Computer-based method for determining the propagation attenuation of a radio frequency signal, computer program and device

Abstract

TITLE: Computer-based method for determining the propagation attenuation of a radio frequency signal, computer program and device. The present invention relates to a method for determining the propagation attenuation of a radio frequency signal and comprising the following steps: loading a digital terrain model, the digital terrain model representing the topography of a terrestrial area; loading a digital surface model of the terrestrial area; generating at least one ray between at least a first point on the terrestrial area and at least a second point, the first point being the position of a transmitter of the radio frequency signal and the second point being the position of a receiver of the radio frequency signal; using each ray to generate respectively a first relief vector from the digital terrain model;use each ray to generate a second vector defining the above-ground characteristic from the digital above-ground model; and calculate, using a trained artificial intelligence algorithm, the propagation attenuation of a radio frequency signal. Figure for the abstract: 2;

Description

Title of the invention: Computer-based method for determining the propagation attenuation of a radio frequency signal, computer program and device

[0001] The present invention relates to a computer-implemented method for determining the propagation attenuation of a radio frequency signal.

[0002] The invention also relates to a computer program comprising software instructions which, when executed by a computer, implement such a process.

[0003] The invention also relates to a data processing system comprising means for implementing such a process.

[0004] Propagation models are known, some recommended by the ITU-R. These are statistical models. They include many input parameters that can be difficult to position. The models referenced by the ITU, 3GPP, and the literature apply only to specific deployment scenarios.

[0005] WO2023214176A1 relates to a method for calculating propagation path loss radio taking into account environmental factors, including buildings, roads, foliage, water surfaces, etc.

[0006] CN110971323B relates to a system for determining the attenuation of a signal.

[0007] Part of the prior art uses systematic approaches (ray tracing, partial differential equation solving) which are extremely costly in terms of computing resources and do not represent the reality of the field because they do not integrate all propagation phenomena.

[0008] Artificial intelligence-based approaches offer perceptron-based solutions (stacking of linear layers).

[0009] The aim of the invention is therefore to propose a method for determining the propagation attenuation of a radio frequency signal with greater quality and to improve coverage predictions.

[0010] To this end, the invention relates to a computer-implemented method for determining the propagation attenuation of a radio frequency signal and comprising the following steps: load a digital terrain model, the digital terrain model representing the topography of a land area; load a digital surface model of the terrestrial zone; generate at least one radius between at least one first point on the land area and at least one second point on the land area, the first point being the position of a radio frequency signal transmitter and the second point being the position of a radio frequency signal receiver; use each radius to generate respectively a first relief vector defining the ground altitude between the first point and the second point from the digital terrain model; use each radius to generate a second vector defining the surface characteristic between the first and second points from the digital surface model; and calculate with a trained artificial intelligence algorithm the propagation attenuation of a radio frequency signal between the first point(s) and the second point(s), the input data of the artificial intelligence algorithm are the first vector(s) and the second vector(s), and the output data of the artificial intelligence algorithm is the propagation attenuation of the radio frequency signal at the second point(s).

[0011] According to other advantageous aspects of the invention, the method comprises one or more of the following features, taken individually or in all technically possible combinations: • each ray is a line of sight between the first point and the second point; • each radius has a bandwidth of the terrain, the bandwidth on the terrain being in particular a width between 5m and 50m; • the first point and / or the second point are on the ground or above the ground; • the first point defines a radio frequency transmitter and / or the second point defines a radio frequency receiver; and / or • The defined characteristic of the overground is coded.

[0012] The invention also relates to a method for generating a cover map comprising the method as defined above.

[0013] The invention also relates to a computer program comprising software instructions which, when executed by a computer, implement a process as defined above.

[0014] The invention also relates to a data processing device comprising means for implementing a process as defined above.

[0015] The invention will become clearer upon reading the following description, given solely by way of non-limiting example, and made with reference to the drawings in which:

[0016] [Fig-1] [Fig.1] schematically shows a system with an emitter, a receiver and a machine learning device;

[0017] [Fig.2] [Fig.2] shows a diagram of a machine learning;

[0018] [Fig.3] [Fig.3] shows a detailed diagram of machine learning using a convolutional neural network;

[0019] [Fig.4] [Fig.4] shows a detailed diagram of machine learning using a seq2seq neural network;

[0020] [Fig.5] [Fig.5] shows a diagram of an embodiment of a method for determining propagation attenuation;

[0021] [Fig. 6] [Fig. 6] shows a data preparation diagram for the artificial intelligence algorithm; and

[0022] [Fig.7] [Fig.7] shows the use of combined vectors for calculating propagation attenuation with the artificial intelligence algorithm.

[0023] Figure 1 schematically shows a system with a transmitter 1, a receiver 3, and a machine learning device 5. Between the transmitter 1 and the receiver 3, the ground surface presents different levels and different aboveground characteristics. Aboveground characteristics include, for example, a body of water (e.g., in the case of a river or lake), a tree, a forest, one or more buildings, etc. When measurement campaigns are carried out, at least one position of a transmitter 1, at least one position of a receiver 3, and the propagation attenuation of a radio frequency signal between the transmitter 1 and the receiver 3 are recorded and fed into the machine learning device 5, for example, via a computer connection.Other characteristics can also be recorded and fed into the machine learning device 5, for example the frequency, the gain of the transmitter and / or receiver antenna 3, the transmitter power.

[0024] The transmitter 1 and / or the receiver 3 may be on the ground or in the air. The position of the transmitter 1 and the position of the receiver 3 contain the geographic position and, in particular, an altitude. For example, the position of the transmitter 1 and the position of the receiver 3 contain geographic coordinates, for example obtained by a satellite positioning system (GNSS), in particular GPS (Global Positioning System), Galileo, Glonass, etc.

[0025] For example the radio frequency signal is between 300 MHz and 5,000 MHz, in particular between 4,000 MHz and 4,500 MHz.

[0026] Fig. 2 shows a diagram of a machine learning system, which is at least partially implemented by the machine learning device 3, which is for example the local server of Fig. 2.

[0027] In step 10, the measured data, in particular a plurality of datasets, are obtained by the machine learning device 3. For example, the dataset is a quadruplet comprising the position of the transmitter 1, the position of the receiver 3, the frequency, and the propagation attenuation of a radio frequency signal. Each dataset may include one or more additional data such as the distance between the position and altitude of transmitter 1 and the position and altitude of receiver 3, the gain of the transmitter antenna, the gain of the receiver 3 antenna and / or the transmitter power.

[0028] For example, during a measurement campaign the transmitter 1 and / or the receiver 3 may change their position and / or altitude.

[0029] In step 20, the machine learning device 3 is adapted to load a digital terrain model, the digital terrain model representing the topography of a land area. Each dataset includes the position of the transmitter 1 and the position of the receiver 3 in that area.

[0030] A digital terrain model is a representation of the topography or altitude of a land area that is readable by a digital calculator or computer.

[0031] In step 20, the machine learning device 3 is adapted to load a digital surface model of the terrestrial area. The digital surface model includes one or more surface features for each point of the terrestrial area.

[0032] In the following, at step 30, for each data set, the process draws a radius between the position of the receiver and the position of the transmitter 1.

[0033] In one embodiment, the radius has a bandwidth of the terrain, the first vector and / or the second vector takes the bandwidth being in particular between 5m and 50m, for example between 10m and 20m. In other words, a slice between the first position of the transmitter 1 and the position of the receiver is taken.

[0034] In step 40, at least one first relief vector defining the ground elevation between the first point (position of transmitter 1) and the second point (position of receiver 3) is defined. For example, the method extracts the list of elevations between the first point and the second point from the digital terrain model, for example using the Bessenham method.

[0035] Also in this step, a second vector defining the surface characteristic between the first point and the second point is defined. For example, in this vector, a number is assigned to each surface characteristic.

[0036] The first vector and / or the second vector take into account the bandwidth of the radius on the ground. For example, to increase the bandwidth, the position of the transmitter and the position of the receiver are incremented in width. In other words, a set of positions over a width of ground between the transmitter and the receiver are taken.

[0037] In step 50 an artificial intelligence algorithm is trained using in particular the first vector and the second vector as input data and the propagation attenuation of a radio frequency signal as output value.

[0038] According to one embodiment, the input data are then concatenated into 2D arrays of size B, K, X, X with B the batch size (i.e. the number of measurements taken), with K the number of possible channels (in the case of two, therefore a first channel with the first vectors defining the topology or the set of ground altitudes between the first point (position of transmitter 1) and the second point (position of receiver 3) and a second channel with the second vectors defining the aboveground characteristic) and X the size of the vectors extracted from the maps.

[0039] In examples, it is also possible to concatenate one or more of the additional data, for example the frequency and / or the distance between the transmitter position and the receiver position 3, for example in the form of an embedding of size X. The input array can thus have a size of B, 4, X, X in the case of two additional data.

[0040] At the end of the learning or training of the artificial intelligence algorithm, an exportable model 60 is obtained.

[0041] Figure 3 shows a detailed diagram of machine learning using a convolutional neural network. In other words, the artificial intelligence algorithm is a convolutional neural network. This artificial intelligence algorithm is used in particular for calculating propagation attenuation between two points, for example between transmitter 1 and receiver 3, and for calculating a coverage map, in particular around transmitter 1.

[0042] In one embodiment, the artificial intelligence algorithm comprises a plurality of convolutional layers 70 designed to create a latent space containing important features. For example, these layers may be convolutional layers, LSTM (Long Short-Term Memory) layers, or GRU (Gated Recurrent Unit) layers. These concatenated layers project the data of size B, 4, X, X into a space of fixed size. In [Fig. 3], in each layer, the first digit indicates the number of inputs and the second digit the number of outputs.

[0043] Following the convolution layers there is a max pooling layer 72 and a plurality of linear neural network layers 74.

[0044] Generally, a neural network comprises an ordered succession of layers of neurons, each of which takes its inputs from the outputs of the previous layer.

[0045] More precisely, each layer comprises neurons taking their inputs from the outputs of the neurons of the previous layer, or from the input variables for the first layer.

[0046] Alternatively, more complex neural network structures can be envisaged with a layer that can be linked to a layer further away than the immediately preceding layer.

[0047] Each neuron is also associated with an operation, that is to say a type of processing, to be carried out by said neuron within the corresponding processing layer.

[0048] Each layer is connected to the other layers by a plurality of synapses. A synaptic weight is associated with each synapse, and each synapse forms a link between two neurons. It is often a real number, which takes on both positive and negative values. In some cases, the synaptic weight is a complex number.

[0049] Each neuron is designed to perform a weighted sum of the value(s) received from the neurons of the preceding layer, each value being multiplied by the respective synaptic weight of each synapse, or connection, between said neuron and the neurons of the preceding layer, and then to apply an activation function, typically a non-linear function, to said weighted sum, and to deliver at the output of said neuron, in particular to the neurons of the next layer connected to it, the value resulting from the application of the activation function. The activation function introduces non-linearity into the processing performed by each neuron. The sigmoid function, the hyperbolic tangent function, and the Heaviside function are examples of activation functions.

[0050] As an optional complement, each neuron is also capable of applying, in addition, a multiplicative factor, also called bias, to the output of the activation function, and the value delivered at the output of said neuron is then the product of the bias value and the value from the activation function.

[0051] A convolutional neural network is also sometimes called a convolutional neural network or by the acronym CNN, which refers to the English term "Convolutional Neural Networks".

[0052] In a convolutional neural network, each neuron in the same layer has exactly the same connection pattern as its neighboring neurons, but at different input positions. The connection pattern is called the convolutional kernel or, more often, the "kernel" in reference to the corresponding English term.

[0053] Figure 4 shows a detailed diagram of machine learning using a seq2seq neural network. This artificial intelligence algorithm is used in particular for calculations of propagation attenuation on a predefined path, for example on a road or railway, especially when the receiver 3 and / or transmitter 1 follows the predefined path.

[0054] This artificial intelligence algorithm uses sequence logic in which vectors of size B, T, K, X, X are used as input data, with K=4, X the size of the first or second vectors, and T the dimension following the sequence of receiver positions 3 (in the case where the transmitter is fixed). Sequences of pairs of positions of transmitter 1 and receiver 3 are used when transmitter 1 is mobile.

[0055] For example, this seq2seq neural network includes decoder-encoder type architectures such as Transformers or Mamba 80 architectures to form a latent space, and we then use these linear layers 82 to predict the propagation attenuation sequence for the rest of the trajectory.

[0056] Figure 5 shows a diagram of an embodiment of a method for determining propagation attenuation. In this embodiment, a coverage map around a transmitter 1 is managed.

[0057] At step 110 a position of a transmitter 1 is obtained.

[0058] In step 120 a digital terrain model, the digital terrain model representing the topography of a land area is loaded and a digital above-ground model of the land area is loaded.

[0059] In the following, at step 130, a plurality of radii between the possible positions of the receiver 3 and the position of the transmitter 1 are generated. The radii are used to extract a vector of the different terrain elevations from the digital terrain model; the extraction of this vector is defined, for example, by the Bessenham method.

[0060] In step 140, for each radius, at least one first relief vector defining the ground elevations between the first point and the second point is generated from the digital terrain model. Also in this step, for each radius, a second vector defining the surface feature between the first point and the second point is generated from the digital surface model. For example, in this vector, a number is assigned to each surface feature.

[0061] The trained artificial intelligence algorithm 150 calculates, in particular by loading a trained model 160 (corresponding to the exportable model 60), to calculate for each ray a value of the propagation attenuation.

[0062] In step 170, a coverage map is generated. This coverage map indicates the propagation attenuation for each position. In this way, the propagation attenuation is calculated for all possible positions within a desired area. For example, the calculated propagation attenuation data is displayed on a map.

[0063] In another embodiment the method calculates the propagation attenuation only between two points, in particular between the transmitter 1 and the receiver 3.

[0064] According to another embodiment the prediction of propagation attenuation is calculated for a trajectory, for example the trajectory of a transmitter and / or the trajectory of a receiver.

[0065] Figure 6 shows a data preparation diagram for machine learning with the artificial intelligence algorithm and for computation with The artificial intelligence algorithm for propagation mitigation. This preparation corresponds, for example, to steps 40 and 140.

[0066] In step 200, a first relief vector defining the ground elevation between the first point and the second point is generated from the digital terrain model. Subsequently, in step 220, the elevation values ​​are digitally coded, if necessary.

[0067] In step 210, a second vector defining the surface feature between the first point and the second point is generated from the digital terrain model. In step 230, each surface feature in this vector is assigned or coded with a number.

[0068] In step 240 the two vectors are combined in the form of an embedding.

[0069] Fig. 7 shows the use of combined vectors for calculating propagation attenuation with the artificial intelligence algorithm.

[0070] In step 240, the combined vectors are used as input data for the artificial intelligence algorithm. If a single propagation attenuation value or a propagation attenuation map is desired, in one embodiment the artificial intelligence algorithm is a CNN or a convolutional neural network. In one embodiment, the numerical propagation attenuation values ​​can be classified into a class, for example, good, medium, or bad.

[0071] If a trajectory of a transmitter or receiver is known, the artificial intelligence algorithm is a seq2seq type network.

[0072] The data processing device, in particular the machine learning device 5, is suitable for implementing a method for determining the propagation attenuation of a radio frequency signal, which will be described above.

[0073] The device is an electronic circuit designed to manipulate and / or transform data represented by electronic or physical quantities in computer registers and / or memories into other similar data corresponding to physical data in register memories or other types of display devices, transmission devices or storage devices.

[0074] As specific examples, the device is implemented in the form of a programmable logic component, such as an FPGA (Field Programmable Gate Array), or an integrated circuit, such as an ASIC (Application-Specific Integrated Circuit).

[0075] Alternatively, when the process is implemented in the form of one or more software programs, i.e., in the form of a computer program, also called a computer program product, it is further capable of being stored on a computer-readable medium, not shown. The computer-readable medium is, for example A readable medium is a device capable of storing electronic instructions and being connected to a computer system bus. Examples of such a medium include optical discs, magneto-optical discs, ROMs, RAM, any type of non-volatile memory (such as FLASH or NVRAM), or magnetic cards. A computer program containing software instructions is then stored on this readable medium.

Claims

Demands

1. A computer-implemented method for determining the propagation attenuation of a radio frequency signal and comprising the following steps: loading (120) a digital terrain model, the digital terrain model representing the topography of a terrestrial area; loading (120) a digital above-ground model of the terrestrial area; generating at least one ray between at least a first point on the terrestrial area and at least a second point on the terrestrial area, the first point being the position of a transmitter (1) of the radio frequency signal and the second point being the position of a receiver (3) of the radio frequency signal; using each ray to generate respectively a first relief vector defining the ground elevation between the first point and the second point from the digital terrain model;use each ray to generate respectively a second vector defining the above-ground characteristic between the first point and the second point from the digital above-ground model; and calculate with a trained artificial intelligence algorithm the propagation attenuation of a radio frequency signal between the first point(s) and the second point(s), the input data of the artificial intelligence algorithm are the first vector(s) and the second vector(s), and the output data of the artificial intelligence algorithm is the propagation attenuation of the radio frequency signal at the second point(s).

2. A method according to any one of the preceding claims, characterized in that each ray is a line of sight between the first point and the second point.

3. A method according to any one of the preceding claims, characterized in that each radius has a strip width of the ground, the strip width on the ground being in particular a width between 5m and 50m.

4. A method according to any one of the preceding claims, characterized in that the first point and / or the second point are on the ground or above the ground.

5. A method according to any one of the preceding claims, characterized in that the first point defines a radio frequency transmitter and / or the second point defines a radio frequency receiver.

6. A method according to any one of the preceding claims, characterized in that the defined overground characteristic is coded.

7. Method for generating a cover map comprising the method according to any one of the preceding claims.

8. A computer program comprising software instructions which, when executed by a computer, implement a method according to any one of the preceding claims.

9. Data processing device comprising means for implementing the method according to any one of claims 1 to 7.

Images