WO2022091456A1

WO2022091456A1 - Method for estimating permittivity of object in radio propagation environment

Info

Publication number: WO2022091456A1
Application number: PCT/JP2021/016385
Authority: WO
Inventors: Viet Hoa NGUYEN; Qianrui LI; Nicolas Gresset
Original assignee: Mitsubishi Electric Corp; Mitsubishi Electric R&D Centre Europe BV Netherlands
Current assignee: Mitsubishi Electric Corp; Mitsubishi Electric R&D Centre Europe BV Netherlands
Priority date: 2020-10-30
Filing date: 2021-04-16
Publication date: 2022-05-05
Anticipated expiration: 2023-04-30

Abstract

A method for estimating permittivity of objects in a radio propagation environment is comprising obtaining geometrical parameters of the radio propagation environment containing at least one object having permittivity to be estimated, obtaining geometrical parameters of rays propagating between a given pair of transmitter and receiver in the radio propagation environment by using the geometrical parameters of the radio propagation environment, generating one set of synthetic permittivity parameters associated with the at least one object, computing channel parameters by means of a geometry-based channel model by using the geometrical parameters of rays and the generated permittivity parameters. In the geometry-based channel model, the channel parameters are in a function of the permittivity and the parameters of rays, obtaining an inverse function of the function by virtue of the synthetic permittivity parameters and the computed channel parameters.

Description

[DESCRIPTION]

[Title of Invention]

METHOD FOR ESTIMATING PERMITTIVITY OF OBJECT IN

RADIO PROPAGATION ENVIRONMENT

[Technical Field]

[0001]

The present invention relates to prediction of radio propagation, especially to method for estimating permittivity of objects in a radio propagation environment.

[Background Art]

[0002]

In wireless systems, prediction of radio propagation allows obtaining the prior knowledge of propagation channel in order to allocate better radio resources, to adapt transmission parameters (modulation, coding schemes, etc), or to precode the transmit signal for better reception et cetera.

[0003]

Ray tracing and ray launching are methods modelling the radio propagation effect by deterministic geometrical approach, based on tracing the electromagnetic rays from a source, i.e. transmitter, to a sink, i.e. receiver. Since the electromagnetic wave propagation is justified by physical interaction between the rays and touched surfaces of the obstacles in a radio propagation environment, a precise channel prediction is at the expense of affined parametrization of the propagation environment. Therefore, a calibration step which involves assigning appropriate parameters to surfaces of the objects in 3D environment model of the radio propagation environment is required. In the art, such calibration is merely based on field measurements. In particular, measurement campaign is executed in the real world counterpart of the 3D environment model so as to obtain the channel statistics such as Root Mean Square Delay Spread (RMS-DS). Among many parameters, surface permittivity values in 3D model are critical for the channel prediction and thus shall be adjusted so as to better reflect the real environment. Therefore, the method in the art usually requires many measurements, and it is also dedicated to the specific given environment configuration and does not work for another environment or in case of a sudden change in the current environment, for example a sudden presence of a new obstacle.

[0004]

To improve the channel prediction, recently, machine learning methods are also considered as a powerful tool for the stochastic channel model calibration, prediction of wireless channel features and acceleration of the high complexity and time consuming ray launching procedure. For example, neural network based channel calibration for a stochastic radio propagation model is proposed. In this case, the model calibration is cast into a regression problem involving mapping of the channel transfer function or impulse response to the model parameters. In addition, in the art, there also exists works suggesting to use the neural network to learn and predict wireless channel features such as path loss and angle of arrival. Some other literature applies machine learning to calibrate the uplink/downlink MIMO channel based on a deep neural network. These works generally exploit neural network to learn the implicit correlation between wireless channel features, for example, the uplink/downlink reciprocity, correlation between Angle of Departure (AoD) and Angle of Arrival (AoA) or spatial correlation of path loss.

[0005]

In addition, in the state of art, learning the propagation channel from a given stochastic channel model is done online. This means that actual measurements or synthetically generation of the channel realization from the given model are used to calibrate the parameters of said model. This is usually done in two steps: In a first step, a large measurement campaign or synthetically generation of the channel realization is performed. In a second step, model parameter is then calibrated and can be used for performance prediction.

[0006]

However, the stochastic based channel model parameter calibration in the art cannot differentiate the contribution of the geometry and the texture of the geometry the influences of which on the channel propagation are jointly estimated for predicting propagation channel. It does not allow the usage of the precise 3D geometry information for the predefined environment, and real-time over-the-air calibration is not possible. It means that if there is a sudden change of the predefined radio propagation environment, such as a factory, for example, in case of a sudden presence of a new object of obstacle, all the previous measurements or synthetically generated channel realizations are not validated anymore for the state of the art method. The measurement campaign or synthetically channel realization generation must be re-performed for the new configuration. All parameters in the parameter set for the stochastic channel model shall be recalibrated with the new training set.

[0007]

In this case, even if the machine learning is used in the art to accelerate the high complexity and time consuming ray launching procedure, the whole raytracing procedure is seen as a black box and will be replaced by a trained neural network. Therefore, real-time over-the-air calibration such as a sudden presence of a new object of obstacle in the radio propagation environment is not possible with the state of the art. For instance, in case of the presence of a new object of obstacle, the measurement campaign or synthetically channel realization generation must be re-performed for the new configuration. All parameters for the neural network shall be adjusted with the new training set.

[0008]

The invention aims to improve these drawbacks, especially to reduce the amount of measurements required to calibrate the channel by virtue of exploiting the geometric 3D environment model.

[Summary of Invention]

[0009]

In this regard, according to one aspect of the invention, it is provided a method for estimating permittivity of objects in a radio propagation environment, the method comprising:

- obtaining geometrical parameters of the radio propagation environment containing at least one object having permittivity to be estimated;

- obtaining geometrical parameters of rays propagating between a given pair of transmitter and receiver in the radio propagation environment by using the geometrical parameters of the radio propagation environment;

- generating one set of synthetic permittivity parameters associated with said at least one object;

- computing channel parameters by means of a geometry-based channel model by using the geometrical parameters of rays and the generated permittivity parameters, wherein, in the geometry-based channel model, the channel parameters are in a function of the permittivity and the parameters of rays;

- obtaining an inverse function of the function by virtue of the synthetic permittivity parameters and the computed channel parameters, wherein, in the inverse function, the permittivity is in the inverse function of the channel parameter; and

- measuring real channel parameters for the given pair of the transmitter and receiver; and

- estimating the permittivity by means of the inverse function by using the measured real channel parameters.

[0010]

With such an arrangement, the present invention proposes to learn the “inverse function” of key parameters, i.e. permittivity, of environment, being related to the propagation channel, first from synthetically generated training data (for a given 3D model, but for any set of associated parameters). In other words, the present invention tackles a more complicated problem of learning an inverse function which is more generic which is the same for many parameters, but has the advantage of using synthetically generated channel realization in order to perform this operation offline. It is then possible to only use the measurements in the learning process for determining the parameters. Alternatively, further measurements are merely used to improve the inverse function.

[0011]

In an embodiment, the step of obtaining an inverse comprises using a machining learning algorithm to obtain the inverse function, wherein the synthetic permittivity parameters and the computed channel parameters are training data for training the machining learning algorithm. In particular, the machining learning algorithm, for example, is neural networks as well as other method which can reflect the behavior of channel when changing the permittivity of objects in the environment (without changing the geometrical characteristic): for example k- nearest neighbors, non-linear regression, etc. With the machine learning algorithm, it is possible to handle an environment comprising a large number of objects with different characteristics in a less time consuming manner.

[0012]

Moreover, for the environment with less complexity, it is also possible to use a look-up table to obtain the inverse function in the step of obtaining an inverse function.

[0013]

Alternatively, the step of obtaining the geometrical parameters of rays comprises using a ray optical modelling algorithm, for example, ray tracing or ray launching. [0014]

Alternatively, the geometrical parameters of rays comprise delay, complex gain, angle of arrival, angle of departure, and/or Doppler shift.

[0015]

Alternatively, the step of obtaining geometrical parameters of the radio propagation environment comprises using geometric measuring instruments, such as LIDAR scanner, depth camera.

[0016]

Alternatively, the channel parameters comprise channel impulse response (CIR), power, delay, signal-to-interference-plus-noise (SINR).

[0017]

Alternatively, the step of generating one set of synthetic permittivity parameters comprises using uniform grid, uniform random, and/or Gaussian random to generate the synthetic permittivity parameters.

[0018]

Alternatively, the transmitter and the receiver are fixed or moving relative to the radio propagation environment, and/or the at least one object is moving relative to the radio propagation environment.

[0019]

Thanks to the abovementioned method of the present invention which is based on the geometric based raytracing model, it generally needs fewer training data compared with the stochastic channel model based method in the art which needs sufficient measurements to achieve statistically significant for the model parameters. In addition, there is an implicit effect of geometry for the predefined environment that contributes to the multi-path channel propagation.

[0020]

Furthermore, the abovementioned method of the present invention can create a digital twin which reflects precisely and instantaneously the radio propagation environment of a predefined environment.

[0021]

In addition, the present invention may also be applied in a situation where a new obstacle object presents in the radio propagation environment for the geometric based raytracing model. Since the texture for the other already existed objects (surfaces) stay unchanged, there is no need to re-calibrate the parameters associated to those objects (surfaces). Only the calibration for the additional parameters set related to the new obstacle object is needed. Therefore, few measurements synthetically generated channel realizations are needed and the calibration time will be shorter.

[0022]

According to another aspect of the invention, it is also provided a method for characterizing materials of objects in a radio propagation environment, comprising:

- obtaining permittivity of the objects in the radio propagation environment using the abovementioned method for estimating permittivity of objects in a radio propagation environment;

- determining the materials of the objects according to the obtained permittivity.

[0023]

Since the learned inverse function in the method for estimating permittivity allows estimating the permittivity of objects in a radio propagation environment, the estimated permittivity can be used to characterize the materials in the environment. This is due to the fact that each material corresponds to a specific value of permittivity. Thus, based on the obtained permittivity, one can deduce the material.

[0024]

According to yet another aspect of the invention, it is further provided a method for predicting channel parameters in a radio propagation environment, the method comprising:

- obtaining geometrical parameters of the radio propagation environment containing at least one object;

- obtaining parameters of propagating between a given pair of transmitter (Tx) and receiver (Rx) in the radio propagation environment by using the geometrical parameters of the radio propagation environment;

- obtaining permittivity of the object in the radio propagation environment using a method according to claim 1 ; and

- computing channel parameters by means of a geometry-based channel model by using the geometrical parameters of rays and the obtained permittivity, wherein, in the geometry-based channel model, the channel parameters are in a function of the permittivity and the parameters of rays.

[0025]

With this method proposed, it allows the channel prediction in a more efficient manner with less measurement and quick calibration.

[0026]

Other features and advantages of the present invention will appear in the description hereinafter, in reference to the appended drawings.

[Brief Description of Drawings]

[0027]

[Fig- 1]

Figure 1 is a flowchart of an exemplary method for estimating permittivity in a radio propagation environment according to the invention.

[Fig. 2]

Figure 2 illustrates a pair of transmitter Tx - receiver Rx in a simple 3D scene, with ray paths generated by ray launching.

[Fig. 3] Figure 3 illustrates a neural network architecture to learn the inverse function of environment in Figure 1.

[Fig- 4]

Figure 4 illustrates permittivities estimated by one embodiment according to the invention.

[Description of Embodiments]

[0028]

Figure 1 shows a flowchart of an exemplary method for estimating permittivity of objects in a radio propagation environment according to the invention, wherein the radio propagation environment comprises at least one given pair of transmitter Tx and receiver Rx, between which electromagnetic rays are transmitted from the transmitter Tx to the receiver Rx. The rays compose channel, and the principle parameters of channel, such as the channel impulse response (CIR), the power, the delay, the signal-to-interference-plus-noise (SINR), etc. can be predicted for carrying out the resource allocation, link adaptation, precoding, etc. in wireless communication. These rays are characterized by important parameters such as delay (traveling time), complex gain (attenuation and phase shift), angle of arrival, angle of departure, Doppler shift. The rays when propagating from transmitter to receiver, encounter several incidents in the environment such as: reflection, refraction, diffraction. The interactions change the direction, the moment, the phase, the polarization state of rays. These changes are affected by the material of obstacles. More specifically, the permittivity of surfaces of the objects in the radio propagation environment decides how the ray behaves after an incident.

[0029]

The present invention uses the geometrical approach to modelling the propagation channel which introduces the approximation of the propagation channel by geometrical rays departing from transmitter and arriving to receiver. There exists several ray optical modelling algorithms which allow approximating the propagation of rays given the 3D model of environment, such as ray tracing, ray launching, etc. In general, these techniques allow tracing each ray from the transmitter for every interaction until it reaches the destination.

[0030]

The geometry-based channel model provides an approximation function of propagation channel in the presence of parameters of 3D model. As discussed above, in the present invention, it proposes to learn an “inverse function” of parameters of radio propagation environment, i.e. permittivity of the objects therein, which is related to the propagation channel of the rays, on the basis of a set of synthetically generated training data.

[0031]

In particular, as shown in Figure 1, the method may comprise the following steps SI to S8.

[0032]

SI: Obtaining a 3D model of the propagation environment, and associating the parameters to at least each face/surface of the objects in the 3D model. For example, if the propagation environment is an office room, the objects can be tables, chairs, walls, etc., in the office room. The geometrical parameters can be the shape, size, locations of the objects in the environment. [0033]

S2: Obtaining geometrical parameters of rays (propagating in the environment), such as delay, complex gain, angle of arrival, angle of departure, and/or Doppler shift of rays, by means of a ray optical modelling algorithm, preferably by using ray tracing or ray launching.

[0034]

S3: Generating at least one set of synthetic training parameters of permittivity associated with each face/surface of the objects in the 3D model. The training parameters can be generated by uniform grid, for example: 1, 1.1, 1.2, 1.3 ... by uniform random, wherein the permittivity is random with equal probability for any value; by Gaussian random, wherein the permittivity is random, however the probability is defined as a Gaussian with a given mean and variance; or by other methods known to the skilled in the art. In addition, this step can be conducted offline, as the training parameters are merely the training data for latter inverse function learning, and thus are not necessary to correspond to a real life situation.

[0035]

S4: Computing estimated channel parameters (such as channel impulse response (CIR), power, delay, signal-to-interference-plus-noise (SINR)), being the output of the approximated function f, from the synthetic training parameters and the geometrical parameters of rays.

[0036]

S5: Learning, from the at least one value being the output of the function f of the synthetic training permittivity parameters and estimated channel parameters, an inverse function f , by means of machine learning or a lookup table. In this step, a learning algorithm can be used to approximate the inverse function by a neural network (or any other method), wherein for each set of training parameters, it is possible to generate channel impulse response from the 3D model, and ray-launcher, for fixed positions of transmitter and receiver, and to feed the neural network with generated channel impulse response, in order to obtain estimated parameters. In addition, it is also possible to train the neural network to reduce the distance measure between the several sets of training parameters and estimated parameters. Furthermore, the obtained or trained neural network is the approximation of the actual inverse function of the original function f, and is regarded as the inverse function f of the function f in the sense of the present invention. [0037]

S6: Measuring at least one value of the propagation channel of for the pair of the transmitter Tx and the receiver Rx, wherein the channel value is, for example, channel impulse response (CIR), power, delay, signal-to-interference- plus-noise (SINR).

[0038]

S7: With the learned inverse function f , evaluating the inverse function whose in wherein the input is related to the channel value being the output of the function f of the synthetic permittivity parameters and the output is related with the parameters to be estimated; and [0039]

S8: Finally, outputting the permittivity parameters of the 3D model. [0040]

In this case, by means of the above steps, it is possible to perform measurements in the “real life” and apply the “inverse function” to the measurements in order to calibrate the propagation model. Therefore, the advantage of such a method is that the “inverse function” is mainly characterized by the geometrical aspect of the problem, and can be learnt offline with “virtual measurements”, i.e. the synthetic parameters. This allows using the measurements for the parameters identification only.

[0041]

With the estimated parameters and the 3D model ray launching, it is then possible to predict the propagation channel for any position of transmitter and receiver. Alternatively, it is also possible to characterize the materials in the environment based on the estimated parameters, because there if a one-to-one correspondence between the permittivity and the material, for example, an estimated permittivity of 1.62 corresponds to gold.

[0042] A more specific and exemplary embodiment implementing the above mentioned method is now described by referring to Figures 2 to 4.

[0043]

In this specific embodiment, the radio propagation environment is simply an empty room composing of 6 facets, i.e. 4 walls, a ceil, and a floor. Assuming that they are made of different materials and thus with different permittivities. The geometrical parameters of the radio propagation environment, i.e. the shape, size of the walls, can be obtained by geometric measuring instruments, such as LIDAR scanner, depth camera.

[0044]

A given pair of transmitter Tx and receiver Rx, Tx-Rx, is arranged in the environment. In this embodiment, they are fixed relative to the environment, whereas they can also be arranged to move relative to the environment.

[0045]

The radio propagation channel between the transmitter Tx and receiver Rx can be modelled in a function f with the following Equation 1 :

[0046] p ray p^th , T_p : delay F* : antenna response, ?* : attenuation, D_p : depolarization matrix i : incident (reflection, refraction, etc), 0* : geometrical property,

p* : permittivity

(Equation 1)

[0047]

Figure 2 shows the pair of transmitter Tx - receiver Rx in a simple 3D scene as mentioned above having ray paths generated by ray optical modelling algorithm, such as ray tracing or ray launching using the measured geometrical parameters of the radio propagation environment. In this embodiment in Figure 2, rays between the transmitter Tx - receiver Rx are generated by ray launching. Accordingly, the geometrical characteristics of every interaction of all rays are obtained. In particular, each ray has several interactions (reflection, refraction, etc) with objects, i.e. walls, in the environment. Ray optical modelling algorithm, for example, NVIDIA OptiX, can trace, i.e. identifying the reflection or refraction or etc, rays for a given 3D environment with the known shape, size, etc of objects in environment, then give the output of geometrical parameters of rays, such as delay (traveling time), complex gain (attenuation and phase shift), angle of arrival, angle of departure, Doppler shift, and etc. These rays are used in Equation 1 to compute the radio channel.

[0048]

In addition to the parameters the rays for a given Tx-Rx as shown in Figure 2, in order to predict the channel by using the function f, i.e. Equation 1, it also needs permittivities. Whereas the permittivities are the parameters to be estimated or obtained. In order to solve such a dilemma, the embodiment according to the invention firstly generates synthetic measurements by:

1) generating ten sets of permittivities for each facet of wall in the environment by uniform grid, uniform random, and/or Gaussian random, as shown in the table 1 below.

[0049]

Table 1 : 10 sets of generated permittivity of six facets

[0050]

2) computing the corresponding channel response parameters, in this embodiment, the power, as shown in table 2 below, thanks to the generated permittivities, the obtained rays, and the above-mentioned Formula 1 , wherein the generated permittivities and the obtained rays are the input of the Formula 1, and the channel response parameters are the output of the Formula 1.

[0051]

Table 2: 10 sets of corresponding channel response (power) for permittivities in Table 1

[0052]

In this sense, the synthetic data are composed of a pair of generated permittivities and corresponding channel response. By this way, it allows to generate the synthetic dataset either on line or off line which represent how the permittivity defines the channel.

[0053]

Afterwards, in this embodiment of the invention, it intends to learn from the dataset an inverse function f in the sense if a channel response is given, how the permittivity can be. In this specific embodiment, it proposes a multilayer perceptron neural network composing of an input layer (7 neurons), 2 hidden layers (10 neurons each), and an output layer (6 neurons), as shown in Figure 3. [0054]

Once the neural network is trained, the value of weights and biases (here 230 weights and 26 biases) are obtained. This network and its associated weights and biases is the learned inverse function f , i.e. the inverse function of Formula 1, wherein the channel response is input, and the permittivity is output. [0055]

Then, real channel response for the transmitter Tx and receiver Rx is measured in online phase. The measured real channel response is fed to the learned inverse function f so as to estimate the permittivities. Figure 4 shows the permittivities estimated in this specific embodiment which are compared with true permittivities.

[0056]

It should be noted that, in addition to the situation where the transmitter and receiver as well as the objects in the environment are arranged in the fixed position, the present invention shall also apply to some other alternative situations where the transmitter, receiver and/or the objects are moving in the environment. [0057]

In a first alternative situation, the transmitter and/or the receiver are moving relative to the environment. In this case, the inverse function evolves when the transmitter and/or receiver is moving, however there exists the correlation between positions of the transmitter and the receiver. The problem of learning the inverse function while the transceiver is moving can be treated by split its trajectory into several pairs of transmitter-receiver positions. Thus, the learning is carried out separately in each pair. The interpolation can be used to deduce the inverse function for the rest of trajectory.

[0058]

On the other hand, as the correlation of inverse function defines the evolution from a position to another, a random process, for example Gaussian process, can be proposed to track this correlation. Regarding the random process, a kernel which characterizes how two input points correlate, should be appropriately proposed, for example power exponential kernel or Matem kernel. This approach offers better catch of inverse function while the transmitter and/or receiver are moving in expense of more complex algorithm.

[0059]

In a second alternative situation, there exist more than one pair of transmitter and receiver. Therefore, there are several positions for transmitter and/or receiver. In this case, the learning of the inverse function is carried out independently in each pair of transmitter-receiver positions, which means several learning process are conducted in parallel.

[0060]

Alternatively, in light of the correlation nature of geometry-based channel model, the learning process of multiple position pairs considers the coherence of inverse function in certain geometrical segments. In other words, the inverse function is being relying not only on parameters of 3D model but also on geographical positions. Therefore, the correlation between terminals that are close to each other is exploited to accelerate the learning of inverse function. [0061]

Alternatively, it is proposed to select the best position of the transmitter and the receiver among several positions to learn the model as fast possible. In particular, after learning the inverse function, it is possible to estimate the speed vs accuracy of channel learning for several transmitter and receiver positions. The selection of best positions which offer the fastest learning and the best accuracy can be performed. [0062]

Preferably, in the same desire of learning faster and more accurate the 3D environment, it is proposed to use a moving device as receiver and then find its best trajectory.

[0063]

In a third alternative situation, the objects in the radio propagation environment are moving, i.e. their positions change, which also includes a situation where some objects are added into or removed from the environment. In this case, the inverse function evolves with this change. However, the function will not completely change since most of objects remains the same. The learning process of inverse function allows to track this difference. In the case of moving object, the permittivity is kept still together with the 3D model is updated. In another scenario, where a new object appears in the 3D scene, one can only learn the new added permittivity.

[0064]

In addition, it should be noted that in exemplary embodiment according to the present invention the generation of training parameters and the compute of approximation function based on the geometry-based channel is done offline. The step of learning the inverse function is also offline. The measurement of true channel is carried out at the receiver, then is fed back to the transmitter or is sent to a third party device. The transmitter afterward estimates the parameters by using offline-learned inverse function and received measurements. The transmitter herein is usually the base station who need the prediction of channel to optimize the wireless communication (resource allocation, precoding, link adaptation, etc).

[0065]

With the estimated permittivities, it is possible to detect the materials of the object in the radio propagation environment on the basis of the estimated permittivities. For example, if the estimated permittivity is 1.62, it can be determined that the associated object in the radio propagation environment is gold. [0066]

Furthermore, it is also possible to predict propagation channel using the estimated permittivities and the geometry-based channel model.

[0067]

In order to predict accurate precise channel parameters using a geometrybased channel model, it is necessary to calibrate the parameters of 3D model with the real scene of the environment. In this sense, every material in the 3D model must be allocated with a permittivity coefficient. The accurate value of model parameters (conductivity, permittivity, etc.) is essential to be fed into the geometry-based channel approximation. Therefore, it is necessary to conduct calibration to adjusting the parameters to approach the output of the propagation channel function f to the real measurement.

[0068]

The calibration process remains specific for each scenario, mainly because the position of calibration in 3D scene is very sensitive for the accuracy. In conventional way, the calibration step is disconnected with the approximation step. Once the parameters are set by the calibration, the approximation function is firmly configured. The validation step involved with this approach becomes more complex since if there is any mismatch of approximation function, the calibration step need to be re-executed with new measurements. Whereas, in the present invention, an inversion of channel approximation function is learnt. This is more generic in the sense that it characterizes the dependence of propagation channel on permittivity of materials in the 3D scene. With the presence of inverse function according to the present invention, the calibration is straightforward by simply introducing the measurement as the input, then obtaining the parameters of environment as output. Moreover, it offers the better dynamism and flexibility since the inverse function can be evolved in one scenario or from one scenario to another on condition that there exists the similarity. [0069]

In light of the above, it is possible to obtain a calibrated channel model with the estimated permittivity in the present invention, so as to predict accurate channel parameters.

[0070]

In addition, in the present invention, by using measurements, the inverse function provides the estimation of parameters of 3D model. Then these parameters is fed to the geometry-based channel approximation to obtain the prediction of propagation channel. This predicted channel can be compared to the measurement to adjust the inverse function. Or in the same scenario, the geometry-based channel model can be updated in order to minimize the difference between true channel and predicted channel. In more complex scheme, both inverse function and prediction function of channel are updated.

[0071]

Moreover, it is known to those skilled in the art, the aforementioned exemplary embodiments according to the present invention can be implemented in many ways, such as program instructions for execution by a processor, as software modules, microcode, as computer program product on computer readable media, as logic circuits, as application specific integrated circuits, as firmware, etc. The embodiments of the invention can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment containing both hardware and software elements. In a preferred embodiment, the invention is implemented in software, which includes but is not limited to firmware, resident software, microcode, etc.

[0072]

Furthermore, the embodiments of the invention can take the form of a computer program product accessible from a computer-usable or computer- readable medium providing program code for use by or in connection with a computer, processing device, or any instruction execution system. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can contain, store, communicate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The medium can be electronic, magnetic, optical, or a semiconductor system (or apparatus or device). Examples of a computer-readable medium include, but are not limited to, a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a RAM, a read-only memory (ROM), a rigid magnetic disk, an optical disk, etc. Current examples of optical disks include compact disk-read-only memory (CD-ROM), compact disk-read/write (CD-R/W) and DVD.

[0073]

The embodiments described hereinabove are illustrations of this invention. Various modifications can be made to them without leaving the scope of the invention which stems from the annexed claims.

Claims

22 [CLAIMS]

[Claim 1]

Method for estimating permittivity of objects in a radio propagation environment, the method comprising:

[Claim 2]

The method according to claim 1 , wherein the step of obtaining an inverse function comprises using a machining learning algorithm to obtain the inverse function, wherein the synthetic permittivity parameters and the computed channel parameters are training data for training the machining learning algorithm.

[Claim 3]

The method according to claim 2, wherein the machining learning algorithm is neural network.

[Claim 4]

The method according to claim 1 , wherein the step of obtaining an inverse function comprises using a look-up table to obtain the inverse function.

[Claim 5]

The method according to any one of claims 1 to 4, wherein the step of obtaining the geometrical parameters of rays comprises using a ray optical modelling algorithm.

[Claim 6]

The method according to claim 5, where in the ray optical modelling algorithm is ray tracing or ray launching.

[Claim 7]

The method according to any one of claims 1 to 6, wherein the geometrical parameters of rays comprise delay, complex gain, angle of arrival, angle of departure, and/or Doppler shift.

[Claim 8]

The method according to any one of claims 1 to 7, wherein the geometrical parameters of the radio propagation environment comprise dimensions and/or shapes.

[Claim 9]

The method according to any one of claims 1 to 8, wherein the step of obtaining geometrical parameters of the radio propagation environment comprises using geometric measuring instruments, such as LIDAR scanner, depth camera.

[Claim 10]

The method according to any one of claims 1 to 9, wherein the channel parameters comprise channel impulse response, power, delay, signal-to- interference-plus-noise.

[Claim 11]

The method according to any one of claims 1 to 10, wherein the step of generating one set of synthetic permittivity parameters comprises using uniform grid, uniform random, and/or Gaussian random to generate the synthetic permittivity parameters.

[Claim 12]

The method according to any one of claims 1 to 11 , wherein the transmitter and the receiver are fixed or moving relative to the radio propagation environment and/or the at least one object is moving relative to the radio propagation environment.

[Claim 13]

The method according to any one of claims 1 to 12, wherein the at least one object is moving relative to the radio propagation environment.

[Claim 14]

Method for characterizing materials of objects in a radio propagation environment, comprising:

- obtaining permittivity of the objects in the radio propagation environment using a method according to claim 1 ;

[Claim 15]

Method for predicting channel parameters in a radio propagation environment, the method comprising:

- obtaining parameters of propagating between a given pair of transmitter and receiver in the radio propagation environment by using the geometrical 25 parameters of the radio propagation environment;