Beamforming process and associated devices

FR3133236B1Active Publication Date: 2024-11-01THALES SA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
FR2022001783
Authority / Receiving Office
FR · FR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2022-03-01
Publication Date
2024-11-01
Estimated Expiration
2042-03-01

AI Technical Summary

Technical Problem

Existing beam forming methods face challenges in achieving accurate and efficient beam formation with broadband signals, particularly in multi-antenna systems, due to the complexity of implementing real delays and equalizing reception channels, which require significant computational resources and memory, especially when forming multiple beams with different bandwidths.

Method used

A method and system for beam forming that utilizes FIR architectures operating in the frequency domain, employing discrete Fourier transforms and inverse transforms to implement delays and equalizations in the spectral domain, reducing computational and memory resources by pooling operations and enabling multiplexing of beams with different bandwidths.

Benefits of technology

This approach significantly reduces computational and memory resources while maintaining accuracy in beam formation, allowing for the simultaneous creation of multiple beams with varying bandwidths, and supports calibration to ensure consistent channel responses.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

Beamforming Method and Associated Devices The present invention relates to a method for forming Q beams by a computer of a beamforming system comprising P receiving channels each suitable for receiving signals from a respective antenna, the method comprising: - passing signals from each antenna into the spectral domain, - implementing operations in the spectral domain, the operations including pointing operations, the pointing operations comprising: - for each receiving channel, a generation operation by duplication of a contribution to each of the Q beams to be generated, - implementing for each contribution an operation of introducing a delay, and - an operation of summing the contributions to a beam to be generated from all the receiving channels, the summation operation being carried out for each beam to be generated.Figure for the abridged version: figure 9.
Need to check novelty before this filing date? Find Prior Art

Description

Description of the invention: Method for forming beams and devices associates The present invention relates to a method for forming Q beams using a beamforming system comprising P receiving channels. It also relates to a computer, a Q beamforming system, a computer program product, and a computer-readable medium involved in implementing the beamforming method. In a multi-antenna system, each antenna receives a set of signals from multiple listening directions with delays that depend on both the angle of the listening direction and the position of the antenna within the antenna system. Ideally, beam formation in direction 9 consists of compensating for the delay differences between antennas to allow a coherent summation of the same signal between the different receiving paths with the dual aim of gaining spatial selectivity and signal-to-noise ratio. An example of such a beamforming system or beamformer is shown in [Fig.1]. The system comprises an array of antennas 1, each connected to a digital receiver chain 2, and a beamforming unit 3 which generates N beams. Whether analog or digital, the realization of a beamformer capable of listening in different directions involves the realization of programmable delays, which, in practice, is very difficult to achieve. Typically, the signals to be listened to are considered to be narrowband around a carrier fo assumed to be known. In this case, "simple" phase shifts allow the formation of the listening beam; if T, is the delay to be compensated on the chain of index k, these phase shifts are equal to h, = 21f0T,. However, as soon as the listening signal bandwidth increases, this method is no longer possible and beamforming involving the realization of "true" delays (as opposed to phase shift, more often referred to by the corresponding English term "true time delay") is then indicated. Thus, the context of the present invention is that of so-called digital receiving chains whose output data are the digitized values ​​of periodic time samples of the signal at a rate noted fzcu. If the delay to be applied to the signal is a multiple of the sampling period T xcu = l / fzcu, the problem is simple since it suffices to delay numerically of an integer number of samples. On the other hand, if the delay to be implemented, measured in sampling periods, includes a decimal part, or even simply a fractional part, the problem is more complicated since it implicitly corresponds to an interpolation of the signal. Indeed, on the one hand, the interpolation filter depends on the delay to be provided (a different filter for each delay) and, on the other hand, for a given precision, the duration of the time response of this filter is greater the closer the band in which this delay is to be provided with precision is to the Nyquist band associated with the sampling (typically inversely proportional to the difference between the Nyquist band and the useful band); it is known that this is linked to the discontinuity at the limits of the sampling band of the response e-”’t associated with a delay 7 when t is not a multiple of the sampling period 1 / fzcu. Furthermore, in any beamforming operation, the quality of the signal degrades when the receiving channels do not exhibit exactly the same response. It is well known that differences in response, both in spectral amplitude and spectral phase, must be corrected prior to achieving good beamforming; the correction data required for the amplitude and spectral phase of each channel is acquired through calibration and is commonly available in the spectral domain. This correction therefore constitutes channel-specific filtering. However, as with the delay, the complex gain corresponding to this correction is generally neither continuous nor continuously differentiable at the limits of the sampling band, so that, again, the duration of the time response of this correction filter is all the greater the closer the band in which we want to ensure this correction with precision is to the Nyquist band associated with the sampling. Furthermore, the bandwidth in which channel formation is desired often depends on the application mode. Thus, in a multimode system, the sampling frequency of the processed signals must, according to the Shannon criterion, be compatible with the largest of the covered bandwidths. In this case, for narrowband modes, maintaining the full input data rate is not desirable; decimation must then be performed, which involves filtering the useful bandwidth. This useful bandwidth can be centered at different points within the sampling bandwidth. Also in this case, it may be desirable to simultaneously form several beams covering the same useful bandwidth. It may also be desirable, in other configurations, to simultaneously form a large number of narrow beams and a smaller number of wider beams. Ideally, it is then possible to carry, by multiplexing, indifferently and simultaneously on a set of output channels clocked at ω, sets of beams of the same band, the new channel carrying K, multiple beams band teplexed fgcy / K, (K, which can vary from one channel to another). It is therefore desirable for a broadband, multimode beamformer to meet several distinct objectives. It is necessary to form a set of beams with given and potentially different useful bands; this implies performing three different filtering operations: equalization of the receive channels, inherently specific to each receive channel; the delay specific to each beam / receive channel pair; and the selection of the useful band, specific to each formed beam. As explained above, these filtering operations can correspond to very long time responses, which are achieved, in practice, in an approximate manner by transverse filters with a priori finite impulse response, all of which are distinct. Such filters are often designated by the acronym FIR, referring to the Anglo-Saxon terminology "Finite Impulse Response." According to the prior art, FIR filters implement operations including the use of time shifts of the signal, gains, and summations.The number of operations is equal to the length of the impulse response of the FIR filter considered (the length being expressed in number of samples). Thus, such a large number of computing resources are involved that they can very quickly become prohibitive. One objective is therefore to reduce the volume of resources involved in the creation of such a trainer. Another objective is to optimize the volume of output information by reducing the rate of each beam according to its useful bandwidth and by multiplexing, on the same output channel, beams of the same bandwidth lower than the channel rate. Another objective is to ensure the accuracy of the calculations by ensuring good calibration between the channels. There is therefore a need for a process that can fulfill at least one of the aforementioned objectives, and ideally all three at once, particularly for broadband beams. To this end, the description describes a method for forming Q beams by a calculator of a beamforming system comprising P receiving channels, each suitable for receiving signals from a respective antenna, each antenna having a respective aperture, Q being an integer greater than or equal to 1 and P being an integer greater than or equal to 2, the method comprising: - the passage through the spectral domain of signals originating from each antenna, - the implementation of operations in the spectral domain, operations including pointing operations, pointing operations including: - for each receiving channel, a generation operation by duplication of a contribution to each of the Q beams to be generated, - the implementation, for each contribution, of an operation to introduce a delay and a possible operation to weight the antenna opening, and - an operation of summing the contributions to a beam to be generated from all the receiving channels, the summing operation being carried out for each beam to be generated. Today beamforming is done with classic FIR architectures operating in the time domain, whereas here we use a FIR architecture working in the frequency domain which was the subject of requests FR 3049131, FR 3102264, FR 3102262 and FR 3102265, cleverly adapted to obtain a substantial gain in resources. Upon reading what follows, the reader will understand that, in the case of beam formation with FIRs working in the frequency domain, two distinct gains are obtained: a gain on the calculation operators (multiplier in particular) and a gain on memory resources through pooling. According to specific embodiments, the Q-beam formation process comprises one or more of the following features, taken individually or in all technically possible combinations: - the operations implemented in the spectral domain also include a channel equalization operation, the equalization operation being implemented before the pointing operations, these operations preferably being carried out multiplicatively. - the operations implemented in the spectral domain also include a band limitation operation, the band limitation operation being implemented after the pointing operations, these operations preferably being carried out multiplicatively. - the process comprises, for each receiving channel, the formation of two calculation channels, the input signal of the first calculation channel being obtained by implementing the offset of the signal received by the antenna by M points, M being an integer greater than or equal to 2 to obtain an offset signal and the calculation of the sum of the received signal and the offset signal, and the input signal of the second calculation channel being obtained by implementing the offset of the received signal by M points to obtain an offset signal and the calculation of the difference between the received signal and the offset signal, this difference being translated in frequency by multiplicative application of a signal eisPim , n being an integer from 0 to M-1. - the transition into the spectral domain is achieved by applying a discrete Fourier transform to M points on each of the computation channels. - the process includes a return step in the time domain, the return step in the time domain involving the application of a discrete inverse Fourier transform to M points on each of the calculation channels, the process comprising a spectral shift operation of the second calculation channel by multiplicative application of a signal e+i«P / M and a summation operation of the two outputs of the two calculation channels. The description also describes a computer, specifically a programmable logic circuit, adapted to implement the operations of a training process as previously described. The description also proposes a Q beam formation system comprising P receiving channels, each capable of receiving signals from a respective antenna, Q being an integer greater than or equal to 1 and P being an integer greater than or equal to 2, the formation system comprising a computer as previously described. The description also describes a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the operations of a process or procedure as previously described. The description also offers a computer-readable medium comprising instructions which, when executed by a computer, lead the computer to carry out the operations of a process or process as previously described. Some features and advantages of the invention will become apparent from the following description, given solely by way of non-limiting example, and made with reference to the accompanying drawings, in which: - [Fig.1] [Fig.1] is a schematic representation of a beamforming system, - [Fig.2] [Fig.2] is a schematic representation of the resources of a long time-response filter architecture controlled in the spectrum, - [Fig.3] [Fig.3] is a schematic representation of an example of a modified architecture compared to the architecture represented in [Fig.2] in which certain resources are shared, - [Fig.4] [Fig.4] is a schematic representation of another example of architecture modified from the architecture represented in [Fig.2] in which certain resources are shared, - [Fig.5] [Fig.5] is a set of two graphs showing the time response of a Hilbert filter (real part at the top and imaginary part at the bottom), - [Fig.6] [Fig.6] is a graph showing the frequency response of a Hilbert filter (real part at the top and imaginary part at the bottom), - [Fig.7] [Fig.7] is a set of two graphs showing the total elimination effect of the second Nyquist band in the spectrum of a Hilbert filter, - [Fig.8] [Fig.8] is a set of two graphs allowing comparison of the Hilbert filter of [Fig.6] (left graph) with the filter required when the input signal is complex to avoid problems related to the discontinuity of the frequency response of the delay at the limits of the sampling band (right graph), - [Fig.9][Fig.10] Figures 9 and 10 are schematic representations of an example of a beam formation system comprising parts A to H, (Fig.9 corresponding to the left part and Fig.10 to the right part), - [Fig.11] [Fig.11] is a schematic representation of parts A to D of the beam formation system of figures 9 and 10, - [Fig.12] [Fig.12] is a schematic representation of parts E to H of the beam formation system of figures 9 and 10, - [Fig.13] [Fig.13] is a schematic representation of the decomposition of an M-point IFFT into a 2-point DFT and an M / 2-point IFFT, - [Fig.14] [Fig.14] is a schematic representation of an example of a beamforming system with multiplexing of 2 beams decimated by 2, - [Fig.15] [Fig.15] shows the modification of the 25-point IFFT for the purpose of multi-plexing 2 beams decimated by 2X, - [Fig.16] [Fig.16] is a schematic representation of an example of a beamforming system with 2% multiplexing of beams decimated by 24, - [Fig.17] [Fig.17] is a schematic representation of a frequency-based recalibration device for the decimated band, - [Fig.18] [Fig.18] is a schematic representation of another example of a beamforming system with multiplexing of 2K beams decimated by 2K, - [Fig. 19] [Fig. 19] is a synoptic representation of beam formation, - [Fig.20][Fig.21] Figures 20 and 21 are a schematic representation of another example of a beamforming system with multiplexing of 2% beams decimated by 24 ([Fig.20] corresponding to the left part and [Fig.21] to the right part), - [Fig.22] [Fig.22] is a schematic representation of an example of a calibration device, - [Fig.23] [Fig.23] is a schematic representation of an example of a time alignment device, - [Fig.24] [Fig.24] shows the modification of the 2N point FFT for the multiplexed processing of 2% decimated beams by 24, - [Fig. 25] [Fig. 25] schematically represents a beamforming system in emission with multiplexing of 2X beams decimated by 24, and - [Fig. 26] [Fig. 26] represents a detail of part of the system of [Fig. 24], and - [Fig.27] [Fig.27] represents a detail of another part of the system of [Fig.24] To facilitate further reading, the description is divided into chapters and sections. However, the chapters and sections are not independent, and the concepts used in one chapter also apply to the others. The following will be discussed in order: | = SOME PRELIMINARY CONCEPTS 1.1 - IMPLEMENTATION OF LONG TIME RESPONSE FILTERS 1.2 —- REALIZATION OF A DELAY 1.3 —- IQ DEMODULATION - HILBERT FILTERING 2 - DESCRIPTION OF EXAMPLES OF TRAINING METHODS 2.1 - GENERAL PRINCIPLE OF THE TRAINING PROCESS 2.2 — DESCRIPTION OF AN EXAMPLE OF A BEAM FORMATION METHOD 2.2.1 - Case of figures 9 and 10 2.2.2 — Generalization 2.3 — DESCRIPTION OF AN EXAMPLE OF A BEAM FORMATION METHOD 2.3.1 - Decimation by 2 (for example, conversion from real to QV) 2.3.2 - Case of bands not between 0 and -1 + 2N-K 2.3.3 - Decomposition into successive partial beam formations 2.3.4 — General case 3— DESCRIPTION OF A CALIBRATION PROCESS 4— TRANSPOSITION OF TRAINING PROCESSES TO BROADCASTING In particular, with reference to paragraph 2.2, a method for forming Q beams by a computer of a beamforming system comprising P receiving channels, each suitable for receiving signals from a respective antenna, each antenna having a respective aperture, Q being an integer greater than or equal to 1 and P being an integer greater than or equal to 2, the method comprising: - the passage through the spectral domain of signals originating from each antenna, - the implementation of operations in the spectral domain, operations including pointing operations, pointing operations including: - for each reception channel, a generation operation by duplicating a contribution to each of the Q beams to be generated, - the implementation, for each contribution, of an operation to introduce a delay and a possible operation to weight the antenna opening, and - an operation of summing the contributions to a bundle to be generated from all receiving channels, the summation operation being performed for each beam to be generated. Depending on specific aspects, the Q-beam formation process includes one or more of the following characteristics, taken individually or in all technically possible combinations: - the operations implemented in the spectral domain also include a channel equalization operation, the equalization operation being implemented before the pointing operations, these operations preferably being carried out multiplicatively. - the operations implemented in the spectral domain also include a band limitation operation, the band limitation operation being implemented after the pointing operations, these operations preferably being carried out multiplicatively. - the process comprises, for each receiving channel, the formation of two calculation channels, the input signal of the first calculation channel being obtained by implementing the offset of the signal received by the antenna by M points, M being an integer greater than or equal to 2 to obtain an offset signal and the calculation of the sum of the received signal and the offset signal, and the input signal of the second calculation channel being obtained by implementing the offset of the received signal by M points to obtain an offset signal and the calculation of the difference between the received signal and the offset signal, this difference being translated in frequency by multiplicative application of an eisP / m signal, n being an integer from 0 to M-1. - the transition into the spectral domain is achieved by applying a discrete Fourier transform to M points on each of the computation channels. - the process includes a return step in the time domain, the return step in the time domain comprising the application of a discrete inverse Fourier transform at M points on each of the calculation channels, the process comprising a spectral shift operation of the second calculation channel by multiplicative application of a signal e*i:P / M and a summation operation of the two outputs of the two calculation channels. The description also includes: a computer, specifically a programmable logic circuit, adapted to put implement the operations of a training process as previously describe. a Q beamforming system comprising P receiving channels each capable of receiving signals from a respective antenna, Q being an integer greater than or equal to 1 and P being an integer greater than or equal to 2, the training system including a calculator such as previously describe. a computer program product comprising instructions which, when the The program is executed by a computer, leading the computer to put into operation works the operations of a process or a process such as previously describe. a computer-readable medium containing instructions which, when they are executed by a computer, leading the computer to implement the operations of a process or a process as previously described. It will also be discussed, with reference in particular to paragraph 2.3, a method for forming Q multiplexed beams on Q' groups of beams indexed by q', q' being an integer varying between 1 and Q°, Q' being an integer greater than 1, and the multiplexing ratio being ke, Φ% being an integer respectively for each value of q', the method being implemented by a computer of a beamforming system comprising P receiving channels each suitable for receiving signals from a respective antenna, each antenna having a respective aperture, Q and P being two integers greater than or equal to 2, the method comprising: - the passage through the spectral domain of signals originating from each antenna, - the implementation of a set of operations in the spectral domain, known as pointing operations, the set of pointing operations comprising: - for each reception channel, a generation operation by duplicating a contribution to each of the Q' groups of beams to be generated, - for each contribution, a decimation operation by 77, and a hold over Ka periods to constitute a group of 2K contributions multiplexed by 9e, - the implementation for each group of contributions of a multiplexed introduction operation by 744' of a set of 7 delays, and a possible set of 7Xr antenna aperture weighting operations, and - an operation of summing the contributions to a beam to be generated, from all the receiving channels, multiplexed for each channel on Q' groups of contributions, the summing operation being carried out for each beam to be generated. Depending on specific aspects, the process for forming Q multiplexed beams includes one or more of the following characteristics, taken individually or in all technically possible combinations: - the operations implemented in the spectral domain also include a channel equalization operation, the equalization operation being implemented before the pointing operations, these operations preferably being carried out multiplicatively. - the operations implemented in the spectral domain also include a band limitation operation, the band limitation operation being implemented after the pointing operations, these operations preferably being carried out multiplicatively. - the process comprises, for each receiving channel, the formation of two calculation channels, the input signal of the first calculation channel being obtained by implementing the offset of the signal received by the antenna by M points, M being an integer greater than or equal to 2 to obtain an offset signal and the calculation of the sum of the received signal and the offset signal, and the input signal of the second calculation channel being obtained by implementing the offset of the received signal by M points to obtain an offset signal and the calculation of the difference between the received signal and the offset signal. - the transition into the spectral domain is achieved by applying a discrete Fourier transform to M points on each of the computation channels. - the process includes a return step in the time domain, the return step in the time domain comprising the application of ak inverse discrete Fourier transforms to M / nks points multiplexed by ke on each of the computing channels, the process comprising a spectral shift operation of the second computing channel by multiplicative application of a signal e[+j floor(n / 2K4 puC Make )], n being an integer from 0 to M-1 and the floor function denoting the integer part function, and a summation operation of the two outputs of the two computing channels. The description also includes: a computer, specifically a programmable logic circuit, adapted to put implement the operations of a training process as previously describe. a system for forming Q multiplexed bundles on Q' groups of sheaves indexed by q', q' being an integer varying between 1 and Q', Q' being a integer greater than 2, Q' being an integer greater than 1, and the mul- rate tiplexage is ok, Æg' is an integer for each value of q', the training system comprising P reception channels, each with its own receive signals from a respective antenna, Q and P being two integers greater than or equal to 2, and the training system including a cal- cultivator as previously described. a computer program product comprising instructions which, when the The program is executed by a computer, leading the computer to put into operation works the operations of a process or a process such as previously describe. a computer-readable medium containing instructions which, when they are executed by a computer, leading the computer to implement the operations of a process or a process as previously described. A calibration method for a beamforming system comprising P receiving channels, each capable of receiving signals from a respective antenna, will also be discussed, particularly with reference to paragraph 3. Q is an integer greater than or equal to 1 and P is an integer greater than or equal to 2. The beamforming system includes a computer capable of implementing the following steps: - the conversion, using spectral analysis, of signals from each antenna within the spectral domain, - the implementation of operations in the spectral domain, the calibration process comprising: - for each receiving channel, the injection into the time domain of a replica of a digital signal used to create a calibration signal, this replica being synchronized with the spectral analysis methods applied to the signals of the receiving channels, the injection being carried out identically for each receiving channel, - for each reception channel, the conversion of this digital signal into the spectral domain by means identical to those applied to the signals of the reception channels, said identical means being synchronized with each other, to constitute an identical reference signal on each reception channel, the conversion being carried out in an identical manner for each reception channel, - the measurement, for each receiving channel, of the reference signal values ​​and the values ​​of the signal propagated on the receiving channel, and - the deduction of the difference between the reference signal and the signal propagated on the receiving channel as a function of the measured values. Depending on specific aspects, the calibration process includes one or more of the following characteristics, taken individually or in all technically possible combinations: - the process comprising, in addition, the following steps: - the generation of a digital calibration signal - the conversion of the calibration signal to analog, - the frequency transposition of the converted calibration signal, - the amplification of the transposed calibration signal, and - balanced distribution across different reception channels. - when several receiving channels are housed on the same electronic card, the constitution of the reference signal is shared in common by all the channels of the same card (only one generation of reference signal per card). - the measured values ​​include the power correlation between the spectra of the signal propagated on a receiving channel and the reference signal. - the measured values ​​include the power spectral density of the reference signal and / or the power density of the signal propagated on the receiving path. - the measured values ​​are obtained by integration over several successive frames. - the deduction step involves estimating the phase of the power correlation between the signal spectrum on a receiving channel and the reference signal, to obtain an estimated phase, the phase of the difference between a receiving channel and the reference signal being the estimated phase. - the deduction step involves estimating the difference in amplitude between a receiving channel and the reference signal by calculating the square root of the ratio between the power spectral densities of the signal on a receiving channel and the reference signal. - The deduction step involves estimating the amplitude difference between a receiving channel and the reference signal by calculating the ratio between the power spectral density of the reference signal and the power correlation between the spectrum of the signal on said receiving channel and the reference signal. This operation is performed at a reduced cost. - the reference signal is noise. The description also includes: - a calibration device for a Q beamforming system comprising P receiving channels, each capable of receiving signals from a respective antenna, Q being an integer greater than or equal to 1 and P being an integer greater than or equal to 2, the calibration device being capable of implementing a calibration process as previously described. - a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to implement a calibration process as previously described. - a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to implement a calibration process as previously described. It will also be discussed, with reference in particular to paragraph 4 and transposed to paragraph 2.2, a method for forming P multi-beam transmission channels from Q beams, the method being implemented by a computer of a beamforming system in transmission comprising P transmission channels each equipped with an antenna, Q being an integer greater than or equal to 1 and P being an integer greater than or equal to 2, the method comprising: - the passage of signals from beams into the spectral domain, - the implementation of operations in the spectral domain, operations including tallying operations, tallying operations including: - a generation operation for each beam of a contribution for each of the P transmission channels by implementing for each contribution an operation of introducing a delay and a possible weighting of the antenna aperture, and - a summation operation of the contributions of an emission channel corresponding to each beam, the summation operation being carried out for each emission channel to be generated. Depending on specific aspects, the process for forming P emission channels includes one or more of the following characteristics, taken individually or in all technically possible combinations: - the operations implemented in the spectral domain also include a channel equalization operation, the equalization operation being implemented after the pointing operations, these operations being implemented in a multiplicative manner. - the operations implemented in the spectral domain also include a band limitation operation, the band limitation operation being implemented before the pointing operations, these operations being implemented multiplicatively. - the process involves, for each beam, the formation of two computing channels, the input signal of the first computing channel being obtained by implementing the offset of the beam signal by M points, M being an integer greater than or equal to 2, to obtain an offset signal and the calculation of the sum of the received signal and the offset signal, and the input signal of the second computing channel being obtained by implementing the offset of the beam signal by M points to obtain an offset signal and calculating the difference between the received signal and the offset signal, this difference being shifted spec-trally by multiplicative application of an eP / M signal, n being an integer from 0 to AM-I1. - the transition into the spectral domain is achieved by applying a discrete Fourier transform to M points on each of the computation channels. - the process includes a return step in the time domain, the return step in the time domain comprising the application of a discrete inverse Fourier transform at M points on each of the calculation channels, the process comprising a spectral shift operation of the second calculation channel by multiplicative application of a signal e*iP / M and a summation operation of the two outputs of the two calculation channels. The description also includes: a computer, specifically a programmable logic circuit, adapted to put implement the operations of a training process as previously describe. a system for forming P multibeam emission paths from Q beams, the training system comprising P emission channels equipped each of an antenna, Q being an integer greater than or equal to 1 and P being a integer greater than or equal to 2, the training system comprising a cal- cultivator as previously described. a computer program product comprising instructions which, when the The program is executed by a computer, leading the computer to put into operation implements the operations of a process for forming P multi-path emission pathways beams from Q beams or a process for forming P channels multibeam emission from Q beams such as previously describe. a computer-readable medium containing instructions which, when they are executed by a computer, leading the computer to implement the operations of a process for forming P multibeam emission paths at starting from Q beams or a process for forming P multi-emission channels beams from Q beams as previously described It will also be discussed, with reference in particular to paragraph 4 and transposed to paragraph 2.3, a method for forming P multi-beam transmission channels from Q beams multiplexed onto Q' beam groups indexed by q', where q' is an integer varying between 1 and Q', Q' is an integer greater than 1, and the multiplexing ratio is ok, Æg being an integer for each value of q', the method being implemented by a computer of a beamforming system in transmission comprising P transmission channels each equipped with an antenna having a respective aperture, Q and P being two integers greater than or equal to 2, the method comprising: - for each group q' of beams, the passage through the spectral domain of the signals composing the multiplexed beam group, to obtain multiplexed spectra, - the implementation of operations in the spectral domain, operations including pointing and interpolation operations comprising: - an operation to create as many replicas of the spectra as there are emission channels to be formed, - for each replica corresponding to each of the P emission channels, an operation to generate a contribution to the corresponding emission channel by implementing, for each contribution, a multiplexed introduction operation by 9K4' of a set of 7% delays, and a possible set of 7; antenna opening weightings, - an integration operation on oÉer points, - an interpolation operation by inserting OK - | Zeros, and - a summation operation of the contributions to an emission channel corresponding to each group of beams, the summation operation being carried out for each emission channel to be generated. Depending on specific aspects, the process for forming P emission channels includes one or more of the following characteristics, taken individually or in all technically possible combinations: - the operations implemented in the spectral domain also include a channel equalization operation, the equalization operation being implemented after the pointing operations, these operations being implemented in a multiplicative manner. - the operations implemented in the spectral domain also include a band limitation operation, the band limitation operation being implemented before the pointing operations, these operations being implemented multiplicatively. - the method comprises, for each beam group, the formation of two calculation channels, the input signal of the first calculation channel being obtained by implementing the shift of the beam signal by M points, M being an integer greater than or equal to 2x9K9 to obtain a shifted signal and the calculation of the sum of the beam group signal and the shifted signal, and the input signal of the second calculation channel being obtained by implementing the shift of the beam group signal by M points to obtain a shifted signal and the calculation of the difference between the beam group signal and the shifted signal, this difference being translated in frequency by multiplicative application of a signal e[+j floor(n / 2Kvye( M / 2%4)], n being an integer from 0 to M-1 and the floor function denoting the floor function, M being an integer greater than or equal to 2x9K4, - the transition into the spectral domain is carried out on each of the calculation channels by applying a discrete Fourier transform to M / aKy points multiplexed by gold. - The process includes a return step in the time domain, the return step in the time domain comprising the application of an inverse discrete Fourier transform at M points on each of the computation channels, the process comprising a spectral shift operation of the second computation channel by multiplicative application of a signal e*isF / M and a summation operation of the two outputs of the two computing methods. The description also includes: a computer, specifically a programmable logic circuit, adapted to put implement the operations of a training process as previously describe. a system for forming P multibeam emission paths from Q multiplexed bundles on Q' groups of sub-bundles indexed by q', q' being an integer varying between 1 and Q', Q' being an integer greater than 2, system beamforming in emission comprising P emission channels each equipped with an antenna having a respective aperture, Q and P given two integers greater than or equal to 2, the formation system also including a calculator as previously described. a computer program product comprising instructions which, when the The program is executed by a computer, leading the computer to put into operation implements the operations of a process for forming P multi-path emission pathways beams from Q multiplexed beams or a P formation process multibeam emission paths from multiplexed Q beams, the the process being as previously described. a computer-readable medium containing instructions which, when they are executed by a computer, leading the computer to implement the operations of a process for forming P multibeam emission paths at starting from Q multiplexed beams or a process for forming P channels multibeam emission from Q multiplexed beams, the process being as previously described. — SOME PRELIMINARY CONCEPTS These preliminary notions are part of the technological background. 1.1 — IMPLEMENTATION OF LONG TIME RESPONSE FILTERS When the time response lengths of a FIR filter are very long, implementing the filtering becomes problematic or even impossible due to the very large number of resources involved, especially if the FIR needs to be programmable. To circumvent this problem, one solution is to perform this operation in the spectrum. This is done by transforming from the time domain to the frequency domain using a Fourier transform, making the filtering operation multiplicative, and then returning to the time domain using an inverse Fourier transform. This allows us to define a generic filter architecture that was the subject of a patent application published under number FR 3 049 131 A1, the content of which is briefly summarized below, with further information contained in the description of this request. In accordance with [Fig.2] which refers to a long time response filter architecture controlled in the spectrum, this architecture makes it possible to realize using FFTs of durations N an impulse response FIR of duration N entirely contained in the positive time domain. As a reminder, a Fourier transform is implemented using a fast Fourier transform, often referred to by the acronym FFT. More precisely, the transition from the time domain to the frequency domain is achieved using an FFT, while the transition from the frequency domain to the time domain is achieved using an IFFT. The acronym IFFT stands for Inverse Fast Fourier Transform. In the FIR of [Fig.2], the coefficients p, represent the spectral density obtained by an FFT at 2N points of the FIR response, the negative times representing N zeros on the 2N values ​​of the time frame: 2N0E OT IST ON AS p,=Xae 5 =S'ae F Pre n=2 Note that, if the time response associated with the filtering is entirely contained in negative times (as happens when filtering is performed on a signal using the conjugate spectrum), then the output recombination between the two channels (even and odd frequencies) is done by subtraction instead of the addition shown in [Fig.2]. We can see that, in this architecture, the block 1 + z* (moving from position 2 to position 3 on [Fig.2]) can be translated at will to any of the positions 1 to 5. It can also be translated to either position a or b but on condition of being replaced by a block 1 — zN. This allows us to define several equivalent architectures, including, in particular, the two that are proposed respectively on [Fig.3] and on [Fig.4] and which allow the zN memory resources to be shared. These FIR architectures are particularly interesting because they allow a considerable reduction in the computing resources required compared to a classic FIR as the number of coefficients increases, typically in a ratio M to log(M); however, this comes at the cost of a certain increase in memory resources (factor 4 for the architectures in Figures 3 and 4 and 5 for the architecture in [Fig.2]). The computing resources required for a response with a maximum duration of N = 2M are: “additions: 3 + 4 x 2 x M = 8M + 3 * multiplications: 4 +4 x M / 2=2M +4 Elements on Radix decomposition To understand the rest of the description, it is necessary to explain some generalities about the decomposition of a discrete Fourier transform (DFT) into Radix (x being the number of factorable points). The discrete N-point Fourier transform of a sequence of N points x, (N being (an integer) is equal to: TE y LES XF ONE nm In the case where N is the product N = N, x N, of two integers N, and N, it can be written: Pope Ne YEARS Barely Na DPSSE SELS COR My date => = your af but =i mafimel There are two ways to decompose the discrete Fourier transform, both corresponding to the cascade of a discrete Fourier transform with N points and a discrete Fourier transform with N points. It is the order in which these discrete Fourier transforms are performed that differs. One of the discrete Fourier transforms involved is weighted by a local oscillator allowing a fine shift of the spectrum; the exponential term enabling this shift is called the "tweedle factor": far SR tweedie factor = e By principle, the fastest-changing output index is m, since it corresponds to the "fast" discrete Fourier transform, i.e., the non-multiplexed one. Thus, the output order of the frequencies is not the natural order: in fact, the m values ​​equal to 0 modulo N are output first, then the m values ​​equal to 1, and so on down to the m values ​​equal to N, - 1, which is always 0 modulo N; this order is called "bit reverse" because it corresponds to reversing the binary representation of the index to obtain the output rank (for a number of points that is a power of 2). The implementation of the inverse discrete Fourier transform, which recovers the frequency data in "bit reverse", is carried out in a dual manner, by reversing the operations. According to the first case, the discrete Fourier transform at N points ("fast") is followed by the Fourier transform at N points ("multiplexed"), which corresponds to: AM LeagltaMiTE Nm Na-mei ONE L ap BIENS The pot cn Meme ON ft A * M2 palme Ro Ta en she Nom No ee EN su _n NE, - E,Ÿ wi FE y mfégem > D Apent de * 3 5° EEE Pen Most _pz EME AE d'e (8 xYe Preuii et AS — — “st re” likes af “= tweedte factor FFT at N points (Radix N,} Re re N-point FFT multiplexed by N, {Radix N.} The Fourier transform is performed as a cascade of a non-multiplexed N-point FFT and a N-point FFT multiplexed by N. The "tweedle factor" is placed at the beginning, before the first stage of the Fourier transform. According to a second case, the discrete Fourier transform at N points ("multiplexed") is followed by the Fourier transform at N points ("fast"), which corresponds to: lol Lai ke get ae Nothing to report X een, 7 => pe 6 PONS LE EF = * a 2 Sat NL La M SSSR $S> fai € a xSe x &_ TE From 1 tn naaaltau ali even Li vweedie = factor FFT at N, points multiplexed by N, (Radix N,} FFT at N, points (Radix N,} The realization is performed this time as a cascade of a discrete Fourier transform at N points multiplexed by N, and a discrete Fourier transform at N points without multiplexing. The "tweedle factor" is placed between the two stages of discrete Fourier transform. Since the first discrete Fourier transform performed is multiplexed, this decomposition accommodates parallelized input data; the multiplexing then becomes parallel processing. Following one of the classical versions of radix2 decomposition, the 2M-point FFT of this sequence is written: Tai 21 t d- = » Prat S —j2. ; ç b,= ae PRES a APPERE AT ST ae PRE m5 nf == Especially : 4-3 H= — NV, 0 M rms MES — Joss) MT <= —ilepAIaés — F1 sema / sé Baup = Ha,ee +fIF D'étypant et ==; n=ù Or f »<{ either : Bomep = FFT, (ae sa}+ (HF FFT, faue Fret where FFT denotes the FFT at M points. It should be noted that this spectral density can also be written in the following two ways, which will simplify its implementation: Banep = FFT la 25919) FFT, fa, 6 TRUE) And Bones = FFT,fl- Va taie È pps nt) Denoting © as the spectral response of the filter, the filtered response is obtained by inverse transformation: dt Si ; ' * = > eager pif dice Saskp A 3263 > 2x = > bles t* Pat > a bompmas® ' —l —l 73 == p=t mm The answer is calculated only for points 0 < k < M-1 (due to the folding inherent in the FFT) to finally obtain: x, > > + set ; e IFF TlPrmep Pine) — 2 where IFFTm denotes the inverse FFT with M points. 1.2 — REALIZATION OF A DELAY It is known that the complex gain corresponding to a delay t is H(f) = e^(-πt) * πt. When attempting to perform this operation on a signal sampled for a delay other than a multiple of the sampling period, one encounters the problem of frequency response folding, which is, in principle, zero when π is not an integer. Therefore, one must seek to implement a filter exhibiting a periodic complex gain of period πt and a response H(f) = e^(-πt) * πt over this period. It can be recalled here that, in principle, shifting a digitized signal xn by a delay t amounts to interpolating the signal x using the Shannon interpolating filter whose time response is sinc(fecut) where sinc(x) = sin(1x) / (7x) and replacing the sample number m with value x' by y, = x(m Tecy - T). Mathematically, this can be written as: 5 pos Hsoas(T5cs — vf] Ÿ pô -im,#=Fx, sir fs ON Tone = T = ETr055 5 MrcelmT ep -1=1) 5 5 Men (M Tzez 7 -KEc5) The time response of the digital filter ensuring the delay t is therefore: El } sin reg (iT5en — 7) _ Si front (AT EE re ir Hfecos UAT2e5 — 7) ï ÎcouT—H In principle, this filter is not feasible since its time response extends from -Φ to +θ. By truncating and weighting its response between -N and +N, we obtain a set of coefficients corresponding to a filter that provides a differential delay t relative to a reference delay equal to N Tzcy. The frequency response of this filter is then the folding of the frequency response of the weighting. In any case, the slow decay of h, with the index n (h, -1 / n) leads to very large lengths when both wide bandwidth (close to the Nyquist band) and high dynamic range (good suppression of image bands) are sought. The equations presented above indicate that this slowness is a direct consequence of the fact that the complex gain g-=% associated with a delay © is discontinuous at the boundaries of the sampling band when t is not a multiple of the sampling period l / fzcu. Similarly, the complex gain corresponding to the equalization corrections of the receiving channels is generally neither continuous nor continuously differentiable at the boundaries of the sampling band, so that, again, the duration of the time response of this correction filter is greater the closer the band in which this correction is to be performed accurately is to the Nyquist band associated with the sampling. It follows from all these observations that this filter leads to an infinite time response h, decreasing slowly with the index n due to the discontinuity at the edge of the band. It is known that weighting a response limits its extent in the dual space; indeed, it is common practice to weight a time response to limit the spectral obstruction inherent in the discontinuity at the boundaries of the sampling domain. Similarly, weighting in the spectral domain, which corresponds to filtering, greatly limits the temporal extent associated with implementing a non-integer delay (in terms of sampling periods), by reducing, through filtering, the band in which the delay will be operational. Thus, the wider the band in which the delay is ensured with precise timing, the greater the length of the time response of the FIR filter realizing it. 1.3 — IQ DEMODULATION - HILBERT FILTERING Furthermore, it is possible to sample real signals on the carrier (and no longer in baseband after analog V / Q demodulation). In this case, by principle, the positive frequencies of the useful signal are strictly contained within one of the bands of The Nyquist sampling frequency (i.e., between k frcy / 2 and (k+1) frcy / 2 where k is arbitrary) is used. After sampling, the reconstruction of the analytical signal then involves a filtering process, known as a Hilbert filter, which eliminates the spectrum of negative frequencies. Typically, as shown in [Fig. 5] representing the time response of a Hilbert filter (real and imaginary parts), the time response of this filter comprises a real part limited to a single central coefficient and an antisymmetric imaginary part decreasing as 1 / n, with one out of every two samples being zero. In doing so, as shown in [Fig.6] which illustrates the frequency response of a Hilbert filter, the complex gain is -6 dB at the zero frequency and at the Nyquist frequency; moreover, the level of rejection of negative frequencies is higher as the duration of the response is greater. If the actual signal does not occupy the entire Nyquist band, or if the complex signal does not occupy the entire sampling band, it is common practice to filter the useful band (denoted BU in [Fig. 6]) occupied by the signal to allow for a subsequent decimation step aimed at bringing the sampling frequency as close as possible to the useful band. In this case, the filter's response resembles that of the Hilbert filter, except that its bandwidth is narrower. 2-DESCRIPTION OF EXAMPLES OF TRAINING PROCESSES 2.1 —- GENERAL PRINCIPLE OF THE TRAINING PROCESS The signal formation process can be applied either directly to a real signal after analog-to-digital conversion, or to a complex signal after baseband conversion, / Q demodulation, and partial decimation. The process is designed to simultaneously perform the various steps: equalization of the receive channels, delay and weighting associated with pointing, and limitation of the occupied bandwidth. For this, the process is carried out by the filtering structures presented in the paragraph entitled "Realization of filters with long time responses", which allows control of the filtering in the spectral domain and makes the operations independent and multiplicative. It is also possible to duplicate the delay operation as many times as necessary to create the contribution of a receiving channel to as many beams as required. It should be noted that it is desirable for the response of the Hilbert filter or the filter limiting the spectrum to the useful band to sufficiently filter out the discontinuities associated with the delay and equalization operations in order to prevent these operations from introducing noticeable artifacts caused by these spectral discontinuities. This is schematically represented in [Fig. 7], which shows the modification of the Hilbert filter for the complete elimination of the second Nyquist band (left graph before modification and right graph after modification). In cases where the input signal is complex and its spectrum occupies almost the entire sampling bandwidth (fecu), filtering is necessary to avoid problems related to the discontinuity of the frequency response of the delay at the sampling bandwidth boundaries. In this case, the filter's time response is a subsampling by a factor of 2 of that of a filter similar to the Hilbert filter, but designed at twice the frequency. Figure 8 illustrates this by showing the filter required when the input signal is complex to avoid problems related to the discontinuity of the frequency response of the delay at the sampling bandwidth boundaries. 2.2 — DESCRIPTION OF AN EXAMPLE OF A BEAM FORMATION METHOD 2.2.1 - Case of figures 9 and 10 It is proposed to exploit the filter architectures presented in figures 2 to 4 as a constituent element of a beamforming system architecture. The beam formation system is a system that forms Q beams, where Q is a non-zero integer. Each beam can thus be identified by an index q, an index which is an integer between | and Q. It should be noted that in a particular embodiment of this scenario, Q can be equal to 1. The beamforming system comprises P receiving chains, each equipped with an antenna. P is an integer greater than or equal to 2. A receiving chain is a set consisting of an antenna and a processing circuit. The term "receiving path" is sometimes used for the processing circuit. This term will be used throughout the rest of the description. Each of the P receiving channels is thus specific to receiving signals from a respective antenna. Furthermore, each of the P reception channels includes means for performing operations that will be described later. In particular, each reception channel includes spectral analysis means; beamforming here has the particularity of involving operations in the spectral domain, and more specifically, beam-pointing operations. Each receiving channel can be identified by an index p which is between | and P. Note: * C,(m) the complex correction to be applied to the reception channel p in order to equalize it with the other reception channels with a resolution equal to fecu(2M), the index m which, modulo 2M, is between 0 and 2M-1, is representative of the frequency, * T,' the delay to be applied on the receiving channel p to form the beam q and, if f is the carrier frequency brought back to the DC after sampling, $,(m) = -27r ( fo + M fecu / (2M) ) T,(# is the corresponding spectral phase, * O9 a possible real weighting coefficient of the antenna aperture, for each index q of the formed beam, the weighting law a, being intended to improve the level of the lobes of the directivity diagram of the antenna thus formed numerically, and "H(m) the complex gain associated with the band limitation specific to each beam formed (Hilbert filter or limitation of the spectrum to the useful band); this complex gain is derived from a 2M point FFT of its causal time response. This can be interpreted as a frequency-dependent matrix operation that relates the vector of signals from the receiving channels to the vector of the formed beams. This operation can be decomposed as the product of three matrices, the first and last of which are diagonal: * [C] the diagonal P x P matrix of equalization corrections for the receiving channels, * [ aeip ] the full Q x P matrix of spectral phases associated with delays and aperture weighting coefficients, and * [ H ] the diagonal matrix Q x Q of the band-limiting filters. If [X] is the vector P x 1 of the power spectral densities (sometimes denoted by the acronym SPD) of the receiving channels and if [Y] is the vector Q x 1 of the power spectral densities of the formed channels, the operation can then be written as: EY Jos =1H]oc x [ ae” Ja: x { C Jr x [ X Jay (matrices and vectors depend on the frequency of an antenna). It should be noted that these three filtering functions, which are applied in a multiplicative manner due to the filtering structure, since they operate in the spectral domain, follow each other in a very particular order, which is that of the matrix decomposition mentioned: "The equalization of the channels [C] is applied first, after the FFTs which form the passage in the spectrum, then * the delay as well as the antenna aperture weighting j gel p, an operation duplicated on each receiving channel as many times as there are beams to be formed, and finally * the specific band limitation [ H ] applied to each beam formed, i.e. after summing the different contributions from all the receiving channels. Figures 9 and 10 describe a beamforming system in the case of- particular where P and Q are equal (here equal to 4), that is to say that it is a 4-beam former from 4 receiving channels. In these figures, for greater readability, the lines become increasingly wider depending on the index p or q considered. The beamforming system is referenced as 10 and has 4 receiving channels 12, 12, 123, 124. Schematically, the training system 10 comprises, for each reception channel 12, an input 14 operating in the time domain, a processing circuit 16 operating in the frequency domain and an output 18 operating in the time domain. Input 14, processing circuit 16 and output 18 implement operations and can be interpreted as computer program products. In particular, they can be implemented at least partially by a computer, such as a programmable logic circuit. Such a circuit is often referred to by the acronym FPGA, an English acronym for "Field-Programmable Gate Array," a network of gates that can be programmed in situ. In another example, they form an application-specific integrated circuit. Such a circuit is often referred to by the acronym ASIC (acronym for "Application-Specific Integrated Circuit"). This means that these elements perform a beamforming function computationally, this function sometimes being referred to by the acronym FFC. The acronym DBF is sometimes used in reference to the corresponding English term "Digital BeamForming". In the example described, the processing circuit 16 has two computing channels 20 and 22 which are each connected to the input 14 by an FFT application unit, 24 and to the output 18 by an IFFT application unit, 26. The two calculation paths 20 and 22 perform similar operations on points having respectively an even index (calculation path 20) and an odd index (calculation path 22). On each channel, there is an equalization unit 28, a delay and weighting unit 30 (shown in larger format on parts A to H of figures 11 and 12), a summation unit 32 and a band limiting unit 34. These notations are only introduced on the first calculation channel 20 of the first reception channel 12, to ensure a certain readability of figures 9 and 10. The equalization unit 28 applies the aforementioned matrix C, with the notation C,(2m) if it is the first lane 20 (referring to the fact that even indices are processed) and the notation C,(2m+1) if it is the second lane 22 (referring to the fact that odd indices are processed). As seen in Figures 11 and 12, the delay and weighting unit 30 applies delay and weighting coefficients as many times as there are bundles to be formed, i.e. 4 here. The summation unit 32 of the first channel 20 sums the contributions obtained after the delay and weighting unit 30 from the first channels 20 of each receiving channel 12. For the case of the first channel 20 of the first receiving channel 12), this means that the summation unit 32 sums the output of the delay and weighting unit 30 of the first channel 20, (the index 1 indicates that it is that of the first receiving channel 12,), the output of the delay and weighting unit 30 of the first channel 20 2 the output of the delay and weighting unit 30 of the first channel 20; and the output of the delay and weighting unit 30 of the first channel 204. Similarly, the summation unit 32 of the second channel 22 sums the contributions obtained after the delay and weighting unit 30 from the second channels 22 of each receiving channel 12. For the case of the second channel 22 of the first receiving channel 12), this means that the summing unit 32 sums the output of the delay and weighting unit 30 of the second channel 22, the output of the delay and weighting unit 30 of the second channel 22, the output of the delay and weighting unit 30 of the second channel 22: and the output of the delay and weighting unit 30 of the second channel 22, This being true for each summation unit 31, it is therefore clear that a relatively complex interconnection 36 is visible between the delay and weighting units 30 and the summation units 32. The band-limiting unit 34 applies the H matrix and the same notations as for the C matrix are used. Output 18 performs a recombination of the computation channels 20 and 22 using a shift unit 38 and an adder 40. It should be noted that, depending on the application, the delay required on the different paths to deorient a beam can exceed one clock period. For example, for an antenna covering a relative bandwidth of 20%, with 32 x 32 (1024) elements, and a sampling frequency covering the entire antenna bandwidth, the maximum delay associated with beam formation in both directions is 2 x 16 x (1 / fy / 2) = 16 (frcu / B) x (B / fo) / fecu = 6.4 / fecu. One might then imagine implementing the integer part of this delay numerically in the time domain, but this is no longer possible once the frequency domain is considered to simplify the simultaneous creation of several distinct beams. This delay must therefore be fully implemented in the spectrum and it must be taken into account in the design of the filters, in the sense that their increased time response of the maximum delay must remain less than N err / Ecou: 2.2.2 — Generalization The elements that have just been described can easily be generalized, which leads to the training process now explained. In a general case, the process of forming Q beams is implemented by a beamforming system comprising P receiving channels, each connected to a respective antenna, where Q is a non-zero integer and P is an integer greater than or equal to 2. For each receiving channel, the process includes the following steps: - reception of a signal received by the antenna, - generation of two calculation channels by summing and subtracting the signal from the receiving channel and that same signal delayed by M periods, - realization on the difference channel of a frequency shift of fcy / 2M by multiplicative application of a signal eir7M, n being an integer from 0 to M-1, M being an integer strictly greater than 2, - Implementation of initial operations on the first computing channel, the initial operations comprising: - the application of a discrete Fourier transform to M points on a signal from the transmitted signal to obtain M points in the spectrum of the signal from the transmitted signal, M being an integer strictly greater than 2, each point of the spectrum of the transmitted signal corresponding to even indices of a spectral analysis of 2*M points of the signal from the transmitted signal and being identified bijectively by an index which is an even number between 0 and 2*M-1, - applying a channel equalization function to the spectrum at M points, to obtain equalized points, - applying a delay to the equalized points in the spectrum to generate Q series of M delayed points (via multiplication with a linear exponential in phase), and - implementation of second operations on the second computing path, the second operations comprising: - the application of a discrete Fourier transform to M points on the signal from the transmitted signal to obtain M points in the spectrum of the signal from the transmitted signal, each point of the spectrum of the signal from the transmitted signal corresponding to odd indices of a spectral analysis to 2*M points of the signal from the transmitted signal and being identified bijectively by an index which is an odd number between 0 and 2#M-1, - applying a channel equalization function to the spectrum at M points, to obtain equalized points, and - applying a delay to the equalized points in the spectrum to generate Q series of M delayed points (via multiplication with a linear exponential) (in phase), - formation of each beam from the series of M delayed points generated, the formation comprising, for each beam: - the implementation of third operations, the third operations including: - the summation of the series of M delayed points coming from each first calculation channel of a reception channel, to obtain summed points, - applying a band-limiting function to the summed points to obtain bounded points, and - applying a discrete inverse Fourier transform to M points on the bounded points, to obtain a first output signal, - the implementation of fourth operations, the fourth operations comprising: - the summation of the series of M delayed points coming from each second calculation channel of a reception channel, to obtain summed points, - applying a bandwidth limiting function to the summed points to be processed to obtain limited points, and - applying an inverse discrete Fourier transform to M points on the bounded points, to obtain a second output signal, and - the frequency shift of the second output signal to obtain a shifted output signal - the recombination of the first output signal and the second translated output signal to obtain the formed beam. In certain specific cases, the process may also exhibit the following properties: - the first and second operations also include the application of an opening weighting of the antenna of the receiving channel considered. - the received signal being sampled at a sampling frequency, the equalization function has a resolution equal to the ratio of the sampling frequency by 2M; in a numbering from 0 to 2M-1 even points are applied to the first calculation channel and odd points to the second calculation channel. - the limiting function is obtained by applying a discrete Fourier transform to 2*M points of a time response of 2*M points exclusively contained in positive times (i.e. zero on the last M points). 2.3 — DESCRIPTION OF AN EXAMPLE OF A BEAM FORMATION METHOD In this other example of a beamforming process, the process also optimizes the volume of output information by reducing the rate of each beam according to its useful bandwidth and by multiplexing, on the same output channel, beams of the same bandwidth lower than the channel rate. Indeed, if the reception band of a given beam is significantly smaller than fecu / 2 for real signals and f:cy for complex signals, it is desirable to decimate the signal by an integer or non-integer factor that can go up to the ratio of the bands; this is only possible in a simple way, of course, if this factor is an integer greater than 2 and preferably a power of 2. 2.3.1 — Decimation by 2 (for example, conversion from real to VO) For example, when the input signal is real, it is common practice to halve the signal after Hilbert filtering in order to reduce the information flow; this is made possible by the fact that negative frequencies have been eliminated from the spectrum. Given this, we will focus on restoring the signal only for even-numbered times at the output of M-point IFFTs. If H is the input of the IFFT and h its output, we can then write the following relationship, since the negative frequencies are eliminated by the filter at a level low enough to be considered zero in terms of instantaneous dynamics: Aa sy 14857 RE sup 25RR G1505E QE Gus EE MS 2.2 RIT ON CN HS h,=2H,e *“=NYHe x = CS H,e > : È This means that the even points, which are the only ones retained, consist of M / 2 IFFTs of only positive frequencies. However, the natural output of the FFT is bit reversed, which, for an M-point FFT, involves alternately outputting positive and negative frequencies, more precisely a point of positive frequency H, followed immediately by a point of negative frequency H,…mn. Restricting the FFT output to only positive frequencies then appears as a decimation of this output by a factor of 2, which means that outputting the spectrum in bit reversed form expresses the duality that exists between subsampling by a factor of 2 in the time domain and positive / negative subbands in the spectrum. This characteristic can then be advantageously used to time-multiply two different beams; this is made possible thanks to the internal architecture of the DFT with M = 2N points forming an IFFT. DFT is the acronym for the English name "Discrete Fourier Transform". Indeed, it is known to advantageously decompose the 2N-point inverse DFT into N successive stages called "radix2," each forming a 2-point inverse DFT (which is identical to the 2-point DFT) and separated by time-varying phase rotation stages called "tweedle factors." In such a decomposition, these 2-point inverse DFTs operate on data whose multiplexing ratio increases along the IFFT (1 for the first stage, 2 for the second, 4 for the third, etc.). Due to the "bit reverse" output, H and Hm+mn are two points in the spectrum, one in the positive frequencies and the other in the negative frequencies, which are delivered successively at the output of the M-point FFT. Therefore, formally, the output of the M-point IFTTT is written as: NX tr, + Hd The = a“ " CMS y mb ie ME + @r,-—H,wr) PS SRE ES ARE MR 405 IE = ° e HR,< Lu n=5 je MF We observe that, indeed, the first step of this M-point IFFT consists of a DFT of the two consecutive points H and Hmn, whose result, sum and difference, also emerges consecutively. This result is then multiplied by a different "tweedle factor" for the two points thus created. We also observe that the two results thus multiplied form two interlaced series of points which each undergo an M / 2 point IFFT, which leads to the scheme in [Fig.13], which illustrates the decomposition of an M point IFFT into a 2 point DFT and an M / 2 point IFFT. As shown in [Fig.13], an M = 2° point IFFT can be considered as the cascade of a non-multiplexed 2-point DFT and an M / 2 = 25! point IFFT multiplexed by 2, which amounts to simultaneously performing 2 distinct M / 2 = 2 N-! point IFFTs on two interlaced point series. This allows, by short-circuiting the 2-point DFT head stage and the tweedle factor stage that directly follows it, the use of an M-point IFFT operator, either as such, or as an M / 2-point IFFT operator multiplexed by 2, that is, operating in multiplexed mode on two interlaced sets of points. It is this capability that can be advantageously implemented to multiplex two distinct beams decimated by 2 onto the same output, as shown in [Fig. 14], which illustrates a beamforming system with multiplexing of two beams decimated by 2. In this case, the digital local oscillator @*#14#2) is adapted to be maintained for 2 periods (multiplexing rate here). Decimation by 2α of the band between 0 and -1 + 2055 By repeating K times the decomposition presented in the previous paragraph, we can exploit the fact that at the output of an FFT with M = 25 points, the points with frequency Hm+irez for 0 appear successively <k <-1 +2k. For example, by decomposing the M / 2 point IFFT in the same way into a 2 point DFT, a "tweedle factor" stage and an M / 4 point IFFT, we can also multiplex 4 series of interlaced points since only the frequencies from 0 to -1 + M / 4 are kept and we can therefore decimate by 4 after applying the anti-aliasing filtering which eliminates the other frequencies. By generalizing this approach, it is possible to consider more generally that an IFFT with M = 2" points is the cascade of an inverse DFT with 2K points not multiplexed, a stage of "tweedle factors" and an IFFT with 2N-K points multiplexed by 2K, which amounts to simultaneously performing 2% IFFT with 2N< distinct points on 25 interlaced point series. Thus, step by step, by short-circuiting the inverse DFT stage with 2X head points and the tweedle factor stage that directly follows it, it is also possible to multiplex 2K series of interlaced points using the same IFFT multiplexed by 2K, provided that only the points with frequencies from 0 to -1 + M / 28 are retained and can therefore be decimated by 2%. The capability described previously for decimation by 2 and multiplexing of 2 beams can thus be extended to decimation by 2% and multiplexing of 2% different beams. One way to achieve this is shown in [Fig. 15], which illustrates the modification of the 2N-point IFTTT for multiplexing 2* beams decimated by 24. As shown in [Fig. 15], it suffices to equip the 2N-point IFTTT with a set of switches that, through appropriate addressing, allow entry into the IFTTT at the location required to perform 2N-point IFTTT processing on 2X multiplexed data. The addressing of the various switches is achieved through the binary description of 25 — 1. Such a block can be integrated into the training system of [Fig. 16] to make it operate with a decimation / multiplexing rate of 2%. The block is indicated by "block [Fig. 15]" in [Fig. 16], which presents such a generalized training system. We will denote bit_reverse(k,K) the function which associates to k the integer given by reversing the bits of the description of k on K bits. It can be noted that bit_reverse(bit_reverse) is the identity function. Furthermore, we will have, for example: * bit_reverse(0,K)=0. * bit_reverse(2#!,K) = 1 and, More generally, bit_reverse(25M,K) = 2M—1 for 1 <m<k, * bit_reverse(2k+1,K) = bit_reverse(2k,K) + 25! for O<k<28! — 1. In general, the band of interest to be kept, consisting of 254 cells, m + kx 25Ketm-—1+(k+1) x 2NX with O <m<28— 1: the case m = k = 0 corresponds to the previous paragraph in which the band is between 0 and -1 + 2N<, <sim=0et k # 0, la bande d’intérêt correspond à l’une des sous-bandes naturelle de fft ; les cases qui composent sortent alors façon périodique mais avec un offset equal to bit_reverse(k,K), As soon as m # 0, the different cells constituting the band of interest no longer exit periodically. Although these three cases lead to differences in the behavior of the data output order, all three operating cases can nevertheless be grouped under a single operating mode. All you need to do is: * read the spectrum only at the instants corresponding to the frequency cells constituting the useful band, that is to say at the indices n such that: Nm = M + k 234 < bit_reverse(n,2N) < Nmax = M = 1 + (k+1) 2N% * maintain the value until the next sampled value, and * resample the signal thus obtained at times nx 25 — 1. An implementation of such an operating mode is represented in [Fig.17] which represents a frequency recalibration device of the decimated band including in particular a counter, a bit reverse unit and two sampler-blockers. This introduces a processing latency of 2% clock cycles. In the example in [Fig. 17], it is assumed that there is no processing latency, as these latencies are potentially accounted for by using devices known to those skilled in the art. To avoid disrupting the synchronization of subsequent operations and resulting in a fixed latency regardless of the multiplexing ratio, this latency can be compensated, for example, but not exclusively, by applying a complementary delay equal to 2N - 2* clock cycles to the resulting signal. This creates a folding of the useful band into the undersampled band [0 - f xcu / 2K]. Indeed, the order of the data output from the FFT is as follows: * first the 2% samples 0 of each of the 28 sub-bands, these sub-bands appearing in the order fixed by bit_reverse(x,2X), then * of the other samples, delivered in successive packets of 2 values ​​each corresponding to the same index in the set of sub-bands, the order of the indices being fixed by bit_reverse(x,2K-N). Therefore, the first sample of interest (i.e., belonging to the useful band) to emerge from the FFT is the sample such that (m + kx 2NX) modulo 27 is zero, that is, sample O of the kth sub-band in the case where m = 0 and of the (k+1)th <e si m # 0. When m = 0, the samples traverse the k?we sub-band; they are all spaced 2K apart. If m ≠ 0, the samples of interest belong either to the kè?"* sub-band or to the (k+1)th, However, it turns out that the FFT output samples are presented in 2% point increments, each increment representing all sub-bands for the same position. data within the sub-band. Therefore, by sampling at the times corresponding to the output times of the 25K points of the useful band and maintaining the value until the next frequency point, we have at the end of a slice, that is to say at t = nx 2K — 1, a frequency sample whose position in its original sub-band is indeed that corresponding to the slice in question. In doing so, the last samples of the kth subband were placed at the end of the band thus created; this corresponds to the folding of the useful band, positioned straddling two subbands, into the undersampled band [ 0 - f:cu / 2% 1. This folding corresponds to a frequency transposition of -(k+1) frcy / 25-%. Thus, the center of the processed band is located at [ 2-N (2m — 1) - 2 ] / frcu / 2 in the undersampled band [ 0 - fzcu / 2% ]. The subsequent evolution of the beam formation system with multiplexing of 2X beams decimated by 2X but now occupying the same band located arbitrarily in the sampling band is shown in [Fig.18], a figure in which each block indicated as "block of [Fig.17]" contains the elements of [Fig.17]. It is then clear that, as will be presented in the following paragraphs, the pointing operations of these 2% multiplexed beams must also be carried out in a multiplexed manner as a set of 2% delays, and as a possible set of 2K antenna aperture weightings. 3.3 — De fig En rm fi fai i ÿ When the number of receiving channels becomes significant, it is not possible to perform all the calculations in a single device, both because of the number of inputs / outputs this implies and for reasons of computing power. In this case, it is necessary to distribute all the calculations across several devices. The natural break in this decomposition of calculations occurs at the level of the summation of the contributions of the different channels, an associative operation by nature which also generally concentrates the data because the number of beams formed is very often much lower than the number of receiving channels, especially since, as soon as the useful bandwidth of a beam is reduced, it is advantageous to perform multiplexing as described earlier in this section.The receiving channels are then grouped into subgroups and the beamforming part that concerns them forms partial beams or subbeams, the additive recombination of which is also carried out in an additional grouping device, which can itself be split into several successive concentration stages if the number of inputs / outputs is too large. This is represented on the one hand in [Fig. 19], which corresponds to a sy- representation on the optics of the operation and on the other hand on figures 20 and 21 which each represent a part of the beam formation system which would thus be obtained (an example is proposed here with four sub-beam formers and two sub-beam combination stages). 2.3.4 - General case The elements that have just been described can easily be generalized, which leads to the training process now explained. In a general case illustrated by [Fig. 19], the process of forming Q beams is implemented by a beamforming system comprising P receiving channels, each linked to a respective antenna, where Q is a non-zero integer and P is an integer greater than or equal to 2. The Q beams are implemented by Q' beam groups calculated in a multiplexed manner by the same operator as described previously. For each receiving channel, the process comprises the following steps: - the establishment of two calculation methods by summing and subtracting the signal from the receiving channel and the same signal delayed by M periods, and - on the second calculation method (difference method), realization of a frequency shift of n / 2M by multiplicative application of a signal e(-jn:1 / M) (or etis#» depending on the notation used, both designating the same operation), n being the current variable ranging from 0 to M-1, - Applying FFTs to M points on each computation channel allows us to move into the frequency domain, and - an application of a multiplicative spectral correction function intended to equalize the reception channels, - by duplication are created the Q channels intended to form the Q' groups of multiplexed bundles also numbered q' between 1 and Q'. - Implementation of operations including possibly periodic sampling followed by resampling, according to the technique in Figure 15, to perform effective decimation by g of these Q' series of points maintained for 2Fs periods, which allows subsequent multiplexing of 7% beams. Such multiplexing makes it possible to transmit several beams formed in a given time and thus increase capacity. All these operations are intended to obtain Q=sum of or on q° ranging from 1 to Q' contributions of the receiving channel to the formation of the Q beams. Another way to present this process would be to use operations as proposed in section 2.2.1 (as shown in Figures 9 and 10). This equivalent description is not repeated below to avoid making this document too long. description. On each of these Q contributions, partially multiplexed into Q” processing channels, each comprising two calculation channels (even and odd frequencies of the spectrum), application of: a linear spectral phase delay and an amplitude weighting. Then, the processing channels from each of the reception channels are summed to form Q partially multiplexed bundles into Q' processing channels. This summation operation can be carried out in accordance with paragraph 2.3.3 by creating intermediate sub-bundles. On each of the two calculation paths of each of the Q” processing paths referenced by the index q': application of a multiplicative band-limiting function. application of an iFFT to M / 7%s" points multiplexed by 7Ks. - on the second calculation channels (odd frequencies), transposition of the spectrum by a frequency shift of n / 2M by multiplicative application of a signal e[+j floor(n / 24a"y1 / ( M / 3K)], n being the current variable ranging from 0 to M-1 and the floor function denoting the integer part function, On each of the referenced processing channels, the summation of the two computational channels ultimately forms the Q partially multiplexed bundles into Q? groups of bundles referenced by the index q' and multiplexed by kr, 3 — DESCRIPTION OF A CALIBRATION PROCESS As mentioned, in any beamforming process, the quality of the signal degrades when the receiving channels do not exhibit exactly the same response. It is also known that differences in response, both in spectral amplitude and spectral phase, are corrected prior to any beamforming. This correction is achieved by the beamforming system via the equalization stage using the c coefficients. These correction values ​​relate to the amplitude and spectral phase of the complex gain of each channel and are generally acquired through calibration. In principle, beamforming does not require that the channels be equalized, that is to say that their responses be brought to the same flat response, but simply that they be recalibrated with respect to each other, that is to say that the complex gain differentials between channels are compensated so that all the receiving channels are brought to the same average response. However, the equalization of lanes, which is therefore not a requirement associated with lane formation, can be justified elsewhere, for example to ensure that the appropriate processing (pulse compression) is carried out with the desired filtering function. Whether for equalization or simple recalibration, the principle of calibration consists of injecting the same test signal into the different channels and comparing the received signal to a reference. Depending on whether the goal is equalization or simple recalibration, the calibration philosophy regarding this reference may vary: "For equalization, which consists of calibrating each channel independently, comparison to the reference must be used to extract the response of each channel; for this, the reference, which is the test signal, must be known, but, since the calibrations of the different channels are independent, it is not required that the same signal be used for all channels, and "For recalibration, only the knowledge of the response differential between the channels is sought, so that knowledge of the calibration signal is not required, since each channel is compared to the other channels; it is sufficient for this to inject, without needing to know it, the same signal into two channels to access the differential between these channels. Regardless of the objective, whether equalization or recalibration, calibration involves comparing two signals. Traditionally, the signals used for calibration are sinusoidal signals, and their use provides gain, amplitude, and phase information to be realigned for each of the frequencies used. But it is also possible to use noise when a spectral analysis is available in each receiving channel; this allows the measurement of the transfer function differential between two receiving channels. For example, a method for determining the phase difference coefficient of two receiving channels, each comprising an analog-to-digital converter, can be considered. Each analog-to-digital converter is clocked at the same sampling frequency. According to industry standards, the receiving channels include, prior to sampling, filtering stages designed to limit the bandwidth of the received signals to a single Nyquist band of the function. sampling. The process includes at least one step of: - injection of a signal at the input of each of the receiving channels, the signal being an optionally white noise signal, to obtain output signals, - obtaining the output signal from each analog-to-digital converter, - spectral analysis, frame by frame, of each output signal, this spectral analysis allowing for a plurality of frequency channels, each channel being identified by an integer, - estimation, by integration over several successive frames, of the correlation in power between at least two channels of the same index from the spectrum of signals from each of the analog-to-digital converters, the phase difference between two receiving channels being carried by the phase of this correlation, and - on each of the reception channels, estimation, by integration over several successive frames of the spectra obtained, of the DSP of the signal, the amplitude differential between the two reception channels corresponding to the ratio of the DSPs thus estimated. The signal to be injected can be obtained by implementing the following steps: - the generation of a digital calibration signal, - the conversion of the calibration signal to analog, - the frequency transposition of the converted calibration signal, - the amplification of the transposed calibration signal, and - balanced distribution across different reception channels. Thus, such a process corresponds to using noise, not necessarily white, but which can be considered as the result of filtering white noise, allows us to trace back to the amplitude and phase differentials via the cross-correlation between channels of the same rank and via the DSP at the output of each channel (which is obtained as the variance of the output of the different channels). The spectral analysis output of one of the two channels can also be replaced by the spectral analysis of a reference digital signal constituting the test signal after digital-to-analog conversion; with prior calibration of the transmission part, this allows the absolute calibration of a channel for the purpose of equalization. In any case, whether the calibration is intended for equalization or recalibration, whether the type of signal used is coherent (sinusoidal) or not (noise), the calculation of correction coefficients requires raw data obtained by correlation and / or partial spectral analysis (a demodulation with low-pass filtering intended to calibrate a single frequency point is similar to a single-channel spectral analysis). Moreover, particularly because beamforming has, in the past, only been practiced in a narrowband application setting which reduces the controls to simple sets of complex gains (phase and amplitude) independent of frequency, calibration has, until now, been carried out in the spectral domain with sinusoidal test signals (one calibration point per working frequency). However, the need to form broadband beams requires that calibration also become broadband, which requires that the calibration signal itself be composed of several frequencies covering the working domain, whether sequentially or instantaneously. Two philosophies therefore exist regarding the nature of calibration. According to one philosophy, the spirit of the old techniques is preserved by staying in the spectral domain and replacing the sinusoidal test signal with a succession of sinusoidal frames (this technique is sometimes also referred to by the corresponding English term "step frequency"). This involves in particular setting up a calibration process sequence involving both an agile generator for the synthesis of the test signal and all the reception channels for the measurement. The sequencer must also handle the storage and transmission of data acquired during the measurement process to a computer. Synchronizing this process also takes into account the stabilization times of each frequency step, as well as the accurate timing of these measurements from one receiving channel to another. This last point is not a problem if all the receiving channels are located on a single circuit, but becomes one when these channels are distributed across multiple circuits. In contrast to this first philosophy lies the use of broadband noise in the sense that this type of signal instantaneously contains all the spectral components useful for calibration. In this case, two approaches are possible: either time-based or frequency-based. The time-domain approach involves calculating the covariance between a reference signal and a set of replicas of the received signal, each shifted by one clock cycle. This approach provides a time-domain response that would need to be deconvolved by autocorrelation of the reference signal, which is only feasible in practice if the signal is white noise. Therefore, this approach does not allow, in particular, for one channel to be used as a reference for another channel in order to measure a differential quantity. Thus, the frequency approach is clearly preferable; this approach is described previously with the aforementioned determination method. This approach describes the possibility of performing a differential measurement between two receiving channels, provided that these channels are equipped with spectral analysis. The technique consists of calculating, through frame-by-frame integration, the power spectral density p,(k) on each channel and the power correlation c,…(k) between channels of the same rank k on each channel; the amplitude / phase difference between channels is then obtained as the normalization of the correlation by one of the DSPs: €Cnm(k) / Pn(k) where Cmn(K) / Pnlk) = Cam(k}* / Pn(k) The advantage of this method is that it does not require any particular sequencing since all frequencies of the spectrum are processed simultaneously. Another advantage is that the noise does not need to be white; it is sufficient that it can be considered the result of filtering white noise, and that all the useful spectral components are present; thus a simple wideband noise generator (noise diode, chain of high gain amplifiers) is sufficient knowing that the analog receiving chains take care, in principle, of limiting its band to the useful band with a selectivity consistent with the need for instantaneous dynamics. It will be easily understood that the 10 training system is adapted to this second philosophy, since it involves a spectral analysis on each reception channel and the resolution required for calibration is, in principle, that of this spectral analysis. Furthermore, the presence of the first multiplier used for channel alignment can be leveraged, via a multiplexer, to perform the product inherent in the correlation calculation. The only additional resource required to ensure calibration is then a multiplexed integrator on M points (M being the length of the FFTs). Figure 22 shows a calibration device for the training system 10, the calibration device making some modifications to the receiving channels to ensure calibration: - a multiplexer M for each channel taking as input the channel equalization function (coefficients C(2m) or C(2m+1) omitting the channel reference) and a reference signal denoted R, and - a DC calibration circuit including a frame-to-frame integrator. These resources are duplicated on the even frequency channel and on the odd frequency channel. The calibration process described above allows for differential calibration between several receiving channels by correlating the FFT outputs. This process is performed autonomously as long as all receiving channels are located on the same circuit. If this is not the case, as has been mentioned, then it is necessary to "pass" at least one reception channel from one circuit to another, which poses both a synchronization problem and increases the need for communication channels between circuits. One solution to overcome this problem is to prefer absolute calibration, relative to a common numerical reference for all receiving channels. For this, we will have two elements: - a noise generator used for synthesizing calibration noise; the signal from this generalizer, or a decimation of this signal (it is desirable that the noise be generated at a higher rate so as not to be degraded by its transmission chain), constitutes the reference signal, and - a duplication of this reference generator on each of the circuits allowing the formation of sub-beams from a subgroup of receiving channels; the use of the replicas thus formed eliminates the need to broadcast the reference signal to all the circuits, which simplifies interconnections and thereby makes each circuit completely autonomous. To facilitate synchronization between different circuits, care will be taken to ensure that the calibration noise generated by this generator is synchronous with the FFT frames. A second aspect of the calibration process involves substituting a synchronization signal for the FFT signal during the calibration phase. Using the data path to carry this signal, which restores synchronization, allows it to be naturally directed to the grouping blocks. If grouping occurs via several sub-blocks, such a synchronization signal must be regenerated in each sub-block for the next, and finally, for the last sub-block, for the final block that performs filtering and time return. The calibration process as described therefore requires no specific resources in terms of real-time signal transmission: neither distribution of the common reference, nor exchange of signals between the different subbeam formation circuits. A very particular aspect of noise calibration processes via spectral analysis lies in the fact that the time lag between the measured channel and the reference signal tends to bias the measurement; this is due to the fact that the coherence between the two signals disappears at the edge of the frame. Thus, the main effect of this partial decorrelation is a multiplicative amplitude bias equal to 1 —T| / T where t is the time lag between the two signals and T the duration of the spectral analysis, here 2N;r7 / frcu- Other, less significant effects can also occur due to decentering, such as localized ripples primarily at the edge of the band. Even though this bias is similar for all receiving channels as soon as calibration is performed using a reference signal (and not between channels) and therefore does not impact the channel differential, it is still advisable, to reduce this effect, to perform calibration in two steps: "A first, short step is used to 'rough out' the result in terms of delay: an order of magnitude for this delay is determined from the average phase slope, then rounded to the nearest multiple of 1 / f:c. From this initial measurement, it is possible to digitally realign the signals using a realignment device placed upstream of the processing, and "a second step, longer and more precise, not or very little biased because it deals with signals that are time-recalibrated. This implies, of course, that a time-referencing device must be introduced at the beginning of the processing, before entering the spectral domain. Such a device, the schematic of which is shown in [Fig. 21], comprises an adjustable delay line, directly controlled by the result of the first calibration phase (delay rounded in number of sampling periods). This device is directly controlled by the delay to be applied, which can be either positive or negative between -29! and -1 + 291. When the delay to be applied is zero, i.e., in the initial position, the absolute delay applied by the device is 29"; the delays are applied relative to this absolute reference delay. Inverting the most significant bit allows this delay to be centered within the multiplexer's excursion. It should be noted that the calculation of the amplitude differential between the measured channel and the reference signal can also be done by calculating the square root of the ratio of their respective power densities. This has the advantage of not being biased in amplitude but, on the contrary, has the disadvantage of requiring an additional dedicated computing resource to estimate the DSP of the signal of the measured channel. By reformulating what has just been described in a similar way to what was done previously for the training process, this section deals with a calibration process for the training system comprising: 1) Exploiting the first part of the architecture described above (see previous sections) to operate: - a spectral analysis with 2M points distributed across two computing channels, respectively handling the even and odd frequencies of this analysis, and - a multiplicative function in the spectral domain, 2) a replication of the calibration signal generator on each of the circuits allowing the formation of sub-beams from a subgroup of receive channels, this generation being synchronized with the spectral analysis of the point |, 3) spectral analysis of this calibration signal by a device identical to that of point 1. 4) on each of the reception channels, the calculation of the power correlation between the spectrum of a reception channel and that of the calibration signal and the DSP of the calibration signal, and 5) the calculation of the correction to be applied to the track as a ratio between the DSP of the calibration signal and the power correlation, or, alternatively, the exploitation of only the phase of this ratio and the ratio of the DSP of the reference signal and the signal of the measured track. Since the replication of each calibration signal is synchronized with the spectral analysis of point 1 for each receiving channel, by construction, the calibration process ensures that the receiving channels are recalibrated with each other. 4— TRANSPOSITION OF TRAINING PROCESSES TO BROADCASTING Beam formation at emission, which could more accurately be called The formation of emission paths in a multibeam context is generally based on the same principles as those outlined for reception. We have several groups of 2% signals sampled at fçc# / 2% and multiplexed, the 2% sampling / multiplexing rank being specific to each group and each signal being intended for a transmission beam produced by an array of antennas. These different signals are complex and are included, by principle, in the undersampled band (i.e. [0 - frcy / 2*]); their transmission therefore requires that they be interpolated to be brought back to fecy on each transmission channel. However, due to the potential subsampling by 2K, an ambiguity of a multiple of fzçy / 2K, which is a fundamental ambiguity, remains regarding the actual position of these signals within the final sampling band [0 - f:cu]. Furthermore, it is possible to take advantage of the beamforming operation to perform a frequency shift specific to each beam group. In parallel with beamforming at the receiver, the formation of the various beams consists of applying certain filters to the signals: - an interpolation adapted to the different signals, then - the application of delays corresponding to the different beam-transmission channel pairs, accompanied by antenna weighting and followed by beam transposition, then - additive groupings intended to form emission pathways, then - the application of equalization corrections to the transmission channels. All of these operations correspond to the reversal of the operations performed in beam formation at reception. In this context, it is the FFT that is modified to adapt to the multiplexing rate of the input data; this modification, which is the dual of the modification made to the IFFT on [Fig.15], is the subject of [Fig.24]. The proposed multibeam emission path formation system is schematically represented in [Fig.25]. This system includes, in particular: * Partial channel formation for the multiplexed beams corresponding to the so-called "dual block of the [Fig. 17]" block, which is described in [Fig. 26]: this operation, which can be interpreted as a transposition, consists of integrating all the multiplexed beams that have been previously weighted and delayed. This operation is accompanied by sampling and resetting at the end of each slice (a 2% sample slice corresponding to a frequency of 2X multiplexed beams) and holding for the duration of one frame. "The interpolation (x2%) at fcu of the partial channels thus formed is described in [Fig. 27]: this step consists of inserting zeros for all ex- frequencies outside the useful bandwidth. Using the same notation as in the section dedicated to beamforming at the receiver, if n is the current index, the useful frequencies correspond to num < bit_reverse(n,2N) < Nyax WITH Nyin = M + k 2NX and nyax =m-1 + (k+1) 2N-K, Considering all the elements described so far, and again reformulating, a first embodiment of a corresponding process is a process for forming P multibeam emission channels from Q partially multiplexed beams on Q' beam groups indexed by q' (between 1 and Q') for multiplexing ratios equal to ak with Q = E,2 < The process includes, for each group of beams, the following steps: - generation of two channels by summing and subtracting the signal and the same signal delayed by M periods, - realization of a frequency shift of na(MnE4) by multiplicative application of a signal e[-jnx / (M / 24)], n being an integer from 0 to M / 3%-1, M / 7% being an integer strictly greater than 2, - implementation of operations on each of the calculation channels, the operations comprising: - the application of a discrete Fourier transform to M / aKe points multiplexed by aka to obtain Make points in the spectrum, each point of the spectrum corresponding to even indices of a spectral analysis to 2* M / gKy points of the signal resulting from the formed signal and being identified bijectively by an index which is an even number between 0 and 2* M / 9K-1, - applying a band-limiting function to points in the spectrum to obtain limited points, - the formation by duplication of as many replicas as there are emission channels to be formed (if this number is large, this replication can extend over several different computing boards), for each of the replicas formed on each of the computing channels and for each multiplexed beam: - the application of the delay to help form the emission path associated with the aftershock, - the application of amplitude weighting, - the partial formation of the emission pathway by integration of the oKs multiplexed points and maintenance over oKg periods ([Fig.26]), - the application of the interpolation process, which may be accompanied by a frequency shift through resampling and insertion of O ([Fig.27]), and At this stage, we have Q' in the spectral domain for each channel to be formed partially formed pathways. The process then includes summing the signals from these partial Q' channels to form a transmission channel available on a first computing channel for even frequencies of the spectrum and on a second computing channel for odd frequencies of the spectrum. For each of the calculation methods of each path formed: application of frequency correction for equalization of emission channels, application of an IFTTT to M points, on the second calculation method, performing a linear transposition of the spectrum, The process then involves a final additive grouping operation of the two calculation channels to obtain the time signal of the channel formed. Conclusion The described methods enable the execution, through a single device, of all operations associated with beamforming at reception and multibeam channel formation at transmission. To achieve this, they utilize specific FIR filter architectures that are driven within the spectrum. Thanks to the shift to the frequency domain, these architectures reduce the realization of any delay to a simple multiplication by a complex exponential with linear phase; they simultaneously ensure: - the filtering associated with the delay function, - at reception, Hilbert filtering for 1 / Q demodulation if the input signal is real and / or filtering of the useful band associated with the formed beam; at transmission, interpolation filtering in the case of narrowband beams, and - filtering associated with equalization of the receiving or transmitting channels. In this architecture, a pass through the spectrum is performed, and all the aforementioned filtering operations take place within this domain; this pass through the spectrum is carried out: at reception, at the level of each reception channel, the return in time taking place at the level of each beam formed, at the transmission, at the level of each beam, the time return is carried out at level of each path formed. When the useful bandwidth is smaller than the sampling bandwidth, it is possible, thanks to the inherent nature of FFT and IFFT architectures which consists of a decomposition into successive stages of radix-2, to multiplex several beams occupying the same useful bandwidth; this is done naturally via a simple adaptation of this architecture. At the receiver, the output of such a beamformer therefore makes it possible to have in- unlike a band beam of fçcy / 2 OR fgcy Depending on whether the input signal is real or complex (i.e., from an I / Q demodulation process with decimation by 2) or from 2% multiplexed band beams of f:cu / 2K. By multiplying the formation operations and output paths, it is also possible to simultaneously have on different outputs a set of beams as described but presenting different bands on each output n with the relation B, x N, = fecy / 2 WHERE frcy Depending on the type of input signal, B, being the band of the beams carried by the output n and N, their number. In transmission, this type of architecture is applicable in a dual manner to transmission with a few adaptations associated with the fact that beams are multiplexed at reception and de-multiplexed at transmission. Since this beam and / or channel formation is very wideband, the present methods are suitable for any receiving (respectively transmitting) system comprising several digital receiving chains whose combination allows improved listening in one direction (respectively several transmitting chains whose combination allows simultaneous transmission of several beams in different directions).< / e> < / k>

Claims

Demands

1. Method for forming Q beams by a computer of a system of beam formation comprising P own receiving channels each to receive signals from a respective antenna, each antenna having a respective aperture, Q being an integer greater than or equal to ] and P being an integer greater than or equal to 2, the process comprising: - the passage into the spectral domain of signals originating from signals coming from each antenna, - the implementation of operations in the spectral domain, the operations including time-tracking operations, time-tracking operations including: - for each reception channel, a generation operation by du- application of a contribution to each of the Q beams to be generated, - the implementation for each contribution of an operation introduction of a delay and a possible weighting operation of the antenna opening, and - an operation to sum the contributions to a bundle to be generated from all reception channels, the summoning operation being performed for each beam to be generated.

2. A training method according to claim 1, wherein the operations implemented in the spectral domain include, in In addition, a lane-levelling operation, the levelling operation being implemented before the timekeeping operations.

3. A training method according to claim 1 or 2, wherein the operations implemented in the spectral domain include, in In addition, a bandwidth limitation operation, the bandwidth limitation operation strip being implemented after the spotting operations.

4. Forming method according to any one of claims 1 to 3, in which the process comprises, for each receiving channel, the formation of two computational pathways, the input signal of the first computing channel being obtained by the setting implementing the offset of the signal received by the antenna by M points, M being an integer greater than or equal to 2 to obtain a shifted signal and the calculation of the sum of the received signal and the offset signal, and the input signal of the second computing channel being obtained by the implementing the shift of the received signal by M points to obtain a signal offset and the calculation of the difference between the received signal and the offset signal, this difference being translated into frequency by multi-application plicative of a signal eir#M, n being an integer from 0 to M-1.

5. A training method according to claim 4, wherein the passage in the spectral domain is achieved by applying a transform of Discrete Fourier transform with M points on each of the computational channels.

6. A training method according to claim 4 or 5, wherein the the process includes a step of returning to the time domain, the step returning to the temporal domain involving the application of a discrete inverse Fourier transform at M points on each of the calculation methods, the process involving a shift operation spectral of the second computational path by multiplicative application of a signal e+inPi / M and a summation operation of the two outputs of two ways of calculating.

7. Calculator, in particular programmable logic circuit, suitable for implement the operations of a training process according to one any of the claims | to 6.

8. Q-beam forming system comprising P receiving channels each one capable of receiving signals from an antenna respective, Q being an integer greater than or equal to 1 and P being an integer greater than or equal to 2, the training system includes a calculator according to claim 7.

9. Product computer program comprising instructions which, when the program is executed by a computer, they drive it to implement the operations of a process according to any claims 1 to 6.

10. Computer-readable media including instructions which, When executed by a computer, they lead the computer to to implement the operations of a process according to any one of the demands | to 6.