EP1518441B1 - Device and method for suppressing a feedback - Google Patents

Abstract

The invention relates to a device for suppressing a feedback in an environment containing a microphone (10) and a loudspeaker (13), comprising a device (20), for embedding a test signal in a loudspeaker signal, a microphone signal, or a modified microphone signal, preferably using a psychoacoustic marking threshold by means of a pseudo-noise test signal, a device (50) for determining a property of a transmission path in the environment between the loudspeaker (13) and the microphone (10), using the embedded test signal and the microphone signal, a filter (40), for filtering the loudspeaker signal, to give a filtered loudspeaker signal, whereby the filter may be adjusted with regard to the filter characteristics thereof by the determining device (50) to match the same to the properties of the transmission path and a device (30) for subtracting the filtered loudspeaker signal from the microphone signal to give a modified microphone signal in which the feedback as a result of the loudspeaker is reduced. The feedback suppression concept achieves an effective feedback suppression without a loss in audio quality in particular without detracting from the artistic work of an artist.

Description

The present invention relates to audio playback systems, and more particularly to audio playback systems in live environments.

In typical rock concerts, there is a high degree of dynamics in that, for example, the singer moves a lot on stage. The same often applies to the guitarist. On the other hand, however, the speakers are statically arranged in such a display environment. Therefore, it is not enough that the singer and his microphone, as well as the guitarist and his microphone mounted on the guitar sometimes get a little closer to the speakers and sometimes a little further away from the speakers. While the case is unproblematic where a microphone is far from a speaker, the case where a microphone is placed very close to a speaker is very problematic. After a high gain is present in the signal path from the microphone to the loudspeaker, coupling the loudspeaker signal into the microphone causes the microphone / loudspeaker system to begin to oscillate. Such a vibration manifests itself as feedback at a certain frequency. It always occurs when the amplitude and phase conditions are met. The particular phase condition that is currently best fulfilled sets the frequency, which is typically relatively high, so that feedback is perceived as loud whistling. This whistling is not only unpleasant for the listener, but also for the artists.

In terms of signal theory, a strongly time-variable channel exists from one or more loudspeakers to one or more microphones.

Known feedback suppression techniques mix audible feedback sounds into the microphone and use filters to suppress budding feedback.

Alternative feedback cancellation techniques use a so-called pitch-shifting technique to shift the feedback to inaudible parts of the spectrum, thus avoiding stable feedback sounds.

While the first solution requires a short feedback to trigger suppression, the other approach in some cases causes a strange sound, e.g. makes singing and voicing difficult for artists.

Especially in multi-channel systems, the two aforementioned feedback suppression solutions are very problematic if not impracticable.

US Patent Publication US 2002/0064291 A1 discloses a feedback suppression concept for a hearing aid having a filter construction in the feedback path which models the feedback from the hearing aid loudspeaker to the microphone of the hearing aid. In particular, Fig. 9 shows an embodiment in which a test signal (sample 954) is injected into the signal to be amplified by the amplifier 118 through the upstream summer, which signal is then sent through the loudspeaker 120 as an "embedded signal" , is received via the feedback path 124 and simultaneously inserted through the filter setting blocks 956, 958, 936.

It is important to note, however, that the amplitude of the probe signal, as indicated in section [0066] on page 6 of the document, is increased when the input signal level is low, while the amplitude of this additional signal is reduced when the input signal level is high ([ 0066], lines 10 and 11).

US Pat. No. 6,347,148 B1 discloses a method and a device for feedback reduction in acoustic systems, again mentioning hearing aids as the field of application. The input signal is subjected to a spectral estimation in block 24 and a subsequent psychoacoustic evaluation to calculate the masking threshold. With this masking threshold, an output signal of a noise source in the filter 28 is then spectrally shaped. However, this signal is not used to set a feedback rejection filter. Instead, using this psychoacoustically spectrally stained test signal, feedback detection is performed by means of blocks 32, 34 and 36, as set forth in column 5, line 11. In response to a feedback detection, the transfer function of the forward path H ₂ (f) is then frequency-selectively adjusted, for example. This document performs no filter adjustment or subtraction. Instead, it is determined whether there is a feedback situation based on a level comparison. In response to such feedback detection, as has been stated, a transfer function of a forward path is then varied. This variation is frequency selective. Thus, a particular frequency band of the transmission path is typically attenuated so that there is no feedback. Of course, this leads to audible artifacts that will both irritate viewers and irritate the artist.

The object of the present invention is to provide an improved concept for suppressing feedback.

This object is achieved by a device according to claim 1, a method according to claim 14 or a computer program according to claim 15.

The present invention is based on the finding that effective feedback suppression can be achieved by processing a microphone signal which is a superposition of a wanted signal and a feedback signal originating from a loudspeaker or a plurality of loudspeakers prior to mixing the feedback component is subtracted from the microphone signal so that only the wanted signal remains after the subtraction.

Regardless of whether the feedback signal component is large in the case of an unfavorable channel, that is, if the microphone is very close to the loudspeaker, or small in the case of a favorable channel, that is, if the microphone is relatively far away from the loudspeaker, the feedback signal component will be preferred continuously removed from the microphone signal. For this purpose it is necessary to synthetically determine the feedback signal component at the microphone.

According to the invention, a marking operation is carried out for this purpose so that the signal emitted by the loudspeaker can be detected. This is achieved in that either in the microphone signal after subtraction or in the microphone signal before the subtraction or in the signal after mixing and amplification, ie in the z. B. still digitally present playback signal for a speaker, a test signal is embedded.

Further, according to the present invention, means for detecting a characteristic of a transmission channel from the speaker to the microphone or directly for feedback loop from a microphone back to itself using the received microphone signal which is a superposition of the feedback signal and the wanted signal, and using the known test signal that has been embedded.

A preferred way to determine the nature of the transmission channel in the environment between the loudspeaker and the microphone is to perform a cross-correlation between the microphone signal and the test signal. For example, the cross-correlation directly provides the impulse response of the channel between the considered one Speakers and the microphone considered. Alternative channel determination methods can also be used.

Using the detected characteristic of the transmission channel, a filter is set which filters the loudspeaker signal to obtain a filtered loudspeaker signal. In other words, the time-variant channel from the speaker to the microphone is to a certain extent "simulated" in order to synthetically calculate the feedback signal fed into the microphone, so that it is ready for the subtraction device.

The present invention performs optimal feedback cancellation when the channel changes only slowly. This is very often the case at concerts, given the movements made by human artists. Even if an artist makes a very fast movement, this fast movement does not take much time, so that a short, fast movement is followed by a slower movement or even a pause. The system according to the invention is capable of suppressing feedback not only newly at the beginning of the "settling" but also during the transient, in such a way that a possibly already started feedback is still suppressed, as it originated. can be subtracted out.

On the other hand, a fast movement often also causes the channel to change back to "good", so that the microphone moves away from the channel a little farther, which in turn means that feedback that may be occurring may be reduced without feedback suppression. The requirement for a temporally constant channel is therefore very low in the suppression concept of the present invention.

In the preferred embodiment of the present invention, the test signal is a pseudo noise sequence can be easily, quickly and inexpensively produced with little effort, for example, using feedback shift register and, if such a shift register is provided in several places, is readily reproducible. In particular, multiple shift register means to generate such a pseudorandom sequence may be initialized with the same seed or "seed". It is known that pseudo-noise sequences look like noise, but usually have a relatively large period. The buzzing appearance of a pseudo-noise sequence is reflected in the frequency domain in that the pseudo-noise signal has a white spectrum, such that all frequencies are equally strong. If the dynamics of the microphone signal are reasonably known, this white pseudo-noise signal can be mixed in directly if it is ensured that the level of the pseudo-noise signal mixed in is relatively small and leads to noises that are audible, or merely minor audible interference.

In order to improve the effectiveness of the feedback suppression, i. to improve the channel simulation, the test signal, whether or not it is a pseudo-noise signal, is derived using a microphone signal which has already been freed of its feedback component or by using one of the amplified microphone signal, that is, the loudspeaker signal psychoacoustic masking threshold.

Addition of the thus evaluated test signal to the microphone signal or to the loudspeaker signal causes the embedded test signal to be inaudible to the listener, so that the listener will not notice the constantly running feedback suppression procedure.

In other words, in this case, the feedback suppression has no negative consequences in terms of Audience quality perceived by the viewer. On the other hand, for an effective suppression, ie for the most accurate determination of the impulse response of the channel between the speaker and the microphone, ie for accurate simulation of the feedback component, a test signal with the highest possible energy in the speaker signal is desirable. The maximum energy is achieved without sacrificing audio quality when the test signal is a pseudo-noise signal, that is over the entire relevant frequency range, and is psychoacoustically weighted to be below the masking threshold of the loudspeaker signal. In signal components of the loudspeaker signal with a high masking effect, the test signal is therefore represented with a high energy, while in signal components of the loudspeaker signal with a low masking effect, for example in tonal audio components, the test signal is represented with relatively little or no energy, to the effect that no audio quality losses for the listener arises.

It should be noted that in the case where the microphone is not immediately in front of the loudspeaker, rather loud loudspeaker signal passages are problematic. Due to the fact that in such loud speaker passages the acoustic masking threshold is normally relatively high, such problematic loudspeaker signal components also contain a considerable amount of test signal energy, which immediately results in more precise and thus more efficient channel determination and hence feedback suppression. The concept preferred for the present invention of using pseudo-noise test signal in conjunction with a psychoacoustic weighting of the pseudo-noise test signal thus leads to the fact that precisely in the case where a well-functioning feedback suppression is used, as in If loud signals, even a good channel determination with high signal-to-noise ratio can be performed. The good feedback needed in such a case is also provided according to the invention.

The present invention is particularly suitable for multi-channel environments where multiple microphones and multiple speakers are present. The use of different test signals embedded in the individual microphone signals, which are preferably mutually orthogonal, and the use of a cross-correlation device for the determination of each relevant channel results in that the optimum feedback proportion for each microphone can be calculated. This results in a flexible and to the individual microphone signals exactly matched feedback suppression takes place, since each channel is simulated individually.

It can be seen that in the case where a plurality of microphones and a plurality of speakers are provided at different locations, the computing power for channel determination may preferably become considerable using cross-correlation. However, this is not a problem since a typical amplification system, such as a loudspeaker. In such a setting, some digital signal processors for calculating the channel characteristics and suppressing the feedback components will not be significantly significant in terms of the overall cost of the system.

On the other hand, the present invention provides efficient feedback cancellation without negative consequences for the listeners on the one hand, and especially for the artists on the other hand, with typically almost negligible costs relative to the overall system. In particular, value is placed on the fact that the artists are not disturbed in their artistic expression, such that they z. For example, in the case of pitch shifting, the signals perceived by the performer hear a different pitch when they were sung by the artist, for example. Although nuances in pitch shifting will suffice for this well-known feedback suppression, these are still annoyances for the artist, who are likely to limit his artistic expression. On the other hand, it is precisely the artist who ultimately determines which facility must be provided for him. A market acceptance of the inventive concept is therefore to be expected because the feedback suppression concept according to the invention does not bother the artist and even allowed him a maximum freedom of movement, so that he must fear without unwanted feedback tones, the entire stage space can use for artistic expression, regardless of whether it comes close to a feedback-endangered speaker component or not.

Depending on the embodiment, the test signal can be embedded directly in the loudspeaker signals, ie before the analog / digital conversion and acoustic reproduction. In this case, the adaptation to the psychoacoustic properties of the loudspeaker signal will be best, since the psychoacoustic model of the loudspeaker signal will be directly meaningful to what a viewer hears or not.

An embedding in the loudspeaker signal also has the advantage that in fact transmission functions from each loudspeaker to each microphone can be individually simulated and used for feedback suppression. This alternative according to the invention leads to a better sound quality for the listener, but requires higher computing power, for example if three microphones and three loudspeakers already exist, nine different transmission channels have already been determined with regard to properties, typically with FIR filters and have to be used for subtraction , wherein before the actual subtraction of the total feedback signal nor an addition of the three single simulated feedback signals supplied in the case described by three speakers must be performed.

Another alternative of the present invention is to embed the test signal in the modified microphone signal, that is to say after the subtraction, that is to say before the microphone signals are mixed and amplified in order to obtain an embedding signal. The embedding signal is simultaneously used to be filtered and to supply the filtered signal to the subtractor. The psychoacoustic model is here preferably calculated on the basis of the modified microphone signal in order to obtain the masking threshold for optimum embedding.

However, the psychoacoustic masking threshold information may also be derived from the individual loudspeaker signals and applied to the appropriate embedding means prior to blending / amplification so as to provide better control of the test signal.

As has been stated, the test signal should not be audible on the one hand and be present with the highest possible energy on the other hand. If a psychoacoustic model is derived from a signal that does not directly correspond to the loudspeaker signal, but only approximates, the energy of the embedded test signal is kept below the psychoacoustic masking threshold by a certain margin, which does not impair the quality of the audio but degrades it Signal / noise ratio in the transmission channel determination and thus could lead to a poorer feedback suppression.

On the other hand, in this case, not so many channels have to be calculated, so that this alternative has a shorter computation time can be used and therefore can be used inexpensively, especially in smaller reproduction systems or minimal reproduction equipment.

Again alternatively, the test signal may be inserted into the microphone signal prior to the feedback fraction subtraction. If the feedback proportion is calculated accurately, the embedded test signal will survive the feedback fraction subtraction relatively "unscathed", such that this case can be considered similar to the case where the test signal is already embedded in the modified microphone signal.

Fig. 1a

eine bevorzugte Ausführungsform der vorliegenden Erfindung in einer Mehrkanalumgebung mit Einbettung auf der Mikrophonseite;

Fig. 1b

eine alternative Ausführungsform des erfindungsgemäßen Rückkopplungsunterdrückungskonzepts mit Einbettung auf der Mikrophonseite;

Fig.2

eine alternative Ausführungsform der vorliegenden Erfindung mit Einbettung auf der Lautsprecherseite;

Fig.3

ein Prinzipdiagramm eines Obertragungskanals; und

Fig.4

eine schematische Zusammenfassung der Vorgehensweise zur Berechnung einer Impulsantwort des in Fig.3 gezeigten Übertragungskanals unter Verwendung einer Kreuzkorrelation.

Preferred embodiments of the present invention will be explained below in detail with reference to the accompanying drawings. Show it:

Fig. 1a

a preferred embodiment of the present invention in a multi-channel environment with embedding on the microphone side;

Fig. 1b

an alternative embodiment of the inventive feedback suppression concept with embedding on the microphone side;

Fig.2

an alternative embodiment of the present invention with embedding on the speaker side;

Figure 3

a schematic diagram of an Obertragungskanals; and

Figure 4

a schematic summary of the procedure for calculating an impulse response of the transmission channel shown in Figure 3 using a cross-correlation.

1 shows a preferred embodiment of the present invention in a multi-channel setting, in which a plurality of microphones 10, 11, 12 and a plurality of loudspeakers 13, 14, 15 are arranged. Between the microphones on the microphone side and the loudspeakers on the loudspeaker side, there is arranged a signal processing device 16, which is any sound system which, among other things, can also perform mixing or amplification of the sound signal fed by the microphones.

Signals from the three loudspeakers 13, 14, 15 are superimposed on each microphone and form a feedback signal f _i (t) for each microphone. The loudspeaker signals of the loudspeakers 13, 14, 15 are transmitted via a free space transmission channel 17 which can be defined such that from the three loudspeakers to the first microphone a first transmission channel h _{1 is} defined, that from the three loudspeakers to the second microphone 11 a second transmission channel h _{2 is} defined, that of the three speakers to the third microphone 12, a third transmission channel h _{3 is} defined.

In the exemplary embodiment shown in FIG. 1a, a test signal is respectively embedded in a modified microphone signal using an embedding device 20, 21, 22 in order to obtain a respective embedding signal for each microphone channel at the output of the device 21, 21 and 22. In particular, a first test signal p _{1 is} embedded in the modified microphone signal of the first microphone 10 in order to obtain a first embedding signal. In the modified microphone of the signal of the second microphone 11, a second test signal p _{2 is} embedded to obtain a second embedding signal. Finally, a third test signal p _{3 is} embedded in the modified microphone signal of the third microphone 12 to obtain a third embedding signal.

In order to come from a microphone signal at the output of the respective microphone 10, 11, 12 to a respective modified microphone signal, a subtractor 30, 31, 32 is further associated with each microphone. The subtracting means is arranged to subtract a simulated feedback component, which in the ideal case is equal to the feedback component f _i (t) received by a microphone, from the microphone signal. Thus, in the ideal case, a modified microphone signal is present at the output of the respective subtraction device 30, 31, 32, which corresponds to the original useful signal s ₁ (t), s ₂ (t) or s ₃ (t).

To simulate the feedback components, each microphone is assigned its own channel simulation filter 40, 41, 42, wherein the first simulation filter 40 is designed to have the same channel impulse response h ₁ (t) as shown in block 17, wherein in Fig. 1b not only the free space channel is assigned to the presentation in block 17, but also the transfer function by the block mixture / amplification 16. At this point it should also be pointed out that the simulated channel impulse response also already includes the necessary delay.

Similarly, the second channel simulation filter 41 is configured to have the same channel impulse response h ₂ (t) as outlined in block 17 (including mixing / amplification). Finally, the third simulation filter 42 is designed to have the same channel impulse response h ₃ (t) as indicated in block 17 (including mixing / amplification).

The channel impulse responses for setting the simulation filters 40, 41, 42 are determined in respective devices 50, 51, 52 for determining a characteristic of a transmission channel. For this purpose, the first means 50 for determining obtains the test signal which has been fed into the modified microphone signal of the microphone 10. Analogously, the second device 51 is provided for determining the Test signal p ₂ , which has been used in the means 21 for embedding. Finally, means 32 for determining the third microphone receives the same test signal p ₃ which has been fed into the modified microphone signal of the third microphone.

In a preferred embodiment of the present invention, the three test signals p ₁ , p ₂ , p _{3 are} each pseudo-noise sequences that are orthogonal to each other so that they are cross-correlated with the one in the devices 50, 51, 52 for detection respective test signal p ₁ , p ₂ , p _{3 can be distinguished} from the provided with the other test signals modified microphone signals and thus emitted speaker signals.

A cross-correlation eg of the microphone signal of the first microphone 10 with the pseudo-noise sequence p ₁ will cause the modified microphone signals provided with the pseudo-noise sequences to be out-correlated by the second and the third microphone, so that only the actual one of the first Microphone signal to be subtracted feedback component, which is problematic in terms of generating a feedback is subtracted. It should be noted that typically, if in the device 16 not significant microphone / speaker allocation changes are made in short time intervals, feedback signals from the other two microphones 11, 12 are not critical here, since such feedback signals in the signal processing path from the first microphone to the three loudspeakers 13, 14, 15 are uncritical of feedback generation.

In the embodiment of the present invention shown in Fig. 1a, the embedding signal of this microphone channel is further used and filtered for filter parameter calculation for each microphone channel. Specifically, the filter 40 for generating the filtered signal to be supplied to the device 30 receives the embedding signal on Output of the device 20 is supplied. Accordingly, the filter 41 is fed with the embedding signal from the device 21. In addition, the filter 42 is fed with the embedding signal from the device 22.

It should be noted that the embodiment shown in FIG. 1 a merely subtracts the signal that is problematic for feedback. So far problematic for a feedback via the first microphone is only the (former) signal from the first microphone, which is (later) coupled again. In this case, it does not matter from which speaker (s) the first microphone signal is reproduced. The channel calculated by correlation of the first microphone signal with the first test signal corresponds to a "loopback", ie a round trip from the microphone, via the mix / amplification, one or more loudspeakers and the free space channel back to the microphone (including the transmission characteristic of the actually used microphone) , It should also be noted that the determined impulse response h ₁ "automatically" also includes the occurred in the feedback loop delay, so that no further precautions must be taken for this. Furthermore, in this case, the situation is transparent in that the psychoacoustic masking threshold of the signal fed into the embedding device can be used for spectral coloring.

Alternatively, a loudspeaker signal could also be fed back and fed into the filter. Depending on the main figure of a microphone on a speaker, the assignment is such that the speaker signal 13 is filtered and returned to the first microphone 10, in principle arbitrary. If the dominant assignment of the first microphone is more to the loudspeaker 2, the loudspeaker signal of the loudspeaker 14 would be fed back via the simulation filter 40 to the first microphone. The assignment of the loudspeaker signals to the microphones is therefore merely exemplary in FIG. 1 a and may also vary from time to time depending on the mixture in the signal processing device 16.

The alternative embodiment of the present invention shown in Fig. 1b differs from the embodiment shown in Fig. 1a in that loudspeaker signals are fed back instead of embedding signals, and in that the signals of the various loudspeakers 13, 14, 15 in one Summing means 23 are summed, and that then the loudspeaker sum signal with the corresponding different simulation filters 40, 41, 42 is filtered to produce the three synthesized feedback components which are supplied to the respective subtractors 30, 31, 32, as shown in FIG. 1b is shown. In this embodiment, it is assumed that superimpose the loudspeaker signals of all speakers in the transmission channel 17 and lead to, for example, a resulting feedback signal f ₁ (t), which consists of signal components of the first, second and third loudspeaker modified by a corresponding definable transfer function. For the transmission of the sum signal of the three loudspeakers, which are superposed in the free space transmission channel, to the first microphone, a first transmission function h _{1 is} defined. For the transmission of the sum signal to the second microphone 11, a transfer function h _{2 is} defined and, finally, a resulting transfer function h _{3 is} defined for the transmission of the sum signal to the third microphone 12.

These transfer functions h ₁ , h ₂ , h ₃ are again determined in the devices 50, 51, 52 preferably by cross-correlation with the corresponding pseudo Neuseh sequence p ₁ , p ₂ or p ₃ assigned to a specific microphone. The execution of the subtraction devices 30, 31, 32, the embedding devices 20, 21, 22 and the simulation filter 40, 41, 42 is designed as in the embodiment described with reference to FIG. La.

In the following, the exemplary embodiment shown schematically in FIG. 2 will be discussed. In contrast to the exemplary embodiments shown in FIGS. 1a and 1b, the embedding of the test signal does not take place on the microphone side but on the loudspeaker side. Thus, not only three different channels but also nxm different channels can be defined, where n is a number of loudspeakers greater than or equal to 1, and where m is a number of microphones greater than or equal to 1. By correlating the output of the first microphone 10 with the first test signal p ₁ , the channel from the loudspeaker 1 to the first microphone M1, designated h ₁₁ , can be calculated. By correlating the output of the first microphone 10 using the second pseudo-noise sequence p ₂ , the channel from the second loudspeaker 14 to the first microphone 10, designated h ₁₂ , can be calculated. Analogously, by correlating the microphone signal of the first microphone 10 with the third pseudo-noise sequence p _3, the channel from the loudspeaker LS3 to the first microphone M1, designated h ₁₃ , can be simulated.

Analogously, the procedure for the output signals of the microphones 11 and 12, as indicated by the means 50, 51, 52 for determining. The devices 50, 51, 52 are thus able to calculate for the channel of each speaker to each microphone its own channel transfer function with which each individual loudspeaker signal can be convolved, which takes place in the simulation filters 40, 41, 42, then eg within the subtraction means 30, 31 or 30 or in an upstream block from the three channel output signals for each microphone to calculate the resulting feedback component by addition in order to arrive at a resulting feedback component. This is then from the fed into a respective microphone feedback signal f _i (t) is subtracted to arrive at a modified microphone signal for each microphone in which each channel has been selectively considered.

Depending on the embodiment, a means 50 for determining may be executed in a completely parallel manner, so as to calculate the channel impulse responses h ₁₁ , h ₁₂ and h ₁₃ simultaneously, as it were. However, the corresponding device could also be designed in series, in which case a buffer is preferred with respect to an optimal temporal synchronicity of the three channels h ₁₁ , h ₁₂ , h ₁₃ . However, at the expense of some error, such an intermediate storage could be dispensed with, such that the three related impulse responses from each of the speakers 13, 14, 15 to the first microphone 10 are not related to the same time period, but to successive periods, which however, it is harmless if the signals in an environment do not change too fast, relative to the time required for the correlation.

Likewise, the filtering means 40, 41, 42 may be serially or parallel, with a parallel embodiment providing the best results, such that for each possible channel of the possible channels in Fig. 2 a separate individual simulation filter is provided such that the filtering means 40, for example, actually comprises three individual simulation filters whose filter coefficients are set using the corresponding channel impulse response h ₁₁ , h ₁₂ , h ₁₃ . The addition of the three simulated feedback components from each loudspeaker to a resulting feedback component could therefore also occur in the filter device 40 immediately after the calculation of the corresponding impulse responses and the convolution of the loudspeaker signals with these impulse responses. In the embodiment shown in FIG. 2, as well as in the embodiments shown in FIGS. 1a and 1b, the three test signals p ₁ , p ₂ , p ₃ should be as orthogonal to each other as possible. This condition is easily and safely achieved by pseudo-noise sequences, and this property is also not lost by psychoacoustic filtering of the test signals prior to embedding.

In the embodiment shown in Fig. 2, it should be noted that a speaker signal is the signal actually heard by a listener. Therefore, in view of inaudible embedding of the test signals in the loudspeaker signals, the embedding will be best performed if the loudspeaker signals are used to calculate the psychoacoustic masking thresholds.

Thus, in the exemplary embodiment shown in FIG. 1b, a psychoacoustic model based on the respective loudspeaker signals 13, 14, 15 could also be calculated and used for embedding in the corresponding microphone signals in the devices 20, 21 and 22, respectively. Thus, in the psychoacoustic model, reinforcements that take place between a microphone and a loudspeaker in the device 16 can easily be taken into account. However, if a considerable addition / subtraction or other processing of the microphone signals is carried out in the device 16, for example in the case of a mixing process, so that a loudspeaker signal not only essentially reproduces the output signal of a single microphone, but reproduces the output signals of several microphones, an embedding of a Test signal using the psychoacoustic masking threshold inaccurate. This is because on the one hand not a single loudspeaker signal can be used to calculate the psychoacoustic masking threshold, and on the other hand, it is not immediately possible to use a microphone to calculate the psychoacoustic masking threshold. However, after the mixing in the mixing console 16 is deterministic, it is preferred in such a case to have a psychoacoustic masking threshold corresponding to that To calculate a loudspeaker signal in which, when the loudspeaker signal is the combination of a plurality of microphone signals, the test signals of several microphones are embedded to different degrees or equal to one another, the test signals, but overall the psychoacoustic masking threshold of a loudspeaker signal substantially follow, so that embedding with maximum energy is achieved while at the same time no or only negligible audio quality losses are effected.

The following summarizes how the impulse response h (t) of a channel is determined by cross-correlation. For this purpose, the channel is subjected to a time-discrete test signal p (t). On the output side, the channel outputs a received signal y (t) which, as is known, corresponds to the convolution of the input signal and to the channel impulse response. For the following explanation of a procedure for determining the cross-correlation with reference to FIG. 4, a matrix notation is used. By way of example, a channel impulse response with only two values h ₀ and h _{1 is} assumed without restriction of generality. The channel impulse response h ₀ , h ₁ can be written as a channel impulse response matrix H (t) having the band structure shown in Fig. 4, with the remaining elements of the matrix filled with zeros. In addition, the excitation signal p (t) is written as a vector, it being assumed here that the excitation signal has only three samples p ₀ , p ₁ , p ₂ without limiting the generality.

It can be shown that the convolution illustrated in FIG. 3 corresponds to the matrix-vector multiplication illustrated in FIG. 4, so that a vector y results for the output signal. The cross-correlation can be written as the expected value E {...} of the multiplication of the output signal y (t) by the conjugate-complex-transposed excitation signal p ^{* T.} The expected value is calculated as a limit value for N against infinity over that in FIG. 5 shown summation of individual products for different excitation signals p _i . The multiplication and subsequent summation yields the cross-correlation matrix shown at the top left of Fig. 4, which is weighted by the rms value of the excitation signal p, represented by σ _p 2. For example, to directly obtain the channel impulse response h (t), the first row of the channel impulse response matrix is taken, whereupon the individual components are divided by σ _p ² to directly obtain the individual components of the channel impulse response h ₀ , h ₁ .

If a spectrally colored excitation signal is used instead of a white excitation signal p (t), then the spectral coloring can be represented by a digital filtering, the filter being described by a filter coefficient matrix Q. In the equation shown in the last line in FIG. 4, the correlation matrix H also results on the output side, but now still weighted with the expected value over Q x Q ^H. By dividing the individual impulse response coefficients h ₀ , h ₁ by the expected value over Q x Q ^H , ie by taking into account the color filter, for example in the device 50 for determining a property of the transmission channel of Fig. 1a, 1b or 2 can directly the channel impulse response with respect to their individual Components are determined.

It should be noted that the cross-correlation concept for calculating the impulse response is an iterative concept, as can be seen from the expected expectation summation approach illustrated in FIG. The first multiplication of the reaction signal with the conjugate-complex-transposed excitation signal already provides a first still very rough estimate for the channel impulse response, which becomes better and better with each further multiplication and summation. If the entire matrix H (t) is calculated by the iterative summation approach, it turns out that the elements set to zero in FIG Band matrix H (t) gradually go to zero, while in the middle, so the band of the matrix, the coefficients of the channel impulse response h (t) remain and take certain values. Once again, it should be noted that it is not necessary to calculate the entire matrix. It is enough, only z. For example, to calculate a row of the matrix H (t) to obtain the total channel impulse response.

It should be noted at this point that the concept according to the invention is not restricted to the procedure for calculating the cross-correlation described with reference to FIG. 4. All other methods for calculating the cross-correlation between a measurement signal and a response signal can also be used. Other methods of determining an impulse response in lieu of cross-correlation may also be used.

It should be noted at this point that the pseudo-noise sequences used should be dimensioned with respect to their length depending on the expected impulse response of the considered channel. Thus, impulse responses with a length of a few seconds are conceivable for larger acoustic environments. This fact must be accounted for by selecting an appropriate length of the pseudo-noise sequences for correlation.

Depending on the circumstances, the method according to the invention can be implemented in hardware or in software. The implementation may be on a digital storage medium, in particular a floppy disk or CD with electronically readable control signals, which may interact with a programmable computer system such that the method is executed. In general, the invention thus also consists in a computer program product with program code stored on a machine-readable carrier for carrying out the method according to the invention when the computer program product runs on a computer. In other words, the invention can thus as a Computer program with a program code for performing the method can be realized when the computer program runs on a computer.

It should be pointed out again that the inventive concept can be used for any number of microphones and any number of speakers. This means, of course, also that the inventive concept can already be used advantageously for a speaker and a microphone. This is immediately apparent from FIGS. 1a, 1b and 2, when the second and the third microphone 11, 12 as well as the second and the third loudspeaker 14, 15 are ignored and also the blocks addressed by these signals are thought away.

At this point it should also be noted that the embedding of the test signal does not necessarily have to be done in the modified microphone signal or the loudspeaker signal, but that an embedding of the test signal in the microphone signal in front of the corresponding subtraction device can take place, although the embedding of the test signal after the Subtraction device is preferred. This is due to the fact that in the case of a not so favorable channel impulse response calculation and thus in the case of a not particularly precisely synthesized feedback component, the embedded test signal by the subtraction of a not exactly matching feedback component u. U. could be damaged, which should lead to a further complication of the channel simulation by the devices 50, 51, 52.

Thus, in preferred embodiments of the present invention in a multi-channel setting, a non-audible wideband signal is embedded in each microphone signal. This signal is adaptively adapted in terms of its spectral envelope to the recorded sound, whereby a psychoacoustic model can be used, which in principle may be arbitrary and can be calculated based on time domain data or else based on frequency domain data. As a broadband signal, a pseudo-noise sequence is preferred, since in such a sequence, orthogonality between several sequences can be readily achieved.

For each microphone, the pre-embedded signal is compared to the pseudo-noise signal and used to calculate the acoustic characteristics of all loudspeakers to the corresponding microphone. As a comparison operation, a cross-correlation is preferred, which, when the iterative algorithm shown in Fig. 4 is used, can be calculated computationally-unpredictably with an arbitrarily scalable accuracy. In particular, the scalability provides the ability to provide a fast, but coarser, calculation for specific situations, such as a rock group that has a lot of motion on the stage, while for other application scenarios, such as those in the art. a rock group in which the performers of one are static, e.g. Scaling to a greater number of iteration values could be performed because the individual channels are less time-variant.

Using an appropriate channel, an inverse filter is applied to suppress unwanted components. The inverse filter is realized according to the present invention by the simulation filters and the corresponding associated subtraction devices. The use of microphone signals allows spectrally shaped PNS signals to be stored so as to avoid interfering with original sound signals, and to compute a spectral shaping psychoacoustic model only once and not in the corresponding means for determining it again must be calculated.

Alternatively, as illustrated with reference to Figure 2, a unique PNS signal is embedded in the signal from each speaker. This embedding on speaker side approach allows the measurement of a path from each speaker to each microphone. A suppression filter is used separately for each loudspeaker, which achieves better sound quality, but at the expense of higher computational effort, but is unlikely to be particularly significant in terms of the overall cost of medium to large size sound systems.

Claims (15)

1. An apparatus for suppressing feedback in an environment where a microphone (10) and a loudspeaker (13) are located, comprising:

   a means (20) for embedding a test signal into a loudspeaker signal, a microphone signal or a modified microphone signal to obtain an embedding signal, wherein the microphone signal is output from the microphone and wherein the loudspeaker signal is input in the loudspeaker; wherein the means (20) for embedding is formed to spectrally color the test signal by using a psychoacoustic masking threshold, so that the embedded signal is essentially inaudible;

   a means (50) for determining a characteristic of a transmission channel in the environment between the loudspeaker (13) and the microphone (10) by using the test signal and the microphone signal;

   a filter (40) for filtering the loudspeaker signal or the embedding signal to obtain a filtered signal, wherein the filter is adaptable to be adapted to the characteristic of transmission channel with regard to its filter characteristic in response to the means (50) for determining; and

   a means (30) for subtracting the filtered signal from the microphone signal to obtain the modified microphone signal, in which a feedback is reduced.
2. The apparatus of claim 1, wherein the test signal is a pseudo-noise signal.
3. The apparatus of one of the previous claims, wherein the means for determining is formed to perform a cross correlation by using the test signal and the microphone signal to calculate a channel impulse response as characteristic of the transmission channel.
4. The apparatus of claim 3, wherein the means (30) for subtracting is adapted to perform a sample wise subtraction in the time domain.
5. The apparatus of claim 3 or 4, wherein the filter (40) is a digital filter whose coefficients can be adjusted such that an impulse response of the filter corresponds to the channel impulse response within a predetermined deviation threshold.
6. The apparatus of one of the previous claims,
   wherein several microphone signals can be supplied from several microphones (10, 11, 12),
   wherein an individual means for embedding (20, 21, 22) a test signal is provided for every microphone signal,
   wherein every means for embedding (20, 21, 22) is fed with a different test signal to generate an individual embedding signal from every microphone signal, wherein the test signals are orthogonal to one another within a deviation threshold;
   wherein a means (50, 51, 52) for determining is provided for every microphone signal which is each formed to determine a channel impulse response of a channel from a microphone via one or several loudspeakers back to the microphone, and
   wherein an individual filter (40, 41, 42) is provided for every microphone signal to filter the embedding signal to obtain a filtered signal and to feed the filtered signal to a means for subtracting (30, 31, 32) for this microphone signal.
7. The apparatus of one of claims 1 to 5,
   wherein a plurality of loudspeakers (13, 14, 15) and a plurality of microphones (10, 11, 12) are provided,
   wherein an individual means (20, 21, 22) for embedding the test signal into the modified microphone signal is provided for every microphone signal,
   wherein every means (20, 21, 22) for embedding is fed with a different test signal, wherein the test signals are orthogonal to one another,
   wherein an individual means (50, 51, 52) for embedding is provided for every microphone signal, which is each formed to obtain a channel impulse response based on a sum (23) of signals of the loudspeaker to the corresponding microphones (10, 11, 12) and by using a test signal associated for this microphone, and
   wherein it is provided for every microphone signal to filter the sum of loudspeaker signals with the filter, which has an impulse response, which has been determined by using the test signal associated to an examined microphone signal, and to supply it to means (30) for subtracting for this microphone signal.
8. The apparatus of one of claims 1 to 5,
   wherein a plurality of loudspeakers (13, 14, 15) and a plurality of microphones (10, 11, 12) are present,
   wherein an individual means (20, 21, 22) for embedding a test signal into a respective loudspeaker signal is provided for every microphone signal,
   wherein every means (20, 21, 22) for embedding is fed with a different test signal, wherein the test signals are orthogonal to one another within a deviation threshold,
   wherein a means (50, 51, 52) for determining is provided for every microphone signal, which is each formed to calculate channel impulse responses for channels from every loudspeaker to the microphone, wherein the test signal embedded into the loudspeaker signal for the loudspeaker is used for a channel from a loudspeaker to a microphone, and
   wherein a plurality of filters is provided for every microphone signal, which is equal to a number of loudspeakers to filter every loudspeaker signal with an corresponding filter for a microphone signal, and to sum filtered loudspeaker signals from every loudspeaker to obtain a resulting synthesized feedback signal and to feed the resulting synthesized feedback signal to a means (30) for subtracting for this microphone signal.
9. The apparatus of one of the previous claims, further comprising:

   a means (16) for converting one or several modified microphone signals into one or several signals from which the loudspeaker signals are derived.
10. The apparatus of claim 9, wherein the means (16) for converting is formed to perform mixing and/or amplification of modified microphone signals.
11. The apparatus of claim 1, wherein the means for embedding a test signal is formed to embed the test signal into the loudspeaker signal, and
    wherein the means (20) for embedding is further formed to perform embedding by using a psychoacoustic masking threshold of the loudspeaker signal.
12. The apparatus of claim 1, wherein the means (20) for embedding is formed to embed the test signal into the modified microphone signal, and
    wherein the means (20) for embedding is further formed to evaluate the test signal prior to embedding with a psychoacoustic masking threshold of the microphone signal.
13. The apparatus of claim 1, wherein a plurality of microphones and a plurality of loudspeakers are present, wherein further a mixing means (16) for mixing two or several modified microphone signals is present to generate one or several loudspeaker signals, and
    wherein the means (20) for embedding is formed to perform embedding of several test signals into several microphone signals such that a resulting energy of the embedded test signals results under consideration of mixing, so that the resulting energy of the embedded test signals is in a signal for a loudspeaker below a psychoacoustic masking threshold of a loudspeaker signal for this loudspeaker.
14. A method for suppressing feedback in an environment where a microphone (10) and a loudspeaker (13) are located, comprising:

    embedding (20) a test signal into a loudspeaker signal, a microphone signal or modified microphone signal to obtain an embedding signal, wherein the microphone signal is output from the microphone and wherein the loudspeaker signal is input into the loudspeaker, wherein the means (20) for embedding is formed to spectrally color the test signal by using a psychoacoustic masking threshold, so that the embedded signal is essentially inaudible;

    determining (50) a characteristic of a transmission channel in the environment between the loudspeaker (13) and the microphone (10) by using the test signal and the microphone signal;

    filtering (40) the loudspeaker signal or the embedding signal to obtain a filtered signal, wherein the filter is adaptable to be adapted with regard to its filter characteristic through the characteristic of the transmission channel; and

    subtracting (30) the filtered signal from the microphone signal to obtain the modified microphone signal wherein a feedback is reduced.
15. A computer program with a program code which effects the method of claim 14 when it is run on a computer.

Images