Location based noise reduction system

Abstract

The present disclosure relates to a device for suppressing noise at a defined location, comprising: a plurality of sensors 6 for detecting reference signals correlated with the noise; a plurality of acoustic output means 7 for the acoustic output of sound signals for suppressing the noise, wherein the acoustic output means are arranged in the vicinity of the defined location; a plurality of acoustic input means 8 for detecting error signals, wherein the acoustic input means are arranged in the vicinity of the defined location; a processing unit 2 which receives and processes the reference signals and the error signals and, based on these, generates control signals for the acoustic output means and outputs them to them;wherein the processing unit generates the control signals by means of adaptive control signal filters 10, and the adaptation of the control signal filters 10 is carried out by means of a method for reducing an error function based on the error signals; wherein at least one control signal is generated based on a plurality of reference signals; wherein at least one control signal filter 10 generates a control signal based on processing in the frequency domain, and the at least one control signal filter 10 is implemented by a plurality of parallel-arranged control signal sub-filters 10', each of which is adapted separately from one another in the frequency domain.

Description

[0001] The present disclosure relates to the reduction of noise at defined locations, for example at vehicle seats, where vehicles are understood to mean all types of means of transport for people on land, water or in the air.

[0002] The reduction of noise and unwanted ambient noise is a central development task in vehicle development and research. In addition to passive, structural measures, active noise reduction systems (Active Noise Control / Cancellation, ANC systems) can also be used due to the increasing digitalization and networking of vehicle systems. The solution approaches differ between global implementations for the low-frequency range and local, (seat-)oriented systems for a defined location that cover an extended frequency range.

[0003] Active noise reduction systems (ANC systems) can make a significant contribution to reducing background noise in vehicles. Background noise in vehicles can be caused, for example, by engine noise, rolling noise, or wind noise. At low speeds, engine and tire noise dominate in the frequency range up to 500 Hz; in the frequency range up to 1000 Hz, the airborne noise component dominates due to tire-road contact. A general distinction is made between local, seat-oriented systems and global systems that cover the entire vehicle interior. Global systems can be used effectively in a frequency range below 300 Hz (engine, rolling noise). However, they are not efficient at higher frequencies. In contrast, local, seat-oriented systems can achieve background noise reduction even at higher frequencies.The aim is to eliminate noise in the ears of passengers through destructive interference using anti-noise. This typically involves loudspeakers as actuators, for example, in the headrests, microphones at various installation locations, a control unit, and sometimes additional reference sensors at favorable positions in the vehicle (engine block, chassis, etc.).

[0004] The majority of applications are based on an adaptive filter to generate the anti-noise. The filter can generally be adapted in the time or frequency domain using a (Fast) Fourier Transformation (FFT). Depending on the sensor configuration and sampling rate, adaptation in the time domain results in considerable computational effort, which conventional signal processors often cannot process in real time. While adaptation in the frequency domain results in less computational effort, the block processing of the filter results in a significant time delay in practice, which is why it is usually not suitable for ANC systems.

[0005] Often, a simple SISO (single-input, single-output) system is used for noise reduction. However, a single reference signal cannot establish a sufficiently coherent relationship with the sound in the vehicle interior. This results in significantly less noise reduction compared to a system with multiple reference sensors and secondary speakers to generate the anti-noise (multiple-input, multiple-output (MIMO)).

[0006] Another problem is that microphones for capturing the error signal cannot be placed directly at the user's ears.

[0007] This document addresses the technical problem of presenting an improved, seat-oriented ANC system for vehicles of all types. This problem is solved by the features of the independent claims. Advantageous embodiments are described in the dependent claims.

[0008] According to a first general aspect of the disclosure, a novel partitioned-block adaptive feedforward filtered-reference least-mean-square (FxLMS) algorithm in the frequency domain is described for a signal configuration with multiple reference sensors and multiple loudspeakers (multiple-input-multiple-output (MIMO)) to generate the anti-noise. It is also possible to implement the MIMO configuration using a dual-MISO system (two coupled multiple-input-single-output (MISO) systems). Each of these MISO systems assumes a main transmission path between loudspeakers (LS) and ears (e.g., left LS to left ear and right LS to right ear). In conjunction with a corresponding hardware platform and partitioned operation in the frequency domain, this enables extended parallel processing. The convergence properties are also improved while ensuring the real-time capability of the filter.

[0009] Another general aspect of the disclosure is the combination of an ANC algorithm with the Remote Microphone Technique (RMT) in the frequency domain as a partitioned-block implementation to address the problem that microphones for capturing the residual error signal cannot usually be placed directly at a user's ears. The sound field at the listening position is determined from nearby microphones and previously measured and estimated transmission paths. One aspect of the system presented here is the combination of the ANC system with the RMT in the frequency domain as a partitioned-block filter. In this way, the required computing power for signal processing can be reduced and its real-time capability can be improved.For estimating the error signal at the listening location, a special variant of the remote microphone technique (delayed-RMT) is proposed to ensure the causality of the system with a low delay. With appropriately selected parameters and an FxLMS-based antinoise generation using a partitioned-block approach, synergy effects between the two subsystems can be exploited, thus saving computing power while maintaining the same performance.

[0010] The achieved space-based noise reduction can be used in all means of transport in which passengers (users) are located at specific, defined locations, such as vehicles such as cars, trains, buses, airplanes, ferries, etc. However, the proposed approach for reducing unwanted noise is not limited to these examples. It can be applied more generally to situations in which people are located at specific locations in a room (e.g., in theater or cinema seats) and can be exposed to sound through individual acoustic reproduction devices.

[0011] According to a first general aspect, a device for suppressing noise at a defined location, for example the listening location of a user, is disclosed.

[0012] The device comprises: a plurality of sensors for detecting reference signals x[n] correlated with the noises; a plurality of acoustic output means, for example loudspeakers, for the acoustic output of sound signals to suppress the noises, wherein the acoustic output means are arranged in the vicinity of the defined location; a plurality of acoustic input means, for example microphones, for detecting acoustic error signals m[n], wherein the acoustic input means are arranged in the vicinity of the defined location; and a processing unit which receives and processes the reference signals x[n] and the acoustic error signals m[n] and, based on these, generates control signals u[n] for the acoustic output means and outputs them to them.

[0013] The processing unit generates the control signals u[n] using adaptive control signal filters. The control signal filters are adapted using a method for reducing an error function based on the acquired acoustic error signals m[n]. The control signal filters can be adapted, for example, using the feedforward filtered-reference least-mean-square (FxLMS) algorithm. This algorithm attempts to reduce a quadratic error function step by step, iteratively, by selectively adapting the filter parameters using a gradient method.

[0014] At least one control signal is generated based on a plurality of reference signals and by at least one control signal filter based on processing in the frequency domain. The at least one control signal filter is implemented by a plurality of parallel-arranged control signal sub-filters, each of which is adapted separately in the frequency domain.

[0015] In embodiments of the invention, the processing unit comprises a transformation device that transforms the reference signals x[n] into the frequency domain. The transformation can be performed, for example, using a Discrete Fourier Transformation (DFT) and generate corresponding spectra X of the reference signals. One possible implementation of the DFT is a Fast Fourier Transformation (FFT), although other spectral transformations can also be used.

[0016] Preferably, the transformation takes place block by block for a block b of length L of consecutive reference signals and generates a corresponding spectrum X b .

[0017] The at least one control signal filter can filter the reference signals x[n] in the frequency domain by multiplying spectral representations X of the reference signals with respective spectral representations W of the control signal subfilters. The at least one control signal can be generated based on the outputs of the control signal subfilters.

[0018] In a further embodiment of the invention, a spectral representation of the at least one control signal in the frequency domain can be formed for individual frequency bins by adding the contributions of the control signal sub-filters, wherein values ​​for the spectral representation X bpof temporally shifted blocks bp of the reference signals x[n] can be used. Some of the control signal subfilters can filter values ​​from temporally shifted blocks of the spectral representation of the reference signals, which can be easily retained by buffering the blocks. In an alternative embodiment, the control signal subfilters filter the spectral representation of the current block of the reference signals, and the results of this filtering are added block by block with a time shift to obtain a spectral representation of a control signal for the current block.

[0019] In a further embodiment of the invention, the processing device can further comprise a reference signal filter for filtering the reference signals x[n] in the frequency domain with an estimate of the transmission path Ĝ ebetween at least one acoustic output device and the defined location. In this way, a spectral representation R of the filtered reference signal is generated in the domain of the defined location.

[0020] The processing device can also have a separate adaptation device for each control signal sub-filter. At least one of the adaptation devices can adapt the corresponding control signal sub-filter based on a spectral representation R of the filtered reference signals and a spectral representation E based on the acoustic error signals m[n] detected by the acoustic input means. The spectral representation E can correspond to an estimate of the error at the defined location. At least one of the adaptation devices can process a filtered spectral representation R bp of a temporally shifted block pb of the reference signals x[n]. In other words, the values ​​of temporally shifted blocks of the filtered spectral representation of the reference signals are used for adaptation, which can be easily maintained by buffering the blocks.

[0021] In embodiments, the reference signals x[n] can be processed in blocks. The transformation device can then transform the reference signals x[n] into the frequency domain based on two consecutive blocks, each with L sample values. The control signal sub-filters for filtering a specific reference signal can each have a number of filter coefficients in the time domain that corresponds to the number 2*L of signal values ​​of the reference signal in two blocks. A number L of filter coefficients in the time domain, which corresponds to one block, can be set to zero values. Preferably, the last L values ​​of the filter coefficients are set to zero. In the frequency domain, however, all frequency coefficients are generally not equal to zero; only in the corresponding filter in the time domain is the second half set to zero.In a further embodiment of the invention, a control signal can be formed from a portion of the values ​​of a spectral representation U of the control signal for a block, which is transformed back into the time domain and generated by the control signal subfilter. In this way, cyclic components in the control signal can be removed.

[0022] Furthermore, the causality of the filter coefficients of an adapted control signal subfilter can be taken into account by causing certain filter coefficients to have a value of zero in the time domain. These are preferably those filter coefficients that were previously set to zero and whose values ​​have changed due to the adaptation.

[0023] In exemplary embodiments, the reference signal filter is implemented by a plurality of reference signal subfilters arranged in parallel. At least one reference signal subfilter can filter a temporally shifted block of the reference signals x[n]. Optionally, cyclic components can be removed from the spectral representation R of the reference signals obtained in this way.

[0024] In a special embodiment according to the MISO principle, two acoustic output devices and a defined location with two error signals e[n] are provided. For a specific acoustic output device, each reference signal can be assigned its own control signal filter. The control signal for the acoustic output device can be formed from the sum of the output signals of these control signal filters for all reference signals. The adaptation of these control signal filters can be based on an error signal for the defined location that is assigned to this acoustic output device. Error signals for other defined locations are not taken into account during adaptation. An error signal for a defined location can be determined as described below using acoustic error signals detected by acoustic input devices.

[0025] If the acoustic input devices are located close enough to the defined locations, the adaptation of these control signal filters can be based on the error signals from acoustic input devices exclusively assigned to this acoustic output device. Error signals from other acoustic input devices are not taken into account during adaptation.

[0026] In one embodiment, the device described above serves to suppress noise at a seating position in a vehicle, wherein the sensors are arranged on vehicle devices to detect signals representative of noise generated by these devices (e.g., engine, tires). The acoustic output means and the acoustic input means can be arranged near a seat of the vehicle, preferably on the headrest of the seat.

[0027] In a further embodiment of the invention, the processing unit can comprise a device for estimating an error signal at the defined location based on the acoustic error signals detected by acoustic input means. The defined location is the listening location of a user, and the error signal for the listening location can be estimated in the frequency domain and taking into account an estimate O of the transmission paths between the acoustic input means and the listening location. The spectral representation E of the error signal for the listening location thus determined can then be used to adapt at least one of the control signal subfilters.

[0028] The device for estimating an error signal for the listening location can use transmission filters Ĝ m , Ĝ ewhich model the transmission of a control signal from the acoustic output means to the locations of the acoustic input means and the transmission of the control signal from the acoustic output means to the listening location. At least one transmission filter can be implemented by several transmission sub-filters arranged in parallel, and at least one transmission sub-filter can filter a temporally shifted block of the control signals.

[0029] The spectral representation E of the estimated error signal for the listening location can be formed in the frequency domain for individual frequency bins by adding the contributions of partial errors to the spectral representation of the estimated error signal for the listening location. Values ​​for the spectral representation M of the acoustic error signals acquired by acoustic input devices and values ​​for an estimated spectral representation ¶ mwhich is representative of the control signals after transmission to the locations of the acoustic input means.

[0030] For a partial error, values ​​for the spectral representation M bp of the acoustic error signals detected by acoustic input means for time-shifted blocks bp of the error signals and values ​​for a spectral representation ¶ m,bp , which is representative of the control signals after transmission to the locations of the acoustic input means, can be used for temporally shifted blocks bp of the control signals. Furthermore, the partial error can be based on an observation subfilter O p corresponding to the partial error, wherein the observation subfilter O p at least partially models the transmission paths between the locations of the acoustic input means and the listening location.

[0031] The device for estimating an error signal for the listening location can estimate past error signals at the listening location based on current acoustic error signals acquired by acoustic input devices. The temporal offset between the estimated past error signals at the listening location and the current error signals can be a multiple of the block length used to process the reference signals and the acoustic error signals acquired by acoustic input devices. In this case, the temporal distribution and alignment of the processing units for the adaptive filters and for estimating the error signal match, and the processing units are well coordinated, allowing for resource-efficient calculations.

[0032] According to a further aspect of the invention, a method for suppressing noise at a defined location is disclosed. The method comprises the following steps: detecting a plurality of reference signals x[n] correlated with the noise; detecting a plurality of acoustic error signals m[n] at positions near the defined location; generating a plurality of control signals u[n] for generating a plurality of sound signals for suppressing the noise, based on the reference signals x[n] and the acoustic error signals m[n] and by means of adaptive control signal filters, wherein at least one control signal is generated based on a plurality of reference signals and by means of processing in the frequency domain;Adapting the control signal filters using a method for reducing an error function based on the detected acoustic error signals m[n], wherein at least one control signal filter is implemented by a plurality of parallel-arranged control signal sub-filters, each of which is adapted separately from one another in the frequency domain; and outputting a plurality of sound signals at positions near the defined location to suppress the noise.

[0033] In embodiments, the method may further comprise a transformation of the reference signals x[n] into the frequency domain to generate spectral representations X of the reference signals. Generating the at least one control signal u[n] may comprise filtering the reference signals in the frequency domain by multiplying spectral representations X of the reference signals by respective spectral representations W of the control signal subfilters.

[0034] The control signal subfilters can filter spectral representations X bp of temporally shifted blocks of the reference signals x[n]. The at least one control signal can be generated based on a sum of the outputs of the control signal subfilters. Alternatively, the control signal subfilters can each filter a spectral representation X b of a current block b of the reference signals x[n], and the results of the control signal subfilters can each be added, shifted in time by one block.

[0035] In embodiments, filtering of the reference signals in the frequency domain can be carried out with an estimate Ĝ e the transmission path between at least one acoustic output device and the defined location.

[0036] The adaptation of the control signal subfilters can be performed separately. At least one control signal subfilter can be adapted based on a spectral representation R of the filtered reference signals and a spectral representation E based on detected acoustic error signals. At least one adaptation of a control signal subfilter can be performed based on a filtered spectral representation R bp of a temporally shifted block bp of the reference signals x[n].

[0037] In embodiments of the invention, a back transformation of a spectral representation U of the at least one control signal into the time domain and a selection of a portion of the back-transformed values ​​for a block can be provided. The transformation of the reference signals into the frequency domain can be carried out based on two consecutive blocks. The control signal sub-filters for filtering a reference signal can each have a number of filter coefficients in the time domain that corresponds to the number of signal values ​​of the reference signal in two blocks. A number of filter coefficients in the time domain that corresponds to a block can be set to zero values. The back transformation can generate a vector with the length of two blocks, and the selection of a portion of the back-transformed control signal values ​​can select a block of values ​​from the vector.

[0038] Furthermore, in embodiments, an error signal e[n] at the defined location can be estimated based on the detected acoustic error signals m[n] at positions near the defined location. The defined location is the listening location of a user, and the error signal for the listening location can be estimated in the frequency domain and taking into account an estimate of the transmission paths O between the positions for detecting the acoustic error signals and the listening location. The spectral representation E of the error signal thus determined at the listening location can then be used to adapt at least one of the control signal subfilters.

[0039] A further independent second general aspect of the present disclosure relates to a device for suppressing noise at a defined location, for example the listening location of a user.

[0040] The device comprises: at least one sensor for detecting at least one reference signal x[n] correlated with the noises; at least one acoustic output means, for example a loudspeaker, for the acoustic output of sound signals to suppress the noises, wherein the acoustic output means is arranged in the vicinity of the defined location; at least one acoustic input means, for example a microphone, for detecting at least one acoustic error signal m[n], wherein the acoustic input means is arranged in the vicinity of the defined location; a processing unit which receives and processes the at least one reference signal x[n] and the at least one acoustic error signal m[n] and, based on these, generates at least one control signal u[n] for the at least one acoustic output means and outputs it to the latter;and means for estimating an error signal e[n] at the defined location based on the at least one detected acoustic error signal m[n].;

[0041] The processing unit generates the at least one control signal u[n] using at least one adaptive control signal filter. The at least one control signal filter is implemented by a plurality of control signal sub-filters arranged in parallel, each of which is adapted separately in the frequency domain. The adaptation of the control signal sub-filters is carried out using a method for reducing an error function based on the estimated error signal e[n] at the defined location. The adaptation of the control signal sub-filters can be carried out, for example, using the feedforward filtered-reference least-mean-square (FxLMS) algorithm. This algorithm attempts to reduce a quadratic error function step by step, iteratively, by specifically adapting the filter parameters using a gradient method.

[0042] The device for estimating an error signal for the defined location has transmission filters that model the transmission of the at least one control signal from the at least one acoustic output means to the locations of the at least one acoustic input means and the transmission of the at least one control signal from the at least one acoustic output means to the defined location. At least one transmission filter is implemented by a plurality of parallel transmission subfilters, and at least one transmission subfilter filters a temporally shifted block of the at least one control signal.

[0043] A corresponding method for suppressing noise is also disclosed, comprising the following steps: detecting at least one reference signal x[n] correlated with the noise; detecting at least one acoustic error signal m[n] at one or more detection positions near the defined location; generating at least one control signal u[n] for generating sound signals for suppressing the noise, based on the at least one reference signal x[n] and the at least one acoustic error signal m[n] and by means of at least one adaptive control signal filter, wherein at least one control signal filter is realized by a plurality of control signal sub-filters arranged in parallel, each of which is adapted separately from one another in the frequency domain; estimating an error signal e[n] at the defined location based on the at least one acoustic error signal m[n];Adapting the control signal subfilters using a method for reducing an error function based on the estimated error signal e[n] at the defined location; and outputting one or more sound signals at one or more output positions near the defined location to suppress the noise.

[0044] The estimation of an error signal for the defined location is performed using transmission filters that model the transmission of the at least one control signal from the at least one output position to the at least one detection position and the transmission of the at least one control signal from the at least one output position to the defined location. At least one transmission filter is implemented by a plurality of parallel transmission subfilters, and at least one transmission subfilter filters a temporally shifted block of the at least one control signal.

[0045] The device and method according to the second general aspect of the invention can be combined as desired with the features of the first aspect. Thus, the configurations and elements of exemplary embodiments of the first aspect presented above can be applied to the second aspect.

[0046] In general, the features described above can be combined in a variety of ways, even if such a combination is not explicitly mentioned. In particular, features described for a method can also be used for a corresponding device, and vice versa.

[0047] In the following, exemplary embodiments of the invention are described in more detail with reference to the schematic drawings. In the drawings: Figs. 1 schematically shows an embodiment of an overall system for reducing unwanted noise in certain locations; Figs. 2schematic of the main paths and processing of the signals of the noise reduction system; Figs. 3 schematically an embodiment of the dual-MISO approach in the time domain; Figs. 4 schematic representation of the decomposition of a filter into several subfilters (partitions) for partitioned-block and overlap-save processing; Figs. 5 schematic 2 possible block diagrams for the realization of a partitioned-block filter; Figs. 6 schematic of the overlap-save procedure for a single block of input data; Figs. 7 schematic of a possible block diagram for the partitioned-block MISO FxLMS algorithm for one channel; Figs. 8 schematically the principle of RMT; Figs. 9 schematically shows an embodiment of a delayed RMT in conjunction with an FxLMS algorithm; and Figs. 10 a method for suppressing noise according to an embodiment.

[0048] The exemplary embodiments described below are non-limiting and purely illustrative. For illustrative purposes, they may include additional elements that are not essential to the invention. The scope of protection is to be determined solely by the appended claims.

[0049] The following examples enable a reduction of unwanted noise for vehicle occupants in certain seating positions.

[0050] Fig.1schematically illustrates an embodiment of an overall system 1 for reducing unwanted noise in specific locations, such as the seats of a vehicle. In contrast to many known noise reduction applications that use headphones to emit the anti-noise, loudspeakers are used here that are not located at the user's ear, but are mounted at a certain distance from the user at a suitable location, e.g. a seat headrest. This means that the cross paths of the transmission path between loudspeakers and ears can also be effective and, under certain circumstances, there is a high position dependency of the listening position. Also, the interference signal cannot be recorded directly at the listening position; it must be derived from microphones in the best possible proximity.

[0051] This document addresses an active, seat-oriented noise reduction system (ANC system) for a vehicle for passenger transport. The exemplary embodiment for an overall system can comprise: an active headrest per seat with several, preferably optimally arranged and aligned loudspeakers 7, which may have a specific embodiment such as gradient loudspeakers, one or more microphones 8 for detecting the disturbing noises in the headrests (and / or in the vehicle interior), a zonal control unit 3 per seat for generating the anti-noise signal, a unit 4 per seat for detecting the position and orientation of the head, several reference sensors 6 (e.g.Accelerator sensor) on the chassis and / or engine for deriving the interference signal, a central control module 5 for coordinating the individual systems, for processing the signals from the reference sensors, for recording additional information by networking with a vehicle data bus or other sensors and a local data bus 9 for networking all control units. Figs. 1 illustrates the schematic structure of the overall system and the local units 2, each of which is assigned to a seat.

[0052] The goal of the noise reduction system is to provide an improved, seat-oriented ANC system for passenger vehicles. The overall system is designed as an adaptive feedforward system. For this purpose, signals from several reference sensors 6 located on the chassis and engine are filtered using one or more (digital) filters to generate a control signal for noise reduction. These filters are adaptively adjusted via the reference signal and the error signal. A filter is understood to mean one or more filters for various reference signals and for generating the control signals for the loudspeakers. The filters, their adaptation, and the other signal processing described below can be carried out using a digital processor, for example, a digital signal processor (DSP).For this purpose, the zonal control unit 3 and / or the central control module 5 can have one or more processors as well as a memory and input / output elements. For the parallel processing of the signals, as enabled by the present disclosure, multiple processors or a processor with multiple processor cores can be provided.

[0053] Figs. 2 schematically shows the essential paths and processing of the electrical (or digital) and acoustic signals of an embodiment of the noise reduction system. With the vector x[n] are the reference signals of the sensors 6 for the sampling time n. The vector one] indicates the noise reduction control signals for the speakers 7. Under r[n] are filtered reference signals for the adaptation of the control filter(s) w (reference symbol 10). of [n]the interference signal is detected at the listening position (user’s ear) and with dm [n] the interference signal at the microphone position (e.g. at a neck rest). The signals from microphones 8 are vectored m[n] Finally, e[n] the error signals at the listening position, as they result from the sum of the effect of the control signals at the listening position (via the transmission path ge or G e ) and the interference signal of [n] The transmission distance between loudspeakers 7 and microphones 8 is denoted by gm and G m, respectively. The symbols pe and pm represent the unknown transmission distances from the noise generators 13 (such as engine, tires, wind) to the listening location and the microphones, respectively. Estimated values ​​are marked with a ^ symbol, for example, ĝ e is the estimated transmission distance between loudspeaker and listening location and its [ n] for the estimated error signals. Also shown is one (or more) reference signal filters 11 for generating the filtered reference signals r[n] from the reference signals x[n] by means of a digital filter that determines the estimated transmission distance ĝ e between loudspeaker and listening position. Finally, 12 denotes the Remote Microphone Technique (RMT), which calculates the estimated error signal its [ n ] at the listening location from the microphone signals m[n] and the control signals one] For this purpose, the estimated transmission distances ĝ e and ĝ m , as well as an estimated transmission distance or opt between the positions of the microphones and the listening location. An adaptation device 18 is provided for adapting the control filters 10, which, among other things, uses an adaptation step size µ.

[0054] The overall system can be divided into two parts: (i) the adaptive adjustment of the filter(s) for noise reduction and (ii) the estimation of the error signal at the listening position. First, the principle of a partitioned-block MIMO / MISO ANC system is explained.

[0055] One aspect of the present invention addresses the adaptive adjustment of the filters 10 for noise reduction. These filters can generally be implemented in both the time domain and the frequency domain using partitioned-block processing. This takes into account that at the sampling time n multiple reference signals x [ n ] = [ x 0 [ n ] x 1 [ n ], ..., x N x-1 [ n ]] T< , several loudspeakers 7 in the headrests or in the room near the listening position with their signals u[n] = [ u 0 [ n ], u 1 [ n], ..., u N U-1 [ n ]] T< and several error signals e [ n ] = [e 0[n], e1[n], ..., e N e-1 [ n ]] T< must be processed at the positions where the noise level is to be reduced.

[0056] The reference signals x [ n ] are recorded by 6 (acceleration) sensors on the vehicle. The required number and position depend on the vehicle model and the available computing power and data bus width and cannot be generically determined. Tools such as multiple coherence or iterative optimization / search algorithms can be used to efficiently locate suitable positions. These reference signals are processed by a control filter. w 10 to obtain the signals for the individual loudspeakers 7. This is usually implemented as a transversal filter (or FIR filter). Overall, a MIMO implementation requires a set of NU × NX Control filters 10 with each NW filter coefficients are required. In principle, the number of filter coefficients for the individual control filters can be different. i -th coefficient of the filter for the k -th reference signal and the m -th speaker can be used as wm,k,i The control signal for the m -th loudspeaker 7 is thus u m n = ∑ k = 0 N x − 1 ∑ i = 0 N W − 1 w m , k , i x k n − i .

[0057] For the adaptation, so-called filtered reference signals are required. These are obtained by filtering the reference signals x[n] using the reference signal filters 11 with the measured or estimated transmission paths. Ĝ e ( z) between all loudspeakers and all considered listening positions. Modeled as an FIR filter with N ge Taps can be j -th coefficient between the m -th loudspeaker and the position of the l -th error signal as ĝ e,l,m,j (Listening location l ), the ^-operator indicates an estimate of the transmission distance. All N x Reference signals are now available with all NE × NU Transmission lines filtered for the k -th reference signal, the m-th loudspeaker and the l -th error position this corresponds to r l , m , k n = ∑ j = 0 N g e − 1 g ^ e , l , m , j x k n − j .

[0058] The adaptation of the control filters 10 is typically performed using the adaptive feedforward Filtered-Reference Least-Mean-Square (FxLMS) algorithm. The term "feedforward" refers to the fact that the reference signals precede the interference signal at the listening position, which requires time to filter the signals. In conventional adaptive Least-Mean-Square (LMS) filters, the filter output is added directly to the interference signals. In an ANC application, however, the filter output is subjected to the non-ideal behavior of the loudspeakers and the acoustic transmission path to the listening position. This would prevent a conventional LMS filter from converging in most cases. However, the FxLMS approach utilizes the principle of superposition of linear and time-invariant systems, and the reference signal for adaptation is filtered precisely with this transmission path, or an estimate / model of it.The adaptation is performed in the time domain by changing the filter coefficients to reduce the square error at the listening position. This is done in the adaptation device 18 with . w m , k , i = 1 − ρ w m , k , i − μ ∑ l = 0 N E − 1 r l , m , k n − i e l n , where µ an adaptation step size and ρ E [0; 1) represents a so-called "leakage" coefficient, which can prevent instability and oversized filter coefficients over longer adaptation periods. With the special case ρ = 0, a conventional FxLMS algorithm can be described without leakage.

[0059] To ensure optimal adaptation, the algorithm can be implemented as a normalized FxLMS (NFxLMS). The adaptation step size is normalized with the estimated (instantaneous or recursively filtered) signal energy of the filtered reference signals.

[0060] A special form of the MIMO approach is discussed below. As a rule, two listening positions or error signals can be assumed – namely the sound at a person's two ears. If exactly two loudspeakers are used for active noise reduction, whose cross paths to the ears are negligible (i.e., if they have an orientation and design such that, for example, only the path from the left loudspeaker to the left ear is relevant and the path from the left loudspeaker to the right ear is negligible), two parallel MISO systems can be used for ANC. Both systems have the same reference signals at the input, but each generates the output signal for only one loudspeaker.

[0061] Each of the two control filters w 0 and w 1 is responsible for a main path (left speaker to left ear, right speaker to right ear) and generates a control signal u 0 [n ] or u 1 [ n ], which is reproduced by the corresponding loudspeaker. The corresponding NW Filter coefficients in the time domain thus correspond, for example, for the k-th input (reference signal) and the 0-th output w 0, k = [ w 0, k ,0 , w 0, k ,1 , ... , w 0,k,NW -1 ] T< As already mentioned, for each main path (0 or 1) N x Filters are provided, each processing a reference signal for this path. The calculation is identical to the MIMO case already discussed.

[0062] For the dual MISO design, the cross paths (right speaker to left ear and left speaker to right ear) are not considered separately for adaptation by the algorithm, since in the usual resting position of the head there is sufficient attenuation compared to the main paths and they can therefore be neglected. Compared to the MIMO approach, a significantly smaller number of 2 • N x filtered reference signals. With the channel index l = [0; 1] EN +< for the distance from loudspeaker 0 to the location of the error signal 0 or loudspeaker 1 to the location of the error signal 1, the filtered reference signals are r l , k n = ∑ j = 0 N g e − 1 g ^ e , l , l , j x k n − j .

[0063] This corresponds to ĝ e,l,l the transmission paths from left loudspeaker to left ear, or right loudspeaker to right ear. The adaptation of the filter coefficients in the time domain in the adaptation devices 18-0 and 18-1 for the two main paths is now carried out with w l , k , i = 1 − ρ w l , k , i − μr l , k n − i e l n , The changed indexing results from the fact that for the Dual-MISO implementation m = l This significantly reduces the computational and memory requirements compared to the full-fledged MIMO approach. A block diagram for an embodiment of the Dual-MISO approach in the time domain is shown in Figs. 3 where the filtered reference signals are represented as vector r l [[ n ] = [ r l ,0 [ n ] , r l, 1 [ n ], ..., r l,N x -1 ] T< are shown. Acoustic paths are marked with dashed lines.

[0064] In the Dual-MISO approach, as a special case of the MIMO approach, the adaptation of the control filter w 0 10-0 only based on the filtered reference signal ro generated by the reference signal filter 11-0 (taking into account the estimated transfer function ĝ e ,0,0 between loudspeaker 0 and listening location 0) and the error signal e 0 at the listening location 0. The same applies to control filters w 1 10-1. In Figs. 3 In addition (those not considered for the Dual-MISO approach) acoustic transmission paths g e ,0,1 and g e ,1,0 for the cross paths. However, in the case of the general MIMO approach, the cross paths must be taken into account in the adaptation, as explained above, even if they do not make a significant contribution. This affects the conditioning and adaptation speed of the entire algorithm.

[0065] To reduce the computational effort for long filter lengths, a conversion in the frequency domain can be carried out using FFT and overlap-save methods. For simplicity, only the general formulations for the l -th output channel or loudspeaker.

[0066] In order to reduce the latencies caused by processing in the frequency domain and to reduce them to a real-time range, the adaptive filters are NP shorter subfilters, so-called partitions. The system is based on block processing, the block length L in samples is determined by the system configuration and largely determines the latency. The current block of the input signal in the time domain is concatenated with the previous block and C = 2 L -point DFT or FFT is transformed into the frequency domain. For the b -th signal block and the k-th reference signal results in X k,b = FFT {[x[bL − 2L], x [ bL - 2 L + 1], ... , x [ bL - 1]] T< }, where the relationship to the previous time index is n = bL In this notation, all (discrete) frequency coefficients or frequency bins for an input channel or reference signal k and signal block b to X k,b summarized, however, refers X b [ κ ] = [ X 0,b [k] ,X 1,b [κ] , ... , X N x -1, b [kappa]] T< to all transformed reference signals for a single frequency bin κ In the following, symbols in capital letters indicate the frequency range.

[0067] The partitioned control filters in the frequency domain correspond to a decomposition in the time domain into LTaps length, zero-padded to also the length L (ie to L zeros added) and transformed into the frequency domain. This is Figs. 4 for a reference signal. The figure shows the decomposition of a filter w l of length N w in N p subfilters (partitions) w l,0 to w l ,Np-1 for partitioned-block and overlap-save processing. In each case, L=N w / N p of the N w filter coefficients of the entire filter in the time domain are assigned sequentially to the subfilters. One can also see the addition of the filter coefficients in the time domain by appending L zeros, so that each subfilter has 2 L filter coefficients. The relationship between the filter coefficients in the time domain and the filter spectra W is also evident using the Fourier transform. l , p .

[0068] In general, this corresponds to the p- ten sub-filter, for the k-th reference signal and the l -th output channel W l , k , p = FFT w l , k , pL , w l , k , pL + 1 , … , w l , k , pL + L − 1 , 0 , … , 0 ︸ L T , or for a frequency bin κ and all input channels W l,p [ κ ] = [ W l, 0 ,p [ κ ] ,W l ,1, p [ κ ] , ... , W l,N x -1 ,p [ κ ]] T< . The buffered reference signal blocks of the previous blocks are now multiplied bin-wise in the frequency domain with the corresponding partition to U ˜ l , b κ = ∑ p = 0 N P − 1 W l , p T κ X b − p κ = ∑ p = 0 N P − 1 ∑ k = 0 N x − 1 W l , k , p κ X k , b − p κ .

[0069] The control signal for block b and a frequency bin κ is formed from the sum of the partial signals of the individual partitions for the frequency bin κ, whereby the partial signals are formed by the scalar product of the vectors of the spectra of the reference signals and the corresponding partial filter. The spectrum for the error signal is calculated separately in the frequency domain for each frequency bin. The calculation is carried out separately for the individual partitions of the control filter, and these partial results are then added for all partitions. In this case, time-shifted values ​​for the spectra of the reference signals X k,bp The time shift bp corresponds to the consecutive number of the partition p. This calculation can be performed in parallel on DSPs with multiple cores, which can significantly increase efficiency.

[0070] Figs. 5schematically illustrates the partitioned-block approach and shows two possible block diagrams of a partitioned-block filter in the frequency domain. z - L < a delay of L Samples or a signal block. The control filter 10 is divided into a single parallel sub-filter 10'. The filter 10 can be implemented with either a pre-delay (delay elements 14 at the input data) or a post-delay (delay elements 14 at the sub-filter outputs; right in the image). The result is identical, neglecting rounding errors. When using the "predelay", the Fourier-transformed blocks for a reference signal X b are LSamples, or a signal block, are delayed and processed with the corresponding subfilter for the output channel / (multiplied as specified above). It should be noted that the delays can be implemented by simply buffering the values ​​of the previous blocks, since with a block length of L, all transformed blocks essentially overlap by 50%. Therefore, only the most recent input block needs to be transformed, and transformed blocks from previous points in time are buffered.

[0071] The partial results of the individual sub-filters are then added in the frequency domain to obtain the spectrum Ũ l , b the control signal for this output channel land block b. In post-delay processing, the subfilters 10' are applied to the undelayed reference signal spectra. Subsequently, the partial results of the subfilters are delayed on the filter output side and before addition. Mathematically, both implementation options are equivalent, and the spectrum of the control signal is calculated using the formula mentioned above.

[0072] The above-mentioned operation for calculating the control signals in the frequency domain, when transformed back to the time domain, corresponds to a cyclic convolution, not a preferred linear convolution. Signal components of the result that exceed the block size are thus shifted back to the beginning of the block and overlap with the actually desired payload data.

[0073] These signal components with cyclical portions in the result are generally invalid and do not match the linear convolution. However, a usable result can be calculated using the overlap-save algorithm. Two blocks of the input signal (i.e., 2L samples) are transformed into the frequency domain. The first part of a filter has a length of L coefficients in the time domain. Before transformation into the frequency domain, L zeros are added at the end (zero padding). This allows part of the filter result to be kept free of cyclic artifacts. The control signal is now calculated by element-wise multiplication in the frequency domain. The usable time-domain signal is now contained in the last L samples of the inversely transformed block [ u 0 [ bL ] ,u 0 [ bL + 1], ... , u 0 [ bL + L ]] T< = P y * IFFT { Ũ 0 ,b}, where the matrix P y = [ 0 L×(CL)< I L×L< ] only causal components without cyclic artifacts are taken into account. 0 a matrix of zeros and I the identity matrix with the specified dimensions. This process is described in Figs. 6 for a single signal block of the input data. Two blocks x b-1 and xb of the input (reference) data with their respective lengths are shown. L. As already mentioned, the FFT uses the transformation length C = 2 L. This input data is cyclically convolved in the partitioned filters. As already described in Figs. 4As shown, the filter coefficients in the time domain are supplemented with L zeros at their end up to length C. The result of the cyclic convolution is a block ub for the control signal of length C, where the valid signal data is located in the second half of the block of length L, while cyclic parts occur in the first half of the block, which can be ignored. In this way, one obtains for the block xb of length L a control signal ub of the same length.

[0074] The adaptation of the control filters is carried out analogously in the frequency domain. The filtered reference signals can also be determined using the partitioned-block approach. For this, the filter Ĝ e,l,l as described above, are first divided into subfilters. Thus, the filtered reference signals are R ˜ l , b κ = ∑ p = 0 N P Ge − 1 G ^ e , l , l , p κ X b − p κ calculated, where N PGe , one of N p The number of partitions can be different. The partition length L is however identical.

[0075] The control filter can now be adapted directly with the just-calculated result of the filtered reference signals in the frequency domain using an "unconstrained" approach, i.e., without considering boundary conditions. Cyclic components in the spectrum of the filtered reference signals are not removed, which leads to a certain amount of "noise" in the results, which can, however, be quite acceptable. For better adaptation without cyclic effects, a "constrained" approach can also be used to consider boundary conditions. In this case, the resulting spectrum of the filtered reference signals is transformed into the time domain, with only the last L valid samples are used (see explanations for Figs. 6 ), connected to the previous block, and transformed back into the frequency domain. For the k -th input channel this results in R l , k , b = FFT r l , k bL − 2 L , r l , k bL − 2 L + 1 , … , r l , k bL − L − 1 , P y IFFT R ˜ l , k , b T T .

[0076] For the adaptation itself, the error signals are zero-padding before the samples (i.e. inserting L zeros at the beginning of a block of error signals) as E l , b = FFT 0 , … , 0 ︸ L , e l bL − L , e l bL − L + 1 , … , e l bL − 1 T transformed into the frequency domain. By setting the first L values ​​of a block of error signals to zero and using the conjugate complex filtered reference signals, a cross-correlation between the filtered reference signals and the error signal is formed for the adaptation. The adaptation of the control filters thus results in W ˜ l , p κ = W l , p κ − μ R l , b − p ∗ k E l , b κ , where (·)* represents the complex conjugate realization. The adaptation of the spectrum of a partition pThe control filter is applied in the frequency domain separately for each frequency bin in which a portion (µ) of a correction term determined by multiplying between a spectral representation of the filtered reference signals and a spectral representation of the error signals is applied (subtracted), whereby the spectral representation of the filtered reference signals for a temporally previous block and the spectral representation of the error signals of the current block b The time shift bp for the spectral representation of the error signals corresponds to the consecutive number of the partition p In this way, a very efficient adaptation of the control filters can be achieved.

[0077] As a rule, it is advisable to pay attention to the causality of the filter coefficients when adapting in the frequency domain. This means that the "zero-padded" (i.e., zero-filled) parts of the filter coefficients (see Figs. 4 and 6 ) drift away from zero over time. This can be corrected so that the overlap-save algorithm delivers valid output data. To do this, the filters are transformed into the time domain, all zero-padded components are set back to zero, and then transformed back with W l , p = FFT 1 , … , 1 ︸ L 0 , … , 0 ︸ L T ⊙ IFFT W ˜ l , p , where ⊙ corresponds to an element-wise (Hadamard) multiplication of vectors or matrices. This need not be performed during each computation block and / or for all partitions, but can also be performed sequentially, e.g., one partition per computation block in round-robin scheduling. Alternatively, during adaptation, only the causal part of the change W l , p κ = W l , p κ − μFFT 1 , … , 1 ︸ L 0 , … , 0 ︸ L T ⊙ IFFT R l , b − p ∗ κ E l , b κ be used.

[0078] Analogous to the time domain calculation, a normalized adaptation coefficient can also be selected here, whereby the signal energy can be estimated in the frequency domain.

[0079] Figs. 7 shows a possible block diagram for the partitioned-block MISO FxLMS algorithm for one channel. The block diagram is for an output channel of the partitioned-block FxLMS algorithm with pre-delay implementation and unconstrained filtered reference signals. For clarity, the adaptation path has been shown in dashed lines.

[0080] The figure shows schematically the processing of the reference input signals x[n] by means of a blockwise C-point FFT (or DFT) in a transformation unit 15, the filtering of the transformed input signal X b for the block b with a partitioned-block filter 10 and the inverse transformation via an inverse FFT in an inverse transformation unit 16, which also contains the last LValues ​​of the inversely transformed control signal ũ l,b for the output channel / and selects block b as its output. The control signal u[n] is then sent to the loudspeaker 7, which generates the anti-noise. The filtering of the transformed input signal X b with the sub-filters W l,p 10' is carried out block by block in the frequency domain, as in connection with Figs. 5 explained. The spectrum X b of the input signal for block b is filtered by the subfilter W l,o, the spectrum of the previous block X b-1 is filtered by the subfilter W l,1, etc. The delays by L samples are indicated by the delay blocks 14. Finally, the partial results of the individual subfilters are added in the frequency domain to obtain the spectrum Ũ l , b of the control signal for the output channel / and the block b to obtain.

[0081] For the adaptation of the sub-filters by means of adaptation device 18, the spectrum of the input signal for block b is compared with the spectrum of the estimated transfer function G̃ el,l between speakers of the output channel l and the associated listening position / in the reference signal filter 11 to obtain the filtered spectrum R l ,b of the reference signal. This is used together with the spectrum E l ,b of the error signal for the output channel / for adapting the first, undelayed sub-filter W l , 0 as shown above. The transformation block 17 for the error signal el [n] also performs a blockwise C-point FFT or DFT after a block of L Error signal values ​​by prefixing L Zeros were added to a size of C values. The filtered spectrum R l,b-1 of the reference signal of the previous block b-1 is combined with the spectrum E l ,b of the error signal for adapting the first delayed sub-filter W / , 1 etc. For a MIMO case, the adaptation would be similar, but apart from the higher number of control filters and reference signals, all error signals must also be taken into account for all control filters and processed accordingly.

[0082] In summary, partitioned-block processing offers a number of practical advantages, particularly for processing in the frequency domain. It enables the processing of high filter orders with comparatively low resource consumption and low latency, and parallelization is also feasible. However, to effectively utilize these advantages, hardware with a correspondingly optimized DFT implementation and a block length of at least 64 samples is advantageous.

[0083] In the following, the implementation of a delayed RMT for virtual sensing as a partitioned-block filter and connection with a partitioned-block FxLMS filter is explained.

[0084] The second aspect of the disclosure concerns the combination of FxLMS filtering with the Remote Microphone Technique (RMT) in the frequency domain as a partitioned-block approach. Error signal estimation using RMT is essentially a MIMO approach, meaning that multiple error signals at the listening locations are estimated based on multiple microphone signals. The ANC system and FxLMS filtering can be implemented as MIMO or dual-MISO. If both the FxLMS algorithm and RMT operate with the same block size, an efficient implementation with low latency can be achieved. For better understanding, the description begins in the z-domain; the definition in the frequency domain will be given later.

[0085] The RMT is used to estimate the error signal at the listening position. For a high-performance ANC algorithm with a broad frequency range, the most accurate error signal at the listening position is required. However, sensors cannot usually be placed directly in or on a user's ears without negatively affecting comfort. Instead, N m Monitoring microphones placed near the ears (e.g. on headrests or a seat) pick up both the primary noise dm [ n ] = [ d m, 0 [ n ] , d m, 1 [ n ], ..., d m,N m -1 [ n ]] T< at the microphone positions, as well as the reproduced control signals u [ n ] = [ u 0 [ n ] ,u 1 [ n ]] T< , exposed to the acoustic transmission path G m z = G m , 0,0 z G m , 0,1 z ⋮ ⋮ G m , N m − 1,0 z G m , N m − 1,1 z between the m-th loudspeaker and the two error or listening positions (0,1).

[0086] Thus, the signal at the microphones is M z = M 0 z , M 1 z , … , M N M − 1 z T = D m z + G m z U z .

[0087] Conversely, if a measured or estimated transmission distance Ĝ m ( z ) and the control signals in addition to the microphone signals are known, the primary noise D ^ m z = M z − G ^ m z U z be appreciated.

[0088] Via a matrix of so-called observation filters O (z) can be recalculated to the noise signal at the listening position. These filters can be interpreted as a transfer function between the monitoring microphones and the listening positions (0,1). Thus, the primary noise signal at the listening positions is D ^ e z = D e , 0 z , D e , 1 z T = O z M z − G ^ m z U z By adding the control signals with the corresponding transmission path to the listening position, the estimated error signal at the listening positions is E ^ z = E ^ 0 z , E ^ 1 z T = O z M z − G ^ m z U z + G ^ e z U z appreciated.

[0089] An embodiment of a block diagram of the RMT 12 for one channel is shown schematically in Figs. 8 The task of the RMT is to detect the error signal its [ n ] at the listening position using the microphone signals m[n] and the control signals u[n] (see also Figs. 2 ). For this purpose, the control signal u[n] is filtered with the filter block 21 with the estimated transmission path Ĝ m ( z ) and with the filter block 22 with the estimated transmission distance Ĝ e ( z ) filtered. Filter block 26 is similar in the z-range to block 11 for the estimated transmission distance ĝ e in Figs. 2 , which generates the filtered reference signals r[n], is applied here to the control signals u[n]. Block 20 represents the transmission path G m (z)between loudspeaker 7 and microphone 8. The estimated interference signal d̂ m ( n) at the location of the microphone 8 is filtered by means of filter block 23 with the observation filter O (z) filtered to obtain the estimated noise signal d̂ e ( n ) at the listening location, which is finally added to the estimated effect of the control signal.

[0090] To calculate the two error signals at both ears of the user, both the MIMO and MISO approaches use as many or all microphone signals as possible. The microphones do not have to be clearly assigned to one ear and positioned accordingly. However, with the MISO method, the control filter adaptation only works with the estimated error signal for the respective channel. In other words, the control filter(s) (or the corresponding control subfilters) that generate(s) the control signal for a loudspeaker for a listening location (the user's left or right ear) are adapted based on the estimated error signal for this listening location. As already mentioned, however, the cross-paths between loudspeakers and listening locations (ears) are neglected for the adaptation of the control filters.

[0091] For efficient calculations in the frequency domain, partitioned-block processing in the frequency domain is also recommended. In a first step, the spectra of the control signals at the listening position are b -th block Y ^ e , b κ = ∑ p = 0 N P Ge − 1 G ^ e , p κ U b − p κ certainly. U b speaks the C -point transforms of the control signals (ie as above 2 concatenated blocks with length L = C / 2). The spectrum for this intermediate value is calculated in the frequency domain separately for each frequency bin. As above, a subfilter Ĝ e ,p is the spectrum of a time-shifted (previous) block of control signals The bp The time shift bp for the spectral representation of the control signals corresponds to the consecutive number of the partition p. This allows for very efficient calculations. The same applies to the spectra of the control signals at the microphone positions. Y ^ m , b κ = ∑ p = 0 N P Gm − 1 G ^ m , p κ U b − p κ , where N PGm the number of partitions of Ĝ m corresponds to.

[0092] Also at THE e,b and THE m,b, constraints can be taken into account if necessary by transforming them into the time domain, only the last L = C / 2 samples can be retained, concatenated with the result of the previous block in the time domain, and transformed back again. This way, cyclic effects can be eliminated.

[0093] With these calculated intermediate results buffered for several blocks, the error signal estimation can now be implemented using RMT as E ^ b κ = ∑ p = 0 N P O − 1 O p κ M b − p κ − Y ^ m , b − p κ + Y ^ e , b κ , where the observation filter in N PO Partial filter is broken down and divided into the frequency range as O p The blocked and transformed microphone signals M b These were analogous to You b composed of two consecutive input blocks in the time domain before a C -point FFT. Similar to above, The p [ κ ] and M b [ κ ] to get the corresponding values ​​for a single frequency bin κ . The error signal for block b and a frequency bin κis formed from the sum of the partial error signals of the individual partitions for the frequency bin, whereby the partial error signals are formed by matrix multiplications of the spectra of the control signals and the microphone signals with the spectra of the various transfer functions. The calculation of the spectrum for the error signal is carried out separately in the frequency domain for each frequency bin. As above, the calculation is carried out separately for the individual partitions O p of the observation filter, and these partial results are then added for all partitions. In this case, time-shifted values ​​for the spectra of the microphone signals M bp and for the intermediate size ¶ m,bp which estimates a spectrum representative of the control signals after transmission to the microphone locations. The temporal shift bp corresponds to the consecutive number of the partition pFor a partial error of a partition, values ​​for the spectrum of the microphone error signals and values ​​for a spectrum representative of the transmission of the control signals to the microphone locations, for temporally shifted blocks of the error signals or control signals, and the corresponding partition of the observation filter are used. This allows for a very efficient calculation.

[0094] To take additional boundary conditions into account, ITS b how the other (intermediate) quantities are transformed into the time domain, cyclic components are zeroed and transformed back again.

[0095] The obvious advantage of this partitioned-block approach, besides higher computational efficiency, is that ITS b from the RMT estimation, with correspondingly identical parameters for sampling rate, block length L, and DFT order C, corresponds to the input format for the FxLMS algorithm. If these parameters match, the spectra can be directly imported, especially in an "unconstrained" estimation, without computationally intensive additional frequency transformations and signal buffering. Likewise, additional, delay-inducing signal blocking is eliminated, thus reducing processing latency.

[0096] The Observation Filter O(z) can usually be determined from signal statistics in both the time and frequency domains. The definition in the frequency domain is discussed below. For this purpose, the primary interference signals are used in measurement scenarios. d e [ n ] at the listening position, for example with an artificial head or binaural microphones, and at the monitoring microphones d m [ n] recorded. After a transformation into the frequency domain, the respective auto- or cross-power density spectra between all monitoring microphones S d m d m κ = E D m κ D m H κ , or between all monitoring microphones and the listening positions S d m d e κ = E D e κ D m H κ be appreciated. E ⋅ the expected value and (·) H< the Hermitian of a complex matrix. On the identity matrix I and a regularization β, an optimal observation filter in the frequency domain with O opt κ = S d m d e κ S d m d m κ + β I − 1 be calculated. Regularization β >_ 0 may be required to ensure stable and well-conditioned results. Note that the optimal observation filter depends on both the head position and the primary noise signal itself. Therefore, different filter sets can be calculated and selected during operation for satisfactory performance and for different driving conditions and head positions.

[0097] With the above-presented variant of RMT, those signal components can be causally estimated that first reach the monitoring microphones and then the listening position; acausal components that first reach the microphones cannot be estimated. The so-called delayed RMT takes this into account. It does not attempt to determine the current error signals from the currently measured signals, but rather compares past error signals with E ^ Δ z = z − Δ E ^ z = z − Δ O z M z − G ^ m z U z + z − Δ G ^ e z U z = O Δ z M z − G ^ m z U z + z − Δ G ^ e z U z , The subscript Δ describes a version of a signal or filter delayed by Δ samples. This can be implemented particularly easily and resource-efficiently if Δ in samples is a multiple of the block length L corresponds to Δ = sL and s EN +< . With this definition, the partitioned-block definition can be E ^ b − s κ = ∑ p = 0 N P O − 1 O Δ , p κ M b − p κ − Y ^ m , b − p κ + Y ^ e , b − s κ The error signal for the past block bs and a frequency bin κis formed from the sum of the partial error signals of the individual partitions for the frequency bin, whereby the delay by Δ = sL at the microphone is determined by means of O Δ, p [ κ ] is taken into account for the frequency bin. O Δ,p [ κ ] corresponds to the observation subfilter for the p-th partition delayed by Δ samples. The delayed subfilter is defined as a separate variable, since the delay is already taken into account in the filter design.

[0098] The observation filter for the delayed RMT O Δ , opt κ = S d m d e , Δ κ S d m d m κ + β I − 1 is measured via a "delayed" cross-power density spectrum S d m d e , Δ κ = E D e , Δ κ D m H κ calculated. For this purpose, only the signals at the listening position are delayed by Δ samples before transformation into the frequency domain.

[0099] With delayed RMT, it should also be noted that the filtered reference signals for adapting the FxLMS algorithm must also be delayed by Δ samples to achieve synchronicity with the error signals. This is another reason to choose the delay as a multiple of the block size. Alternatively, the delay for delayed RMT and the FxLMS algorithm can already be included in the estimated transmission distances. Ĝ e This also includes L Different delays can be implemented efficiently.

[0100] Figs. 9 shows an embodiment of a schematic structure of the delayed-RMT in conjunction with an FxLMS algorithm. In addition to the Figs. 8Delay elements 25 are provided in the known filter blocks 20, 21, 22, which represent a delay of Δ samples. The delay elements 25 are provided in the path for considering the control signals u[n] at the listening position with the delay Δ and in the path for generating the filtered and delayed reference signals r[n- Δ]. Unlike in Figs. 8 generates the delayed observation filter O Δ ( z ) 24 the delayed estimated noise signal d̂ e (n - Δ ) at the listening location. For adaptation, the delayed estimated error signal is obtained by adding this value to the delayed effect of the control signals at the listening location. its [ n - Δ]. Filter block 26 corresponds in the z-range to block 11 in Figs. 2 for the estimated transmission distance ĝ e and generates the filtered reference signals r[n].

[0101] When combining the delayed RMT with the previously described processing using partial filters, the Figs. 7 shown adaptation path directly before or after the filter block 11, which generates the filtered reference signals for the adaptation, to insert a delay element 25, which generates the delay by Δ samples in order to align the signals for the adaptation devices 18 correctly in time. As can be seen from Figs. 9 can be seen, the estimated error signal is also its or its spectrum ITS delayed by Δ samples.

[0102] The following describes the required signal processing steps for an exemplary application in a vehicle. The proposed method is particularly suitable for noise reduction in a vehicle. In this application, the defined location is the listening location of a user, for example, a passenger of the vehicle sitting in a seat. The listening location can also be understood as the positions of the user's two ears. Of course, this is only one possible scenario, and the proposed devices and methods can generally be used for noise reduction in a specific location.

[0103] Figs. 10shows steps for a method 100 for suppressing noise at a listening location according to an embodiment of the present disclosure. Some of the following steps can also be performed in a different order or in parallel.

[0104] In step 110, several reference signals x correlated with the noises are detected by sensors 6 conveniently arranged on the vehicle, for example, for detecting engine, road, and / or wind noise. The reference signals x can be preprocessed in a central control module 5 and then further processed in one or more local units 2 to generate the respective control signals for the listening location assigned to the corresponding local unit 2. Such a configuration is suitable for noise reduction for multiple listening locations / passengers of a vehicle.

[0105] In step 120, the reference signals x are transformed into the frequency domain in order to generate spectral representations X (spectra) of the reference signals, for example by means of an FFT or DFT.

[0106] In step 130, the reference signals are filtered in the frequency domain with an estimate of the transmission path Ĝ e between the loudspeakers 7 and the listening location. This filtering can be achieved with a digital filter by multiplying the spectra of the reference signals X with the spectrum of the transmission path Ĝ e and generates the spectra R of the filtered reference signals. Usually, two loudspeakers 7 are used, each assigned to one of the user's ears and arranged accordingly near the respective ear (e.g., in a seat headrest).

[0107] In step 140, several acoustic error signals m are detected at positions near the listening location. For example, several error signals are detected by microphones 8 arranged on or near a headrest of a seat. Step 140 can be performed in any order to steps 110-130 or even in parallel.

[0108] In step 150, an error signal e at the user's listening location is estimated based on the error signals m detected by the microphones 8. The listening location can be the user's two ears, and in this case, the error signal at the listening location comprises two partial signals e 0 for the left ear and e 1 for the right ear. The error signal for the listening location is preferably estimated in the frequency domain and taking into account an estimate of the spectrum O of the transmission paths between the microphones 8 and the listening location.

[0109] Step 160 relates to generating a plurality of control signals u for generating sound signals by means of the loudspeakers 7 for suppressing the noise, based on the reference signals x and the error signals e and by means of adaptive control signal filters 10. For example, two control signals u are generated for two loudspeakers. The control signals are generated based on a plurality of reference signals x and by means of processing in the frequency domain. The control signal filters 10 can be implemented as digital filters in order to generate the control signals by digitally filtering the reference signals x by multiplying spectra X of the reference signals by respective spectra W of the control signal filters 10. The processing is carried out in blocks of length L of consecutive reference signal values.

[0110] The control signal filters 10 are implemented by a plurality of control signal sub-filters 10' arranged in parallel. The p-th control signal sub-filter 10' filters the spectrum X bp of a block of reference signals shifted in time by p blocks. The control signals are generated based on a sum of the outputs of the control signal sub-filters 10'. Alternatively, the control signal sub-filters 10' can each filter a spectrum X b of a current block of reference signals, and the results of the control signal sub-filters 10' are each added, shifted in time by one block.

[0111] The transformation of the reference signals x into the frequency domain is carried out based on two consecutive blocks. The control signal subfilters 10' for filtering a specific reference signal xl each have a number L of relevant filter coefficients w l,i of the entire control signal filter in the time domain, after which L zeros are inserted (see Figs. 4 ). Thus, the number of filter coefficients per subfilter corresponds to the number C of signal values ​​of the reference signal in 2 blocks.

[0112] The method may further comprise a back-transformation of the spectra U of the control signals into the time domain and a selection of a portion of the back-transformed values ​​for a block. The back-transformation generates a vector with a length of 2 blocks, and the selection of a portion of the back-transformed control signal values ​​selects a block of values ​​from the vector. In the overlap-save method, the last L values ​​from the block of values ​​are selected.

[0113] In step 170, the control signal filters 10 are adapted using a method for reducing an error function based on the detected error signals. The feedforward filtered-reference least-mean-square (FxLMS) algorithm can be used here, which reduces a quadratic error function step by step, iteratively, by targeted adaptation of the filter parameters using a gradient method. The parallel-arranged control signal sub-filters 10' are each adapted separately in the frequency domain, with the spectrum E of the error signal e for the listening location being used to adapt the control signal sub-filters. The control signal sub-filters 10' are adapted based on spectra R of the filtered reference signals and on spectra E of the error signals at the listening location. The adaptation of a portion of the control signal sub-filters 10' is based on a filtered spectrum R bp of a time-shifted block of the reference signals.

[0114] Step 180 includes outputting the sound signals by means of the loudspeakers 7 and based on the control signals u to suppress the noises at the listening location.

[0115] As already mentioned, the signal processing can be performed at least partially in the frequency domain. The above-mentioned configurations are only examples of possible configurations of the processing steps and can be modified in many ways. Those skilled in the art will recognize such variations of the inventive approach to noise suppression after studying the present disclosure.

[0116] The above description of exemplary embodiments includes a large number of details that are not essential to the invention defined by the claims. The description of the exemplary embodiments serves to understand the invention and is purely illustrative and should be understood without limiting the scope of protection. Those skilled in the art will recognize that the described elements and their technical effects can be combined with one another in various ways to create further exemplary embodiments covered by the claims. Furthermore, the described technical features can be used in devices and methods, for example carried out by programmable devices. They can be implemented in particular by hardware elements or by software. As is known, digital signal processing is preferably implemented by specially designed signal processors.Communication between individual components of the described device can be wired (e.g., via a bus system) or wireless (e.g., via Bluetooth or Wi-Fi). Protection is also expressly claimed for a computer-implemented implementation and the associated program or machine code in the form of data storage media or in a downloadable representation.

Claims

1. A device for suppressing noise at a defined location, comprising: a plurality of sensors (6) for detecting reference signals correlated with the noise; a plurality of acoustic output means (7) for the acoustic output of sound signals for suppressing the noise, wherein the acoustic output means are arranged near the defined location; a plurality of acoustic input means (8) for detecting acoustic error signals, wherein the acoustic input means are arranged near the defined location; and a processing unit (2) which receives and processes the reference signals and the acoustic error signals and, based on these, generates control signals for the acoustic output means and outputs them to them;wherein the processing unit generates the control signals by means of adaptive control signal filters (10), and the adaptation of the control signal filters (10) is carried out by means of a method for reducing an error function based on the detected acoustic error signals; wherein at least one control signal is generated based on a plurality of reference signals; wherein the at least one control signal is generated by at least one control signal filter (10) based on processing in the frequency domain, and the at least one control signal filter (10) is implemented by a plurality of parallel-arranged control signal sub-filters (10'), each of which is adapted separately from one another in the frequency domain.; 2. Device according to claim 1, wherein the processing unit (2) has a transformation device (15) which transforms the reference signals into the frequency domain, and wherein the at least one control signal filter (10) performs a filtering of the reference signals by multiplying spectral representations of the reference signals with respective spectral representations of the control signal sub-filters (10'), and the at least one control signal is generated based on the outputs of the control signal sub-filters (10').

3. Device according to claim 1 or 2, wherein a spectral representation of the at least one control signal in the frequency domain is formed for individual frequency bins by adding the contributions of the control signal sub-filters (10'), wherein values ​​for time-shifted blocks of the spectral representation of the reference signals are used.

4. Device according to one of the preceding claims, wherein the processing unit (2) has a reference signal filter (11) for filtering the reference signals in the frequency domain with an estimate of the transmission path between at least one acoustic output means (7) and the defined location and an adaptation device (18) for each control signal sub-filter (10'), wherein at least one adaptation device (18) adapts the corresponding control signal sub-filter (10') based on a spectral representation of the filtered reference signals and a spectral representation based on the acoustic error signals detected by the acoustic input means (8), and at least one adaptation device (18) processes a time-shifted block of the filtered spectral representation of the reference signals.

5. Device according to one of claims 2 to 4, wherein the processing of the reference signals takes place in blocks, the transformation device (15) transforms the reference signals into the frequency domain based on 2 consecutive blocks in each case and the control signal sub-filters (10') for filtering a reference signal each have a number of filter coefficients which corresponds to the number of signal values ​​of the reference signal in 2 blocks, wherein in the corresponding time domain a number of the filter coefficients which corresponds to a block is set to zero values.

6. Device according to one of the preceding claims, wherein a control signal is formed from a part of the values ​​of a spectral representation of the control signal for a block, which is transformed back into the time domain and generated by the control signal sub-filters (10').

7. Device according to one of the preceding claims, wherein the causality of the filter coefficients of an adapted control signal sub-filter (10') is taken into account by causing certain filter coefficients in the time domain to have a value of zero.

8. Device according to one of claims 4 to 7, wherein the reference signal filter (11) is realized by a plurality of reference signal sub-filters arranged in parallel and at least one reference signal sub-filter filters a time-shifted block of the reference signals, wherein cyclic components in the spectral representation of the reference signals obtained in this way are optionally removed.

9. Device according to one of the preceding claims, wherein for each acoustic output means (7) a control signal filter (10) is assigned to a reference signal and the control signal for the acoustic output means (7) is formed based on the sum of the output signals of these control signal filters (10), and wherein the adaptation of these control signal filters (10) is based on an error signal for the defined location which is assigned to this acoustic output means (7), and error signals for other defined locations are not taken into account in the adaptation.

10. Device according to one of the preceding claims for suppressing noise at a seating position in a vehicle, wherein the sensors (6) are arranged on devices of the vehicle to detect signals which are representative of noises generated by these devices, wherein the acoustic output means (7) and the acoustic input means (8) are arranged in the vicinity of a seat of the vehicle, preferably on the headrest of the seat.

11. Device according to one of the preceding claims, wherein the processing unit (2) has a device (12) for estimating an error signal at the defined location on the basis of the acoustic error signals detected by acoustic input means (8), wherein the defined location is the listening location of a user and wherein the estimation of the error signal for the listening location takes place in the frequency domain and taking into account an estimate of the transmission paths between the acoustic input means (8) and the listening location and the spectral representation of the error signal thus determined for the listening location is used for the adaptation of at least one of the control signal sub-filters (10').

12. Device according to claim 11, wherein the device (12) for estimating an error signal for the listening location has transmission filters (21, 22) which model the transmission of the control signal from the acoustic output means (7) to the locations of the acoustic input means (8) and the transmission of the control signal from the acoustic output means (7) to the listening location, wherein at least one transmission filter (21, 22) is realized by a plurality of transmission sub-filters arranged in parallel and at least one transmission sub-filter filters a time-shifted block of the control signals.

13. The device according to claim 12, wherein the spectral representation of the estimated error signal for the listening location in the frequency domain is formed for individual frequency bins by adding the contributions of partial errors to the spectral representation of the estimated error signal for the listening location, wherein values ​​for the spectral representation of the acoustic error signals and values ​​for a spectral representation representative of the control signals after transmission to the locations of the acoustic input means (8) are used.

14. Device according to claim 13, wherein for a partial error, values ​​for the spectral representation of the acoustic error signals and values ​​for a spectral representation which is representative of the control signals after transmission to the locations of the acoustic input means (8), for time-shifted blocks of the acoustic error signals or the control signals and an observation sub-filter corresponding to the partial error are used, wherein the observation sub-filter at least partially models the transmission paths between the locations of the acoustic input means (8) and the listening location.

15. Device according to one of claims 11 to 14, wherein the device (12) for estimating an error signal for the listening location estimates past error signals at the listening location on the basis of current acoustic error signals detected by acoustic input means (8), wherein optionally the time offset between the estimated past error signals at the listening location and the current acoustic error signals can be a multiple of the block length when processing the reference signals and the acoustic error signals.

16. A method for suppressing noise at a defined location, comprising: - detecting (110) a plurality of reference signals correlated with the noise; - detecting (140) a plurality of acoustic error signals at positions near the defined location; - generating (160) a plurality of control signals for generating a plurality of sound signals for suppressing the noise, based on the reference signals and the detected acoustic error signals and by means of adaptive control signal filters (10), wherein at least one control signal is generated based on a plurality of reference signals and by means of processing in the frequency domain;- adapting (170) the control signal filters (10) by means of a method for reducing an error function based on the detected acoustic error signals, wherein at least one control signal filter (10) is implemented by a plurality of parallel-arranged control signal sub-filters (10'), each of which is adapted separately from one another in the frequency domain; and - outputting (180) a plurality of sound signals at positions near the defined location to suppress the noise.

17. The method according to claim 16, comprising - a transformation (120) of the reference signals into the frequency domain in order to generate spectral representations of the reference signals; wherein the generation (160) of the at least one control signal comprises filtering the reference signals by multiplying spectral representations of the reference signals by respective spectral representations of the control signal sub-filters (10'); wherein the control signal sub-filters (10') filter temporally shifted blocks of the spectral representations of the reference signals and the at least one control signal is generated based on a sum of the outputs of the control signal sub-filters (10'), or the control signal sub-filters (10') each filter a spectral representation of a current block of the reference signals and the results of the control signal sub-filters (10') are each added temporally shifted by one block.

18. The method according to claim 16 or 17, comprising - filtering (130) the reference signals in the frequency domain with an estimate of the transmission path between at least one acoustic output means (7) and the defined location; wherein the adaptation (170) of the control signal filters (10) is carried out separately for each control signal sub-filter (10'), at least one control signal sub-filter (10') is adapted based on a spectral representation of the filtered reference signals and a spectral representation based on detected acoustic error signals, and at least one adaptation of a control signal sub-filter (10') is carried out based on a time-shifted block of a filtered spectral representation of the reference signals.

19. The method according to one of claims 17 to 18, comprising - a back transformation of a spectral representation of the at least one control signal into the time domain and a selection of a portion of the back-transformed values ​​for a block, wherein the transformation of the reference signals into the frequency domain is carried out based on two consecutive blocks in each case, the control signal sub-filters (10') for filtering a reference signal each have a number of filter coefficients which corresponds to the number of signal values ​​of the reference signal in two blocks, in the corresponding time domain a number of the filter coefficients which corresponds to a block is set to zero values, the back transformation generates a vector with the length of two blocks and the selection of a portion of the back-transformed control signal values ​​selects a block of values ​​from the vector.

20. The method according to one of claims 16 to 19, comprising - an estimation (150) of an error signal at the defined location based on the detected acoustic error signals at positions in the vicinity of the defined location, wherein the defined location is the listening location of a user and wherein the estimation of the error signal for the listening location is carried out in the frequency domain and taking into account an estimate of the transmission paths between the positions for detecting the acoustic error signals and the listening location, and the spectral representation of the error signal thus determined at the listening location is used for the adaptation of at least one of the control signal sub-filters (10').

Images