BE1023212B1 - Method and device for measuring the direction of RF sources - Google Patents

Abstract

Method for determining the direction of a radiofrequency source from signals received by an antenna array 111 and by performing a search for a minimum of the standards of predefined vector projections of antenna response models in a vector space generated by NP eigenvectors of a correlation matrix of the received signal, the NP vectors corresponding to the lowest NP eigenvalues, N being the number of antennas and P the number of source (s) transmitting simultaneously (P = 1 for a source ). The calculations are mainly carried out on the processor of a graphics card.

Description

TECHNICAL FIELD [0001] The invention relates to the detection of the direction and / or the location of radio frequency sources.

STATE OF THE ART [0002] Direction finding and / or radiofrequency source location (DOA- direction of arrival) methods are known. These sources are, for example, RFID tags.

[0003] Among these methods, the MUSIC (multiple signal classification) algorithm is one of the most popular. Nevertheless, its real-time resolution requires a lot of computing power. This is why resolution methods in parallel computation have been developed. For example, article 1: "Parallel implementation of the wideband DOA algorithm on single core, multicore, GPU and IBM cell BE processor" by Majid M.W., Schmuland T.E., and Jamali Μ. M. published in Science Journal of Circuits, Systems and Signal Processing, 2013, 2 (2), pp29-36, compares resolution rates on a single-core central processing unit (CPU), on a 4-core CPU, on a GPU graphical processing unit and on an IBM processor. The MUSIC resolution on GPU is the fastest.

Nevertheless, the method used in this article 1 suffers from a significant lack of precision (see for example Article 2: "Direction-of-arrival estimation for closely coupled dipoles using embedded pattern diversity" by Liu Y., Xiong X Chen S., et al., Published in Antennas & Propagation (ISAP), 2013 Proceedings of the International Symposium on, vol.01, pp.467-469). Indeed, the authors make the assumption that the antennas receiving the signals of the radiofrequency sources are decoupled which is not the case in reality. Article 2 shows that the error can reach 2 ° for two antennas. In general, the method used for radiofrequency source localization involves more than two antennas, which further complicates the coupling but increases the accuracy. The resolution of the decoupled problem, although faster, only solves an approximated problem that is simpler than reality, and therefore does not make it possible to precisely locate radio frequency sources in real time, which is problematic. Summary of the Invention [0005] The invention is defined by the independent claims. The dependent claims define preferred embodiments of the invention.

In a first aspect, the invention relates to a method for measuring the direction of a radiofrequency source.

One of the aims of the invention is to reduce the calculation time required to determine the direction of a radio frequency source. This reduction in the calculation time makes it possible to determine the direction of many radio frequency sources in real time.

The method for determining the direction of a radiofrequency source comprises the steps: a. provision of a computer with a graphics card, a radio frequency module comprising at least one antenna array and a controller. The at least one antenna array comprises N antennas, N being strictly greater than one, and the controller is configured for processing the radio frequency signals received by said antennas, converting these radio frequency signals into digital signals. The radio frequency module is connected to the computer via a connection. b. receiving a radiofrequency signal from each of the N antennas of the antenna array, converting N sampled values at the same instant in each of the N radio frequency signals into N digital signals by the controller of the radio frequency module, storing said digital signals in a vector column and transmission of the column vector comprising the N digital signals to the computer. vs. calculating a correlation matrix obtained by multiplying the column vector comprising the N digital signals by the conjugated transposed column vector and spectral decomposition of this correlation matrix to obtain eigenvectors of noise.

The method is characterized in that it further comprises the steps of: d. obtaining predefined vectors of antenna response models. e. predefined vector projections of antenna response patterns respectively in the vector space generated by the noise vectors and calculation of the norms of these projections. These calculations are performed by the processor of the graphics card. f. searching for an extremum standard, preferably a minimum, and the predefined vector corresponding to this standard and obtaining the direction of the radiofrequency source corresponding to the direction of the predefined vector.

The calculation of the direction of a radiofrequency source and made shorter thanks to the use of the graphics card. The architecture of the processor of the graphics card makes it possible to distribute the computations (parallel computations) which greatly reduces the computation time, many operations being performed simultaneously rather than sequentially.

[0010] Preferably, the method comprises the conversion of N * M, M being strictly greater than one, values sampled at M instants in each of the N radio frequency signals received by the antennas, in N * M digital signals by the controller of the radiofrequency module. These N * M digital signals are then stored in M column vectors. These vectors are transmitted to the computer and a correlation matrix is calculated as the arithmetic mean of the instantaneous correlation matrices obtained by multiplying each of the M column vectors comprising the N digital signals by each of the M conjugated transposed column vectors. The average of several signals reduces the effects of noise and improves the signal.

The predefined vectors of antennas response models are important since they make it possible to find the direction of a radiofrequency source. They can be obtained by discretizing a unitary half-sphere into a grid of points according to two angles corresponding to the direction of a unit vector coming from the center of the sphere and pointing towards one of the points of the grid. The discretization of this space makes it possible to define the precision of the search. Then, for each pair of angles, the space of the polarizations of the incoming wave is discretized into a discrete set of points according to two complex values. The discretization of the space of the polarizations makes it possible to make reliable the computation vis-à-vis the multiple paths ("multipaths"). Finally, for each pair of angles and polarizations, the frequency response of the antenna array is simulated and / or measured. This simulation and / or measurement is stored in the searched vectors.

Preferably, the calculations carried out in parallel on the processor of the graphics card for determining the direction of a radiofrequency source are made by copying the predefined vectors of response models of the antennas and eigenvectors of noise in a global memory. of the graphics card. Then, separate subsets of predefined vectors of antenna response patterns and noise vectors are copied to a local memory of each computing unit. Each computing unit then calculates the projection standards of the predefined antennas response model vectors belonging to the subset copied to its local memory and projected into the vector space generated by the noise vectors. The results can be copied to the local memory and an extremum, preferably local minimum, can be searched. Then the standards are stored in the global memory. Calculations can be repeated. The global memory copy of the predefined antennas response model vectors is generally performed only once.

[0013] Preferably, the search for an extremum, preferably an overall minimum, is performed in parallel calculation on the processor units of the graphics card by copying subsets of extremums, preferably minimums, stored in memory. global memory in the local memory of each computing unit, looking for the extremum, preferably minimum, stored in each local memory and copying these results into global memory. These calculations are repeated until the extremum, preferably minimum, global.

Preferably, the conversion of the sampled values in each of the radio frequency signals received by the antennas into digital signals is performed by an analog / digital converter included in the radiofrequency module.

Preferably, the calculation of a correlation matrix of the digital signal is performed by a programmable logic circuit included in the radiofrequency module.

[0016] Preferably, the controller is a microcontroller that transmits the correlation matrix to the computer via a connection. These last two features make it possible to speed up the search for the direction of a radiofrequency source and to limit the data rate on the connection.

Preferably, the calculations are performed only for a subset of predefined vectors of antennas response models, said subset being chosen from the directions close to a last known direction of the radiofrequency source and taking into account maximum speed of the source. This maximum speed can be predefined or evaluated statistically based on the history of the last known directions.

[0018] Preferably, the predefined antennas response model vectors {Hk (n, τ, ε)) are normalized. This reduces the effects of noise and minimizes the impact on the search for the minimum standard.

Preferably, the radio frequency sources are RFID tags, wifi, bluetooth, zigbee, IEEE 802.15, IEEE 802.11, IEEE 802.16, IEEE 802.20, or IEEE 802.22. Preferably, the tags use the polarization diversity broadcast.

Preferably, obtaining the direction of a radiofrequency source is obtained in a time interval of between 100 ns and 10 ms, preferably between 250 ns and 1 ms, even more preferably between 400 ns and 600 ns for a precision between 1 ° and 3 °.

Preferably, the calculations performed by the processor of the graphics card may be performed by a programmable logic circuit having a plurality of calculation units.

In a second aspect, the invention consists of a method for locating a radiofrequency source. This method is based on the method of finding the direction of a radiofrequency source. It comprises obtaining a first direction of a radiofrequency source with respect to a first antenna array, obtaining a second direction of the same radiofrequency source with respect to a second antenna array. Once the two directions are known, we search for the intersection of the lines defined by the first and second directions with respect to the distance and the relative orientation between the radiofrequency modules that are known. Finally, we locate the position of the radiofrequency source with respect to the first and second antenna arrays by trigonometry.

According to a third aspect, the invention relates to an apparatus 100 for finding the direction of at least one radiofrequency source. This apparatus 100 comprises at least one radiofrequency module, a controller. The radio frequency module comprises an antenna array comprising a plurality of antennas and the controller is configured for processing the radio frequency signals and converting the radio frequency signals into digital signals. The apparatus 100 also includes a computer with a graphics card, connections for connecting the at least one radio frequency module to the computer, and implements one of the methods described above.

Preferably, the antenna array of the apparatus 100 comprises between 2 and 24 antennas, preferably between 4 and 16 antennas, even more preferably the network comprises 8 antennas.

[0025] Preferably, the controller of the at least one radio frequency module comprises a programmable logic circuit.

Brief Description of the Drawings [0026] These aspects as well as other aspects of the invention will be clarified in the detailed description of particular embodiments of the invention, with reference to the drawings of the figures, in which: FIG. 1 shows the components necessary for the implementation of the method for measuring the direction and / or location of a radio frequency source according to the invention and an apparatus according to the invention in connection with a radiofrequency source; FIG. 2 shows a diagram of a method for determining the direction of an RF source according to the invention; FIG. 2a shows a diagram of an exemplary method for simulating predefined vectors of antenna response models; FIG. 3a shows an example of the architecture of the processor of a graphics card; FIG. 3b shows a diagram of an example of processor calculation of a graphics card for determining the direction of an RF source; FIG. 4 shows a diagram of an exemplary method for locating a radiofrequency source.

The drawings of the figures are not to scale. Generally, similar elements are denoted by similar references in the figures. The presence of reference numbers in the drawings can not be considered as limiting, even when these numbers are indicated in the claims.

DETAILED DESCRIPTION OF PARTICULAR EMBODIMENTS OF THE INVENTION [0027] FIG. 1 shows an example of the components necessary for the implementation of the method 200 for measuring the direction and / or location of a radio frequency source 101 (RF). according to the invention an apparatus 100 according to the invention. These components comprise: at least one RF source 101, at least one RF module 110, connections 102, a computer 120 provided with a graphics card 130. The RF module 110 comprises: an antenna array 111 comprising N antennas 112, N being strictly greater than one, a controller 113 configured for processing the RF signals received by said antennas 112 and for converting these RF (analog) signals into digital signals.

Preferably, the source or sources RF 101 are sources of RFID tags, wifi, bluetooth, zigbee, IEEE 802.11, IEEE 802.15, IEEE 802.16, IEEE 802.20, or IEEE 802.22. For example, the sources may be Texas Instruments cc2530 tag, operating at 2.4 GHz with a bandwidth of 1 MHz. For this type of tag, the signal will be brought back to baseband, that is to say that the 1 MHz band centered at 2.4 GHz will be recentered around 0.

Preferably, the RF module 110 comprises at least one analog / digital ADC converter capable of sampling the analog signal to transform it into a digital signal. Preferably, the RF module 110 also includes a (micro) controller 113 for processing the digital signal and sending it to the computer 120 via a connection. Preferably, the RF module 110 also comprises an FPGA programmable logic circuit programmed to perform certain steps of the method according to the invention. For locating an RF source 101, at least 2 separate antenna arrays 111 each comprising N antennas 112 may be used. Preferably, each antenna array 111 comprises between 2 and 24 antennas 112, or, preferably, between 4 and 16 antennas 112 or, ideally 8 antennas 112. Preferably, these antennas 112 are positioned equidistantly angularly on a disk, for example radially.

Preferably, the RF module 110 and the computer 120 are interconnected by connections 102 with or without wires, preferably these connections 102 are ethernet or wifi network connections. Alternatively, any connection 102 may be used, for example: usb, serial, parallel, coaxial, telephone.

Preferably, the computer 120 is a PC x86-64 computer or equivalent equipped with a graphics card 130. Preferably, the graphics card 130 comprises a dedicated RAM memory of at least one GB and a GPU whose frequency is at least 500 MHz. Preferably, the GPU comprises at least 4 calculation units, for example 4 multiprocessor streaming for an Nvidia card. For example, the NVIDIA® Maxwell ™ GPU series can be used.

The computer 120 may optionally be connected to the internet and provide remote access via a local interface, web or via the cloud.

Alternatively to the computer 120, a programmable logic circuit comprising a plurality of computing units and preferably having an architecture comparable to that of a GPU, could be used. Preferably, the number of calculation units of said circuit should be greater than 4.

Figure 2 shows a diagram of the steps of a method 200 for determining the direction of an RF source 101 according to the invention. The first steps, S201 to S203, consist in the provision of the physical components described above and necessary for the implementation of the method. Step S201 comprises providing a computer 120 equipped with a graphics card 130. Step S202 comprises providing an RF module 110 comprising at least one antenna array 111 each comprising N antennas 112, N being strictly greater than one, a controller 113 configured for: processing the RF signals received by said antennas 112, converting these RF signals into digital signals. Step S203 includes connecting the RF module 110 to the computer 120 via a connection 102.

Step S210 consists of receiving an RF signal from the RF source 101 and, possibly tainted noise. Since an RF signal is, by definition, a function varying in time, this signal is considered at a time t. For example, the noise may be due to noise from other sources, noise due to reflections and diffractions of the initial signal on different surfaces, interference with the atmosphere, RF noise picked up by antennas 112 and thermal noise due to electronic circuits. The RF signal is picked up by each of the N antennas 112 of the antenna array 111 of the RF module 110. Thus, N RF signals (one per antenna), γ t), are obtained.

Step S211 comprises sampling the N RF signals at the same time t, to obtain N values. These N values are then converted into N digital signals, yit

These signals are then stored in a column vector y whose components are yt,

[0037] Preferably, each antenna 112 receives a signal. The N signals received by the antennas 112 are sampled simultaneously at M different times, M being strictly greater than one. We thus obtain N * M values and we convert these N * M values into N * M digital signals. These digital signals are then stored in M column vectors ys,

This conversion can be performed by the controller 113 of the RF module 110 or, alternatively, by one or more ADC analog / digital converters.

Step S212 comprises the transmission of the column vector (s) y ouys,

to the computer 120 via a connection. For example, this transmission is provided by the (micro) controller 113 of the RF module 110.

Step S220 comprises calculating a correlation matrix C. This matrix is calculated by multiplying a column vector y by the vector yh, complex conjugate transposed of the column vector y.

Preferably, M instantaneous correlation matrices are calculated by multiplying M columns respectively.

by the M vectors

corresponding to the complex conjugates transposed of the column vectors ys. The correlation matrix C being equal to the arithmetic average of the M instantaneous correlation matrices:

The calculation of the correlation matrix C can be carried out by the central processor of the computer 120 (CPU), or by the processor of the graphics card 130 (GPU) or before transmission to the computer 120, by a logic circuit programmable FPGA or by a (micro) controller of the RF module 110.

Step S221 comprises the spectral decomposition of the correlation matrix C into its eigenvectors vo0 <i <N. This decomposition comprises the search for eigenvalues and eigenvectors of the correlation matrix C. Preferably, this step is performed on the CPU of the computer 120. Alternatively, this calculation can be performed on the GPU of the graphics card 130, or on a programmable logic circuit comprising a plurality of calculation units. The spectral decomposition can be performed via a Householder matrix transformation, then via the use of one or more eigenvector resolution algorithms such as for example the QR algorithm, or the Hessenberg solver. In the case where the RF sources emit successively, the eigenvectors vt corresponding to the N-1 eigenvalues, with N the number of antennas are called eigenvectors of noise and generate a vector space called noise vector space. Preferably, there are N-P noise vectors corresponding to the N-P lowest eigenvalues with N the number of antennas and P the number of sources (s) emitting simultaneously.

Step S230 includes obtaining predefined vectors of antenna response patterns. A vector

comprises the transfer functions in the frequency domain of the antennas 112 for an arrival direction n and a polarization of the source wave (τ, e), data. The direction is determined by two angles a, β and the polarization by two complex values (τ, ε). All of these predefined vectors

are, for example, simulated and / or measured beforehand. Figure 2a shows a diagram of an exemplary method 240 for simulating predefined vectors of Hk antenna response patterns. This method includes: a. S241: the discretization of a unitary half sphere into a point grid, each point being defined by two angles α, β. The center of the sphere corresponding to the center of the antenna array 111; b. S242: in each of the points of the grid, the two-dimensional space of the polarizations (τ, e) is discretized. We therefore have a 4-dimensional parameter space; vs. S243: in each of the points of this parameter space, the frequency response of the antenna array 111 is simulated; d. S244: These responses are stored in predefined antennas response pattern vectors

where k is the number of points in the parameter space.

For example, for a precision of 3 °, the two angles α, β can be discretized according to 60 * 120, and for 4 * 4 polarization values per pair of angles, 115200 predefined vectors of antennas response models are obtained.

For a precision of 1 ° and 8 * 8 polarization values per angle pair, we obtain 4147200 predefined vectors of antennas response models

Alternatively or additionally, it is possible to measure the impulse response of the antenna array 111 for known quadruplets (α, β, τ, ε). These measurements can, for example, make it possible to refine the results of the simulations. Preferably, the predefined response model vectors of the Hk antennas are normalized to limit the effects of digital errors.

A property of the method used for the invention is that the predefined vector of antenna response models Hk corresponding to the direction of the RF source 101 is orthogonal to the space generated by the noise vectors vt of the matrix. C. The step S231 therefore comprises the projection of the predefined vectors of antenna response models

in the space generated by the noise vectors

of the correlation matrix C. In other words, we calculate the Hermitian scalar products

for each of the predefined response model vectors Hk antennas. These calculations are performed by the GPU of the graphics card 130.

Alternatively, the calculation of Hermitian scalar products Hk.v £ can be performed for a subset of predefined vectors of Hk antenna response models. This subset of predefined vectors of antenna response models Hk is chosen according to the last known direction and comprises predefined vectors of response models of antennas Hk close to this last known direction. For example, one can choose to take the predefined response model vectors Hk antennas corresponding to an angular region of 45 ° to 45 ° around the last known position, or preferably 20 ° to 20 ° or 10 ° on 10 °. It is also possible to take into account a maximum speed of movement of the RF source. This maximum speed can be defined on the basis of the last known directions or, alternatively, it can be evaluated statistically on the basis of the history of the last known directions or with respect to the average speed of the RF sources observed. Based on this maximum speed, a larger angular region is defined for a larger speed and a smaller angular region for a smaller speed.

Step S232 comprises calculating the standard of the projections calculated in step S231. These calculations are performed by the GPU of the graphics card 130.

Step S233 includes the search for a standard extremum, preferably minimum. In theory, the image vector of the projection of the predefined vector of antenna response patterns of the desired direction in the noise vector space is zero, or, equivalently, this vector is orthogonal to each eigenvector of noise. Nevertheless, because of the numerical inaccuracy of floating-point calculations and the inaccuracy of measurements and / or simulations, these Hermitian scalar products are not perfectly equal to 0 and therefore the minimum standard must be sought. In summary, one can write the search for this minimum as:

It should be noted that the previous equation represents the minimum of the standard squared and not the minimum of the norm. However, since the square root is a monotonic function, it is equivalent to search for the minimum in this way.

The step S234 comprises obtaining the direction of the RF source 101. Once the minimum found, we find the predefined vector of antenna response models having led to this minimum and thus we obtain the direction of the RF source 101 which corresponds to the quadruplets of parameters (α, β, τ, έ) of said predefined vector of antenna response models.

Figure 3a shows an example of the architecture of a graphics card 130 and its GPU processor. The card comprises a global memory 331, communication buses 332 and the GPU comprising a plurality of calculation units 340. Each calculation unit 340 comprises a local memory 341 and a plurality of arithmetic and logical units (ALU) and a a plurality of floating point calculating units (FPUs) and a plurality of registers 342.

FIG. 3b shows a diagram of the method for calculating the determination of the direction of an RF source 101 on the GPU 310 of the graphics card 130. This method is carried out in parallel computation on the units of calculation of the GPU. of the graphics card 130.

The first step S311 comprises the copying of the predefined vectors of antenna response patterns.

in the global memory of the graphics card 130. This operation is rather slow and can be performed once at initialization. Preferably, the predefined vectors of antenna response patterns

are partitioned. That is, the memory space they occupy is divided into sub-blocks of identical size comprising at least one predefined vector of antenna response patterns.

The second step S312 comprises copying the eigenvectors of noise vt in the global memory of the graphics card 130. Access to the global memory of the graphics card 130 is generally slow. The noise own vectors vt being few (their size depends on the number of antennas 112 of the antenna array 111, for example 8, alternately, their number depends on the number of antennas minus the number of sources (s) emitting simultaneously), the amount of information to be transferred is nevertheless small.

The third step S313 comprises the copying of distinct subsets of predefined vectors of antenna response patterns.

respectively in the local memory of each computing unit. Preferably, the copied subsets correspond to sub-blocks of identical size comprising at least a predefined vector of antenna response patterns created during partitioning. Preferably, the number of blocks corresponds to a multiple of the number of calculation units.

The fourth step S314 comprises copying the eigenvectors of noise vt into the local memory of each computing unit. These vectors are therefore present in the local memory of all the calculation units.

The fifth step S320 comprises calculating the projection norms of each of the predefined vectors of antenna response patterns in the noise vector space. Projections including the calculation of Hermitian scalar products

Each computing unit of the GPU calculates the standards corresponding to the sub-block of predefined vectors of antenna response models

contained in his local memory.

The sixth step S321 is optional and includes storage in the local memory standards projections. Preferably, a local extremum, preferably a local minimum, is sought in the projection standards stored in local memory. This step makes it possible to avoid copying all the standards from the local memories of the calculation units into the global memory of the graphics card 130. The search for this minimum can be carried out for example, using a sorting algorithm. . Preferably, a "divide and conquer" method will be used in which a set of standards is divided into two equal parts. We then compare the values having the same index of the two sets and we retain only the minimums. This leads to a set that is twice as small, to which the algorithm is reapplied until only one value remains. Note that if the set is odd, we add an infinite value so that its length is even. Alternatively, this search for minimum can be performed by the CPU of the computer 120.

The seventh step S322 comprises the copy of the standards calculated by each calculation unit from the local memories of the calculation units in the global memory of the graphics card 130. In the case where the step S321 is carried out, only the standards minimums are copied.

The eighth step S323 includes the search for the global minimum standard. This search can be performed by the central CPU of the computer 120 or directly on the GPU of the graphics card 130. The latter possibility reduces the number of memory exchanges and speeds up the procedure. This step S323 comprises the copying of subset (s) of extremums, preferably minimums, from the global memory of the graphics card 130 to the local memories of the calculation units, or alternatively of a part of the calculation units. Each unit then searches for the minimum in its subset and copies this value to global memory. The procedure continues until an overall extremum is obtained, preferably a minimum.

[0057] Calculations of hermitian scalar products

performed during the projection of the predefined antennas response model vectors in the noise vector space, for each of the predefined vector models of response of the antennas and their standard by the GPU of the graphics card 130, allow a saving of time calculation that depends on the precision sought. In fact, the larger the number of predefined antennas response model vectors, the greater the accuracy (the parameter space being more discretized) and the greater the computing time. For example, for a precision between 1 ° and 3 °, the calculation time is between 100 ns and 10 ms, preferably between 250 ns and 1 ms, even more preferably between 400 ns and 600 ns. In comparison, the computing time on the CPU of the computer 120 is for the same accuracy is of the order of 30 ms.

According to a second aspect, the invention relates to a method 400 for locating an RF source 101. This method comprises the steps: S410, obtaining a first direction of an RF source 101 relative to a first network of antennas 111 according to the method explained above, S420, obtaining a second direction of the same RF source 101 with respect to a second antenna array 111 according to the method explained above, S430, calculation of the intersection of the lines defined by the first and second directions and S440, location of the source relative to the first and second antenna arrays 111. The distance and the relative orientation between the two networks 111 are known. Preferably, the lines defined by the first and second directions are left and their intersection is defined as the point located midway from the minimum distance between the two lines. By knowing the distance and the relative orientation between the two antenna arrays 111 and the directions of the two lines, it is possible to know, by trigonometry, the position of the RF source 101.

According to a third aspect, the invention relates to an apparatus 100 for finding the direction of at least one RF source 101 comprising: at least one RF module 110 comprising: an antenna array 111 comprising a plurality of antennas 112, a controller 113 configured for: processing RF signals, converting RF signals into digital signals; a computer 120 provided with a graphics card 130; connections 102 for connecting the at least one RF module 110 to the computer 120; said apparatus 100 being configured to execute the method for finding the direction of at least one RF source 101 described above. Preferably, the apparatus 100 will comprise two antenna arrays 111 for finding the location of at least one RF source 101. Preferably, the antenna array (s) 111 comprise between 2 and 24 antennas 112, preferably between 4 and 16 antennas 112, even more preferably the network comprises 8 antennas 112. Preferably, the controller 113 of the at least one RF module 110 comprises a programmable logic circuit and / or a microcontroller.

The present invention has been described in relation to specific embodiments, which have a purely illustrative value and should not be considered as limiting. In general, the present invention is not limited to the examples illustrated and / or described above. In particular, the invention also relates to combinations of the technical characteristics of the embodiments described above. The use of the verbs "to understand", "to include", "to include", or any other variant, as well as their conjugations, can in no way exclude the presence of elements other than those mentioned. The use of the indefinite article "a", "an", or the definite article "the", "the" or "I", to introduce an element does not exclude the presence of a plurality of these elements. The reference numerals in the claims do not limit their scope.

In summary, the invention can also be described as follows: Method for determining the direction of a radiofrequency source from signals received by an antenna array 111 and performing a search for a minimum of the standards of the predefined vector projections of antenna response models in a vector space generated by NP eigenvectors of a correlation matrix of the received signal, the NP vectors corresponding to NP the lowest eigenvalues, N being the number of antennas and P the number of sources (s) emitting simultaneously (P = 1 for a source). The calculations being mainly made on the processor of a graphics card.

Claims (14)

claims A method (200) for determining the direction of a radio frequency source (101) comprising the steps of: a. providing (S201) a computer (120) with a graphics card (130); b. providing (S202) a radio frequency module (110) comprising: at least one antenna array (111) each comprising N antennas (112), N being strictly greater than one, a controller (113) configured to : processing the radio frequency signals received by said antennas (112), converting these radio frequency signals into digital signals; vs. connecting (S203) the radio frequency module (110) to the computer (120) via a connection (102); d. receiving (S210) a radio frequency signal from each of the N antennas (112) of the antenna array (111); e. converting (S211) N sampled values at the same instant in each of the N radio frequency signals received by the antennas (112) into N digital signals by the controller (113) of the radio frequency module (110) and storing said digital signals in a vector column (y); f. transmitting (S212) the column vector (y) comprising the N digital signals to the computer (120); boy Wut. calculating (S220) a correlation matrix (C) by multiplying the column vector (y) comprising the N digital signals with the conjugated transposed column vector (yh); h. spectral decomposition (S221) of the correlation matrix (C) to obtain eigenvectors of noise {vt); characterized in that the method further comprises the steps of: i. obtaining vectors (S230) predefined antennas response models (H (nk, rk, ek)); j. projections (S231) of predefined vectors of antenna response models (H (nk, Tk, ek)) respectively in the vector space generated by the noise vectors (^), these projections being made by the graphics card processor ; k. calculating the standards (S232) of the results of the projections obtained in step i, this calculation being carried out by the processor of the graphics card (GPU); l. searching for an extremum standard (S233), preferably minimum, and the predefined vector (// (nfc, Tfc, efc)) corresponding to this standard; m. obtaining the direction (S234) of the radiofrequency source corresponding to the direction of the predefined vector found in step I. The method for determining the direction of a radiofrequency source (101) according to claim 1, wherein steps e to g comprise: a. conversion of N * M, M being strictly greater than one, values sampled at M moments in each of the N radio frequency signals received by the antennas (112), in N * M digital signals by the controller (113) of the radiofrequency module and storing said digital signals in M column vectors (ys); b. transmitting the M column vectors (ys) each comprising N digital signals to the computer; vs. calculating instant correlation matrices by multiplying each of M column vectors (ys) comprising the N digital signals by each of the M conjugated transposed column vectors (ysh), and calculating the correlation matrix (C) as the arithmetic mean of the matrices instantaneous correlations. A method (240) for determining the direction of a radio frequency source (101) according to any one of the preceding claims, wherein the predefined vectors of antenna response patterns (H (nk, Tk, ek)) are obtained by realizing the following steps: a. discretizing (S241) a unit half-sphere into a grid of points according to two angles (α, β) corresponding to the direction of a unit vector from the center of the sphere and pointing to one of the points of the grid; b. discretizing (S242) the polarization space of the arrival wave into a discrete set of points according to two complex values (τ, e) and for each of the pairs of angles {α, β) of a direction; vs. simulation and / or measurement (S243) of the frequency response of the antenna array (111) for each pair of angles (α, β) and of polarizations (τ, e); d. obtaining and storing (S244) the searched vectors {H (nk, zk, ek)). A method (310) for determining the direction of a radiofrequency source (101) according to any one of the preceding claims, wherein steps j and k of claim 1 are performed in parallel computation on the computing units (340) of FIG. Graphics Card Processor (GPU) by performing the steps of: a. copying (S311) predefined vectors of antenna response patterns {H (nk, rk, ek)) into a global memory (331) of the graphics card (130); b. copying (S312) eigenvectors of noise (vt) into the global memory (331) of the graphics card (130); vs. copying (S313) distinct subsets of predefined vectors of antenna response patterns (H (nk, rk, ek)) respectively into a local memory (341) of each computing unit (340); d. copying (S314) noise vectors (vt) into the local memory (341) of each computing unit (340); e. calculating (S320), by each computing unit (340), prediction standards of the predefined antennas response pattern vectors (H (nk, rk, ek)) belonging to the corresponding subset defined in step c in the vector space generated by the noise vectors (vt); f. storing (S321) the result of step e in the local memory (341) and searching, by the computing unit (340), for a standard extremum, preferably minimum, among the results obtained in step e ; boy Wut. storing (S322) standards in the global memory (331); steps b to g being repeatable, step f being optional, and in which step I of claim 1 is optionally performed in parallel computation on the computing units (340) of the graphics card processor (GPU) by performing the steps of: h. subset copy of extremums, preferably minimums, stored in global memory in step g in the local memory of each computing unit (340); i. searching for the minimum preferably extremum, stored in each local memory (341) and copying these results in global memory (331); steps h and i being repeated until the extremum, preferably minimum, global (S323). A method for determining the direction of a radio frequency source (101) according to any one of the preceding claims, wherein the controller (113) of the radio frequency module (110) comprises a programmable logic circuit, a microcontroller and at least one analog converter. / digital and wherein steps e to g of claim 1 consist of: a. converting sampled values in each of the radio frequency signals received by the antennas (112) into digital signals by the at least one analog / digital converter; b. calculating a correlation matrix (C) of the digital signals by the programmable logic circuit; vs. transmission, by the microcontroller, of the correlation matrix (C) to the computer via a connection (102). The method according to any one of the preceding claims, wherein steps i to I of claim 1 are performed only for a subset of predefined vectors of antenna response patterns {H (nk, Tk, ek)), said subset being chosen from directions close to a last known direction of the radiofrequency source (101) and taking into account a maximum speed of the source; this maximum speed can be predefined or evaluated statistically based on the history of the last known directions. The method of any of the preceding claims, wherein predefined vectors of antenna response patterns (H (nk, Tk, ek)) are normalized. 8. Method according to any one of the preceding claims, wherein the radio frequency sources (101) are RFID tags, wifi, bluetooth, zigbee, IEEE 802.15, IEEE 802.11, IEEE 802.16, IEEE 802.20, or IEEE 802.22. 9. Method according to any one of the preceding claims, in which the direction of a radiofrequency source (101) is obtained in a time interval of between 100 ns and 10 ms, preferably between 250 ns and 1 ms, even more preferably between 400 ns and 600 ns for an accuracy between 1 ° and 3 °. 10. Method according to any one of the preceding claims, wherein the calculations performed by the processor of the graphics card (GPU) are made by a programmable logic circuit having a plurality of calculation units. 11. Method (400) for locating a radiofrequency source (101) by performing the following steps: a. Obtaining (S410) a first direction of a radiofrequency source (101) according to a method for determining the direction of a radiofrequency source (101) according to any one of the preceding claims with respect to a first antenna array (111) ; b. obtaining (S420) a second direction of the same radiofrequency source (101) according to a method for determining the direction of a radiofrequency source (101) according to any one of the preceding claims with respect to a second antenna array (111) ; vs. calculating (S430) the intersection of the lines defined by the first and second directions determined in steps a and b relative to the distance and the relative orientation between the radio frequency modules (110); d. locating (S440) the position of the radiofrequency source with respect to the first and second networks (111). Apparatus (100) for finding the direction of at least one radiofrequency source comprising: a. at least one radiofrequency module (110) comprising: an antenna array (111) comprising a plurality of antennas (112), a controller (113) configured for: processing radio frequency signals, converting radio frequency signals to digital signals ; b. a computer (120) having a graphics card (130); vs. connections (102) for connecting the at least one radio frequency module (110) to the computer (120); characterized in that it is configured to perform a method according to claims 1 to 12. 13. Apparatus (100) according to claim 12 wherein the antenna array (111) comprises between 2 and 24 antennas (112), preferably between 4 and 16 antennas (112), even more preferably the network comprises 8 antennas (112). Apparatus (100) according to any of claims 12 or 13 wherein the controller (113) of the at least one radio frequency module (110) comprises a programmable logic circuit.