FR2983335B1 - METHOD AND DEVICE FOR CONTROLLING AN ACTIVE NOISE REDUCTION SYSTEM - Google Patents

Abstract

The method is applicable to a vehicle (12) in which an active noise reduction system (15) has one or more speakers (13) for producing one or more noise-canceling output signals and a or a plurality of microphones (11) for receiving one or more feedback signals which quantify the noise reduction obtained. To carry out the method, the vehicle (12) comprises a control device (23) which comprises return verification means arranged to establish a return diagnosis by analyzing at least one return signal (u) in combination with activation means (24) and neutralization of the active noise reduction system (15) controlled to activate respectively disable the active noise reduction system following a diagnosis of positive feedback, respectively following a negative return diagnosis.

Description

The invention relates to a method for treating the occurrence of a malfunction in an Active Noise Control (ANC) system.

More specifically, the invention relates to a method for controlling an active noise reduction system, for example in a passenger compartment having one or more loudspeakers for producing one or more sound output signals that oppose noise and a or more microphones for capturing one or more feedback signals that measure the noise reduction achieved. The invention also relates to a device and a computer program for implementing said method.

Active noise reduction is an old subject. Henri Coanda thus described in 1930 a method of protection against noise in his patent FR722274 consisting in obtaining a zone of silence by interference of sound waves. The idea was good but not so obvious to realize. A little later in 1934, Paul Lueg limited his patent US2043416 to the simplified case of a pipe in which he placed a speaker downstream of a sound source with a microphone placed between the source and the speaker for capture the sound to be removed before it reaches the speaker.

Despite numerous scientific developments since that time, particularly in the field of signal processing, the use of this invention, however old, struggles to become democratized, in other words to become accessible to a wide audience. The disadvantage is that, in spite of a strong current control of theoretical noise reduction, the slightest defect or unforeseen event on the equipment or the environment of a device for reducing sound waves per generation appropriately out of phase, has consequences contrary to the desired effect, namely those of adding the energy of the phase-shifted sound waves to those of the source instead of cutting them off, with the effect of increasing rather than reduce the noise.

Among the many attempts since then to try to overcome these inconveniences, there may be mentioned, for example, that disclosed in US Pat. No. 7,334,064, which relates to microphone monitoring to stop the production of phase-shifted sound waves when the signals delivered by the microphone retain the same sign for a predetermined duration.

This recent solution as other known solutions, is not comprehensive enough to provide complete satisfaction.

In order to overcome the drawbacks of the prior art, the subject of the invention is a method of controlling an active noise reduction system, remarkable in that it comprises: a step of activating the system of noise reduction; active noise reduction; a return verification step which establishes a return diagnosis by analyzing at least one return signal and which allows a re-execution of the activation step following a positive feedback diagnosis; a neutralization step which is executable following a negative feedback diagnosis and which deactivates the active noise reduction system.

In particular, the return verification step includes a low detection substep which calculates an operating return criterion and renders the return diagnosis negative when at least one return signal does not satisfy a minimum value of the return criterion. operating.

More particularly, the operating return criterion corresponds to a plurality of variations of the feedback signal over a given time interval.

Advantageously, the return verification step comprises a high detection sub-step which renders the return diagnosis negative when at least one feedback signal is not less than a maximum amplitude value.

Preferably, the control method comprises an output verification step which establishes an output diagnostic by analyzing at least one output signal, which allows a re-execution of the step following a positive output diagnostic and which makes the neutralization step executable as a result of a negative output diagnostic so as to deactivate the active noise reduction system.

Specifically, the output verification step includes a high detection substep that negates the output diagnostic when at least one output signal is not less than a maximum amplitude threshold.

Also particularly, the output verification step includes a substep of amplitude limiting of the output signal to a maximum threshold.

Particularly advantageously, the method comprises a moderation step which allows a re-execution of the activation step following a diagnosis of negative output or feedback until a sufficient number of an output diagnostic or negative feedback occurrence, which triggers the disabling step when there is a sufficient number of output diagnostic or negative feedback occurrences.

More preferably, the control method comprises a state monitoring step of an environment of the active noise reduction system that allows activation of the active noise reduction system only when said environment is in a state compatible with said activation. The invention also relates to a device for controlling an active noise reduction system, particularly in a cabin, comprising one or more loudspeakers for producing one or more sound output signals that oppose noise and a or more microphones for capturing one or more feedback signals that quantify the noise reduction obtained. Remarkably, the control device comprises: - return verification means arranged to establish a return diagnosis by analyzing at least one feedback signal; and activation and deactivation means of the activated noise reduction system for activating or respectively deactivating the active noise reduction system following a positive feedback diagnosis, respectively following a diagnosis of negative return.

In particular, said return verification means are arranged to calculate an operating return criterion and to make the return diagnosis negative when at least one return signal does not satisfy a minimum value of the operating return criterion.

Advantageously, said return verification means are arranged to make the return diagnosis negative when at least one feedback signal is not less than a maximum amplitude value. The invention further relates to a computer program, comprising program code means for performing all or part of the steps of the method which is the subject of the invention, when said program works on a computer, in particular on a computer. embedded calculator. The invention particularly relates to a computer program product, comprising program code means, stored on a computer-readable medium, for implementing the method according to the invention, when said program product runs on a computer. The subject of the invention is in particular a vehicle in which an active noise reduction system in a passenger compartment comprises one or more loudspeakers for producing one or more noise-canceling output signals and one or more pickup microphones. one or more feedback signals which quantify the noise reduction obtained, the vehicle comprising a control device according to the invention. The invention will be better understood using exemplary embodiments of a device according to the invention with reference to the accompanying drawings, in which: FIG. 1 is a schematic view of a system on which the invention is applied; Figure 2 shows process steps according to the invention; Figures 3 and 4 are details of steps of the method according to the invention; Figures 5 to 7 are behavioral curves obtained in certain steps of the method.

Figure 1 shows a motor vehicle referenced 12 which comprises an active noise reduction system (ANC) which some of the elements shown below are shown outside the vehicle 12 simply for ease of explanation.

One or more microphones 11, generally three, each picks up an Umic error signal at a predetermined delivery point of the vehicle 12. Each delivery point is predetermined during the design and testing phases of a vehicle prototype in such a manner. to achieve optimum noise reduction at the ears of the driver and preferably each occupant. Thus the microphones can be placed closer to the driver's head and each occupant or at locations of the vehicle geometrically correlated with noise amplitude nodes of noise in the vicinity of the occupants' heads.

One or more boxes or speakers 13 whose number generally varies between four and five, receive uhp control signals from an electronic device 14 shown in dashed lines.

The electronic device 14 comprises an electronic card 15 which houses an ANC algorithm for active noise reduction, an audio device 16, for example radio, which can also independently include multimedia and / or navigation functions and a mixer 17. The algorithm hosted by the electronic card 15 generates a signal uc for each speaker 13 to reduce or eliminate the noise in the cabin 20 of the vehicle 12. The noise that is to be suppressed is a noise frequency or spectrum of frequencies determined by a measurable or calculable RPM rotation of a vehicle member such as for example but not necessarily of a power train 21. Sounds coming from other sources such as for example the tread of the wheels 22 on the roadway or a sound emission device, could then also be reduced or eliminated provided that they are transmitted at the same frequency or at the same time. One or more frequencies of the controlled spectrum. The audio device 16 generates a signal ur for one or more of the speakers 13 to generate sound in the passenger compartment 20 of the vehicle 12, for example that of a radio broadcast, the music of a multimedia support or messages sound navigation, vehicle operating status or telecommunication.

The electronic card 15 and the audio device 16 may be contained in the same housing of the electronic device 14 or in separate boxes. The ANC algorithm can also be hosted in an on-board computer shared with the audio device 16 in the form of one or more programs executable by the on-board computer.

The mixer 17 generates uhp signals which result from the signals uc and ur to be routed each to that of the speakers 13 which are intended uc signals and ur respectively combined two by two.

The electronic card 15 is connected to a vehicle signal multiplexing bus 18, for example a CAN (Controller Area Network) bus, in order to read in real time various information on the state of the vehicle comprising in particular the engine speed. RPM (acronym for Revolutions Per Minute) expressed in revolutions per minute. This information can also be received wireline from point to point. The advantage of the multiplexing bus 18 is to make accessible to the ANC algorithm real time information on many other states of the vehicle such as the opening state of a door or window or the engine running condition. The advantage of the on-board computer for carrying out the functions of the electronic card 15 is to provide a great flexibility in the progressive operation of the states available on the bus 18.

The electronic card 15 or the on-board computer also receives a supply voltage Va of the electronic components from a battery 19 of the vehicle 12. The ANC algorithm uses, in a manner known in the other way, the theoretical and practical knowledge of active noise reduction material without the need here to expose all the details.

It is simply recalled that currently there are essentially two types of ANC algorithms for active noise reduction, feedforward algorithms and feedback-based algorithms, usually referred to as feedback loops. reaction.

As for example in French in the article by Emmanuel Friot "An introduction to active acoustic control" of October 6, 2005, especially on pages 16 to 21 and 38 to 41, or exposed in English in the article by SJ Elliot " December 2008, especially on pages 12 to 14, the predictive action consists mainly of enslaving a loudspeaker sound to a noise reference, measured by a first microphone or estimated depending for example on a motor speed. A second microphone is used to adapt in real time the transfer function of the servo so as to obtain a noise cancellation measured by the second microphone.

As for example stated in the international patent application WO2010136661, the retroactive action consists mainly in slaving a loudspeaker sound on an error with respect to a zero reference, as measured by a microphone. A favored noise frequency is used to adapt the transfer function of the servocontrol so as to obtain cancellation of the error at said favored frequency. The adaptation of the transfer function is modifiable so as to follow the changes in the preferred frequency that can change, for example depending on the rotational speed of the engine. The invention can be used for both the predictive and retroactive action algorithms and for any other type of ANC algorithms for active noise reduction, notably by producing a sound well at a noise frequency that results from the engine rotation speed. . Nevertheless, a retroactive action type algorithm is preferred. The inventors have noted that the predictive algorithm requires constantly adapting the transfer function to any variation measured by the second microphone. But these adaptations require calculations that are heavy to perform in real time. The inventors have noted that, on the other hand, retroactive algorithms do not require the transfer function to be constantly adapted to any variation measured by the microphone which conventionally acts as a feedback function with a given transfer function in a closed control loop. The inventors have thus found it advantageous to adapt the transfer function to a favored frequency which corresponds, for example, to the fundamental frequency or to the second harmonic of the noise generated by the motor. It is then possible to pre-calculate adaptation parameters of the transfer function for a series of favored frequencies and then quickly select in operation the adaptation of the transfer function which corresponds to the frequency favored as a function of the speed of transmission. motor rotation observed in real time.

The transfer function involves one or more frequency response functions between the uhp signal (s) sent on the loudspeakers 13 and the Umic-o signal (s) measured on the error microphones 11.

It is possible to evaluate a frequency response function at a value FRF (f) as a function of a frequency f by sending a signal UHp (f) to or to one of the loudspeakers 13 in the absence of noise and by collecting the Umicro signal (f) from each microphone 13, using the formula: FRF (f) = umicro (f) / uhp (1)

The frequency response functions depend on the audio reproduction system comprising speakers and loudspeakers 13, the acoustic cavity formed by the passenger compartment 20 of the vehicle 12 and the error microphone or microphones 13 including the sensitivity.

It will be noted that the electronic device 14 represented in FIG. 1 comprises a device 23 for controlling the active noise reduction system, in particular the electronic card 15.

The device 23 comprises on the one hand one or more output couplers for transmitting to the mixer 17 one or more sound output signals ue to the speaker or speakers 13 to oppose the noise in the passenger compartment 20.

The device 23 further comprises one or more input couplers connected to the microphone or microphones 11 for receiving one or more feedback signals for transmission to the ANC algorithm, for example hosted on the electronic card 15 to quantify the reduction of noise obtained. The device 23 may also include an input coupler connected to the battery 19 to observe the voltage level and one or more input couplers connected to the bus 18 to read the engine speed RPM and different states of the vehicle 12 relative to example when opening the doors or windows, the vehicle load acceleration, deceleration, uphill or downhill.

Return verification means are arranged in the device 23 to establish a return diagnosis by analyzing at least one Umic- back signal.

The return verification means may perform several analyzes of the return signal.

A first possible analysis consists in calculating an operating return criterion C1, in other words non-zero return specific to the use of the operations performed by the algorithm ANC. The existence of a return signal which does not satisfy a minimum value of the operating feedback criterion C1 then renders the diagnosis of feedback negative.

A second possible analysis consists in comparing each Umic return signal with a maximum amplitude value so that the return diagnosis is made negative as soon as a UmiC feedback signal is not lower than this maximum value.

Control verification means are arranged in the device 23 to establish an output diagnosis by analyzing all or part of the uc signals output from the ANC algorithm, each intended for one of the loudspeakers 13.

Here again, the control verification means can perform several analyzes of the output signal by means of tests.

A first test consists in verifying that the amplitude of the control signal uc of each of the loudspeakers 13 does not exceed a threshold value. The threshold value is different for each of the loudspeakers because the control signals sent to each of the loudspeakers are different. On the other hand, the threshold value depends on the control frequency associated with one or more harmonics, in particular with the harmonic 2 of the engine speed. This threshold value is tabulated and entered by the manufacturer at the time of manufacture of the vehicle. The control signal sent to the loudspeakers is a sinusoidal type signal or a periodic signal composition in a predetermined frequency band. The test consists in verifying, at a given engine speed, that the maximum of this control signal never exceeds a threshold value given in the table. If the signal has a value that exceeds the threshold value, then in this case, the amplitude of the signal is replaced by this threshold value so as to limit the amplitude. This operation ensures that the amplitude of the signal emitted by the loudspeakers never exceeds a certain threshold.

The device 23 thus sends the mixer 17 a controlled signal ue associated with each control signal uc so that the amplitude of the signal ue is equal to the amplitude of the signal uc as long as the amplitude of the signal uc is less than the value threshold and which is equal to the threshold value when the amplitude of the signal uc is greater than or equal to the threshold value. Other tests may be considered to complete the failure detection system, for example a frequency test and a phase test between the signals. It is not absolutely necessary to implement these other tests because the system has shown satisfactory behavior in a large number of situations during tests carried out as part of the development of the invention.

With regard to the detection of a failure on the frequency, the following test can be imagined and easily implemented. The control signal sent to the loudspeakers is a sinusoidal type signal whose frequency must be identical to that of the control relative to the harmonic 2 of the engine speed. It is then a question of verifying that the two signals have the same frequency. One way of realizing it is to calculate a time inter-correlation coefficient Cup between the control signal u (x1, t) of the loudspeaker xi at time t and the sinusoidal signal p (xi, t + τ) which has a frequency identical to the control frequency at time t + τ, represented by the following expression:

where E [] is the mathematical expectation operator.

The detection of a failure on the phase between the signals is more delicate because it is essentially the ANC algorithm.

More specifically, the device 23 comprises means 24 for activating and neutralizing the active noise reduction system, comprising, for example, the electronic card 15. The means 24 are driven firstly to activate the active noise reduction system. following a positive return diagnosis. The means 24 are controlled on the other hand to deactivate the active noise reduction system following a negative feedback diagnosis.

The device 23 for controlling the active noise reduction system may be implemented by means of an FPGA-type electronic circuit (acronym for Field-programmable gate array in English for programmable gate array), ASIC

(acronym for English Application-Specifies Integrated Circuit for "Application Specific Integrated Circuit") or other.

Advantageously, the device 23 is made using the resources of the on-board computer or a multimedia computer of the vehicle shared with the ANC algorithm, the radio and the mixer.

The memory of the computer used then contains a computer program, comprising program code means for performing all of the steps of the method set out below, when said program is running on a computer.

The program can thus be loaded on the computer at the factory in the form of a computer program product, comprising program code means, stored on a computer-readable medium, for implementing the method set out below. , when the program product is running on a computer.

An exemplary method according to the invention is now explained with reference to FIG.

First, a failure tree of the Active Noise Control System (ANC) is documented to determine the various possible causes of failure. On the basis of this analysis, a self-diagnosis procedure is established, preferably broken down into three levels of diagnosis: - Level 1: limitation of the amplitude of the control signals sent to the loudspeakers. - Level 2: use of information from the external environment to the ANC system.

Level 3: A posteriori detection based on tests on the ANC signals of exit and entry, in particular of return.

The process starts in a step 100 when the active noise reduction system, powered up, receives a positive voltage Va of the battery 19. The step 100 triggers a step 101 which consists of checking different states of the vehicle that allow or not to activate the ANC algorithm under the conditions for which it is intended.

The states can be binary as for example purely illustrative and non-exhaustive, door open or closed, glass closed or lowered, engine started or stopped.

It will be understood from the recalls given above on the predictive and counter-reaction algorithms that the effectiveness of the ANC algorithm is conditioned by a well-tuned agreement of FRF frequency response functions in the passenger compartment of the vehicle. These frequency response functions are disturbed by any modification of the cabin space, including the changes that result from an opening on the outside. It is therefore considered in step 101 that a correct state (OK abbreviated) of the opening of the vehicle corresponds to a closed position of the passenger compartment of the vehicle.

We have also seen above that the noise that is to be suppressed is a frequency or frequency spectrum noise determined by a measurable or calculable RPM rotation of a vehicle member, in particular of the power train 21 It will be understood that it is difficult to agree on a zero frequency in the absence of rotation of the motor. It is therefore considered in step 101 that a correct state (abbreviated OK) of the engine speed corresponds to an RPM value greater than a positive lower limit.

The states can also be quantified on a continuous or discrete scale when they relate to a temperature in the passenger compartment or to a vehicle occupancy rate.

To detect whether a state is correct or not, its value is read on the vehicle communication bus. A diagnostic variable Diag (E) is set when all the states E of the vehicle necessary for proper operation of the algorithm ANC are correct. The variable Diag (E) is set to zero as soon as a state E is not correct.

A step 103 is triggered if all states are detected correct in step 101 and a step 102 is turned on otherwise, in other words if at least one state is detected incorrectly. Step 102 is to block any activation of the ANC algorithm, in other words not to enable or disable the ANC algorithm. Step 102 loops back to step 101 so that step 103 can be started as soon as all states are detected correct in step 101. Step 103 is to enable activation of the ANC algorithm other terms to activate or maintain the activation of the ANC algorithm.

An output verification step 110 and a return verification step 130 are started from step 103. Step 110 consists in establishing an exit diagnostic by analyzing at least one output signal, in particular by verifying that the uc signals each for a loudspeaker 13 conform to pre-established amplitude, frequency and / or phase criteria for suppressing the noise in the passenger compartment 20 of the vehicle 12.

An example of step 110 according to the invention is described in more detail with reference to FIG. 4.

In step 110, a sub-step 111 is cyclically engaged as long as the ANC algorithm is activated, for example purely illustrative and nonlimiting after each activation of step 103. In the sub-step step 111, an index j is initialized to zero to designate the first speaker 13 of a list containing a quantity Qhp of loudspeakers 13.

For each execution cycle of substep 111, a substep 112 is repeated as many times as there are monitored speakers, in other words substep 112 is executed to verify the set. output signals of index j, j ranging from zero to Qhp-1 · A run cycle typically, but not necessarily, corresponds to a sampling period on the inputs or to a date of the fundamental frequency of the noise to be canceled.

The verification of an output signal in the sub-step 112 as illustrated in FIG. 4, consists in ensuring that the amplitude | uc (j) 1 of the control signal of the index loudspeaker 13 j, is less than a predetermined useuii threshold value.

The value useuii is typically predetermined during tests on a prototype vehicle so as to cancel the noise under optimal conditions with a minimum of margin above the noise actually observed. The sound power of the noise and therefore the useuii value are generally a function of the load and the rotational speed of the engine, in other words the RPM regime. Different tests are then carried out for different RPM engine speeds during tests on the prototype vehicle. The adjusted useuü values are then stored in an associative data structure indexed by the possible RPM engine speed values. The same possible value of RPM engine speed can index several useuu values, each specially adjusted to one of the speakers 13 in relation to its range of octaves and its position in the vehicle. The data structure thus obtained is then duplicated on data carriers, for example readable by the on-board computer or any other computer of the vehicle so as to be usable on vehicles of the same type at the output of the production line. The assurance on the amplitude of the noise canceling signal below the threshold value is intended in particular to prevent the cancellation signal from being greater than the noise to be canceled, with the opposite and disastrous effect of amplifying it. The amplitude test can be performed in real time in different ways.

A simple way is to compare the absolute value of the signal uc (j) at each sampling period with the threshold value. Indeed any exceeding of the useuii value by the signal ucG) indicates an amplitude greater than Useuii · It is reasonable to consider that an amplitude greater than the threshold value causes a crossing by at least one sample of the signal uc (j) when the sampling frequency is significantly higher than the Nyquist-Shannon frequency.

A more elaborate way is to raise the sampled signal squared and then filter it by a first-order filter with a time constant substantially greater than the period of the sampled signal. It will be recalled that the squaring of a first signal of sinusoidal nature generates a second signal comprising a DC component equal to the half-amplitude of the first signal and a sinusoidal component with a frequency double that of the first signal. The filtering of the double-frequency sinusoidal component with a sufficiently high time constant then leaves only the continuous component which it is sufficient to multiply by two to reproduce the amplitude of the first signal. It is also possible to take a second threshold value equal to half of the threshold value considered in the preceding paragraph and thus directly compare the output of the filter with the second threshold value without having to double the filtered signal. This second way, of course, with respect to the first way, has the advantage of not requiring a sampling frequency much higher than the Nyquist-Shannon frequency to guarantee the detection of at least one peak above the value threshold when the amplitude is greater. A sampling frequency at least equal to the Nyquist-Shannon frequency is then sufficient because it reproduces the entirety of the properties of the signal and consequently of its squaring. This second way also has another advantage which is that of filtering any inadvertent exceeding of the threshold value which could result for example from a parasite or a momentary accumulation of sound energy of several signals and not a real overshoot. amplitude.

Another more elaborate way is to find in the output signal, the amplitude of a component of the frequency signal ωο / 2π corresponding to the expected frequency of the noise generated by the engine for a given RPM regime.

This third way of verification of the amplitude is particularly adapted to the signal uhp from the mixer 17. In fact the signal comprises not only the frequency of the signal uc theoretically equal to ωο / 2π but also the frequencies of the signal ur which scan at least the spectrum of frequencies audible by the human ear. The signal may also contain the frequency ωο / 2π and this with an amplitude such that, added to that of the signal uc, a total amplitude greater than the threshold value of the amplitude is obtained, with the risk of saturating the regulation of the ANC algorithm and consequently to cause non-linearities detrimental to the adaptive parameterization.

This third way of verifying the amplitude does not bring any particular advantage to the verification of the amplitude of the single signal uc whose frequency is, unless the ANC algorithm itself is malfunctioning, equal to the frequency expected ωο / 2π. In this case, the verification should not focus on the amplitude but more on the frequency itself, for which we will see a possible method below.

The third way of verifying the amplitude is implemented by calculating an abscissa B and an ordinate C relating to an amplitude module A by means of the following formulas:

Whatever the manner of checking the amplitude chosen in the sub-step 112, a sub-step 113 is executed if the amplitude is not less than the amplitude threshold value and a sub-step 114 is executed if the amplitude is less than the amplitude threshold value useuü

In the sub-step 113, the amplitude of the signal is limited to the useuii value and a Diag variable (S) previously initialized to 1 for example in the sub-step 111, is set to zero.

In sub-step 114, the amplitude of the signal is maintained at its value and the variable Diag (S) retains its previous value.

Thus, it is necessary for the amplitudes of all the verified output signals to be lower than their respective threshold values for the Diag (S) variable to be maintained at 1 and all that is required is an output signal amplitude greater than its threshold value. to set the Diag (S) variable to zero. The implementation example illustrated in FIG. 4 corresponds to a sequential execution of the method in which a substep 115 increments the index j following each execution of the substep 113 or of the substep 114 to perform sub-step 112 as long as the index j is less than Qhp- A substep 117 of the end of the cycle is started as soon as a substep 116 detects that the index j reaches the number QHP of signals output to check. Other implementations of the method than that illustrated in FIG. 4 are possible. For example, in a multitasking or logic circuit implementation, a parallel execution of Qhp verification tasks each of an output signal, sets to 1 or 0 a distinct Diag (S) variable for each signal. A synthesis task or an AND gate then makes the product of all the Diag (S) variables to obtain a final value equal to 1 if all the checks are positive or a final value equal to 0 as soon as a check is negative.

Steps 112, 113 and 114 of Fig. 4 illustrate a check on a maximum allowable amplitude of output signal. Other output signal checks with or without correction are possible.

In particular, it is possible to check whether the frequency of an output signal U2 (t) and close or not to the control frequency f = α> ο / 2π by means of an inter-correlation coefficient CuiU2 (x) of the following way.

An ui (t) sinusoidal signal of frequency f = ωο / 2π is arbitrarily generated.

Three mathematical expectations E [] are calculated using the following approximate formulas:

The inter-correlation coefficient Cuiu2 (t) between the two signals ui (t) and u2 (t) is then calculated using the following formula:

The correlation coefficient Cuiu2 (x) thus obtained is compared with a threshold value coefficient, for example equal to 0.9.

If the inter-correlation coefficient Cuiu2 (x) is not greater than the threshold coefficient, the variable Diag (S) is set to zero to signify that the frequencies of the two signals ui (t) and u2 (t) are not sufficiently close to each other.

In particular, it can be verified whether the phase of an output signal uc (t) varies over time. Knowing the period T equal to the inverse of the frequency f (ü () / 27i of the signal uc (t), the values of the signal uc (t) are compared at two times separated by a period T.

For example, the difference u'c (t + T) is evaluated:

A non-zero difference then makes it possible to detect a phase variation.

For example, the difference u "c (t + T) is still evaluated:

A non-zero difference then makes it even better to detect a variation in phase variation. Those skilled in the art will of course take the usual precautions in terms of filtering to avoid unwanted phase variation detections so as to position the variable Diag (S) to zero in the event of detection of known phase variation.

The variable Diag (S) is then evaluated in the substep 117 after having made all the checks provided on all the output signals to be checked. The positive escapement of sub-step 117 when the variable Diag (S) is at 1 corresponds to the positive escape of step 110 and the negative escape of substep 117 when the variable Diag (S) is at 0 corresponds to the negative exhaust of step 110. The positive exhaust of step 110 leads to a continuation of the process without taking particular action so as to allow a re-execution of step 103 as a result of a positive output diagnostic, for example as illustrated in Figure 2 by looping back to step 101 when Diag variable (S) is equal to 1. The negative exhaust of step 110 triggers a test step 120 occurrences of the variable Diag (S) equal to zero so as to make a step 104 of neutralizing the ANC algorithm executable when the occurrences of negative output diagnostics are considerable. Step 104 then essentially consists in deactivating the active noise reduction system, preferably at least until a new start of step 100 by powering up after powering off, or even until further control is performed. in after sales service. Step 120 essentially consists in detecting whether the Diag (S) non-1 escaped variable of step 110 reflects a failure that occurs over a period Tdiag = o that is too long. The duration Tdiag = o is considered too long if it exceeds a predetermined duration Tseuil during prototype phases of the vehicle and possibly adjustable to sensitize or desensitize the ANC deactivation in the presence of failures. We will see later in the description a possibility

to determine a duration Tseuil common to the outputs and returns. When a duration Tseuil equal to a duration Ts (S) specific to the outputs is determined, the following test is carried out on a duration Tdef (S) of zero diagnostic proper to the outputs:

Step 120 can be implemented in different ways.

A first way is simply to increment a counter at each occurrence of a zero value of the variable Diag (S) for an observation duration equal to Tseuil counted from the first occurrence of the variable Diag (S) to zero. If the observation time is reached or exceeded without the contents of the counter corresponding to a number of all zero occurrences of the variable Diag (S), the counter is reset. This first way, if it has the merit of being simple, is not entirely satisfactory because each reset of the counter causes erasure of past events in terms of failure detection.

Advantageously, a second way is to apply a first order filter with a time constant a on the variable DIAG = Diag (S) to obtain a filtered variable DF:

Figure 5 shows a temporal evolution curve of the variable DIAG in the upper part and a temporal evolution curve of the variable DF in the lower part.

When at a time to, the variable DIAG initially at the value 1, passes to the value 0, the variable DF up to the value 1, decreases with the time constant a until a threshold ε of zero detection is reached at a time t2 provided that the variable DIAG remains at zero.

If at a time ti posterior to the instant to and which precedes the moment t2, the variable DIAG returns from the value 0 to the value 1, the variable DF then to a value lower than 1, increases with the time constant a until the variable DIAG returns to zero at a time t3. When, at time t3, the variable DIAG passes to the value 0, the variable DF then at a value greater than that reached at time ti decreases with the time constant a until it reaches the threshold ε of detection null to

a moment t6 clearly greater than the instant t2. We note here that the rise of the variable DIAG to 1 during a duration Î3 - ti, causes a crossing of the threshold ε of detection zero with a delay te - Î3 much higher than the duration t3 - ti during which the variable DIAG is ironed at 1.

Although the probability of such a phenomenon occurring is very small, one solution may be to lower the time constant a or to raise the threshold ε of zero detection.

Figure 6 shows a time evolution curve of the DF variable when the filter is applied only if the variable DIAG is at zero and the variable DF remains frozen if the variable DIAG is at 1.

This second way of filtering the zero crossings of the variable DIAG is well adapted to the example of implementation of the method illustrated in FIG. 2. The filter is applied in step 120 only in case of escape from the step 110 with a null value of the variable DIAG equal to the variable Diag (S). As long as step 110 loops back to step 101 with a unit value of the variable Diag (S), the variable DF is not changed in step 120.

When at a time ti posterior to the instant to and which precedes the instant t2, the variable DIAG returns from the value 0 to the value 1, the variable DF then to a value lower than 1, retains this value until the variable DIAG returns to zero at a time Î3. When at the moment Î3, the variable DIAG passes to the value 0, the variable DF then to a value equal to that reached at the instant ti, decreases with the time constant a until the threshold ε of detection zero at a moment U greater than the instant t2 but very close to the instant t3. We note here that the rise of the variable DIAG to 1 during the duration Î3 - ti, causes a crossing of the threshold ε of detection zero with a delay U - t3 significantly lower than the duration t3 - ti during which the variable DIAG is ironed at 1. It is thus avoided that the DIAG variable returns to 1 do not inadvertently hide zero crossings whose occurrences indicate a system failure.

However, it is observed in FIG. 6 that if the variable DIAG remains at 1 for a duration γ-ti that is significantly greater than ti-to, the variable DF persists in crossing the threshold ε of detection zero at a time t8 greater than the instant t7 but nevertheless as close to the instant t7 as the instant t4 is the instant of the moment Î3. We note here a memory effect of the zero crossings of the variable DIAG which, even of negligible durations in front of those of passages to 1 of this same variable DIAG, have the effect of making the variable DF fall below the threshold ε of detection null by a short transition to zero DIAG variable after a time more or less long and completely unpredictable.

Figure 7 shows a temporal evolution curve of the variable DF when the filter is applied if the variable DIAG is at zero with a time constant a of value lower than if the variable DIAG is at 1.

This third way of filtering the zero crossings of the variable DIAG poses no particular difficulty of realization because the variable DIAG being a binary logical variable, it suffices to type the time constant a in the form of parameter instantiated at a first low value when the variable DIAG is at 0 and at a second high value when the variable DIAG is at 1. The application of the filter during the ascent of the variable can be carried out in a step not represented in FIG. 2 but which is situated easily between step 110 and step 101.

When at time ti, the variable DIAG returns from the value 0 to the value 1, the variable DF then to a value less than 1, increases until the variable DIAG returns to zero at a time Î3. When at the moment Î3, the variable DIAG passes to the value 0, the variable DF then to a value greater than that reached at time t1 but lower than that reached at the same instant in FIG. 5 because of the constant of higher value time, decreases with the time constant a of low value until crossing the threshold ε of detection zero at a time t5 greater than the instant ti but lower than the instant t6 of FIG. note here that the rise of the DIAG variable to 1 during the duration Î3 - ti, causes a crossing of the threshold ε of detection zero with a delay ts - ti much lower than the duration t6 - ti of Figure 5. In comparison with the first way of filtering, it is thus more easily avoided that upward movements of the variable DIAG do not inadvertently hide zero crossings whose occurrences testify to a system failure. Compared to the second way of filtering, it is also avoided that descents to 0 of the variable DIAG do not inadvertently hide a predominance of the passages to 1 whose occurrences testify to an acceptable functioning of the system because the time constant has, well that of high value at the ascent, leads to a duration tg - ti of reaching the threshold ε of null detection higher than the duration tg -ti when the duration Χη - ti is clearly greater than the duration ti - to.

A negative escape of step 120, as long as the negative diagnosis has not persisted long enough, loops back to step 101 at the rate of the signal samplings. Step 130 consists in establishing a return diagnosis by analyzing at least one return signal, in particular by verifying that the UmiC signals coming from each of a microphone 11, comply with pre-established criteria of amplitude, frequency and / or or phase according to the expected noise suppression in the passenger compartment 20 of the vehicle 12.

An example of step 130 according to the invention is described in more detail with reference to FIG.

In step 130, a substep 131 is cyclically engaged as long as the ANC algorithm is activated, for example purely illustrative and not limiting, following each activation of step 103. step 131, an index i is initialized to zero to designate the first microphone 11 of a list containing a Qmicro quantity of microphone 11.

For each cycle of execution of the substep 131, a substep 132 is repeated as many times as there are controlled microphones, in other words the substep 132 is executed to check all the signals. output of index i, i ranging from zero to Qmicro-1 · A run cycle typically, but not necessarily, corresponds to a sampling period on the inputs or to a date of the fundamental frequency of the noise to be canceled.

Verification of an output signal in sub-step 132 as illustrated in FIG. 3, is to ensure that the amplitude | umiC (i) | of the signal picked up by the microphone 11 of index i, is below a predetermined threshold value umaxi.

The umaxi value is typically predetermined during tests on a prototype vehicle in agreement with a sound level bearable by the human ear. The sound power of the noise and therefore the Umaxi value are generally a function of the rotational speed and the engine load, in other words the RPM regime. Different tests are then carried out for different RPM engine speeds during tests on the prototype vehicle. The adjusted Umaxi values are then stored in an associative data structure indexed by the possible RPM engine speed values. The same possible value of RPM engine speed can index several values u, ii;, vi. each specially adapted to one of the microphones 11 in relation to its sensitivity and position in the vehicle. The data structure thus obtained is then duplicated on data carriers, for example readable by the on-board computer or any other computer of the vehicle so as to be usable on vehicles of the same type at the output of the production line. The assurance on the amplitude of the signal captured below the maximum value, is intended in particular to avoid a feedback effect with the opposite and disastrous effect of unpleasantly amplifying the noise. The amplitude test can be performed in real time in different ways.

A simple way is to compare the absolute value of the umicro (i) signal at each sampling period with the threshold value. Indeed any exceeding of the Umaxi value by the signal umicro (i) indicates an amplitude greater than the value

Umicro (i) · fl is possible to perform a finer analysis of the maximum value on the microphone signal. Indeed, the error microphone 11 measures a broadband noise, in other words a noise spectrum over a wide frequency range. The active reduction algorithm aims to remove only the noise component at the expected frequency of being canceled. This filtering operation is implicitly performed in the body of the algorithm. But it is possible to perform it also by a simple operation.

It is recalled that, in general, the signal measured by the microphone can be decomposed into a Fourier series. fl is written as:

or :

with T: umicro signal period (t)

The coefficients of this series are given by the relations:

In the case where We want to extract the component of the signal at the control frequency ΐ = ω0 / 2π, we only consider the component which is written in the form

Which is rewritten again in the following form

with B, C calculated using the following relationships:

The total amplitude A of the signal at the frequency f is then such that

. This amplitude varies according to the engine speed A. In general, we have A = A (RPM).

In this case, the test on the simple instantaneous module of the signal mentioned in the preceding paragraph is replaced by a test which consists in checking, at a given RPM engine speed in real time, that the amplitude A (RPM) measured by the microphone is less than a threshold amplitude defined for this same engine speed:

This second maximum amplitude test is more interesting than the first maximum amplitude test explained above because an anomaly, in other words a threshold overshoot, is detected only at the control frequency f, for a RPM regime. given. It should be borne in mind that the error microphone 11 measures at each moment a pressure value which is a mean value while the pressure fluctuation is the superposition of all the frequency contributions as expressed by the Fourier decomposition . But only the fluctuation of pressure at the control frequency interests us. Therefore, if the level of

The pressure measured by the error microphone results from wide-band excitation or mono-frequency excitation but at a frequency different from the control frequency, the first maximum amplitude test detects an error while the second test of maximum amplitude does not detect it. But the purpose of active noise reduction is to remove only the component of the signal that is at the control frequency. The detection of a fault only makes sense at this frequency.

Whatever the manner of checking the maximum admitted amplitude which is chosen in the sub-step 132, a sub-step 133 is executed if the amplitude is less than the amplitude threshold value and a sub-step 134 is executed. if the amplitude is not less than the Umaxi amplitude threshold value ·

In the sub-step 133, it is verified that the microphone 11 of index i is operational, in other words that it actually captures a non-zero signal. At each instant t of passage in step 133, a criterion Clmic (i) is calculated which measures a plurality of variations of the signal Umic (i, t) discretized by time intervals Δt on the N samples preceding the current t in other words, over a period of observation T = NAt, using the formula:

We then check the criterion Clmic (i)> Cl min · If no signal is measured by the microphone, the value of Clmic (i) is very low. The value of the lower limit Cl min is also determined during the tests of the prototype system. The advantage of this criterion is that it does not depend on the average value of the signal and that it is quite simple to calculate.

In the exemplary implementation of the method illustrated in FIG. 3, a substep 136 is executed if the criterion Clmic (i)> Clmini is found and the substep 134 is executed in the opposite case.

In step 134, the variable Diag (R) is set to zero so that a negative response to one of the verification steps 132 or 133 is sufficient to set the variable Diag (R) to zero and the variable Diag (R) remains at 1 if and only if all the responses to verification steps 132 and 133 are positive.

The implementation example illustrated in FIG. 3 corresponds to a sequential execution of the method in which the substep 136 increments the index i following each execution of the substep 133 or of the substep 134 to perform sub-step 132 as long as the index i is less than Qmicro · A substep 138 of the end of the cycle is started as soon as a substep 137 detects that the subscript i reaches the number Qmicro of signals back to check. Other implementations of the method than that illustrated in FIG. 3 are possible. In particular, the order of execution of steps 132 and 133 can be reversed or their execution can be done in parallel. For example still in a multitasking or logic circuit implementation, a parallel execution of Qmicro verification tasks each of a feedback signal, sets to 1 or 0 a distinct Diag (R) variable for each signal. A synthesis task or an AND gate then makes the product of all Diag (R) variables to obtain a final value equal to 1 if all the checks are positive or a final value equal to 0 as soon as a check is negative.

The variable Diag (R) is then evaluated in the substep 138 after having made all the checks provided on all the return signals to be checked. The positive escape of the substep 138 when the variable Diag (R) is at 1 corresponds to the positive escape of the step 130 and the negative escape of the substep 138 when the variable Diag (R) is at 0 corresponds to the negative exhaust of step 130. The positive exhaust of step 130 leads to a continuation of the process without taking particular action so as to allow a re-execution of step 103 as a result of a positive return diagnosis, for example as illustrated in FIG. 2 by looping back to step 101 when the variable Diag (R) is equal to 1. The negative exhaust of step 130 triggers a test step 140 occurrences of the variable Diag (R) equal to zero so as to make executable step 104 of neutralization of the algorithm ANC when the occurrences of negative feedback diagnosis are considerable. Step 140 essentially consists in detecting whether the non-escaped Diag (R) variable of step 130 reflects a failure that occurs over a too long duration Tdiag = o. The duration Tdiag = o is considered too long if it exceeds a predetermined duration Tseuil during the prototype test phases of the vehicle. The duration Tseuil is possibly adjustable to sensitize or desensitize the deactivation ANC in the presence of failures. We will see later in the description a possibility to determine a duration Tseuil common to the outputs and returns. When a duration Tseuil equal to a duration Ts (R) specific to the outputs is determined, the following test is carried out on a duration Tdef (R) of zero diagnostic proper to the outputs:

Step 140 can be implemented in different ways on the model of those of step 120.

A negative escape of step 140, as long as the negative diagnosis has not persisted long enough, loops back to step 101 at the rhythm of the signal samplings.

As we have just seen, a step 120 distinct from the step 140 makes it possible to determine a threshold duration Ts (S) that is customized for the occurrences of the negative output diagnostics and a threshold duration Ts (R) that is customized for the occurrences of the diagnostics. negative output of different values.

If the values of the threshold durations Ts (S) and Ts (R) are identical, steps 120 and 140 can be grouped in a single step in which the occurrences of exit and return diagnostics are nevertheless considered separately or are considered at the same time. In other words, in the latter case, a negative diagnosis is considered an occurrence regardless of whether it is an exit or a return. The invention just described meets the objective of preventing, in use, that the ANC system causes inconvenience to the user when the system becomes defective, for example because a microphone or speaker is down. We have seen how the invention makes it possible to prevent the loudspeakers of the vehicle from emitting a loud sound when the system fails. The invention thus makes it possible to detect the appearance of the fault in a very short time and to find an adequate correction.

The detection is robust in that it does not fire unexpectedly.

This simple solution makes it possible to deal with all the failures identified in the fault tree. Since it is general enough not to depend on the particular origin of the failure, it has the potential to deal with other unidentified failures so far.

Claims (13)

A method of controlling an active noise reduction system having one or more loudspeakers (13) for producing noise-canceling sound output signal (s) and one or more microphones (11) for sensing noise. or return signals which measure the noise reduction obtained, characterized in that it comprises: - a step (103) of activation of the active noise reduction system; - a return verification step (130): - which establishes a return diagnosis by analyzing at least one feedback signal in a low detection substep (132) which calculates an operating return criterion (C1) and which makes negative the return diagnosis when at least one return signal does not satisfy a minimum value of the operating return criterion (C1); and - which allows re-execution of step (103) following a positive feedback diagnosis; a neutralization step (104) which is executable following a negative feedback diagnosis and which deactivates the active noise reduction system. 2. Control method according to claim 1, characterized in that the criterion (C1) corresponds to a plurality of variations of the feedback signal over a given time interval. 3. Control method according to one of claims 1 to 2, characterized in that the step (130) of return verification comprises a sub-step (133) of high detection which makes the diagnosis of return negative when at minus a feedback signal is not less than a maximum amplitude value. 4. Control method according to one of the preceding claims, characterized in that it comprises: - an output verification step (110) in which an output diagnostic is established by analyzing at least one output signal, which authorizes a re-execution of the step (103) following a positive output diagnosis and which makes the neutralization step (104) executable as a result of a negative output diagnostic so as to then deactivate the system active noise reduction. 5. Control method according to claim 4, characterized in that the output verification step (110) comprises a high detection substep (112) which renders the output diagnosis negative when at least one output signal is not less than a maximum amplitude threshold. 6. Control method according to one of claims 4 or 5, characterized in that the step (110) of output verification comprises a sub-step (113) amplitude limitation of the output signal to a maximum threshold. 7. Control method according to one of the preceding claims, characterized in that it comprises a moderation step (120, 140) which authorizes a re-execution of the step (103) following an exit diagnosis or negative return until there is a sufficient number of occurrences of output diagnostics or negative feedback, especially as long as said diagnosis is not negative for a sufficient duration and which triggers the step (104) of neutralization when there is a sufficient number of occurrences of diagnostic output or negative feedback, especially as soon as said diagnosis is negative for a sufficient duration. 8. Control method according to one of the preceding claims, characterized in that it comprises a step (101) for monitoring the state of an environment of the active noise reduction system that allows the activation of the reduction system. active noise only when said environment is in a state compatible with said activation. 9. Apparatus (23) for controlling an active noise reduction system (15) having one or more loudspeakers (13) for producing one or more noise-canceling output signals, one or more microphones (11) for sensing one or more feedback signals which quantify the noise reduction obtained, and activation means (24) and neutralization of the active noise reduction system (15) driven to activate or deactivate the reduction system respectively active noise after a diagnosis of positive feedback, respectively following a diagnosis of negative feedback, characterized in that it comprises: - return verification means arranged to calculate a criterion (Cl) of operative return and to establish a return diagnosis by analyzing at least one feedback signal (umjC) so as to render the return diagnosis negative when at least one return signal does not satisfy a minimum value of the return (Cl) operating return era. 10. Device (23) according to claim 9, characterized in that said return verification means are arranged to make the return diagnosis negative when at least one feedback signal (umiC) is not less than a maximum value. in amplitude. A computer program, comprising program code means for performing all of the steps of any one of claims 1 to 8, when said program is running on a computer. A computer program product, comprising program code means, stored on a computer readable medium, for implementing the method of any one of claims 1 to 8, when said program product is operating on a computer program product. computer. A vehicle (12) in which an active noise reduction system (15) in a passenger compartment (20) has one or more speakers (13) for producing one or more sound output signals which oppose noise and one or more microphones (11) for receiving one or more return signals which quantify the noise reduction obtained, characterized in that it comprises a control device (23) according to one of claims 9 to 10 mounted so as to control the active noise reduction system.