades of neural network layers, including transformer layers utilizing multi-head attention modules, convolutional layers, fully connected layers, and pooling layers. The WESI tokenization provides (i) effective pre-processing techniques to mitigate the effects of noise and hardware impairments on CSI(t), (ii) data fusion solutions to seamlessly combine CSI (t) and ESI(t) and (iii) dimensionality reduction to offer a compact representation of the input while preserving the key features of WESI(T) relative to the original input size. FIG. 3D illustrates a Wireless Environment State Information, WESI, sequence predictor block 356 implemented at a Transmitter Node, TX, configured to be used in a wireless communication network in accordance with an implementation of the disclosure. The wireless communication network is configured to utilize Artificial Intelligence, Al, to perform air interface algorithms. The WESI sequence predictor block 356 is implemented at a transmitter node of the wireless communication network. The WESI is derived from sensors and prior data available at a network node, a receiver node and the transmitter node of the wireless communication network. The WESI includes two main components: (i) channel state information, CSI, and (ii) time-synchronized environment state information, ESI. The transmitter node of the wireless communication network is configured to obtain channel state information, CSI, at a global time t corresponding with internal sampling index ( ), CSI ( t). The channel state information includes information on channel properties. The transmitter node obtains environment state information, ESI, at the global time t corresponding with internal sampling index ( ), ESI(t). The environment state information includes information on environmental properties. The transmitter node integrates a Wireless Environment State Information, WESI, tokenizer block to tokenize the CSI and ESI together into a state information token (x(t)) up to the global time (t).

[0086] The WESI sequence predictor block 356 includes a WESI token sequencing block 358 and a WESI token predictor block 360.

[0087] The WESI token sequencing block 358 takes the past sequence of state information tokens (i.e., WESI tokens) derived by the WESI tokenization block at the transmitter node up to the global time (t) (풳(t) = {x(t)}) as input and crops the set of state information tokens up to the global time (t) (풳(t)) to a cropped set of state information tokens 풳K(t) = {x(t;f_l), x(t)} including K samples / elements up to the time (t) in a global time reference domain. The WESI token sequencing block 358 applies resampling and interpolation to the cropped set of state information tokens (풳K(t)) to equally-spaced sample past sequence of state information tokens yM[ ] comprising M equally-spaced state information tokens sampled in an internal sampling domain, where M denotes is predictor memory (i.e., the length). The cropped set of state information tokens (JCK(t)) include K state information tokens and the sample set (yM[ ]) includes the M last samples ({y[T — M + l],y[T — M + 2],...,y[T]}) at the RE granularity.

[0088] The WESI token predictor block 360 takes the sequence yM[ ] obtained from the WESI token sequencing block 358 as input in the internal sampling domain. The WESI token predictor block 360 includes a WESI predictor 362 and a First-In-First-Out, FIFO, buffer 364. The WESI token predictor block 360 predicts the sequence of future tokens y[P + P] iteratively using P + L — 1 iterations. At iteration 0, the WESI predictor 362 takes as input sequence yM[P] and predict token y[P + 1], Then, at iteration k, 0 < k < P + L + 1, the FIFO buffer 364 of capacity M collects the concatenation of the M — k last samples ({y[ℓ — M + k + 1],...,y[ℓ]}) and the already predicted token sequence {y[P + 1],...,y[P + fc]}, providing as input to the WESI predictor 362 the M samples {y[ℓ — M + k + 1],...,y[ℓ],y[ℓ + 1],...,y[ℓ + k]}. After (P + L — 1) iterations, the WESI predictor 362 outputs the predicted future sequence of state information tokens ŷ[ℓ + P] = {ŷ[ℓ + P ŷ[ℓ + P + 1], ŷ[ℓ + P + L - 1]}.

[0089] FIG. 4 illustrates an exemplary Artificial intelligence, Al, based air interface algorithm implemented in an air interface apparatus in accordance with an implementation of the disclosure. The predicted future sequence of state information tokens 402 (ŷ[ℓ + P]) containing predicted WESI tokens is given as input to the WESI Structured Representation Block 404 including D WESI feature representation learning modules, each one corresponding to a one of D properties contained in the WESI property list 408 (property i ->

[0090]

[0091] [P + P]). FIG. 4 explicitly shows a WESI feature 1 representation learning module 404A, a WESI feature 2 representation learning module 404B, a WESI feature i representation learning module 4041, a WESI feature D — 1 representation learning module 404D, and a WESI feature D representation learning module 404C.

[0092] A possibly different subset of the set of air interface representation inputs (Z[P + P]) are combined to generate an air interface representation input to each one of the A Al-based air interface algorithms 406 to make decisions on transmission parameters for communication. For instance, in FIG. 4, the output of the WESI feature 1 representation learning module 404 A represented as Z(1)[ℓ + P] and the WESI feature 2 representation learning module 404B represented as Z(2)[ℓ + P] combines with the output of the WESI feature D — 1 representation learning module 404D represented as

[0093]

[0094] [ℓ + P] to generate an air interface representation input and communicates the air interface representation input to an Al-based air interface algorithm 406A to make decisions on transmission parameters for communication. The output of the WESI feature i representation learning module 4041 represented as Z(i)[ℓ + P] generates and sends an air interface representation input to an Al based air interface algorithm 406C to make decisions on transmission parameters for communication. The output of the WESI feature D representation learning module 404C represented as Z(D)[ℓ + P] and WESI feature D — 1 representation learning module 404D represented as

[0095]

[0096] [ℓ + P] are combined to generate an air interface representation input and communicates the air interface representation input to the Al based air interface algorithm 406B to make decisions on transmission parameters for communication.

[0097] FIG. 5 illustrates an architecture for Structured Wireless Environment State Information, WESI, representation in accordance with an implementation of the disclosure. The WESI architecture shows a predicted future sequence of state information tokens 502 (ŷ[ℓ + P]) containing predicted WESI tokens, a WESI property list 504, and a WESI structured representation block 506.

[0098] The WESI property list 504 includes D WESI Properties (explicitly in the figure, a WESI Property 1, a WESI Property i, and a WESI Property D). Each WESI property specifies a representation dimensionality such as an output dimension, a representation size, and a time-frequency span of property validity associated with the corresponding WESI feature representation learning module.

[0099] The WESI structured representation block 506 explicitly shows a WESI feature 1 representation learning module 506A, a WESI feature i representation learning module 5061, and a WESI D representation learning module 506D. The predicted future sequence of state information tokens 502 (ŷ[ℓ + P]) is transferred to the D WESI feature representation learning modules for performing representation learning on the D WESI properties specified in the WESI property list 504 and to generate a set of air interface representation inputs. Such air interface representation input is provided as input to an Al-based air interface algorithm. For instance, for the predicted future sequence of state information tokens y[P + P], containing the L predicted WESI tokens at minimum time granularity for a transmission block starting at an internal sampling index T + P, the WESI structured representation block 506 obtains the air interface representation input (i.e., WESI structured representation) as Z[ℓ + P] = {Z(1)[ℓ + P],..., Z(i)[ℓ + P],..., Z(D)[ℓ + P]}, where Z(i)[ℓ + P] denotes the (th individual feature, which corresponds to the (th WESI property specified in the WESI property list 504 for the transmission block starting at an internal sampling index T + P at its specified time and frequency granularity as shown in FIG. 6.

[0100] FIG. 6 illustrates how a Wireless Environment State Information, WESI, structured representation / air interface representation input features are mapped into the time-frequency grid associated with a transmission block in accordance with an implementation of the disclosure. For example, the WESI structured representation of feature 1 602 includes a time granularity of 4 and a frequency granularity of F / 4. Hence, the WESI representation learning of feature 1 is {z(1)1,1[ℓ + P], z(1)2,1[ℓ + P],..., z(1)4,1[ℓ + P], z(1)1,L / 4[ℓ + P],..., z(1)4,L / 4[ℓ + P]}. Similarly, the WESI structured representation of feature D 604 includes a time granularity of L RE and a frequency granularity of F / 2 REs. Hence, the WESI representation learning of feature D 604 is {

[0101]

[0102] z^ + p], z^y + p]}. FIG. 7 illustrates how ithfeature of a Wireless Environment State Information, WESI, structured representation / air interface representation are mapped into the time-frequency grid associated with a transmission block in accordance with an implementation of the disclosure. The WESI structured representation of feature i 702 has time granularity τi= L / Li, where L is the total number of REs in a resource frame, and frequency granularity λi= F / Fiwhere F is total number of REs in a resource block. Then, the WESI structured representation is obtained as

[0103]

[0104] [ℓ + P] = {z(i)1,1[ℓ + P], z(i)2,1[ℓ + P],..., z(i)f,1[ℓ + P], z(i)1,t[ℓ + P],..., Z

[0105]

[0106] z(i)F,L[ℓ + P]}, where z(i)f,t[ℓ + P] ∈ is the feature vector characterizing the (th WESI property for time frequency block containing F;x Ltresource elements starting at the resource element at position (f — 1)Fi+ 1 of the resource block at position (j — 1)Li+ 1 in the time of the transmission block.

[0107] FIG. 8 illustrates an exemplary view of an Air interface algorithm’s decisions granularities in accordance with an implementation of the disclosure. A set of A Al-based air interface algorithms 802 take decision on the transmission parameters for a transmission block to be used for communication. Each air interface algorithm 802 has a time granularity and a frequency granularity to take decisions on non-overlapping time frequency blocks inside the transmission block. The time and frequency granularities of a WESI structure representation required by an air interface algorithm 802 are not smaller than the time and frequency granularities of the algorithm’s decisions.

[0108] FIG. 9 is a flow diagram that illustrates a method for an air interface apparatus configured to utilize artificial intelligence to perform air interface algorithms in a wireless communication network in accordance with an implementation of the disclosure, to perform air interface algorithms in the wireless communication network. At step 902, the method that includes air interface apparatus obtains a channel state information, CSI, at a global time (t) corresponding with internal sampling index (ℓ). The channel state information includes information on channel properties. At step 904, an environment state information, ESI, at the global time (t) is obtained. The environment state information includes information on environmental properties. At step 906, the CSI and ESI is tokenized together into a state information token (x(t)). At step 908, a future sequence of state information tokens (ŷ[ℓ + P]) is predicted for a future internal sampling index ( + P) based on the sequence of past state information tokens until the global time t ({x(t)}). At step 910, a set of air interface representation inputs (Z[ℓ + P]) is generated for the future internal sampling index ( + P) by performing representation learning on at least some of the channel properties and environmental properties and the predicted future sequence of state information tokens. Each property on which representation learning is performed on provides one air interface representation (property i -> Z(i)[ℓ + P]). At step 912, the air interface algorithm(s) is performed by utilizing Artificial Intelligence, Al, air interface algorithm(s) based on the air interface representation inputs (Z[ℓ + P]).

[0109] The method employs wireless environment state information, WESI, which contains CSI enriched by time-synchronized ESI. The ESI is a representation of the state of the environment obtained by fusing the information captured by different sensors such as cameras, lidars, weather-related sensors, GPS sensors, and other prior information such as geographical maps, sitespecific hardware information at a network node, a receiver node, and a transmitter node. The method employs representation learning techniques to determine a WESI representation model which is easier predicted than raw CSI. The method allows for independent management of the life cycle of wireless environment state information, WESI, representation related to the Al models for wireless functionality. The method employs / deploys less complex and less scenario or wireless-propagation dependent Al-based algorithms, which requires less online training and fine-tuning and less frequent Al model updates. The method efficiently addresses the challenge of tokenizing Wireless Environment State Information by combining the CSI and the ESI. This process leverages an Al-based air interface, replacing conventional algorithms with Al models that utilize CSI to enhance WESI performance. The raw CSI contains information about the frequency-domain propagation channels between each transmit and receive antenna pair over the entire or partial resource bandwidth. By utilizing Al-driven models, the method establishes an efficient WESI tokenization procedure. It applies advanced pre-processing techniques to reduce the effects of noise and hardware impairments on CSI(t) and employs a data fusion mechanism to seamlessly integrate CSI(t) and ESI(t) to reduce communication cost.

[0110] In an aspect, a computer program product including program instructions for performing the method as described above is provided, when executed by one or more processors in an air interface apparatus.

[0111] FIG. 10 is an illustration of a computer system in which the various architectures and functionalities of the various previous implementations may be implemented. As shown, the computer system 1000 includes at least one processor 1004 that is connected to a bus 1002, wherein the computer system 1000 may be implemented using any suitable protocol, such as PCI (Peripheral Component Interconnect), PCI-Express, AGP (Accelerated Graphics Port), Hyper Transport, or any other bus or point-to-point communication protocol (s). The computer system 1000 also includes a memory 1006.

[0112] Control logic (software) and data are stored in the memory 1006 which may take a form of random-access memory (RAM). In the disclosure, a single semiconductor platform may refer to a sole unitary semiconductor-based integrated circuit or chip. It should be noted that the term single semiconductor platform may also refer to multi-chip modules with increased connectivity which simulate on-chip modules with increased connectivity which simulate on-chip operation, and make substantial improvements over utilizing a conventional central processing unit (CPU) and bus implementation. Of course, the various modules may also be situated separately or in various combinations of semiconductor platforms per the desires of the user. The computer system 1000 may also include a secondary storage 1010. The secondary storage 1010 includes, for example, a hard disk drive and a removable storage drive, representing a floppy disk drive, a magnetic tape drive, a compact disk drive, digital versatile disk (DVD) drive, recording device, universal serial bus (USB) flash memory. The removable storage drives at least one of reads from and writes to a removable storage unit in a well-known manner.

[0113] Computer programs, or computer control logic algorithms, may be stored in at least one of the memory 1006 and the secondary storage 1010. Such computer programs, when executed, enable the computer system 1000 to perform various functions as described in the foregoing. The memory 1006, the secondary storage 1010, and any other storage are possible examples of computer-readable media.

[0114] In an implementation, the architectures and functionalities depicted in the various previous figures may be implemented in the context of the processor 1004, a graphics processor coupled to a communication interface 1012, an integrated circuit (not shown) that is capable of at least a portion of the capabilities of both the processor 1004 and a graphics processor, a chipset (namely, a group of integrated circuits designed to work and sold as a unit for performing related functions, and so forth). Furthermore, the architectures and functionalities depicted in the various previous-described figures may be implemented in a context of a general computer system, a circuit board system, a game console system dedicated for entertainment purposes, an application-specific system. For example, the computer system 1000 may take the form of a desktop computer, a laptop computer, a server, a workstation, a game console, an embedded system.

[0115] Furthermore, the computer system 1000 may take the form of various other devices including, but not limited to a personal digital assistant (PDA) device, a mobile phone device, a smart phone, a television, and so forth. Additionally, although not shown, the computer system 1000 may be coupled to a network (for example, a telecommunications network, a local area network (LAN), a wireless network, a wide area network (WAN) such as the Internet, a peer-to-peer network, a cable network, or the like) for communication purposes through an I / O interface 1008.

[0116] It should be understood that the arrangement of components illustrated in the figures described are exemplary and that other arrangement may be possible. It should also be understood that the various system components (and means) defined by the claims, described below, and illustrated in the various block diagrams represent components in some systems configured according to the subject matter disclosed herein. For example, one or more of these system components (and means) may be realized, in whole or in part, by at least some of the components illustrated in the arrangements illustrated in the described figures.

[0117] In addition, while at least one of these components are implemented at least partially as an electronic hardware component, and therefore constitutes a machine, the other components may be implemented in software that when included in an execution environment constitutes a machine, hardware, or a combination of software and hardware.

[0118] Although the disclosure and its advantages have been described in detail, it should be understood that various changes, substitutions, and alterations can be made herein without departing from the spirit and scope of the disclosure as defined by the appended claims.

Claims

1. CLAIMS1. An air interface apparatus (210, 314) configured to utilize an artificial intelligence (212) to perform air interface algorithms (316), wherein the air interface apparatus is characterized in that the air interface apparatus is further configured to obtain channel state information, CSI, at a global time (t) corresponding to an internal sampling index ( ), wherein the channel state information comprises information on channel properties,3.obtain environment state information, ESI, at the global time (t) corresponding to the internal sampling index ( ), wherein the environment state information comprises information on environmental properties,4.tokenize the CSI and ESI together into a state information token (x(t)),5.predict a future sequence of state information tokens (ŷ[ℓ + P]) for a future internal sampling index (F + P) based on the past sequence of state information tokens until the global time t ({x(t)}),6.generate a set of air interface representation inputs (Z[ℓ + P]) for the future internal sampling index (F + P) by performing representation learning on at least some of the channel properties and environmental properties and the predicted future sequence of state information tokens, wherein each property on which representation learning is performed on provides one air interface representation input (property i -> Z(i)[ℓ + P]), and then to7.perform the air interface algorithms utilizing an artificial intelligence air interface algorithms (316) based on the air interface representation inputs (Z[ℓ + P]).

2. The air interface apparatus (210, 314) according to claim 1, wherein the predicted future sequence of state information tokens (ŷ[ℓ + P]) for the future internal sampling index (T + P) comprises L’ predicted tokens (y) and the set of air interface representation inputs (Z[ℓ + P]) for the future internal sampling index (T + P) comprises a set of representation input for property i (Z^), wherein each representation input for property i (Z(i)[ℓ + P]) comprises a set of vectors (z(i)f,j(ℓ + P)) corresponding to an individual feature of the state information (CSI, ESI) for a transmission block starting at an internal sampling index (T + P), wherein each vector is for a time index ( / ') and frequency index ( ) at a specified time and frequency granularity.

3. The air interface apparatus (210, 314) according to claim 2, wherein the time granularity is r;= L / L,- and the frequency granularity is 2.,- = F / Ft, wherein the set of interface representations comprises all vectors for all features (i = 1, —, D), all time indices (j = 1, Lt) and all frequency indices ( = 1,, Fi).

4. The air interface apparatus (210, 314) according to any preceding claim, wherein the air interface apparatus is further configured to11.predict the future sequence of state information tokens (ŷ[ℓ + P]) for a future internal sampling index (F + P) up to an end transmission sampling index (F + P + L — 1) and to12.generate the set of air interface representation inputs (Z[ℓ + P]).

5. The air interface apparatus (210, 314) according to claim 4, wherein the air interface apparatus is further configured to predict the token sequence (ŷ[ℓ + P]) as sampled for the whole resource frame starting at internal sampling index F + P, in which transmission is taking place.

6. The air interface apparatus (210, 314) according to any preceding claim, wherein the predicted token sequence (ŷ[ℓ + P]) is sampled at an internal sampling rate at minimum time granularity.

7. The air interface apparatus (210, 314) according to any preceding claim, wherein a property list (310, 504) (channel and / or environmental) is defined by a15.a Representation Dimensionality,16.a Representation Size, and17.a Time-Frequency span.

8. The air interface apparatus (210, 314) according to any preceding claim, wherein the ESI is a representation of the state of the environment obtained by fusing information captured by sensor(s) environment information at a node.

9. The air interface apparatus (210, 314) according to any preceding claim, wherein the air interface apparatus is further configured to perform representation learning on the at least some of the channel properties and environmental properties related to the air interface algorithms to be performed.

10. The air interface apparatus (210, 314) according to any preceding claim, wherein the channel state information, CSI, and the environment state information, ESI, are time-synchronized for the global time (t).

11. The air interface apparatus (210, 314) according to any preceding claim, wherein the CSI and ESI are measured at a transmitter node (324).

12. The air interface apparatus (210, 314) according to any preceding claim, wherein the air interface apparatus is further configured to perform an air interface algorithm utilizing the artificial intelligence air interface algorithms (316) based on one or more elements (Z(i)) of the set of air interface representation inputs (Z).

13. The air interface apparatus (210, 314) according to any preceding claim, wherein the combination of the channel state information, CSI, and the environment state information, ESI, is grouped as a Wireless Environment State Information, WESI, wherein the channel and environmental properties are WESI properties.

14. The air interface apparatus (210, 314) according to any of claims 1 to 12, wherein the air interface apparatus is a controller configured to be used in a wireless communications node.

15. The air interface apparatus (210, 314) according to any of claims 1 to 12, wherein the air interface apparatus is a wireless communications node.

16. A method for an air interface apparatus (210, 314) configured to utilize artificial intelligence (212) to perform air interface algorithms (316), wherein the method is characterized in that the method comprises the air interface apparatus obtaining channel state information, CSI, at a global time (t) corresponding with an internal sampling index ( ), wherein the channel state information comprises information on channel properties,27.obtaining environment state information, ESI, at the global time (t), wherein the environment state information comprises information on environmental properties,28.tokenizing the CSI and ESI together into a state information token (x(t))29.predicting a future sequence of state information tokens (ŷ[ℓ + P]) for a future internal sampling index (ℓ + P) based on the sequence of past state information tokens until the global time t ({x(t)}),30.generating a set of air interface representation inputs (Z[ℓ + P]) for the future internal sampling index ( + P) by performing representation learning on at least some of the channel properties and environmental properties and the predicted future sequence of state information tokens, wherein each property on which representation learning is performed on provides one air interface representation input (property i ->31.

32. [ℓ + P]), and then33.performing the air interface algorithm(s) utilizing artificial intelligence air interface algorithm(s) based on the air interface representation inputs (Z[ℓ + P]).

17. A computer program product comprising program instructions for performing the method according to claim 16, when executed by one or more processors in an air interface apparatus (210, 314).

Images