Method and apparatus for user equipment to improve transmission efficiency

Abstract

Embodiments relate to a method (100) for a user equipment. The method (100) for a user equipment for improving transmission efficiency on a radio channel comprises obtaining (110) a prediction environment model and predicting (120) a channel dynamics of the radio channel based on the prediction environment model. Furthermore, the method (100) comprises receiving (130) a reference signal to measure a channel property of the radio channel and calculating (140) the channel dynamics of the radio channel based on the reference signal. The method (100) further comprises determining (150) a deviation between the predicted channel dynamics and the calculated channel dynamics and adjusting (160) a transmission parameter based on the deviation to improve the transmission efficiency on the radio channel.

Description

Technical Field

[0001] This disclosure relates to the field of wireless communications. Embodiments relate to methods for improving transmission efficiency for user equipment, methods for network entities, devices, vehicles, and computer programs, and more particularly, but not exclusively, to concepts for improving transmission efficiency on radio channels (e.g., adjusting transmission parameters). Background Technology

[0002] The development of 5G has garnered increased attention from the automotive industry, a vertically integrated manufacturer anticipating the most advanced features of the next generation of wireless communication. Among 5G's key innovations, wide-spectrum possibilities (currently licensed up to 28 GHz bands – the first millimeter-wave bands for mobile use), enhanced support for high-mobility scenarios, and new mechanisms for guaranteeing and predicting the quality of service (QoS) experienced have been established as key functions to support an increasingly connected transportation ecosystem.

[0003] KR 20 200 095 137 A discloses a method for wireless communication environment adaptation using deep learning, comprising the following steps: periodically receiving in-phase and quadrature samples from a wireless access environment, and extracting channel characteristic information, including signal-to-noise ratio information, channel state information, and received signal strength indication, from the received in-phase and quadrature samples. Furthermore, the method includes learning a training dataset from the extracted current channel characteristic information using a recurrent neural network to classify the data, generating training results, and using a prediction dataset classified from the channel characteristic information extracted at the current time and the generated training results to predict channel state information at future time points.

[0004] CN 107 276 698 A discloses a method comprising monitoring a first target beam, decoding a beacon of the first target beam, and extracting channel state information of the beacon of the first target beam if the beacon of the first target beam cannot be successfully decoded. Furthermore, the method includes accumulating the channel state information of the beacon of the first target beam into spatial channel state information, and obtaining an optimal estimated angle of arrival (Angle of Arrival) based on a preset channel and the spatial channel state information. The method also includes monitoring a beam along the direction of the optimal estimated Angle of Arrival (Angle of Arrival), and after detecting a second target beam, decoding the beacon of the second target beam, wherein the second target beam is a beam supported by a randomly selected antenna array along the optimal estimated Angle of Arrival (Angle of Arrival).

[0005] CN 112 364 769 A discloses a WiFi-based crowd counting method, which includes collecting channel state information data containing human movement information by a user equipment and denoising the channel state information data using a Butterworth bandpass filter. Furthermore, the method includes removing data redundancy and reducing noise by employing a principal component analysis (PCA) dimensionality reduction algorithm. The method also includes performing pattern recognition on selected channel state power information, estimating direction, and normalizing selected amplitude and phase information. Additionally, the method includes performing feature extraction by fusing features from the source and target domains using an HFA method to obtain a trained model for predicting the number of people.

[0006] Transmission efficiency can depend on several conditions, such as the pilot rate used to estimate Channel State Information (CSI). Obtaining CSI from two communication nodes, both using antenna arrays with beamforming capabilities, is a challenging task because the required pilot signaling increases significantly with the number of antenna elements, potentially leading to unwanted signaling overhead. For example, the pilot rate used to estimate CSI can generate unwanted traffic loads on the radio channel, resulting in reduced transmission efficiency. Consequently, improvements to transmission efficiency on the radio channel may be necessary.

[0007] Therefore, it has been found that adjusting transmission parameters based on the discrepancy between the predicted and calculated channel dynamics can improve transmission efficiency, for example, by reducing pilot rates and thus data traffic. For instance, the predicted channel dynamics can be matched with the calculated channel dynamics, thereby reducing pilot rates. For example, the environmental parameters used to calculate the channel dynamics may be satisfactory, and the CSI can be derived from this environment. Thus, transmission efficiency can be improved, for example, by reducing signaling overhead. Summary of the Invention

[0008] The example provides a method for improving transmission efficiency on a radio channel for a user equipment. The method includes obtaining a predictive environment model and predicting channel dynamics of the radio channel based on the predictive environment model. Furthermore, the method includes receiving a reference signal to measure the channel characteristics of the radio channel and calculating the channel dynamics of the radio channel based on the reference signal. The method further includes determining the deviation between the predicted channel dynamics and the calculated channel dynamics and adjusting transmission parameters based on the deviation to improve transmission efficiency on the radio channel. Thus, the transmission parameters can be adjusted depending on the predictive environment model. Furthermore, CSI can be estimated / determined based on the predictive environment model by predicting the channel dynamics. For example, if the predicted channel dynamics match the calculated channel dynamics, the transmission parameters can be adjusted to obtain information about the CSI, for example, by reducing signaling overhead by decreasing the reference signal rate, reporting content, reporting rate, etc.

[0009] In one example, if the deviation is below a threshold, the transmission parameters can be adjusted by reducing the reference signal rate and / or the reference signal content. This allows for improved transmission efficiency, for example, by reducing the reference signal rate used to determine the CSI.

[0010] In one example, the method may further include, if the deviation is higher than a threshold, re-predicting the channel dynamics, redetermining the deviation between the re-predicted channel dynamics and the calculated channel dynamics, and if the redetermined deviation is lower than the threshold, adjusting the transmission parameters; otherwise, repeating the re-prediction and redeterminism until the deviation is lower than the threshold. Thus, the prediction of channel dynamics can be improved, for example, by using artificial intelligence (AI).

[0011] In one example, prediction and / or re-prediction can be based on a machine learning model trained with the predicted and computed channel dynamics. Thus, the AI ​​can be trained in an improved manner using data available / determinable during normal operation. For example, a training dataset can be generated / edited based on the predicted and computed channel dynamics obtained during normal operation.

[0012] In one example, the method may further include receiving a dataset for training or initializing a machine learning model. This allows the AI ​​to be trained using an increased dataset, which can increase the reliability of the AI.

[0013] In one example, the adjustment of the transmission parameters can be performed by reducing channel quality reports. This reduces signaling overhead.

[0014] The example involves a vehicle capable of performing the methods described above, where information about the environment can be obtained by using one or more sensors on the vehicle to determine and / or receive information about the environment. In this way, data for training the AI ​​can be obtained in a simple manner.

[0015] The example relates to a method for network entities to improve transmission efficiency on radio channels used for communication with user equipment. The method includes receiving a predictive environment model, predicting channel dynamics of the radio channel based on the predictive environment model, receiving a reference signal to measure the channel characteristics of the radio channel, calculating the channel dynamics of the radio channel based on the reference signal, determining the deviation between the predicted channel dynamics and the calculated channel dynamics, and adjusting transmission parameters based on the deviation to improve transmission efficiency on the radio channel. Thus, the transmission parameters can be adjusted depending on the predictive environment model. Furthermore, CSI can be estimated / determined based on the predictive environment model using the predicted channel dynamics. For example, if the predicted channel dynamics match the calculated channel dynamics, the transmission parameters can be adjusted to obtain information about the CSI, for example, by reducing signaling overhead by decreasing the reference signal rate, reporting content, reporting rate, etc.

[0016] In one example, the transmission parameters can be adjusted by reducing the user equipment's report content, report rate, and / or reference signal rate. This reduces signaling overhead.

[0017] In one example, the method may further include re-predicting the channel dynamics if the deviation is higher than a threshold, redetermining the deviation between the re-predicted channel dynamics and the calculated channel dynamics, and adjusting the transmission parameters if the redetermined deviation is lower than the threshold; otherwise, repeating the re-prediction and redeterminism until the deviation is lower than the threshold. In this way, the prediction of channel dynamics can be improved, for example, by using artificial intelligence (AI).

[0018] In one example, prediction and / or re-prediction can be based on a machine learning model trained with the predicted and computed channel dynamics. Thus, the AI ​​can be trained in an improved manner using data available / determinable during normal operation.

[0019] The example further provides an apparatus including one or more interfaces configured to communicate with a network entity or user equipment. The apparatus further includes a processing circuitry configured to control the one or more interfaces and perform the methods described above for user equipment and / or network entities.

[0020] The example further provides vehicles that include the equipment described above.

[0021] The example further relates to a computer program having program code that, when executed on a computer, processor, or programmable hardware component, performs the methods described above. Attached Figure Description

[0022] The following are some examples of devices and / or methods described only by way of example and with reference to the accompanying drawings, in which:

[0023] Figure 1 Examples of methods for user equipment are shown;

[0024] Figure 2 Examples of methods for network entities are shown; and

[0025] Figure 3 A block diagram of the device is shown. Detailed Implementation

[0026] Some examples will now be described in more detail with reference to the accompanying drawings. However, other possible examples are not limited to the features of these embodiments described in detail. Other examples may include modifications to features, as well as equivalents and alternatives to features. Furthermore, the terminology used herein to describe certain examples should not limit other possible examples.

[0027] Throughout the description of the accompanying drawings, the same or similar reference numerals refer to the same or similar elements and / or features, which may be implemented in equivalent or modified forms while providing the same or similar functions. For clarity, the thickness of lines, layers, and / or areas in the drawings may also be enlarged.

[0028] When 'or' is used to combine two elements A and B, it is to be understood as disclosing all possible combinations, namely only A, only B, and A and B, unless otherwise explicitly defined in individual cases. As alternative wording for the same combination, 'at least one of A and B' or 'A and / or B' can be used. This is equivalent to combinations of more than two elements.

[0029] If a singular form such as “a,” “an,” or “the” is used, and the use of a single element is not explicitly or implicitly defined as mandatory, then other examples may use several elements to achieve the same functionality. If a function is described below as being implemented using multiple elements, then other examples may use a single element or a single processing entity to achieve the same functionality. It should be further understood that the terms “comprising” and / or “including”, when used, describe the presence of the specified features, integers, steps, operations, processes, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, integers, steps, operations, processes, elements, components, and / or groups thereof.

[0030] Figure 1An example of method 100 for a user equipment is shown. Method 100 for a UE is used to improve transmission efficiency on a radio channel. Method 100 includes obtaining 110 a predicted environment model and predicting 120 channel dynamics of the radio channel based on the predicted environment model. Furthermore, method 100 includes receiving 130 a reference signal to measure channel characteristics of the radio channel and calculating 140 channel dynamics of the radio channel based on the reference signal. Method 100 also includes determining 150 the deviation between the predicted channel dynamics and the calculated channel dynamics and adjusting 160 transmission parameters based on the deviation to improve transmission efficiency on the radio channel. For example, transmission parameters can be any parameters related to channel state reporting, such as reference signal rate, reporting rate, reporting content, etc. Thus, transmission efficiency can be improved, for example, by reducing the reference signal rate, reporting rate, reporting content, etc., used to determine CSI.

[0031] The UE can communicate with network entities (e.g., base stations) in a mobile communication system. For example, the UE and the network entity can communicate in / via the mobile communication system. The mobile communication system may include multiple transmission points and / or base stations operable to communicate with the UE via radio signals. In one example, the mobile communication system may include the UE and the network entity.

[0032] Network entities can reside in fixed or stationary parts of a mobile communication system. Network entities can correspond to remote wireless heads, transmission points, access points, macro cells, small cells, microcells, picocells, femtocells, metropolitan area cells, etc. The term small cell can refer to any cell smaller than a macro cell, such as a microcell, picocell, femtocell, or metropolitan area cell. Furthermore, a femtocell is considered smaller than a picocell, and a picocell is considered smaller than a microcell. A network entity can be a wireless interface of a wired network that enables the transmission and reception of radio signals to a UE (such as a UE). These radio signals may conform to, for example, radio signals standardized by 3GPP, or generally conform to one or more of the systems listed above. Thus, network entities can correspond to NodeBs, eNodeBs, BTSs, access points, etc.

[0033] Mobile communication systems can be cellular. The term "cell" refers to the coverage area of ​​radio services provided by a transmission point, remote unit, remote head, remote radio head, communication device, UE, network entity, or NodeB / eNodeB. The terms "cell" and "base station" are used synonymously. A wireless communication device (e.g., UE) can be registered to or associated with at least one cell (e.g., network entity). For example, it can be associated with a cell (e.g., network entity) to enable the exchange of data between mobile devices connected to or linked within the network and the associated cell's coverage area using dedicated channels, connections, or links.

[0034] Typically, a UE is a device capable of wireless communication. However, specifically, a UE can be a mobile UE, for example, a UE suitable for being carried by a user. For example, within the meaning of the corresponding communication standard used for mobile communication, a UE can be a user terminal (UT) or user equipment (UE). For example, a UE can be a mobile phone, such as a smartphone, or another type of mobile communication device, such as a smartwatch, laptop computer, tablet computer, or autonomous augmented reality glasses. For example, the UE and network entities can be configured to communicate in a cellular mobile communication system. Thus, the UE and network entities can be configured to communicate in a cellular mobile communication system, such as in a sub-6 GHz cellular mobile communication system (covering the frequency band between 500 MHz and 6 GHz) or in a millimeter wave cellular mobile communication system (covering the frequency band between 20 GHz and 60 GHz). For example, the UE and network entities can be configured to communicate in a mobile communication system / cellular mobile communication system. Typically, a mobile communication system may correspond, for example, to one of the mobile communication networks standardized by the 3rd Generation Partnership Project (3GPP), where the term mobile communication system is used synonymously with mobile communication network. Mobile communication systems can correspond to, for example, 5G, LTE, LTE-A, HSPA, UMTS or UMTS Terrestrial Radio Access Network (UTRAN), eUTRAN, GSM or EDGE enhanced data rate GSM evolution network, GSM / EDGE radio access network (GERAN), or mobile communication networks with different standards, such as the WIMAX network IEEE 802.16, and generally orthogonal frequency division multiple access (OFDMA) network, time division multiple access (TDMA) network, code division multiple access (CDMA) network, wideband CDMA (WCDMA) network, frequency division multiple access (FDMA) network, space division multiple access (SDMA) network, etc.

[0035] Furthermore, the UE / network entity may be adapted to communicate via non-cellular communication systems, or configured to communicate via non-cellular communication systems, such as via device-to-device vehicle communication systems (e.g., according to the IEEE 802.11p standard (IEEE 802.11p standard for vehicle communication)) or via wireless local area networks (e.g., according to IEEE 802.11a, IEEE 802.11b, IEEE 802.11g, IEEE 802.11n, IEEE 802.11ac, or IEEE 802.11ax, also known as Wi-Fi 1 to Wi-Fi 6(E)). Specifically, the UE and network entities may be adapted to communicate in the frequency band between 5 GHz and 7.1 GHz or configured to communicate in the frequency band between 5 GHz and 7.1 GHz, which covers communication in the 5 GHz band (for WiFi in the 5 GHz band), the 5.9 GHz band (for vehicle communication according to the 802.11p standard), and between 5.9 GHz and 7.1 GHz (for WiFi in the 6 GHz band).

[0036] The connection between the UE and the network entity can be a wireless connection, such as a millimeter-wave connection via a mobile communication system (e.g., using a carrier frequency of at least 20 GHz), or it can be performed at a lower carrier frequency, such as using a carrier frequency of up to 7.5 GHz. For example, the wireless connection between the UE and the network entity can be initiated using protocols of a mobile communication system, or using a short-range communication system, such as via a wireless local area network as outlined above.

[0037] As can be clearly seen from the examples above, although communication between the UE and network entities occurs via the mobile communication system, additional and / or alternative communications between the UE and network entities (e.g., if the network entity is a vehicle) can occur via a vehicle communication system. This communication can be implemented directly, for example, via device-to-device (D2D) communication. This communication can be implemented using the specifications of a vehicle communication system. Examples of D2D are direct communication between vehicles, also known as vehicle-to-vehicle (V2V) or vehicle-to-everything (V2X), car-to-car, and dedicated short-range communication (DSRC). Technologies for implementing such D2D communication include 802.11p, 3GPP systems (4G, 5G, NR and higher), etc.

[0038] A predicted environment model can be obtained by receiving and / or determining information. For example, the UE can download a map of a city center area and combine location data determined by the UE (e.g., using a Global Positioning System, GPS) and / or provided by the user with the map to determine information about the UE's environment. This environmental information can be combined with other information to determine the predicted environment model. For example, the predicted environment model can be determined by the UE's processing circuitry (e.g., a central processing unit). The UE's user can, for example, provide the UE with information about a planned journey using navigation software for a vehicle. Thus, the UE can generate a predicted environment model based on the planned route. For example, the UE can predict the journey duration based on traffic information (e.g., which can be received by a base station), and based on the journey duration, the UE can generate the predicted environment model, for example, according to the UE's location on the route.

[0039] Optionally or alternatively, the UE may obtain information about the environment in real time, for example, through measurement, or receive real-time measurement data, which can be used to determine a predictive environment model. For example, the UE may obtain information about the speed of another vehicle in the environment, and thus the predicted environment model may be based on the speed of the other vehicle and the information about the environment.

[0040] Optionally or alternatively, the UE may receive the predicted environment model via another communication device (e.g., infrastructure, another UE in the UE's environment, network entity, etc.).

[0041] For example, the UE can receive information from moving obstacles (e.g., another vehicle) and / or stationary obstacles (such as infrastructure, e.g., traffic lights) that moving obstacles may disappear within a predefined time (e.g., during the red phase of a traffic light) (e.g., the vehicle may depart). Thus, the predicted environment model can predict such changes in the environment.

[0042] For example, information about the environment may include dynamic data about moving obstacles in the environment (e.g., another vehicle or bus), such as the speed of the moving obstacle. The UE can then determine a predicted environmental model based on this information; for example, the UE can predict environmental changes by observing the speed of the moving obstacle to determine the predicted environmental model. In principle, an obstacle can be an object in the line of sight between the UE and the network entity and / or an object in a location that causes a non-negligible change in the electromagnetic field at the UE.

[0043] Predicting the channel dynamics of radio channels 120 based on a predictive environment model can be performed by the UE's processing circuitry (e.g., a central processing unit). The predicted channel dynamics depend on the predicted environment model. For example, if the predicted environment model does not show environmental changes within a predefined time period (e.g., multiple CSI reporting times), the predicted channel dynamics may be lower than the environmental changes shown by the predicted environment model within the predefined time period. For example, if no environmental changes are predicted within the predefined time period, the value of the predicted channel dynamics indicating changes in channel dynamics can be zero, while if environmental changes are predicted and thus changes in channel dynamics are predicted, the value can be greater than zero. For example, if the value of the predicted channel dynamics is zero, CSI reporting can be reduced or even omitted within the predefined time period. For example, as the value of the predicted channel dynamics increases, the need for CSI reporting may increase, which can result in a higher CSI reporting rate for higher predicted channel dynamics.

[0044] Optionally or alternatively, the predicted channel dynamics may depend on the complexity of the predicted environment model, such as the density of obstacles in the predicted environment model. For example, a user equipment (UE) may be located in an open area or a city center area. An open area may result in a less complex environment model for the predicted UE than a model for the predicted UE in a city center area (e.g., an open area may have very few obstacles, resulting in low obstacle density, while a city center area may have many obstacles, resulting in (more) high obstacle density). Thus, even if both models indicate that the environment will not change in the future, the predicted channel dynamics (values) may increase for a city center area compared to an open area, because the likelihood of change in a city center area may be greater than in an open area. For example, more parameters in the predicted environment model (e.g., for obstacles) may increase the predicted channel dynamics (values), for example, leading to a need to increase the CSI reporting rate.

[0045] The 130 reference signal received to measure channel characteristics can be received by one or more sensors of the UE. These sensors can be of a wide variety of UE sensors, such as radar sensors, lidar sensors, automotive radar, ultrasonic sensors, or vision-related sensors (such as cameras or infrared sensors). The reference signal used to measure channel characteristics can be any suitable signal, such as a pilot signal.

[0046] The channel dynamics of the 140 radio channels, calculated based on reference signals, can be performed by the UE's processing circuitry (e.g., a central processing unit). For example, AI can calculate the channel dynamics.

[0047] The deviation between the predicted channel dynamics and the calculated channel dynamics can be determined by the UE's processing circuitry (e.g., central processing unit).

[0048] Improving transmission efficiency on radio channels by adjusting 160 transmission parameters based on deviation can be based on reducing the pilot signal rate, reporting rate, reporting content used to determine CSI, and / or on adjusting beam parameters.

[0049] One or more transmission parameters can define / cause changes in channel strength over time and / or frequency. For example, the selection of one or more transmission parameters of a radio channel can depend on large-scale fading effects and / or small-scale fading effects. For example, the selection of one or more transmission parameters of a radio channel can depend on coherence time (Doppler spread, respectively), frequency coherence (Doppler spread or coherence bandwidth, respectively), channel noise, channel attenuation, and coverage of the UE (and, respectively, network entities). For example, transmission parameters can be spatial angle, spatial domain, transmission power, beamform, selected antenna among multiple antennas used for beamforming, or repetition rate reported by CSI. Thus, by adjusting transmission parameters, signal transmission from the UE can be improved at the receiver (e.g., network entity). For example, for a predicted environment model with high obstacle density, transmission power can be increased and / or beamforming can be performed in a manner where fewer obstacles can interact with the signal, which can increase the received signal strength at the network entity.

[0050] Optionally or alternatively, the transmission parameters can be parameters related to channel state reporting (e.g., CSI), such as the reference signal rate, reporting rate, and reporting content of the CSI. For example, the transmission parameters could be the pilot signal rate used to determine the CSI. Thus, transmission efficiency cannot be increased simply by improving the signal (e.g., by beamforming), but rather by reducing the reporting of channel states (e.g., by reducing the pilot signal rate).

[0051] In one example, if the deviation is below a threshold, the transmission parameters can be adjusted by reducing the reference signal rate. For example, the pilot signal rate, reporting rate, and reporting content of the CSI may be reduced. By reducing the pilot signal rate, signaling overhead can be reduced, and thus transmission efficiency can be improved. For example, the UE can notify another communication device (e.g., a network entity) about the adjustment of transmission parameters, such as the reduction of the pilot signal rate.

[0052] Optionally, the UE can use signal parameters (such as predicted QoS, strength, etc.) to maintain transmission parameters. For example, the UE can reduce the pilot signal rate, and thus it may be unaware of channel changes. However, such changes in the channel can affect signal parameters; for example, the predicted QoS may decrease, which could lead to unsatisfactory radio channel connectivity. Therefore, if the predicted QoS, strength, etc., fall below a quality level / threshold, the UE can increase the pilot signal rate back to its initial value. In this way, the UE can have a control mechanism to prevent undesirable degradation in radio channel communication.

[0053] The threshold can be predefined and independent of the environment / predictive environment model. This allows transmission efficiency to be improved in the same way for all use cases. For example, the threshold can depend on the environment / predictive environment model. For environments with high-density obstacles (e.g., downtown areas), the threshold can be lowered compared to environments with low-density obstacles (e.g., open areas) because even lower deviations can still result in unsatisfactory signal transmission for higher-density obstacles. Thus, the threshold can be adapted to the environment / predictive environment model.

[0054] In one example, method 100 may further include re-predicting the channel dynamics if the deviation is higher than a threshold, and redetermining the deviation between the re-predicted channel dynamics and the calculated channel dynamics. Furthermore, method 100 may include adjusting the transmission parameters if the redetermined deviation is lower than the threshold, otherwise repeating the re-prediction and redetermination until the deviation is lower than the threshold. Thus, the prediction 120 of the channel dynamics can be improved by re-predicting the channel dynamics. Since the reference signal received 130 and the channel calculated 140 can reflect the true channel dynamics with better accuracy than the predicted channel dynamics, the predicted channel dynamics can be corrected relative to the calculated channel dynamics. For example, AI can be used to predict 120 of the channel dynamics, and this can be achieved, for example, by using machine learning to train the AI, which can improve the future prediction 120 of the channel dynamics.

[0055] In one example, the re-prediction can be based on a new prediction environment model. This allows for the expansion of the dataset, for example, for machine learning applications like AI.

[0056] In one example, prediction and / or re-prediction can be based on a machine learning model trained with the predicted channel dynamics and the computed channel dynamics.

[0057] The UE can be configured to provide information about the calculated and / or predicted channel dynamics as input to a machine learning model. Machine learning refers to algorithms and statistical models that computer systems can use to perform specific tasks without explicit instructions, relying instead on models and inference. For example, in machine learning, data transformations inferred from the analysis of historical and / or training data can be used instead of rule-based data transformations.

[0058] Machine learning models are trained using training input data, such as calculated and / or predicted channel dynamics. Many different methods can be used to train machine learning models. For example, supervised learning, semi-supervised learning, or unsupervised learning can be used. In supervised learning, a machine learning model is trained using multiple training samples, where each sample may include multiple input data values ​​and multiple desired output values; for example, each training sample is associated with a desired output value. By specifying both the training samples and the desired output values, the machine learning model “learns” which output value to provide based on input samples similar to those provided during training. Semi-supervised learning can also be used in addition to supervised learning. In semi-supervised learning, some of the training samples lack corresponding desired output values. Supervised learning can be based on supervised learning algorithms, such as classification algorithms, regression algorithms, or similarity learning algorithms. In unsupervised learning, input data can be supplied (only), and structures in the input data can be discovered, for example, by grouping or clustering the input data to find commonalities.

[0059] In at least some examples, reinforcement learning (or a derived concept) is used to train machine learning models. In reinforcement learning, one or more software actors (referred to as "software agents") are trained to take actions in an environment. Rewards are calculated based on the actions taken. Reinforcement learning is based on training one or more software agents to select actions that increase the cumulative reward, resulting in software agents that become better at their given tasks (as demonstrated by the increased reward). This can be the case in various examples of this disclosure—the machine learning model may be trained to provide predicted channel dynamics, or the machine learning model may be trained to output information about predicted channel dynamics based on information about the predicted environment model provided at the input of the machine learning model, the calculated channel dynamics, and / or the deviation between the predicted channel dynamics and the calculated channel dynamics.

[0060] Reinforcement learning can be used to train machine learning models. During training, various methods can be tried and evaluated based on a reward function, which is used to calculate the cumulative reward. Based on the calculated reward, the machine learning model can be modified to perform actions that have already resulted in a higher cumulative reward, thus causing the machine learning model to iteratively improve in predicting channel dynamics. For example, the trained reward function can be based on the deviation between the predicted channel dynamics and the calculated channel dynamics.

[0061] For example, a machine learning model can be an artificial neural network (ANN). An ANN is a system inspired by biological neural networks, such as those found in the brain. An ANN consists of multiple interconnected nodes and multiple connections between nodes, known as edges. There are typically three types of nodes: input nodes that receive input values, hidden nodes that are (only) connected to other nodes, and output nodes that provide output values. Each node can represent an artificial neuron. Each edge transmits information from one node to another. The output of a node can be defined as a (non-linear) function of the sum of its inputs. The node's input can be used in the function based on the "weights" of the edges or nodes that provide the input. The weights of nodes and / or edges can be adjusted during the learning process. In other words, training an artificial neural network can include adjusting the weights of the nodes and / or edges of the artificial neural network, for example, to achieve a desired output for a given input. In at least some examples, a machine learning model can be a deep neural network, for example, a neural network including one or more layers of hidden nodes (e.g., hidden layers), preferably including multiple layers of hidden nodes. In some examples, a machine learning model can be a pointer network.

[0062] Typically, a machine learning model is used by applying input data to the input of the model and by using the output data provided at the model's output. In an artificial neural network (ANN), input data is provided at the ANN's input nodes, then transformed based on the weights of the edges between the ANN's nodes, and finally output by the ANN's output nodes. In this case, previously obtained information is provided as input to the machine learning model. For example, an ANN may include one or more of embedding layers, (masked) attention layers, pointer network layers, and decoding layers to transform data between the input and output layers. In other words, a machine learning model may include one or more embedding layers (e.g., one or more neural network layers). For example, one or more embedding layers may be based on non-linear activation functions, such as the Mish activation function or based on parameter-corrected linear units (PReLU). The UE is configured to provide information about the distances between multiple driving destinations as input to the machine learning model. For example, the UE may be configured to use / execute the machine learning model to influence the transformation between the machine learning model's input and output.

[0063] Training machine learning models requires significant effort; therefore, the ability to reuse machine learning models for different problem sizes can reduce overall training time. In the various examples disclosed herein, machine learning models can be applied to predict / re-predict channel dynamics of radio channels.

[0064] The UE can be configured to receive training input data for training a machine learning model. The training input data includes training information about the predicted channel dynamics and / or the calculated channel dynamics. The UE is configured to train the machine learning model using a reinforcement learning algorithm and a reward function. The machine learning model is trained to output information about the channel dynamics of the radio channel based on the training information about the predicted channel dynamics and / or the calculated channel dynamics at the machine learning model input. The reward function is based on the deviation between the predicted channel dynamics and the calculated channel dynamics. The UE is configured to provide the machine learning model.

[0065] For example, training input data may be obtained via an interface (e.g., the UE's interface). Training input data may be obtained from a database, a file system, a data structure stored in computer memory, or from the UE's sensors. Training input data includes training information about the predicted channel dynamics and the calculated channel dynamics. The term "training information" may simply indicate that the corresponding data is suitable (e.g., designed for) training a machine learning model. For example, training information may include information about the predicted and calculated channel dynamics, representing the corresponding data to be processed by the machine learning model, for example, to obtain a machine learning model suitable for the task at hand (e.g., suitable for evaluating the predicted channel dynamics for advantage).

[0066] For example, a machine learning model can output information about the predicted channel dynamics provided at the model's input, training information, the deviation between the predicted channel dynamics and the calculated channel dynamics. In other words, the machine learning model can be provided with training input data representing multiple scenarios of predicted and calculated channel dynamics, and is given the task of finding the predicted channel dynamics that have the smallest deviation from the calculated channel dynamics. The machine learning model can be iteratively fine-tuned to iteratively improve the prediction of channel dynamics. To determine which predicted channel dynamics is superior, a reward function can be used. Typically, the reward function can be a function that provides an objective measure of the quality of the predicted channel dynamics. For example, training can be performed with the aim of improving the results of the reward function.

[0067] Typically, a reward function can have many components, such as factors that determine the quality of the predicted channel dynamics. For example, a reward function might be based on the deviation between the predicted channel dynamics and the calculated channel dynamics. For instance, a reward function could reward predicted channel dynamics with small deviations and penalize predicted channel dynamics with larger deviations.

[0068] Reinforcement learning can be used to train machine learning models. As described above, in reinforcement learning, one or more software agents (called "software agents") are trained to take actions in an environment. One or more software agents can assign a favorability rating to the predicted channel dynamics. A reward is calculated based on the actions taken (e.g., the favorability rating) and the resulting deviation between the predicted and calculated channel dynamics. In other words, a reward function can be used to calculate the reward in reinforcement learning-based training of a machine learning model. Reinforcement learning is based on training one or more software agents to select actions that increase the cumulative reward, resulting in software agents that become better on their given tasks (as demonstrated by the increased reward). In the example, the determination of the favorability of the predicted channel dynamics can be repeated. Between repetitions, the machine learning model can be adjusted based on the ratings that have already resulted in high (or low, depending on the implementation) rewards.

[0069] For example, a machine learning model can be trained by repeatedly performing a set of training tasks (or method steps) (e.g., at least twice, at least five times, at least ten times, at least 20 times, at least 50 times, at least 100 times, at least 1000 times). For example, a machine learning model can be trained by repeatedly feeding training input data into the machine learning model, determining the predicted channel dynamics based on the machine learning model's output, evaluating the deviation between the predicted and calculated channel dynamics using a reward function, and adjusting the machine learning model based on the evaluation results.

[0070] One repetition of the above task can be called an "epoch" in reinforcement learning. An epoch is when the entire training input dataset is passed forward and backward through the machine learning model once. Within an epoch, multiple batches of training input data can be fed into the machine learning model to determine the advantage of the predicted channel dynamics and be evaluated. For example, to keep the problem size small, the training data can be subdivided (e.g., "batched") into multiple batches, which can be fed separately to the machine learning model. Each of the multiple batches can be fed separately into the machine learning model.

[0071] Based on training input data, a machine learning model can provide outputs, such as predicted channel dynamics. In reinforcement learning-based training of a machine learning model, one or more software agents (which may act based on the machine learning model) can be assigned a favorable position to the predicted channel dynamics based on a reward function to increase or maximize the reward. In the present case, the predicted channel dynamics may occur based on the output of the machine learning model, for example, not within the machine learning model. Therefore, the reward function can be based on the deviation between the predicted channel dynamics and the calculated channel dynamics, which in turn is based on the output of the machine learning model. This favorable assignment can be evaluated using the reward function. In other words, a software agent can be assigned a favorable position to the predicted channel dynamics.

[0072] As already mentioned, a machine learning model may include one or more embedding layers, which may be based on a non-linear activation function, such as PReLU or the Mish activation function. For example, one or more embedding layers (which may be neural network layers) may be trained as part of reinforcement learning training. Similarly, a machine learning model may include attention layers, which may also be trained as part of reinforcement learning training.

[0073] In one example, method 100 may further include receiving a dataset for training or initializing a machine learning model. For example, the UE may receive the dataset from another UE in its environment, which uses the dataset to predict channel dynamics. The environment of the other UE may be comparable to that of the UE, thereby enabling an additional dataset to provide additional training inputs to the machine learning model.

[0074] In one example, transmission parameter tuning can be performed by reducing channel quality reporting. For instance, the combination of measurements in the CSI report—the Channel Quality Indicator (CQI), the Precoding Matrix Index (PMI), and the Rank Indicator (RI)—can vary. For example, for different CSI reports, depending on the predicted channel dynamics of the radio channel, not all three measurements may be performed; only one or two may be executed. Thus, reporting overhead can be further reduced without reducing the CSI reference signal rate (e.g., pilot signal rate) or otherwise.

[0075] In one example, the UE can be a vehicle. For example, the vehicle may be able to perform method 100. The vehicle can obtain information about the environment by using one or more sensors of the vehicle to determine and / or receive information about the environment. For example, information about the environment can be received from communication devices of a mobile communication system, such as from network entities, vehicles, infrastructure, smartphones, base stations, etc. For example, the one or more sensors can belong to a wide variety of UE sensors, such as radar sensors, lidar sensors, automotive radar, ultrasonic sensors, or vision-related sensors, such as cameras or infrared sensors.

[0076] Information about the environment can be relatively (or entirely) static (e.g., determined by a map that takes into account immobile obstacles (e.g., infrastructure like houses, traffic lights, etc.)) and / or relatively dynamic (e.g., determined by sensor data that takes into account, for example, the speeds of moving obstacles in the environment (e.g., vehicles, pedestrians, cyclists, etc.). For example, the predicted channel dynamics may depend only on the UE's (predicted) speed of movement (e.g., based on a planned route) and static information about the environment (e.g., determined by a map). Thus, the prediction of channel dynamics can be performed with fewer computational resources and no additional sensors are required. This can be advantageous in environments with low obstacle density (e.g., open areas), leading to easier determination of the predicted channel dynamics.

[0077] For example, the predicted channel dynamics may depend on an environment with high-density (moving) obstacles, such as a city center area. Therefore, the prediction of channel dynamics can be improved using measurement data from one or more sensors that reflect a real-time picture of the environment.

[0078] For example, measurement data from one or more sensors can be combined with received information about the environment. For instance, the received information may provide information about immobile obstacles, and the measurement data may provide information about moving obstacles. This simplifies the determination process using one or more sensors (since it may only be necessary to determine moving obstacles) and / or reduces the effort required to use one or more sensors. Optionally or alternatively, data on immobile obstacles can be used to verify the determined immobile obstacles using one or more sensors, or vice versa.

[0079] For example, the UE may be equipped with technology capable of interpreting cooperative sensing data (such as cooperative sensing messages (CPM)) received from other road agents (other vehicles, pedestrians, infrastructure, backend applications), which may be frequently updated periodically or in an event-triggered manner. Furthermore, the UE may be able to fuse data obtained using its sensing devices (one or more sensors) with received data (e.g., received via CPM) to construct an environmental model.

[0080] Further details and aspects are mentioned in conjunction with the embodiments described above and / or below. Figure 1 The examples shown may include those in conjunction with the proposed concepts or one or more of the following examples (e.g., Figures 2-3 One or more optional additional features corresponding to one or more aspects mentioned.

[0081] Figure 2 An example of method 200 for a network entity is shown. Method 200 for a network entity is used to improve transmission efficiency on a radio channel. Method 200 includes receiving 210 to predict an environment model and predicting 220 channel dynamics of the radio channel based on the predicted environment model. Furthermore, method 200 includes receiving 230 a reference signal to measure channel characteristics of the radio channel and calculating 240 channel dynamics of the radio channel based on the reference signal. Method 200 also includes determining 250 the deviation between the predicted channel dynamics and the calculated channel dynamics and adjusting 260 transmission parameters based on the deviation to improve transmission efficiency on the radio channel. The network entity may be a reference... Figure 1 The described UE's counterpart. For example, a network entity might be able to perform actions similar to those referenced. Figure 1 The same method steps as described for the UE.

[0082] Network entities may be able to adjust transmission parameters, for example, based on a predicted environment model received from the UE. Since network entities may not be equipped with sensors, they can rely on information received from another communication device (e.g., the UE). This information can be a predicted environment model, which the network entity can use to predict channel dynamics. Machine learning models, as described above, can be used for this prediction. In this way, the network entity can manipulate computationally intensive machine learning models, which can reduce the UE's energy consumption. Furthermore, the network entity can use machine learning models to predict the channel dynamics of multiple UEs, which can improve the performance of the mobile communication system.

[0083] Alternatively, the network entity may receive only dynamic information about moving obstacles in the environment and determine a predicted environmental model. This allows the network entity to perform the same methods as the UE, in addition to determining information about the environment.

[0084] In one example, the adjustment of the transmission parameters is performed by reducing the report content, report rate, and / or reference signal rate of the user equipment. This reduces signaling overhead.

[0085] In one example, method 200 may further include, if the deviation is higher than a threshold, re-predicting the channel dynamics, redetermining the deviation between the re-predicted channel dynamics and the calculated channel dynamics, and if the redetermined deviation is lower than the threshold, adjusting the transmission parameters; otherwise, repeating the re-prediction and redeterminism until the deviation is lower than the threshold. This allows for the training of machine learning models and / or increases the reliability of transmission parameter adjustments (see, for example, regarding...). Figure 1 (Description of the UE).

[0086] In one example, prediction and / or re-prediction are based on a machine learning model trained with the predicted channel dynamics and the computed channel dynamics. The machine learning model can be used with respect to... Figure 1 The description is the same. In this way, the network entity can perform energy- and / or resource-intensive computational tasks, which may increase the UE's uptime. Furthermore, the network entity can receive multiple datasets from different communication devices connected to it (e.g., other UEs). This improves the training of machine learning models, which in turn improves the prediction of channel dynamics.

[0087] Further details and aspects are mentioned in conjunction with the embodiments described above and / or below. Figure 2 The examples shown may include those in conjunction with the proposed concepts or those above ( Figure 1 ) and / or below (e.g. Figure 3 One or more optional additional features corresponding to one or more aspects mentioned in one or more examples described.

[0088] Figure 3 A block diagram of device 30 is shown. Device 30 includes one or more interfaces 32 configured to communicate with a network entity or user equipment. Device 30 further includes a processing circuitry 34 configured to control one or more interfaces and perform the methods described above for a UE (e.g., see reference 34). Figure 1 (as described) and / or the methods described above for network entities (e.g., references) Figure 2 (Described). For example, device 30 may include a means of transportation. For example, a means of transportation may be a land vehicle, such as a road vehicle, automobile, car, off-road vehicle, motor vehicle, bus, robotaxi, van, truck, or truck. Alternatively, a means of transportation may be any other type of vehicle, such as a train, subway train, boat, or ship. For example, the proposed concept can be applied to public transportation (trains, buses) and future mobility vehicles (e.g., robotaxi).

[0089] For example, device 30 may be a UE, wherein the interface is configured to communicate with a network entity. Alternatively, device 30 may be a network entity, wherein interface 32 is configured to communicate with a UE.

[0090] In one example, the vehicle includes device 30. Therefore, the vehicle may be able to perform reference... Figure 1 The method described in the UE is as follows. For example, a vehicle (e.g., the UE) may use, for example, lidar, radar, cameras, WiFi, Bluetooth, etc., to sense its environment. The UE can then create an environment model based on its perception and may know the antenna settings used, such as directivity, beamwidth, etc. The UE can further use this to estimate CSI, for example, to generate a CSI estimate based on the environment model. This may include discretizing the vehicle's perception with regard to wave propagation characteristics, considering the distance between the transmitter and receiver, considering the speed of both the transmitter and receiver, considering the characteristics of the observed object, considering the antenna settings and Fresnel zone, and in particular objects within that zone and / or combinations thereof. The UE can then estimate the CSI based on pilot signals and may use learning methods (e.g., machine learning) to minimize a cost function by implementing a mechanism. The cost function may include a perception / environment model based on both the CSI estimate and the pilot-based CSI estimate. The learning method for the algorithm that minimizes the cost function at the OEM / vendor / cluster level may use centralized knowledge and joint learning (at the backend), especially distributed learning (at each communication node). For example, the learning method for each UE (at each communication node) could be offline learning of larger pilot sequences and online learning of smaller pilot sequences. Furthermore, the UE can use environment model-based CSI estimation and reduce the overhead associated with accurate pilot-based CSI representation by applying the estimated CSI, thereby reducing signaling (pilots, reference signals). (Refer to...) Figure 2 The network entity described can perform the same method. Alternatively, the network entity (and the UE) can exclude the perception of the environment and may instead receive environmental information from communication devices in the environment, for example.

[0091] like Figure 3 As shown, one or more corresponding interfaces 32 are coupled to a corresponding processing circuitry 34 at device 30. In this example, the processing circuitry 34 may be implemented using one or more processing units, one or more processing devices, any component for processing (such as a processor), a computer, or a programmable hardware component operable with appropriately adapted software. Similarly, the functionality of the processing circuitry 34 may also be implemented in software, which then executes on one or more programmable hardware components. Such hardware components may include general-purpose processors, digital signal processors (DSPs), microcontrollers, etc. The processing circuitry 34 is capable of controlling the interfaces 32 such that any data transfers occurring on the interfaces and / or any interactions that may involve the interfaces can be controlled by the processing circuitry 34.

[0092] In one embodiment, device 30 may include a memory and at least one processing circuitry system 34 operatively coupled to the memory and configured to perform the methods described below.

[0093] In the example, one or more interfaces 32 may correspond to any component used to obtain, receive, transmit, or provide analog or digital signals or information, such as any connector, contact, pin, register, input port, output port, conductor, lane, etc., which allows the provision or receipt of signals or information. The interface may be wireless or wired, and it may be configured to exchange (e.g., transmit or receive signals) information with other internal or external components. One or more interfaces 32 may include additional components used to enable communication between vehicles. Such components may include transceiver (transmitter and / or receiver) components, such as one or more low-noise amplifiers (LNAs), one or more power amplifiers (PAs), one or more duplexers, one or more dual-signalers, one or more filters or filter circuit systems, one or more converters, one or more mixers, correspondingly adapted RF components, etc.

[0094] Further details and aspects are mentioned in conjunction with the above embodiments. Figure 3 The examples shown may include those in conjunction with the proposed concepts or the descriptions above ( Figures 1-2 One or more optional additional features corresponding to one or more aspects mentioned in one or more examples of )

[0095] A particular aspect and feature described in a previous example may also be combined with one or more other examples to replace the same or similar features of the other example, or additionally introduce those features into the other example.

[0096] Examples may conform to or even be included in certain standard specifications, such as those defined by 3GPP. For instance, configuration information may be transmitted using signaling radio bearers, for example, by means of Radio Resource Control (RRC) messages, which are specified as Layer 3 control plane messages in the *.331 series of 3GPP specifications, for example. Physical layer specifications may also be influenced by current embodiments, such as the *.201, *.211, *.212, *.213, *.214, and *.216 series of 3GPP specifications, for example, by means of Doppler delay resolution and other physical layer specifications.

[0097] Examples may further be or involve (computer) programs including program code that, when executed on a computer, processor, or other programmable hardware component, performs one or more of the methods described above. Thus, the steps, operations, or processes of the different methods described above may also be executed by a programmed computer, processor, or other programmable hardware component. Examples may also cover program storage devices, such as digital data storage media, which are machine-readable, processor-readable, or computer-readable and encode and / or contain machine-executable, processor-executable, or computer-executable programs and instructions. For example, program storage devices may include or be digital storage devices, magnetic storage media (such as disks and tapes, hard disk drives), or optically readable digital data storage media. Other examples may include computers, processors, control units, (field-programmable arrays) ((F)PLAs), (field-programmable gate arrays) ((F)PGAs), graphics processing units (GPUs), application-specific integrated circuits (ASICs), integrated circuits (ICs), or systems-on-a-chip (SoCs) programmed to perform the steps of the methods described above.

[0098] It should be further understood that the disclosure of certain steps, processes, operations, or functions in this description or claims should not be construed as implying that these operations necessarily depend on the stated order, unless otherwise expressly stated in the individual case or necessary for technical reasons. Therefore, the foregoing description does not limit the execution of certain steps or functions to a particular order. Furthermore, in other examples, a single step, function, process, or operation may include and / or be decomposed into several sub-steps, sub-functions, sub-processes, or sub-operations.

[0099] If aspects are described relative to an apparatus or system, then those aspects should also be understood as describing the corresponding method. For example, a block, device, or functional aspect of an apparatus or system may correspond to a feature of the corresponding method, such as method steps. Accordingly, aspects described relative to a method should also be understood as describing a corresponding block, element, property, or functional feature of the corresponding apparatus or system.

[0100] If aspects are described relative to an apparatus or system, these aspects should also be understood as describing the corresponding method, and vice versa. For example, a block, device, or functional aspect of an apparatus or system may correspond to a feature of the corresponding method, such as method steps. Accordingly, aspects described relative to a method should also be understood as describing a corresponding block, element, property, or functional feature of the corresponding apparatus or system.

[0101] The appended claims are thus incorporated into the detailed description, wherein each claim may stand alone as a separate example. It should also be noted that while in the claims, a dependent claim indicates a specific combination having one or more other claims, other examples may also include combinations of the subject matter of the dependent claim with the subject matter of any other dependent or independent claim. Such combinations are thus explicitly stated unless stated in individual cases where a particular combination is not anticipated. Furthermore, the features of any other independent claim should also be included, even if that claim is not directly defined as dependent on that other independent claim.

[0102] The aspects and features described relative to a particular example in the previous examples may also be combined with one or more other examples to replace the same or similar features of the other examples, or additionally introduce those features into the other examples.

[0103] List of reference numerals

[0104] 30 devices

[0105] 32-bit interface

[0106] 34 Processing Circuit System

[0107] 100 Methods for User Equipment

[0108] 110 Obtain Predictive Environment Model

[0109] 120 Predicting the channel dynamics of radio channels

[0110] 130 receives reference signal

[0111] 140 Calculate Channel Dynamics

[0112] 150 Determine Deviation

[0113] 160 Adjust transmission parameters

[0114] 200 Methods for Network Entities

[0115] 210 Obtain the Predictive Environment Model

[0116] 220 Predicting the channel dynamics of radio channels

[0117] 230 receives reference signal

[0118] 240 Calculate Channel Dynamics

[0119] 250 Determine Deviation

[0120] 260. Adjust transmission parameters.

Claims

1. A method (100) for user equipment to improve transmission efficiency on a radio channel, comprising: The predicted environment model (110) was obtained; Based on the predicted environment model, predict (120) the channel dynamics of the radio channel; Receive (130) reference signal to measure the channel characteristics of the radio channel; The channel dynamics of the radio channel are calculated (140) based on the reference signal; Determine the deviation between the predicted channel dynamics and the calculated channel dynamics (150); The transmission parameters (160) are adjusted based on the deviation to improve the transmission efficiency on the radio channel; If the deviation exceeds the threshold, the channel dynamics are re-predicted; The deviation between the re-predicted channel dynamics and the calculated channel dynamics is re-determined; as well as If the re-determined deviation is below the threshold, the transmission parameters are adjusted by reducing the reference signal rate and / or reference signal content; otherwise, the re-prediction and re-determination are repeated until the deviation is below the threshold.

2. The method (100) according to claim 1, wherein The re-prediction is based on a new prediction environment model.

3. The method (100) according to claim 2, wherein The prediction and / or re-prediction are based on a machine learning model trained with the predicted channel dynamics and the calculated channel dynamics.

4. The method (100) according to claim 3, further comprising: Receive the dataset used to train or initialize the machine learning model.

5. The method (100) according to any one of claims 1-4, wherein The adjustment of the transmission parameters is performed by reducing the channel quality report.

6. A means of transport capable of performing the method (100) according to any one of the preceding claims, wherein Information about the environment was obtained in the following ways: Use one or more sensors of the vehicle to determine information about the environment; and / or Receive information about the environment.

7. A method (200) for a network entity to improve transmission efficiency on a radio channel used for communicating with user equipment, comprising: Receive (210) predictive environment model; Based on the predicted environment model, predict (220) the channel dynamics of the radio channel; Receive (230) reference signal to measure the channel characteristics of the radio channel; The channel dynamics of the radio channel are calculated (240) based on the reference signal; Determine the deviation between the predicted channel dynamics and the calculated channel dynamics (250); Based on the deviation, the transmission parameters (260) are adjusted to improve the transmission efficiency on the radio channel; If the deviation exceeds the threshold, the channel dynamics are re-predicted; The deviation between the re-predicted channel dynamics and the calculated channel dynamics is re-determined; as well as If the re-determined deviation is below the threshold, the transmission parameters are adjusted by reducing the reference signal rate and / or reference signal content; otherwise, the re-prediction and re-determination are repeated until the deviation is below the threshold.

8. The method (200) according to claim 7, wherein The prediction and / or re-prediction are based on a machine learning model trained with the predicted channel dynamics and the calculated channel dynamics.

9. An apparatus (30), comprising: One or more interfaces configured to communicate with a communication device (32); as well as Processing circuitry (34), the processing circuitry being configured to control the one or more interfaces and: Perform the method according to any one of claims 1-5 and 7-8.

10. A means of transport comprising the device (30) according to claim 9.

11. A computer program product having program code, wherein when the computer program is executed on a computer, processor or programmable hardware component, the program code is used to perform the method (100; 200) according to any one of claims 1-5, 7-8.

Images