FR2999772A1 - METHOD FOR ACOUSTICALLY ACTIVE CONTROL OF MOBILE MICROPHONE (S) NARROW (S) BANDWIDTH (N), CORRESPONDING SYSTEM - Google Patents

Abstract

The invention relates to a method for the active acoustic control of interference noise with a narrow band (s) with a mobile microphone (s) intended to attenuate or even eliminate, in real time, said disturbing noise in a controlled area (9). ) of a working space (1) by generating at least one against-noise by a sound generation means (3) controlled by a control signal U (t) produced at the output of a calculation means receiving a input at least one measurement signal Y (t) coming from an acoustic sensor (6), said calculating means implementing a corrector (11) obtained from an acoustic model obtained following a first preliminary step of identification and a second prior step of synthesis of the corrector (11), at least the corrector being stored in the calculation means for use in real time in a third step of attenuation or real-time suppression of said disturbing noise. According to the invention, the area to be controlled (9) can be brought to move over time in the working space (1) and identification points (8) (... Pi .. in said working space (1), in the first step is identified for each identification point (8) the acoustic model, in the second step is synthesized for each model the corresponding corrector (11), and in the third step, in real time, means are used for selecting (12) at least one of the corrector (11) stored for the production of the control signal U (t) in order to at least attenuate said disturbing noise.

Description

The present invention relates to a method of active acoustic control of noise (s) narrow band (s) with mobile microphone (s) and a control system for such an active control. The active acoustic control systems make it possible to reduce / attenuate or even eliminate disturbing / undesirable noises by means of noise-generating means, for example acoustic transducers of the speaker type, excited by control signals produced by calculation means according to measurement signals obtained by acoustic sensors, for example microphones.

The calculation means implement active control algorithms which are known to exist in two main classes: - feedforward algorithms which require the use of a reference source correlated with the disturbing noise perceived at the level of the error microphone (s). This reference measurement serves to supply a filter whose output is the control signal of the correction loudspeaker (s). The coefficients of the filters are adjusted in real time by means of an adaptive device. The algorithms of the LMS series, for example Fx-LMS, belong to this class. The feedback algorithms or only the measurements of the error microphones are used as the measurement signal at the input of the algorithm, independently of any reference. The disturbing noises that the active control method and system of the present invention propose to reduce, or even suppress, are narrow-band noises and in practice quasi-single-frequency or in the case of several individualized frequencies, these frequencies are in determined frequency ratios, in particular integer multiple (s) as in the case of a basic monofrequency noise and its / its harmonics. These frequencies can be fixed in time or be variable. It has already been proposed a method and an active noise control system in the case where the narrow-band frequency (s) of the disturbing noise are variable with implementation of the control law described in the document W02010 / 136661 in the name of Mr Vau for "Method and device for narrowband noise suppression in a vehicle passenger compartment". Although this control law makes it possible to obtain very interesting results in terms of reduction / suppression of disturbing noise, it nevertheless presents a limitation as regards the positioning of the elements of the system and the size of the zone in which the noise disrupter is controlled. In practice, in the case of an application in an automobile, in which the noise frequencies are low and the passengers do not move in the passenger compartment, this is not a problem. Indeed, we know, see for example Elliot's article "A review of active noise and vibration control in road vehicles" (ISVR technical memorandum No. 981 - University of Southampton), that the zone of silence obtained by active control around An error microphone is spatially limited in a radius of the order of one-tenth of the acoustic wavelength corresponding to the narrow band to be rejected. Thus, in the case of a disturbance noise of low frequency, for example 50 Hz, the corresponding acoustic wavelength is, at normal temperature conditions, about 6.6 m and the radius of the zone of silence of the order of 66 cm. For a narrow band of high frequency, for example 1000 Hz, the zone of silence around the error microphone is thus only about 3.3 cm. Furthermore, in the case of active narrow-band control, in particular by a feedback technique, it can be shown that, in general, the phase margin of the closed loop constituted by the loudspeaker or loudspeakers, of the acoustic system and the error microphone or microphones, can hardly exceed ± 90 ° or is even often, in absolute value, less than 90 °, for the frequency of rejection of narrow-band noise. This means that for a given corrector, calculated from a nominal model obtained at a given point in space, for example by an identification method, the stability of the closed loop can no longer be ensured in a zone of radius exceeding about one quarter of the acoustic wavelength associated with the rejection frequency around the given point. It is therefore conceivable that if the active control device is designed to adequately ensure the rejection at a particular point that is called the nominal position and corresponding to the controlled area, there is a risk of creating instability, including effect Larsen, when said microphone moves away from this nominal position, the risk being even higher than the distance of the microphone from the nominal position increases. These constraints may explain why most active control applications involve low frequencies in the audible spectrum. However, in a number of applications, the person to be protected from disruptive noise by the active control system is likely to be mobile and move. This is for example the case of an operator in a noisy workshop who must move in a given area to perform the tasks assigned to him. In the case of broadband interference noise, the use of a headset (active or passive) can effectively protect the disturbing noises in question. The disadvantage of the helmet is to create an isolation of the operator with respect to acoustic signals of interest such as speech or an alarm siren.

In the specific case of noise disturbance narrow band (s), the wearing of the helmet has the effect of isolating the person who wears useful acoustic signals that are in frequency zones that do not correspond to that of noise disruptive. This has the consequence of making it impossible for the operator to communicate with his external environment, even though the acoustic attenuation of most of the audible spectrum is not necessary. The present invention proposes to solve this problem with an active acoustic control solution, intended to reject only the noise narrow band (s) concerned with implementation of microphone (s) of error disposed (s) nearby ears / each ear of the person to be protected from noise, microphone (s) for example maintained (s) by a device of the type "headband". In addition, as the person to be protected from noise is mobile in a space larger than the size of a zone of silence obtained by active control around an error microphone (the tenth of the wavelength as explained above), calculating means are implemented to take into account the movements of the person carrying the error microphones, thus to take account of the variations of position of the microphones. From a technical point of view, the position variation of the error microphones or microphones causes the variation of the transfer function of the speaker (s) to the microphone (s) of error. Also, the present invention proposes a control law intended to produce, by active control, attenuation, or even suppression, of at least one narrow frequency band of a disturbing noise at a given point in space in relation to at least one error microphone, which simultaneously guarantees a good level of rejection performance as well as a good degree of stability regardless of the location of the microphones or microphones in a given space which exceeds by its dimensions a quarter of the length of wave associated with the rejection frequency. The means of calculating correctors that can be implemented can be, depending on the case, the type of feedback ("feedback") or anticipation ("feedforward").

Basically, the invention implements a Multimodel control law for active noise control with narrow band (s) with mobile error microphones. It makes it possible to synthesize a set of correctors, each corrector allowing a correction for a given point of space and the control law can use a switching of the outputs of the correctors for selection of the most appropriate or can calculate a weighted sum of the outputs of the correctors to produce the control signal. Finally, it can implement two modes of detection of the relevant model / correctors, either with a system of position sensors of the person or the microphones it carries, for example by tags, exteroceptive sensors etc. .., with voluntary addition of a set noise in the calculation loop and selection of the most appropriate model / corrector. Thus, the invention relates to an acoustic active noise control method narrow band (s) intended to mitigate or even eliminate, in real time, said disturbing noise in a determined controlled area of a workspace by generation at least one against-noise by at least one sound generation means in said working space, said sound generation means being controlled by a control signal U (t) produced at the output of a computing means receiving as input at least one measurement signal Y (t) coming from at least one acoustic sensor disposed in said controlled area, said calculating means implementing a corrector obtained from an acoustic model obtained in the workspace following a first preliminary step of identifying said model followed by a second prior step of synthesizing the corrector corresponding to said model, at least the corrector being stored in the calculation means for r use in real time in a third step of implementing real-time attenuation or suppression of said disturbing noise. According to the invention, the area to be controlled can be made to move over time in the workspace and: - it previously determines distinct identification locations distributed in said work space and each corresponding to a point d identification of determined coordinate in said work space, - in the first step is identified for each identification point the acoustic model corresponding to the location of said identification point, - in the second step is synthesized for each model the corrector corresponding and stores at least the correctors corresponding to the identification points in said calculating means, and in the third step, in real time, is implemented means for selecting at least one of the corrector stored for the production of the control signal U (t) to at least attenuate said disturbing noise. It is understood, especially when the selection of the correctors is done on the basis of a positioning, that during the storage of the corrector and possibly models in the calculation means to be able, subsequently, to perform the third step, we keep, associated , the locations of the identification points, in practice their coordinates, in order to be able to make relations between positions in the workspace and the identification points of this workspace. The first prior identification step can be separated from the second prior corrector synthesis step, the templates and locations being stored for recovery in the second step. Alternatively, the first and second steps can be combined: for each identification point, the model is first determined and the corrector of this model is synthesized before moving on to the next identification point. This second solution may be preferable if one can test the model and its synthesized corrector before moving to the third step because if the corrector then proves to be defective during the test it will be easier to re-synthesize the corrector or even to make a new identification of a new model for the identification point in question. Otherwise, if the tests are done at the end of all the proofreader summaries, it will be necessary for a defective corrector to return to the corresponding identification point if a new model is to be identified. In various embodiments of the invention, the following means can be used alone or in any technically possible combination, are employed: -the workspace comprises only a disruptive noise source band (s) narrow (s), -the work space comprises several sources of interference noise narrow band (s), - the first prior step is performed in the means of calculation, - the first prior step is performed in a device separate computer from the calculation means, - the second preliminary step is performed in the calculation means, - the second preliminary step is performed in a computer equipment separate from the calculation means, - the position of the acoustic sensor / s defines the position of the controlled zone in which the disturbing noise is attenuated or even suppressed, the sound generation means are in one or more fixed positions in the working space, the medium ns of sound generation are loudspeakers, - the sound generation means are mechanical vibration generators, - the sound generation means are at the periphery of the working space, - the generation means are at the center of the work area, - the sound generation means are distributed in the work area, - the acoustic sensors are microphones, - the acoustic sensors are mechanical vibration sensors - the microphones are arranged on an accessory carried by a person to protect from noise, - the microphones are arranged near the ears of a person to protect from noise, - the microphones are arranged on a helmet or a headband, - the microphones are arranged on an accessory in relation to a person to be protected from noise, in particular a seat on which the person is seated, - the microphone (s) are placed on a headrest of the seat on which the person sits, - the selection means are controlled by a means for locating the current position of the area to be checked in the work space, the selection taking place as a function of the values of the distance of the current position of the zone to be checked with respect to each of the identification points, the acoustic sensor (s) are associated with active or passive tags detectable by the locating means, said selection means are controlled by comparison means responses of the workspace for a current position of the area to be checked and for each of the correctors synthesized at each identification point, the means for comparing the responses using estimators Ei calculated from the acoustic models identified and producing error signals ci (t) as a function of the control signal U (t) and of the measurement signal Y (t), the selection taking place as a function of the values of the error signals, - the control system is formed of the acoustic sensor (s), the calculation means and the sound generation means, - a setpoint or control noise b (t) is added in the measurement signal Y (t) or in the control signal U (t), where such a target noise is generated by at least one setpoint or control noise generator in said work space and is slaved in variance the measurement signal Y (t), the estimators are chosen from among the Luenberger observers, the Kalman filters, a simple model of the electroacoustic system, possibly augmented by band-pass filters, band-cuts or the like; selection select a single corrector at a time by switching between the correctors to produce the control signal U (t), - preferably, a minimum time is imposed between two switches, a selected corrector must remain selected at least said time This may for example be obtained by shifting the switching moment with respect to the detection of the need to change the corrector. The selection means select one corrector at a time or, then, a set of correctors whose correction outputs are combined together and, preferably, in a linear combination, to produce the control signal U (t), - the selection means selects a single corrector at a time and the switching of a first corrector to a second corrector is clear and is preferably shifted in time with respect to the detection of the need for the correction of the correction if said first corrector has not been selected / used for a determined minimum time, - a selection hysteresis is set implemented by a threshold detection in the selection means, the switching being done only if the action of the new corrector which will have to be c ommute is sufficiently greater than that of the current corrector, - during a change of corrector by the selection means, it imposes a minimum time determined between two switches and / or hysteresis is implemented, 15 - the correctors synthesized are selected from feedback correctors and anticipatory correctors, - the workspace is a room of a building, - the workspace is a part of a room, said part of the room corresponding to a course made by the person to be protected from noise in the room, 20 - the distribution of the identification points in the work space is chosen from a homogeneous distribution, the distances between adjacent identification points being identical - the selection in real time of a corrector is made continuously, the real-time selection of a corrector is made at predetermined times in the course of time, the determined moments in the course of time depend on one or more of the following conditions: detection of a change of position of the area to be controlled by the locating means, exceeding of the noise threshold measured in the measurement signal Y (t), a determined periodicity, a manual control. The invention also relates to an active acoustic control system for attenuating, or even eliminating, in real time, disturbing noise with narrow band (s) in a determined controlled area of a work space, said control system comprising at least one acoustic sensor, calculating means and at least one sound generating means, the calculating means outputting a control signal U (t) as a function of a measurement signal Y (t) coming from at least one acoustic sensor disposed in said controlled zone, the control signal U (t) being intended to cause the sound generation means to generate a counter-noise, said calculation means comprising a corrector obtained from an acoustic model obtained in the workspace following a first preliminary step of identifying said model followed by a second prior step of synthesizing the corrector corresponding to said model. The control system is characterized in that the calculation means comprises a set of correctors corresponding to identification points of the workspace, each corrector having been synthesized from a model identified at an identification point. given work space, said identification points being distributed in said work space, and said calculating means comprises means for selecting at least one of the correctors stored for the production of the control signal U (t) in order to at least attenuate said disturbing noise, and the calculating means is a programmable electronic unit comprising a computer program enabling the operation of said control system according to the method of the invention. More generally, the system comprises material means for carrying out the method of the invention. The invention also relates to a computer medium comprising a computer program for calculating the control system of the invention for implementing the method of the invention. The present invention, without being limited thereto, will now be exemplified with the following description of embodiments and implementations in connection with: Figure 1 which represents a workspace in which a person to 2 represents a diagram of the multi-model switching control law with its noise control loop acting on the acoustic system with selection of a corrector according to a position. current in which the noise must be controlled, Figure 3 which shows a diagram of the switching multimodel control law with its noise control loop acting on the acoustic system 30 with selection of a corrector based on comparisons of the responses of the acoustic system and correctors for each identification point, Figure 4 which represents a diagram of the multimodel control law with its noise control loop acting on the acoustic system with selection of correctors whose outputs are linearly combined with each other, the combination being based on comparisons of the responses of the acoustic system and correctors for each identification point, FIG. 5 which represents a monovariable loop composed of a transfer function system and on which noises are applied, FIG. 6 which represents a block diagram of a variance control of a noise for implementation in a loop of active control with feedback correction, Figure 6bis which shows a block diagram of a variance feedback control of a noise for implementation in an active control loop with anticipatory correction, Figure 7 which represents a diagram of the multi-model switching control law implementing a variance control, the selection of the corrector is made according to comparisons of acoustic system responses and correctors for each identification point, Figure 8 which shows a diagram of the multimodel control law with its noise control loop acting on the acoustic system with selection of correctors of which the outputs are linearly combined with each other and implementing variance control, the combination being based on comparisons of acoustic system responses and correctors for each identification point, and Figure 9 which represents a control loop. noise of the type shown in Figure 3 but in which a feedforward is implemented in place of the feedback correction of Figure 3, and Figure 10 which represents a control loop of the type of that of Figure 9 but with variance servo over Y (t). In order to facilitate the description which follows, it is assumed that the noise to be eliminated is a sound, that the perimeter in which the error microphones will move is a determined perimeter which is hereinafter referred to as "workspace" and that the speakers providing the production of the counter-noise or noises are in fixed positions in this workspace. The active control method of the invention implements signal processing from at least measurements (measurement signals, for example from microphones), to produce counter-noise by means of control signals applied to transducers. (eg speakers). The space in which the active control process operates is essentially analog in nature. Analogous signal processing could therefore be envisaged (analog computing means by linear electronics). However, the treatments / calculations to be performed are relatively complex and it is therefore preferred to use digital means for signal processing. Thus, the processing / calculating means are preferably programmable digital devices, for example computer equipment with a digital signal processor or computer / server processor type with interfaces adapted for converting the analog signals into digital signals and vice versa. . As a result, the initially analog signals from the acoustic system 10 are sampled in time due to digital acquisitions of these analog signals. The processed and produced digital signals are therefore sampled in the digital calculation means. In addition, auxiliary signal conditioning devices (filtering, pre-amplification, amplification ...) can be implemented. In order to present the invention in one of its fields of application, FIG. 1 has been considered a working space 1 having a certain acoustic behavior and comprising one or more sources of noise, a rotary milling machine 2 in this type. example. In the following one considers that one of the machines produces a disturbing noise but the invention can be extended to cases where machines, in different locations, produce alternately or all sets of disturbing noise. This disturbing noise is monofrequential or quasi-monofrequential, with possibly one or more harmonics or non-harmonic lines of the base frequency, and it must be attenuated or better eliminated at the level of the auditory system of a person 4 who can move, displacement that takes place in a determined path or not, in said working space 1. For this calculation means (not shown) calculates a control signal U (t) sent to speakers 3 disposed in the work space and generating a counter-noise that will attenuate / eliminate interference interference noise at the level of the auditory system of the person 4. It is thus creates a controlled area 9 from the point of view of the noise level of the auditory system of the person 4. The calculating means calculates the control signal U (t) as a function of a measurement signal Y (t) coming from a microphones 6 arranged at the level of the auditory system of the person 4, on a tight e head in this example. The controlled area is where the microphone (s) are located, and this is why these should be arranged in relation / close to the hearing system of the person. We will see more precisely later that the controlled area / to be controlled is in fact of reduced size compared to the workspace. The (electro) acoustic behavior of the work space with its noise source and the sound generating means (s) and the acoustic sensor (s) may be modeled and, in the following, will be referred to as " system "(for acoustic system) with numeral 10 in the Figures.

The position of the microphones, or the position of the person in particular if the microphones are in a fixed relationship with the person, in the workspace is determined in real time or near real-time by means of location, cameras 7 in this example. Thus, thanks to the locating means, it is possible to know the current position of the zone to be checked and to follow its displacement in the workspace 1. In FIG. 1, a few identification points P1, P6, etc. have also been represented. as examples of the set of identification points that we have to define in the workspace and whose utility will be explained in the following.

The implementation of the active control method with the calculation of the control signal U (t) in the calculation means passes through the realization of several steps described below: - First step: Identification of transfer function models on a mesh of points in the workspace. Since the working space, in all three dimensions, generally has a size in all its directions that is greater than a quarter of the wavelength of the frequency to be rejected, given the maximum phase margins inherent in any control law, a single identification at a particular point (called identification point) of the space of a model of the acoustic behavior of said space between, on the one hand, the control signal (s) U (t) (U being a scalar in the case of a single correction speaker and a vector in the case of several loudspeakers) intended for the loudspeaker (s) to generate counter-noises and, on the other hand, the measurement Y (t) (Y being a scalar in the case of a single microphone and a vector in the case of several microphones), is not sufficient to at least ensure the stability of a single control law in such a workspace. As a result, several model identifications are performed at a plurality of determined identification points of the space. The identification points are distributed in the workspace according to a specific law. For example according to a homogeneous distribution / equal according to the three (or less) directions of the space (regular mesh). The work space may be limited to a space following the person's journey if the journey is determined and the person does not leave it. As a result, a number of identification points P, 1 <in where n is the total number of said identification points which have been taken care of to determine the positions in this workspace, are determined in the workspace. . It is desirable that these identification points be sufficiently close to each other, in particular taking into account the radius of the stability zone of a control law around a nominal position (see the radius zone mentioned above). about one-quarter of the wavelength, this is for the maximum frequency of the noise to be monitored). In practice, we will use the radius of the stability zone of the control law to determine the distribution step (s) between the identification points. Thus, it can be given as an empirical rule that the distance between two adjacent identification points should hardly exceed one quarter of the acoustic wavelength associated with the maximum frequency to be rejected. In an advanced implementation mode, particularly in the case of a device with two microphones placed near each ear of the person, for example on a device of the headband type, one can take into account the rotation of the head with respect to the reference of the working space, for each position of the person in the space, in particular if the rejection frequencies can be relatively high (for example 1000 Hz, see the radius of the order of the tenth of the previously mentioned wavelength).

In general, the person to be protected from noise does not move up in the work area and therefore it is generally sufficient to distribute the identification points in a horizontal plane, parallel to the floor, and at the height of the auditory system. 'a person. In the following explanations, particularly in relation to the calculation methods, an identification point is defined by its Cartesian coordinates, or in variants, by its angular coordinates, with respect to the reference of the working space. The purpose of the identification is to experimentally determine a discrete linear model sampled at the period Te between the orders U (t) and the measurements Y (t). This model is usually expressed as transfer functions or state representation. The identification has two phases: 1) An experimental phase consisting of sending on U (t) a sufficiently rich signal, called "persistent input" in the field of automatic, intended to excite the acoustic system 10 of the space of work and acquire in data files the values of the components of U (t) and Y (t) at each period Te. 2) A phase of exploitation of the data files thus acquired by means of computer programs based on mathematical optimization techniques to obtain / identify models of the acoustic system 10. The algorithms 35 that can be used are numerous and one can refer to the Ljung reference book "System identification, theory for the user" Prentice Hall 1987. Depending on the method used, models can be obtained which are transfer functions or state representations.

For example, in the experimental phase, a signal of the SBPA type (pseudo-random binary sequence) (PRBS in English) can be used for excitation of the acoustic system 10. For example, in the exploitation phase, one can use the method presented in patent application WO2010 / 136661 already mentioned. This experimental phase and this exploitation phase are carried out for each of the identification points previously determined. Thus, at the end of these two phases, we have, for each identification point Pi, a discrete model dependent on the delay operator (4 / linked to the period Te, each model being noted thereafter mi (q-1.) or simply m, in order to simplify the notations - Second step: Synthesis of correctors for each of the models identified For each model: Li, obtained at the point P ', a synthesizer is synthesized whose goal is to ensure the rejection of a disturbance and its possible harmonics or non-harmonic lines at the frequency or frequencies of the narrow band or bands of noise concerned, while ensuring sufficient robustness of the closed loop so that the stability of the looped system remains assured when the error microphone or microphones are moved around the point P. It is understood that the amplitude of this possible displacement around the point Pi is actually limited compared to the size of the working space of the makes me constraints previously mentioned and which led to consider a set of identification points distributed in said workspace. We can implement the two types of correctors mentioned previously: by feedback ("feedback") or by anticipation ("feedforward (a) case of correctors by feedback (" feedback "): Depending on the nature of the models m, obtained during identification, transfer functions or state representation, the correctors can be expressed as transfer functions, with for example a corrector in two blocks of the R-ST type (see, for example, Landau's "Control of systems ", Hermès 2002) The corrector can also be expressed by means of a representation of state, notably multivariable, that is to say when the number of loudspeakers n ,, or of microphones ny is strictly greater than 1. In this case, the control law can be based for example on a control law by reconstructed status feedback (see for example the document De Larminat "Standard State Control" Hermès 2000). of synthesis one or multivariable feedback corrector, reference may be made to document WO2010 / 136661 already mentioned. The corrector can be further synthesized using Hoe control techniques or predictive control, etc.

The common point of all these methods of synthesis of linear correctors is that they incorporate a model of the perturbation to reject according to the principle of Wonham's internal model (Francis & Wonham "The internai model principle for linear multivariable regulators, Applied mathematics & optimization ", Vol2 No. 2,1975). In the case of a corrector expressed as a transfer function of type RST with the block T = 0, the control law is written: Y (t) s -) and the model of the electroacoustic system is based on a transfer function, that is: Y (t) = u -1 where is the delay operator of a sampling period. A diagram of such a system is given in Figure 5. The rejection of a harmonic perturbation at the frequency fpert (in Hz) is obtained by prespecifying a part of the block S, that is to say that it factorize in the form: S (q) if with: hrs ((7-1) = 1 - fpert - Te) - q-2 The calculation of t is done by solving the following Bézout equation: Si (q1 ) A (q1) R (qi) B (q1) _ P (q1) where F ", q-1) is the characteristic polynomial of the closed loop, chosen by placing poles of said closed loop. a corrector expressed as a state, the corrector is composed of a state observer, said observer being based on the state model of the system augmented by a model of the single-frequency disturbance (s). where the frequencies of the narrow bands are variable, it is possible to implement the control law described in document WO20 / 136661, mentioned above, which presents several possibilities of (b) case of correctors ("feedforward"): The anticipatory structure assumes that we have a reference signal sufficiently well correlated with the disturbing noise. In the case of narrow-band acoustic disturbances, noises are often produced by rotating machines with rotating axes, for which a measurement of the speed / frequency of rotation or the angular position of the rotary axis is possible. In this case, this speed / position measurement can be used to create a sinusoidal reference which, by definition, correlates well with the disturbing acoustic noise generated by the rotating machine. Among the usable algorithms, it is possible in particular to use the FxLMS algorithm, which includes a model of the transfer from U (t) to Y (t). Thus for each model we develop a specific Fx-LMS type corrector including the MI model. Thus, in the case of a feedforward algorithm, there is a reference signal r (t) correlated with the acoustic disturbance. The control law is: Lift ') - G (q-1 r (t) where G (q-1 "t) is a finite impulse response (FIR) filter and whose coefficients are adapted in real time by setting a recursive algorithm (recursive gradient algorithm): G = 0.) (t) + 01 (0 q-1 + 02 (0 .q-2) As well as the vector of the coefficients of G: And: f (t - where is the reference signal r (t) filtered by the transfer function of the model: f (t) (t) - The adaptation law is written: 0 (t + 1) 1 - At 0 1 / where A is a positive scalar, used to set the convergence speed of the algorithm, other correctors are possible, IMC-LMS etc. For these algorithm structures we can refer to Elliot's document, « For example, an algorithm of this kind has been described in Active noise control: A tutorial review by Sin M. KUO et al., Procedures of the IEEE Vol 87 No. 6, June 1999 At the end of this second step, o na n model-specific C1 ... creep decimals ml_mn The first model identification step and the second corrector synthesis stage are carried out before the next step and the results obtained are stored in a memory. a computer for use in the next step, third stage of implementation of the control device, which is carried out in real time. The first step and the second step can therefore be done once and for all for a given workspace that remains unchanged. - Third step: Implementation of the control device Since the microphones can move in the workspace, in practice the person wearing the microphones, it is necessary to know the current position (in real time or near-real) of the microphones. However, two embodiments can be implemented for this third step depending on whether the current position of the microphones / zone to be controlled in the workspace can be known / determined or not. This knowledge / determination of the current position can be obtained by a location means. If the current position can not be obtained, it is then necessary to implement additional means for determining the effect on the response of the acoustic system 10 of the current position of the microphones / microphones in the workspace and make a comparison with the models / correctors that were determined in the first step. In a particularly advanced variant, the two embodiments can be provided and the method / system can switch from one to the other in case one of the two is not operational, for example by that for certain positions. current, the location means is not operational. The first embodiment of this third step assumes that the current position of the microphone or microphones is obtained by means of a means of direct localization of the microphones or, alternatively, indirectly by the location of the person wearing the microphone (s). , the positioning of the microphones on the person being in general stable / fixed. During the first identification step, each model m, (and the corresponding corrector) is associated with a position (Cartesian and / or angular) in the reference of the workspace and whose coordinates are known a priori / determined from this first step. This identification point position in the first step can for example be obtained by the locating means which is normally used in the third step. In this first embodiment, the working space in which the wearer of the microphones is supposed to move is equipped with exteroceptive sensors, for example fixed to a ceiling or walls of said workspace. These sensors can be sonar, infrared, or camera (see Figure 1) or be radio frequency receivers. The person wears a headband or a helmet on which the microphones are fixed and the headband or the helmet is equipped with active beacons (for example: ultrasound, infrared, RF) or passive beacons (reflectors, targets, bar codes). .) by allowing localization by exteroceptive sensors. In variants, the tags can be worn by the person independently of a headband or helmet (eg a barcode on a badge on a work coat that can be read remotely and further allow identification or a system setting). It is understood that if the person has to move in a large workspace, the measurements of the microphones will be transmitted by means of calculation by wireless means (radio or infrared). In the case of a small workspace, it is possible to envisage a wired transmission of the measurements of the microphones. The actual location is typically by a triangulation method, by distance measurement or circular multilateration from angle measurements. The current position calculation methods can be based on time measurements or based on the direction of the signal.

Since the current position of the microphones is known, it is possible to select the corrector or correctors to be used according to this current position. This selection can implement more or less complex methods between the choice of a single corrector or several correctors whose effects (control outputs) can be combined according to various criteria (including weighting).

Thus, depending on the location of the microphone or microphones obtained by the locating means, a selection logic of a corrector c corresponding to the model mi is activated and the command U (t) sent to the speaker (s) is the output of said corrector C. This selection logic can operate by switching a corrector 25 of the entire list to the command U (t): We speak of switching-based multimodel control. The resulting command U (t) may alternatively be a weighted sum of the output of several correctors, this weighted sum can be obtained, for example, by means of fuzzy logic techniques, or else by methods evaluating the output of several correctors. probability of a model to be the correct model on which the corrector provides the best compromise robustness-performance of the control law. u1 (Let: noise-canceling speakers, e vector of the commands applied to the High-35 Let j - the vector of the error microphone measurements, Let be the vector of the commands coming from the corrector C. The proposed control law is of the type :) = tif. U (t) With: = 1 In the case where the control law is performed by switching, the / i 'are Boolean. If, on the other hand, the control logic is made by weights, the y's are positive reals and less than or equal to 1. More precisely, if we set the distance of the error microphone (s) at point P ,. In the switching control law, the index 0- such that p '= 1 is determined such that d, <d, j E [1, n] and o- # j. As a variant, the choice of the model (s) can be made not directly on the comparison of the distances d, but on the comparison of the distances weighted by a coefficient f 1, which is adjusted for example as a function of the robustness of the corresponding corrector, multiplicative on d , or any other function dependent on d ,. In this mode, as shown in Figure 2, there is a clear switch between the commands. The control loop on the acoustic system 10 comprises a number n of correctors 11 synthesized on the n models identified at n identification points and selection means 12 for selecting one of the correctors 11 to produce the control signal U (t ), the correctors receiving the measurement signal Y (t). The selection means 12 comprise a switch 13, a switching logic 14 and a means 15 for processing distances with respect to the identification points di (t) ... dn (t) obtained / calculated from the data provided by the means for locating the current position of the microphones / person. In general, it is necessary to keep correctors, even those not selected, up-to-date, in particular because the system is sampled and the calculations use at each period samples obtained at different sampling times. to be implemented. Thus, it is necessary for the different correctors to be used / evaluated constantly so that they are up to date with each calculation period.

In the case of a feedback control law ("feedback"), for each of the correctors, this means that at each calculation period it receives the control signal U (t) selected at this time. Thus, the correctors are used / evaluated even if they are not selected. On these questions, one can consult: Landau "Control systems" hermès 2002 p.388. In the case of a feedforward, the update relates to the vector 0 (t): for each corrector it is necessary for the vector 0 (t) of all the inactive correctors to be initialized on the vector of the active corrector. It should be noted that it is generally preferable to impose a minimum time between two commutations, which can be obtained by shifting the switching moment with respect to the detection of the need to change the corrector so that the current corrector has been selected. / used for a minimum time before moving to the new corrector. In one variant, this time shift is obtained indirectly by implementing a threshold detection in the selection means, the switching being done only if the action of the new corrector to be switched is sufficiently greater than that of the corrector current, in order to obtain a hysteresis. Indeed, the switching time plays an important role in the stability of the system as explained in the work of ID Landau et al., "Adaptive Control" in Chapter 13. In particular, stability can be guaranteed if the switching time is greater than a threshold. In the case where the /.1, can take a continuous value between 0 and 1, the values can be determined by assigning a weight relative to a proximity measurement of an identified model with respect to the current position of the microphones. of error with respect to the model m, (in fact with respect to the position at which the model concerned was determined during the first identification step). The weight in question must be maximum when the distance to a point P, becomes zero and tend towards 0 when this distance tends towards +: - The range of functions used to calculate the ja, is very broad. We give here a non exhaustive list: 1 d, (.1 = rit e The interest of this second way of proceeding consisting in a continuous variation of the is to avoid the abrupt commutations It is then however preferable that at the time of the synthesis of the correctors, the stability is ensured, which can be done by means of the concept of quadratic stability as presented in the document of Mr. Chadli "Multimodels in automatic" Hermès 2012. - The second embodiment of this third step assumes that the current position of the microphones or microphones can not be obtained by locating means, this second embodiment then implements a corrector selection method (s) based on the similarity of the current acoustic system with respect to In the first embodiment of the third step described above, it is assumed that the current position of the microphone or microphones is known by means of microphone location based, for example, on a method of triangulation. These methods, which are precise, have the disadvantage of being expensive and of inducing a complexification of the system. In this second embodiment of the third step, this locating means will be dispensed with. This second embodiment consists in comparing the behavior of the slave system for the current position of the microphones, with respect to those of the models 1q identified at the identification points P, and to choose the corrector calculated on the basis of the best-adjusted model relative to to the current behavior of the acoustic system 10. This command is called in the literature "adaptive multi-model control" as for example in the document Landau ID and all., "Adaptive Control" in Chapter 13. This second embodiment requires the calculation of Ei estimators, one for each model or corrector determined in the workspace. These estimators are then calculated before the third step once and for all for given models / correctors and these estimators are stored in a computer memory for use in the third step. As a function of the vector / control signal U (t) sent on the loudspeaker (s) and possibly as a function of the vector / measurement signal Y (t) (in practice: the measurement signal Y (t) is considered if uses a state observer (for example Luenberger or a Kalman filter) but in the case where a simple model of the system is used as the estimator, which is equivalent to a zero gain observer, the measurement signal Y (t) is not necessary), estimators E, based on each of the models M1, make it possible to estimate an output Y, which is subtracted from the real output of the system Y to form an error signal = Y - The estimator used It may in particular be an observer of Luenberger, regulated for example by a pole placement technique, of which, in a particular case, the gain may be chosen to be zero and, in this case, the estimator E, is then the model M1. This estimator can also be a Kalman filter, set based on statistical considerations, notably covariance matrices of state and output noise. It should be noted that for calculating the estimators, the mi model used for the synthesis of an observer can also be increased by a model of noises or disturbances, in particular the harmonic perturbation to be rejected. The diagram of the control law with its noise control loop and implementing the estimators is given in FIG. 3. This time, the selection means 12 are controlled by means of comparison of the acoustic responses of the space. for a current position of the area to be checked and for each of the correctors 11 synthesized at each identification point, the means for comparing the responses using E1 estimators 17 calculated from the acoustic models identified and producing signals of error e1 (t) as a function of the control signal U (t) and the measurement signal Y (t). The selection means 12 comprise a switch 13, a switching logic 14 and a means 16 for processing the error signals ei (t). The error signals and from the estimators are processed to express variables gi used for decision making regarding switching. In general the variables pi are calculated from the absolute value or the square of the Et 25 or, more generally, from any even function on the E, calculated values which are subsequently filtered by one or more low-pass filters for example in the following form: (i) Jf = 0 (in the monovariable case) Pt (t) () 'aL ° + fli (in the multivariable case) a being a constant (scalars in the monovariable or matrix case in the multivariable case) 8 being also a constant. The selection of the COE corrector whose control is actually chosen on U (t) among the n correctors Ci is done by evaluating the coefficients! Li by looking for the smallest, that is to say the one who by his behavior shows that the corresponding model / corrector comes closest to the actual / current acoustic system, which means that we choose cr such that: <pi, j E [1, and a # j. As indicated above, it is generally preferable to impose a minimum time between two successive commutations or else to set up a hysteresis. In a variant of this second embodiment of the third step and as shown in FIG. 4, the switch is replaced and the control U (t) is calculated as the weighted sum of the control outputs U (t) of each corrector such as that: U t) = p U.1 (t) +112 'U2 (0 - With: E In this Figure 4, the selection means 12 are such that the error signals of the estimators 17 processed in the means of processing of the error signals 16 make it possible to produce the weighting factors pi (t), p2 (t), .i (t) which are multiplied 18 to the control outputs of the respective correctors 11 and the orders thus weighted are added together 19 between they are used to produce the control signal U (t). Note that in the multimodel command, it is preferable that the setpoint be persistent, that is to say that it has a sufficiently rich content of a point of frequency view, in order to be able to determine between all the pre-established models / correctors i which is closest to the behavior of the current acoustic system. This condition is naturally at odds with the principle of active control where the setpoint on Y (t) is by definition zero. Also, in order to allow this discrimination of the correct corrector or correctors in case of linear combination of their outputs, to produce the control signal, it is necessary to voluntarily add a little noise in the closed loop. This added noise is called "set noise" to differentiate the disturbing noise to be attenuated / suppressed, since it is similar to a kind of "setpoint". This setpoint noise should be as low as possible so that it does not over-add itself excessively to the residual noise (ie the noise remaining when the active control system is operating, thus resulting in the control 35 in the area to be tested) on Y (t). Since the residual noise level, that is to say in a closed loop, is not predictable, since it depends on the level of the disturbing noise, it is necessary to be able to control the level of the setpoint noise added at the level of the noise level. Y (t) by means of a feedback loop adding to the initial loop.

In the following we call this loop "external loop", whose objective is to control the variance of the setpoint noise added. Consider a monovariable loop composed of an input transfer function system U (t) and an output Y (t) output which is subjected to a noise w (t). This noise w (t) is supposed to be modeled as the output of a formant filter attacked at its input by a white noise. The loop contains a monovariable transfer function corrector whose input is Y (t) to which is added an additional "white" noise b (t). We assume w (t) resulting from a filtered white noise e (t) by a forming filter (z-1), for example with a finite impulse response to simplify the calculations and explanations. We have (A, B, R, S and F being the preceding functions in z-1 just like P with P = A.S ± B.R): Y (t) = A.S.F (t) + n ?. b (t) Assuming that the white noises e (t) and b (t) are different and decorrelated and putting C = ASF and N = BR, we have: C (z ') C pz N (z) PP (z -1) where c17 is the spectrum of the signal x (t). This expression is valid only because of the decorrelation between the two white noises e (t) and b (t). Hence the following expression linking the variances 2 E E [13] 2 Where is the norm 2 of a transfer function and E the variance of the signal x (t).

We can note in the previous relation, the linearity that there is between the variance of b (t) and the variance of y (t). In the problem which concerns us consisting in the servo-control of the variance of Y (t), this linearity relation is very useful, since it makes it possible to propose a simple linear control law imposing the variance of b (t) from of the estimated variance of Y (t): E [13 In this last equation E [y] 2 is the estimate of the variance of Y (t), calculated by taking the square of Y (t) and filtering it by a low pass filter.

In static regime, and supposing that the estimate E is indeed equal to Er *, we have the relation: 1 E [3] 1 -1-Kp P The gain Ki, which can be chosen understood for example between -0.5 and 0 so that the variance of Y (t) is increased by the outer loop. The generalization of this external servo loop of the variance to the multivariable case is trivial because Kp becomes a diagonal matrix. The schematic diagram of this variance control, applicable to the case of a feedback control law (Figures 7 or 8), is given in Figure 6.

We find, starting from the left and the measurement signal Y (t), a multiplier of said measurement signal on itself, a low-pass filter making it possible to calculate -E [y] 2 and a calculation module according to the equation -I. followed by the extraction of the square root. The resulting parameter is multiplied to white noise from a variance white noise generator 1 to produce the variance-tuned noise b (t). In the case of a feedforward law, the servocontrol of the variance É- [y] 2 can only be done by injecting a command noise b (t) at the command level. U (t). The control law can also be proportional according to the equation: 1 E 1) 2] _ Kp. Where: 2 is the square of standard 2 of the active model transfer function. The schematic diagram of this variance control, applicable to the case of an anticipatory control law (FIG. 10), is given in FIG. 6 bis.

The scheme of the multi-modal switching control law implementing variance control for feedback correctors is given in Figure 7. In this Figure 7, a single corrector 11 is selected by switching. 13 but it is understood that it is also possible to implement a linear combination of the control outputs of the correctors 30 to produce the control signal U (t) as has been done in the control system in connection with Figure 4 and as is moreover shown in FIG. 8, which also relates to feedback correctors ("feedback". The diagram of the multi-modal switching control law implementing the variance control is given in FIG. 7. In this FIG. a single corrector 11 is selected by switching 13 but it is understood that it is also possible to implement a linear combination of the control outputs of the correctors for produce the control signal U (t) as has been done in the control system in relation to Figure 4 and as is also shown in Figure 8. It should be noted that in the description that has just been made, we have considered correctors using feedback algorithms or only the measurements of the error microphones are used as a measurement signal at the input of the correctors, independently of any reference. However, the correctors that can be used in the context of the invention can also implement feedforward algorithms that require the use of a reference source correlated with the noise perceived at the level of the error microphone or microphones. . This would be reflected in the corresponding Figures of the application by adding a reference signal input to each of the correctors. Thus, for example, there is shown in Figure 9 a noise control loop of the type shown in Figure 3 but in which a feedforward correction is implemented. As a result, a reference signal 21 depends on the frequency of the disturbing noise to be attenuated / suppressed between inputs in each corrector. This reference signal 21 may come from the measurement signal in which one can search for the residual / reappearance of the disturbing noise but, preferably, it comes from a sensor in relation to the source of noise, for example a speed sensor. rotation of a rotating machine. FIG. 10 also shows an implementation variant of the type of that of FIG. 9, but in addition with enslavement of noise variance on Y (t), therefore in the case of feedforward correctors. It will be understood that the methods described above in connection with the active control of acoustic noises with narrow frequency band (s) can be applied to other types of noise, such as mechanical vibrations transmitted in material structures. , for example a motor frame, the wings of an aircraft ... In these latter cases, the noise generators are adapted to the application: for example the loudspeakers are replaced by mechanical vibration generators and microphones by mechanical vibration sensors. Likewise, the working space in which active noise control is desired is not limited to an air space but may also correspond to a volume of water, in particular seawater, for suppressing noise in such a liquid medium . Here again, the noise and sensor generators are adapted to the application, for example by using SONAR type means (generator and receivers adapted to the liquid medium).

Claims (11)

REVENDICATIONS1. Acoustic active noise control method with narrow band (s) for attenuating in real time said disturbing noise in a controlled area (9) of a work space (1) by generating at least one counter-noise by at least one sound generation means (3) in said working space, said sound generation means (3) being controlled by a control signal U (t) produced at the output of a computing means receiving in input at least one measurement signal Y (t) coming from at least one acoustic sensor (6) disposed in said controlled area, said calculating means implementing a corrector (11) obtained from an acoustic model obtained in the workspace (1) following a first prior identification step of said model followed by a second prior step of synthesis of the corrector (11) corresponding to said model, at least the corrector being stored in the calculation means for use e n real-time in a third step of implementing the real-time attenuation or suppression of said disturbing noise, characterized in that the area to be controlled (9) can be caused to move over time in the space of work (1) and: - separate identification locations distributed in said working space are determined beforehand and each corresponding to an identification point (8) (... Pi ...) of coordinate determined in said space of work (1), - in the first step is identified for each identification point (8) the acoustic model corresponding to the location of said identification point, - in the second step is synthesized for each model the corrector (11) corresponding and stores at least the correctors corresponding to the identification points (8) in said calculating means, and in that in the third step, in real time, selector means (12) are used. minus one corrector (11) stored for producing the control signal U (t) to at least attenuate said disturbing noise. 2. Method according to claim 1, characterized in that said selection means (12) are controlled by a locating means (7) of the current position of the area to be checked (9) in the work space (1) , the selection taking place according to the values of the distance from the current position of the area to be checked (9) with respect to each of the identification points (8) (... Pi ...). 3. Method according to claim 2, characterized in that the acoustic sensor (s) (6) are associated with active or passive tags detectable by the locating means (7). 4. Method according to claim 1, characterized in that said selection means (12) are controlled by means for comparing the responses of the working space for a current position of the area to be checked and for each of the correctors (11). ) synthesized at each identification point, the response comparison means using estimators Ei (17) calculated from the acoustic models identified and producing error signals ci (t) as a function of the control signal U ( t) and the measuring signal Y (t), the selection taking place according to the values of the error signals. 5. Method according to claim 4, characterized in that we add a setpoint or control noise b (t) in the measurement signal Y (t) or in the control signal U (t) or is generated such a set noise by at least one setpoint or control noise generator in said work space (1), and in that the measurement signal Y (t) is slaved into variance (20). 6. Method according to one of claims 4 or 5, characterized in that the estimators (17) are selected from Luenberger observers, Kalman filters or a simple model of the electroacoustic system optionally increased bandpass filters, cuts -bands or others. 7. Method according to any one of the preceding claims, characterized in that the selection means (12) select a single corrector at a time or, then, a set of correctors whose correction outputs are combined together and, preferably in a linear combination to produce the control signal U (t). 8. Method according to claim 7, characterized in that during a change of corrector (11) by the selection means (12), it imposes a minimum time determined between two switching and / or hysteresis is implemented. 9. Method according to any one of the preceding claims, characterized in that the correctors (11) synthesized are selected from feedback correctors and anticipatory correctors. 10. Acoustic active control system for attenuating in real time interfering noise at narrow band (s) in a controlled area (9) of a work space (1), said control system comprising at least an acoustic sensor (6), a calculating means and at least one sound generating means (3), the calculating means outputting a control signal U (t) as a function of a measurement signal Y (t) from at least one acoustic sensor (6) arranged in said controlled zone, the control signal U (t) being intended to cause the sound generating means (3) to generate a noise counter, said calculation means comprising a corrector (11) obtained from an acoustic model obtained in the workspace following a first prior identification step of said model followed by a second prior synthesis step of the corrector (11) corresponding to said model, characterized in that the computing means co mporte a set of correctors corresponding to identification points of the workspace, each corrector having been synthesized from a model identified at a given identification point (8) of the workspace (1) said identification points (8) being distributed in said working space, and said calculating means comprises means for selecting at least one of the correctors (11) stored for producing the control signal U (t) in order to at least attenuating said disturbing noise, and in that the calculating means is a programmable electronic unit comprising a computer program for operating said system according to the method of any one of claims 1 to 9. 11. Computer support comprising a computer program for calculating the control system according to claim 10 for carrying out the method of one of claims 1 to 9.