EP1415503A2 - Systeme de traitement sonore comprenant un generateur d'ondes a reponses de directivite et de gradient arbitraires - Google Patents

Systeme de traitement sonore comprenant un generateur d'ondes a reponses de directivite et de gradient arbitraires

Info

Publication number
EP1415503A2
EP1415503A2 EP02767372A EP02767372A EP1415503A2 EP 1415503 A2 EP1415503 A2 EP 1415503A2 EP 02767372 A EP02767372 A EP 02767372A EP 02767372 A EP02767372 A EP 02767372A EP 1415503 A2 EP1415503 A2 EP 1415503A2
Authority
EP
European Patent Office
Prior art keywords
wave
gain
signal
audio processor
output
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
EP02767372A
Other languages
German (de)
English (en)
Inventor
Erik W. Rasmussen
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Rasmussen Digital APS
Original Assignee
Rasmussen Digital APS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Rasmussen Digital APS filed Critical Rasmussen Digital APS
Publication of EP1415503A2 publication Critical patent/EP1415503A2/fr
Withdrawn legal-status Critical Current

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
    • H04R3/00—Circuits for transducers, loudspeakers or microphones
    • H04R3/005—Circuits for transducers, loudspeakers or microphones for combining the signals of two or more microphones
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
    • H04R29/00—Monitoring arrangements; Testing arrangements
    • H04R29/004—Monitoring arrangements; Testing arrangements for microphones
    • H04R29/005—Microphone arrays
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04R—LOUDSPEAKERS, MICROPHONES, GRAMOPHONE PICK-UPS OR LIKE ACOUSTIC ELECTROMECHANICAL TRANSDUCERS; DEAF-AID SETS; PUBLIC ADDRESS SYSTEMS
    • H04R25/00—Deaf-aid sets, i.e. electro-acoustic or electro-mechanical hearing aids; Electric tinnitus maskers providing an auditory perception
    • H04R25/40—Arrangements for obtaining a desired directivity characteristic
    • H04R25/407—Circuits for combining signals of a plurality of transducers

Definitions

  • the present invention relates generally to audio signal processing. More specifically, the present invention relates to an audio processing system that exhibits an arbitrary directivity and gradient response.
  • FIG. 1 a block diagram of a sound processing system 10 is shown.
  • the system 10 includes at least one microphone 12 that picks up sounds from a sound field in which it is located and converts these sounds to electrical signals.
  • a plurality of microphones are depicted and the microphones are numbered from one to NL
  • the electrical signals from the microphones 12 are preferably input to an audio processor 14.
  • the sounds are to be reproduced by one or more output devices 16 such as loudspeakers, earphones, and the like.
  • the sound can optionally pass through transmission channels or additional processing before arriving at the output device 16. It may even be recorded and played back before arriving at the output device 16.
  • the sound field into which the system 10 is placed contains not only the sounds to be picked up, referred to as a utility signal, but also unwanted sounds, referred to as noise or noise signals.
  • noise or noise signals unwanted sounds
  • One is known as static beamforming where the signals from two or more microphones are passed through filters and combined to form a single signal.
  • the resulting signal will show a sensitivity to sounds that depends upon the direction of the sound incidence as compared to the direction of the microphone assembly.
  • the directivity response will take the form of one or more beams. Due to the fact that the lobes of the directivity response have different magnitudes, the beamformer will show a signal to noise improvement when the beam is oriented so that the utility signal falls within the main lobe and the main part of the noise falls outside the main lobe.
  • Static beamformers have the disadvantage that, in order to provide substantial noise reduction under general noise conditions, a large number of microphones are required.
  • adaptive beamforming is achieved when the filters of a beamformer are variable and controlled by an adaptation process. Normally such an adaptation process works to minimize the output signal power.
  • An adaptive beamformer can track noise sources and dynamically adjust the directivity response such that the sensitivity at the direction of the noise incidence is minimized while keeping the sensitivity at the utility direction high.
  • Currently known adaptive beamformers show the disadvantage that they are only capable of tracking a limited number of noise sources, mostly only a single. Furthermore adaptive beamformers work with a fairly large time constant in the adaptation process. Therefore they are only able to track quasi-static noise sources.
  • the present invention uses a different approach to the problem. It uses the general equations for sound fields to analyze the microphone signals and find required properties of one or more components ⁇ or waves contained in the input signals.
  • the desired properties can for example be the direction of sound incidence or the pressure gradient of the impinging waves.
  • the incoming waves are amplified with a gain function based on these properties, that is, the directivity or the gradient.
  • Based on the amplified waves an output signal is generated either by synthesizing the amplified waves or by applying filtering to an input signal combination.
  • the present invention can operate in a number of applications including hearing aids, directional microphones, microphone arrays, silicon microphone assemblies, headsets, hearing protectors, cordless phones, mobile phones, camcorders, personal computers, laptops, palmtops, and personal digital assistants, among others.
  • the present invention is especially suited to work with head worn microphones that pick up the speech signal of the wearer.
  • the present invention offers a substantially improved noise reduction when compared to conventional solutions with comparable sound quality.
  • a sound processing system including at least one microphone, an audio processor, and at least one output device.
  • the audio processor includes an analog beamformer, a microphone equalizer, and an apparent incidence processor.
  • Two different embodiments of the apparent incidence processor are disclosed, that is, a wave generation method and a forward filtering method. Both embodiments use the same principles to estimate the properties of the individual waves of the sound field.
  • the present invention it is possible to implement arbitrary directivity or gradient responses using a small number of microphones only, that is, two or three microphones.
  • the present invention offers improved noise reduction also for environments with many independent noise sources. Furthermore, the present invention works for signals and noises with arbitrary statistics.
  • FIG. 1 is a block diagram of a sound processing system
  • FIG. 2 is a block diagram according to a preferred embodiment of the present invention of the audio processor of FIG. 1;
  • FIG. 3 is a block diagram of the analog beamformer of FIG. 2;
  • FIG. 4 is a block diagram of the microphone equalizer of FIG. 2;
  • FIG. 5 is a block diagram of the microphone equalization updater of
  • FIG. 4
  • FIG. 6 is a block diagram according to a preferred embodiment of the present invention of the apparent incidence processor of FIG. 2;
  • FIG. 7 is a block diagram of the analysis beamformer of FIG. 6;
  • FIG. 8 is a block diagram of the wave parameter estimator of FIG. 6;
  • FIG. 9 is a block diagram of the equation solver of FIG. 8;
  • FIG. 10 is a block diagram of an embodiment of the core solver of FIG. 9 using a table look up implementation with optional approximation;
  • FIG. 11 is a block diagram of the output generator of FIG. 6;
  • FIG. 12 is a block diagram of the statistical evaluator of FIG. 11 ;
  • FIG. 13 is a block diagram of the wave generation gain controller of FIG. 11;
  • FIG. 14 is a pair of polar plots of a set of gain versus direction functions;
  • FIG. 15 is a block diagram of the gain mapper of FIG. 11;
  • FIG. 16 is a block diagram of the signal generator of FIG. 11;
  • FIG. 17 is a block diagram according to another preferred embodiment of the present invention of the apparent incidence processor of FIG. 2;
  • FIG. 18 is a block diagram of the forward filter of FIG. 17;
  • FIG. 19 is a block diagram of the forward beamformer of FIG. 18;
  • FIG. 20 is a block diagram of the adaptor of FIG. 19;
  • FIG. 21 is a block diagram of the forward filter gain controller of FIG. 18;
  • FIG. 22 is a block diagram of the forward filter gain function applier of FIG. 21;
  • FIG. 23 is a block diagram of a multiple output embodiment of the wave generation method
  • FIG. 24 is a block diagram of a multiple output embodiment of the forward filtering method
  • FIG. 25 is a block diagram of a forward filter/output generator
  • FIG. 26 is a block diagram of the wave generator/forward filter gain controller of FIG. 25;
  • FIG. 27 is a block diagram of a single combined mathematical transform processor
  • FIG. 28 is a block diagram of a near field embodiment of the audio processor of FIG. 1;
  • FIG. 29 is a block diagram of the microphone equalizer of FIG. 28;
  • FIG. 30 is a block diagram of the microphone equalization updater of FIG. 29;
  • FIG. 31 is a block diagram of the beamformer of FIG. 28;
  • FIG. 32 is a block diagram of the near field gain controller of FIG. 28;
  • FIG. 33 is a block diagram of the statistical evaluator of FIG. 32;
  • FIG. 34 is a block diagram of an embodiment of the near field gain function applier of FIG. 32;
  • FIG. 35 is a block diagram of an embodiment of the near field gain function applier of FIG 32 using a table look up implementation with subsequent approximation/interpolation;
  • FIGS. 36a and 36b are a pair of graphs of gain function of different widths.
  • the components, process steps, and/or data structures may be implemented using various types of operating systems, computing platforms, computer programs, and/or general purpose machines.
  • devices of a less general purpose nature such as hardwired devices, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), or the like, may also be used without departing from the scope and spirit of the inventive concepts disclosed herein.
  • FIG. 2 a block diagram according to a preferred embodiment of the present invention of the audio processor 14 of FIG. 1 is shown.
  • the audio processor 14 includes an analog beamformer 18, a microphone equalizer 20, and an apparent incidence processor 22.
  • an apparent incidence processor 22 two. different embodiments of the apparent incidence processor 22 will be disclosed, that is, a wave generation method and a forward filtering method. Both embodiments use the same principles to estimate the properties of the individual waves of the sound field.
  • the method of apparent incidence processing involves complex signal operations. It is therefore possible that the processing will be performed with digital techniques.
  • the microphone signals will be converted to digital signals with at least one analog to digital (AID) converter 24 and the output signal will be converted back to an analog signal, if needed, with a digital to analog (D/A) converter 26.
  • AID analog to digital
  • D/A digital to analog
  • the analog beamformer 18 provides analog preprocessing according to conventional techniques of the microphone signals that enables the reduction of the resolution of all but one of the AID converters 24. This can save size and reduce the power consumption. For hearing aids, for example, these properties are highly desirable. Depending on the circumstances however, the analog beamformer 18 may be deleted as unnecessary or too costly. [0017]
  • the method of apparent incidence processing requires, as generally do conventional beamformers, that the microphones 12 of FIG. 1 have sensitivities that are matched.
  • the microphone equalizer 20 equalizes the signals from the microphones 12 according to conventional techniques. With this equalization, the functioning of the processing downstream will still be possible even if the microphones have different sensitivities and even when the microphone sensitivities change over time. Again, depending on the circumstances however, the microphone equalizer 20 may be deleted as unnecessary or too costly.
  • FIG. 3 a block diagram of the analog beamformer 18 of
  • the analog beamformer 18 includes at least one filter 28 and a summing device 30. Each of the output beams of the analog beamformer 18 is derived as the sum of the filtered microphone signals. The variable i indexes the analog beam outputs. Each of the filters 28 will generally be different from microphone to microphone and from beam to beam.
  • the beam amic(l) is formed as the sum of all microphone inputs and the other beams are formed as the difference of a specific microphone signal and a reference microphone signal (Microphone(l)).
  • the generalized beam amic(i) will relate to the microphone signals as follows:
  • each of the filters, Filter(j,i) approximates different time delays of the microphone signals, that is, either inverting or non-inverting.
  • each of the beams amic(i) implements the same directivity using different microphone combinations.
  • the microphones are placed equidistant along a common axis.
  • the analog beamformer is defined according to (3) below. The numbering y of the microphones follows their placement along the common axis with number one being closest to the sound source. NB-1 is commonly referred to as the order of the directivity.
  • Filter(j, i) Filter(j + 1, i + 1) for all j ⁇ Nl - 1, i ⁇ N2 - 1
  • N2 N1-NB
  • FIG. 4 a block diagram of the microphone equalizer 20 of FIG. 2 is shown. While the microphone equalizer 20 could be implemented in the time domain with a FIR or an HR filter, the preferred scheme of FIG. 4 works in the frequency domain.
  • the microphone signals (mic(l) to mic( ⁇ 2)) are first converted to the frequency domain in a plurality of forward transformers 32.
  • the equalization is then accomplished by multiplying, in the frequency domain, the microphone signals with at least one equalization function (MicEq(i)) generated by a microphone equalization updater 34.
  • the equalized signals are finally converted back to the time domain in a plurality of reverse transformers 36.
  • the reference signal is referred to with index one, mic(l).
  • This reference signal is by definition equalized and is thus passed through the processing of the microphone equalizer 20 unaltered.
  • the equalization functions generated by the microphone equalization updater 34 should follow the definition of (4) below. (4) cEg( ⁇ —
  • S(l) is the sensitivity defined as the digital value at the input of the microphone equalizer 20 of FIG. 2 divided by the sound pressure of the reference microphone signal and S(i) is the corresponding sensitivity of the other microphones, respectively. All of the terms of (4) are implicit functions of frequency.
  • FIG. 5 a block diagram of the microphone equalization updater 34 of FIG. 3 is shown.
  • Some of the processing is done in a polar complex, that is, magnitude/phase, format so a rectangular to polar converter 38 and a polar to rectangular converter 40 are provided.
  • the rest of the processing except as noted, uses a rectangular, that is, real/imaginary, format for complex numbers.
  • a phase accumulator 42 and a magnitude accumulator 44 hold the equalization response of the specific microphone (i).
  • the response is updated at regular intervals by accumulating small updates to the phase and magnitude accumulators 42, 44, respectively.
  • the current equalized spectrum, CMIC(l) of the reference channel is divided by the equalized spectrum, CMICft), of the respective channel.
  • the quotient is converted to polar format as phase and magnitude in the rectangular to polar converter 38.
  • the phase of the quotient is analyzed in a zero phase condition detector 46. If the phase indicates that the sound incidence is from a direction and a distance for which the current microphone signal should have the same magnitude sensitivity as the reference microphone signal, then the zero phase condition detector 46 will output a logic one as a ZeroPhase output signal. If the phase condition does not hold, then the analyzer will output a logic zero. When the quotient .magnitude minus one is gated with this ZeroPhase switch signal and scaled with a MagnitudeCoef coefficient, a magnitude update value is obtained that is suitable to use to update the magnitude accumulator 44 for the equalizing function.
  • phase update both the phase and the magnitude of the quotient are analyzed. Under normal conditions, it will generally not be possible to derive any ' information regarding a misfit of the phase of the microphone equalization response. But once in a while, triggered by a specific input from a specific direction, the microphone signals will relate to each other in a way that can only be possible if the equalization response is incorrect.
  • a compute excess phase monitor 48 the signals are monitored and if such an "impossible condition" is found to exist the amount by which the phase is un-natural will be output as an excess phase signal. If the phase conditions are natural, the compute excess phase monitor 48 will output zero. As the excess phase conditions depend upon signal frequency, the compute excess phase monitor 48 estimates the frequency for its use.
  • the phase update signal is scaled with the P ⁇ seC ⁇ e/coefficient.
  • the signal inband detector 50 outputs a logic one if the power in the current frequency band is contributed mainly by input contents of frequencies falling within the band. Conversely, the signal inband detector 50 outputs a logic zero if the contents are due mainly to input at frequencies outside of the band. It is widely known that most time-to-frequency transforms "spill" energy from the source band to neighboring bands due to windowing effects or similar mechanisms. However, only signals within the frequency band should be allowed to influence the equalizing value for a band. This differentiation is possible through the use of the Inband signal.
  • the equalization response will be updated dynamically.
  • the phase and the magnitude of the equalization response will be regulated independently. Updating follow through statistical processes that rely on the noise-like nature of the signals that are most likely to be encountered as inputs to the audio processor, such as speech, machine noise, wind noise, and the like.
  • the update signals will contain large noise components, therefore they are scaled with small coefficients, that is, PhaseCoef and MagnitudeCoef, respectively, such that the adaptation times are slow. The use of the coefficients prevents noise from entering the forward signals through the equalization processes.
  • phase and magnitude accumulators 42 and 44 are divided into static and dynamic parts, where the updates only influence the dynamic parts.
  • the effective equalization response is the product of the static and dynamic parts.
  • the static part of the equalization response is measured with standard measuring techniques once or regularly at the time of production test or at some other convenient times and saved.
  • a forgetting factor is included with the dynamic part of the accumulator.
  • the forgetting factor causes the dynamic response to converge towards zero when no updates are received.
  • means are provided that can save the accumulated equalization response when the audio processor is powered down and read the saved response again when the processor is powered up the next time.
  • the signals mic(i) are all omni directional and the zero phase condition detector 46 is implemented so as to compare the magnitude of the phase with a constant value. If the phase magnitude is smaller than the constant, then a logic one ZeroPhase signal is generated.
  • the signals mic(i) are all omni directional and the compute excess phase monitor 48 generates the phase update signal according to (5) below.
  • f(i) is the frequency as estimated by the compute excess phase monitor 48.
  • a,f and ExcessPhase are all vectors covering the frequency range of the frequency transformation used.
  • the * operation in (5) denotes an element-by-element multiplication and not a vector multiply.
  • d(i) is the physical spacing between Microphone (i) and the reference microphone, Microphone(l).
  • c is the speed of sound
  • ⁇ is a small positive constant.
  • the center for the equalization is estimated as the center frequency of the band i.
  • the frequency, f(i), for the equalization is estimated as in (6) below.
  • k is the frequency band index
  • fc(k) is the center frequency of band k
  • BW(k) is the bandwidth of band k
  • b is a positive constant.
  • f(k,i) fc(k) + ⁇ b ⁇ CMIC(k + 1, i) ⁇ + ⁇ CMIC(k - 1, i) ⁇
  • the signal inband detector 50 for each frequency band evaluates the absolute value of its input signal in the current band and the two nearest neighboring bands. If the current band carries the highest absolute value, then the Inband signal for the current band is generated as a logic one and otherwise it is generated as a logic zero.
  • FIG. 6 a block diagram according to a preferred embodiment of the present invention of the apparent incidence processor 22 of FIG. 2 is shown.
  • the wave generation method is shown. The processing runs in three stages. First, analysis beamformiiig 52 is performed on the equalized microphone signals. Second, the parameters of the incoming sound waves are estimated in a wave parameter estimator 54. Finally, an output generator 56 produces a signal that contains the incoming waves modified in such a way that unwanted waves are attenuated by comparison to the utility waves.
  • FIG. 7 a block diagram of the analysis beamformer 52 of
  • FIG. 6 is shown.
  • the analysis beamformer 52 is similar to the analog beamformer 18 of FIG. 3 described above but- it works in the digital domain.
  • the analysis beamformer 52 generates a plurality of abeam signals and for each signal it includes a plurality of filters 58 and a summing device 60.
  • the analysis beamformer 52 serves, among other functions, to remove unwanted noise from the signal thus enhancing quality of the wave parameter estimation.
  • the analysis beamforming of (7) is combined with the analog beamforming of (1) above. In such a combination, it can then be shown that the output of the analysis beamformer 52 will be estimates of the microphone outputs as described in (8) below.
  • abeam(i) c AD Microphone(i)
  • C ⁇ D is the A/D converter conversion gain.
  • the sequence of the processing is preferably changed so that the analysis beamformer 52 precedes the microphone equalizer 20 of FIG. 2. In this way the microphone sensitivities are directly equalized.
  • each of the filters 58, Filter(j,i) approximates different time delays of the microphone signals, that is, either inverting or non-inverting.
  • the latest embodiment described is changed so that only two of the Filters(j,i) for each beam i are present.
  • each of the beams, abeam(i) implements the same directivity using different microphone combinations.
  • the microphones are placed equidistant along a common axis.
  • the analysis beamformer 52 is defined according to (9) below.
  • the numbering of the microphones follows their placement along the common axis with number one being closest to the sound source.
  • NB-1 is commonly referred to as the order of the directivity.
  • Filter (j, i) Filter (j + l,i + 1) for all j ⁇ N2 - 1, i ⁇ Nl - 1
  • N1 N2-NB [0051]
  • the analysis beamforming is performed in frequency bands.
  • FIG. 8 a block diagram of the wave parameter estimator
  • the wave parameter estimator 54 includes a plurality of analysis filters 62, a plurality of forward transformers 64, a normalizer 66, and an equation solver 68.
  • the analysis filters 62 are optional and when implemented serve to create additional input signals to the equation solver 68 such that the individual components carry different weights. If the input consists of two or more sinusoidal waves of the same frequency, then it will not be possible to distinguish between the waves. However, if the waves carry different frequency content, then it will be possible to distinguish between the waves. Processing the input with filters of different magnitude responses, phase responses, or both creates additional information for the equation solver 68.
  • the equation solver 68 is most efficiently implemented in the frequency domain. Therefore, if it has not been previously performed, the inputs are converted to the frequency domain in the forward transformers 64.
  • the equation solver 68 utilizes mathematical functions. Such functions can be included either through table look-up, Taylor-series approximation, or the like. In any case, the dynamic range of the functions may be limited due to hardware constraints. In order to gain maximal use of a limited mathematical dynamic range, the input signals are normalized in the normalizer 66. However normalization may not be necessary or desirable, in which case the normalizer 66 may be omitted. When implemented, the output from the normalizer 66 carries not only the normalized frequency domain signals but also information about the amount by which the signals have been normalized, an exponent, collected in the output signal BeamExp. This exponent enables the recovery of the absolute values from the normalized values. Each beam input and frequency may be equalized independently but the same exponent may also be used across all beams, frequencies, or both.
  • the analysis filters 62 perform differentiation with respect to time. Note that differentiation with respect to time can be expressed in the frequency domain with the transfer function of (10) below. In (10), is the imaginary unit.
  • D is the order of differentiation
  • the analysis filters 62 use the difference equation of (11) below to approximate first order differentiation with respect to time.
  • Iri (11), « is the sample index and ES is the sampling frequency.
  • a single analysis filters 62 is included to filter the first abeam signals only, that is, abeam(l).
  • the forward transformers 64 are FFT based.
  • the forward transformers 64 are performed with a time domain filterbank.
  • the forward transformers 64 are performed with a time domain filterbank that delivers quadrature outputs from which phase information can be extracted.
  • the microphone equalizer 20 of FIG. 4, the analysis beamformer 52 of FIG. 6, and the analysis filters 62 operate in the same domain.
  • the forward transformers 32 of the microphone equalizer 20 will suffice and the reverse transformers 36 of the microphone equalizer 20 and the forward transformers 64of the wave parameter estimator 54 can be omitted.
  • the normalizer 66 independently for each frequency band, finds the complex component, real or imaginary, of the ABEAM(i) signals, with the largest magnitude. A common exponent for all beams is found using this largest component. All beams are then normalized with the common exponent.
  • the equation solver 68 uses floating point arithmetic. In such a case, the normalizer 66 also converts each of the beams to floating point notation. [0064] Turning now to FIG. 9, a block diagram of the equation solver 68 of FIG.
  • a is the analysis filter index and i the abeam index.
  • the equation solver 68 may include a time domain integrator 76 which is optional. When implemented, it integrates product factors P(il,al)*P(i2,a2) over time. Through the use of the time domain integrator 76, it may be feasible to enhance the analysis especially for any embodiment using time domain filterbanks as the forward transformers 64 of FIG. 8. [0066]
  • the equation solver 68 most importantly includes a core solver 78 which solves the sound field equations.
  • a sound field can be described in several ways. The description can be in the time domain or in the frequency domain, among others. Furthermore, the description can involve a potential field describing both pressure and velocity with the same function or the description can have distinct pressure and velocity equations.
  • the sound field be described primarily in the frequency domain and simplifications will be used when judged feasible.
  • the sound field will be described with two functions.
  • the first function is a complex scalar function. It gives the sound pressure as a function of frequency and position.
  • the second function is a complex vector function. It gives the sound particle velocity as a function of frequency and position.
  • the sound consists of M waves.
  • the waves are not required to be plane waves.
  • Third, each of the waves in the description may in reality consist of the sum of several waves. The principle of superposition holds for sounds up until very high sound pressure levels.
  • each of the waves is quasi-sinusoidal in the sense that within each frequency band of the analysis the energy is mainly due to a single sinusoid only. Thus each wave may consist of several sinusoids, spread over the frequency range.
  • the small dots denote element by element multiplication and the large dots denote inner products
  • c is the speed of sound
  • k is the frequency band index
  • ⁇ m is the angular frequency of the wave m. in is the wave index.
  • x(i) is a vector and is
  • v m is a unit vector in the direction of the sound particle
  • v m is thus the direction of sound incidence of the wave.
  • ⁇ m is the damping
  • df? is the physical distance from the reference microphone, microphone 1, to microphone i.
  • the sound field be described primarily in the frequency domain.
  • the frequency domain is generally the most advantageous domain in terms of complexity and computing costs. Nevertheless in some cases the processing of the present invention is most feasible performed together with other audio processing applications. If such audio processing runs in the time domain it may prove efficient to implement the apparent incidence audio processing in the time domain as well. In the following, the sound field equations will therefore be stated in the time domain.
  • (17) states the sound field equations in the time domain under the assumptions as used in the formulation of (14) above. In (17), p is used to describe the time domain version of P and n is the sample index. (17)
  • (17) can generally not be solved for a single sample measurement only. It is necessary to measure over a number of samples and perform some form of averaging. To enable the solution, new measurement signals are defined in (18) below. In (18), indexes al and a2 may have a value of zero. B is the number of samples to average over and win() is an optional window to weight the measurements with.
  • the product terms in the right hand side of (18) will according to (17) be the product of constants and two cosine terms dependent on time through the sample index n.
  • the product of two cosine results in a sum of two different cosines, namely at the sum angle and the difference angle.
  • the sum and difference angles may be DC terms or AC terms.
  • the integral of the product of two cosines goes to zero for large integration times if the cosines are of a different frequencies. Below, (19) shows what remains of (18) when B is sufficiently large.
  • f c is the center frequency of the individual bands of the frequency transform.
  • the waves are assumed to be plane waves with ⁇ ,
  • the first wave is assumed to impinge from a target direction.
  • a further embodiment solves for two waves.
  • the first, signal, wave is constrained to have a direction from within a certain tolerance around a target direction.
  • the second, noise, wave is constrained to have a direction from outside the tolerance field.
  • a still further embodiment solves for M waves.
  • the solutions are not directly constrained but they are ordered so that the waves impinge further away from the target direction the higher the wave number is.
  • the parameters to solve for may include the following: A m - the wave amplitude (note that A m is a complex number incorporating phase information) v m - the direction of sound incidence of the wave
  • the first technique is referred to as direct solving. Under special conditions it is possible to solve (14) above directly using arithmetical methods. Such direct solving yields the wanted parameters as mathematical functions of the input signals to the core solver 78.
  • Formulate an initial guess. Compute an error. Compare computed error to a predetermined limit.
  • the error is found by subtracting the actual measurement from the results that are obtained by inserting the current solution in the right hand side of (14).
  • the error can be expressed in a mean square sense but it can also be taken as maximal of any of the measurements. However, the error can also be expressed as the relative difference between two successive solutions.
  • the wanted directivity response is symmetric around a given axis.
  • the wave parameters are solved using Newton-Rhapson iteration. Parameter error functions are defined as in (22) below.
  • the iteration stops when the mean square e ⁇ or nerr is smaller than a given value.
  • the e ⁇ or cost function, nerr is defined as the maximal relative difference between the last two iterative solutions as in (24) below.
  • Wave iteration may be said to include the following steps:
  • the microphone configuration is axis-symmetric and the e ⁇ or cost function is defined as in (22).
  • the parameter ranges are set to [0 ..
  • full parameter scan is combined with iteration.
  • Parameter ranges and grids are set up using a coarse grid. All parameter combinations on the coarse grid are used as starting guesses for an iteration leading to a "local solution" for the wave parameters.
  • the local solution with the lowest e ⁇ or cost function is chosen as the co ⁇ ect solution.
  • the fourth technique for equation solving is refe ⁇ ed to as solution screening/optimizing for minimal power.
  • the methods described above may yield more than one possible solution for the sound field equations for a given a set of measurements.
  • Measurement noise from sources such as A/D conversion, microphones, etc. can be sources of ambiguity but the system may also be underestimated.
  • the sources of underestimation are that the number of microphones used is not large enough to solve for the number of waves assumed or that the sound in reality consists of more waves than solved for.
  • the chosen solution may not be the co ⁇ ect one, but even if the co ⁇ ect solution is not chosen the strategy results in a system gain for noise components equal to or lower than the noise gain that would have been the result if the co ⁇ ect solution was to be used.
  • a specific embodiment uses the full parameter scan method. All parameter sets for which the e ⁇ or cost function is equal to or lower than the minimal cost function encountered plus a threshold are deemed possible solutions. The solution with the lowest P tot , (26), is chosen.
  • the fifth technique for equation solving is refe ⁇ ed to as solving for a subset.
  • knowledge of the full set of wave parameters is not necessary in order to control the wave gain.
  • a subset only will suffice. This may simplify the task of solving the sound field equations.
  • the wave damping is of interest.
  • Two microphones are used in this embodiment and it is assumed that the sound field consists of a single wave coming from a direction on the microphone axis. In this case, the sound field equations can be simplified as shown in (27) below.
  • a first implementation involves the use of conventional computer software.
  • application software include mathematical programs such as Mathematica, Matlab, Maple, and the like.
  • circuit simulators may be used under certain conditions.
  • An embodiment of the system includes a standard computer architecture to do parts of the acoustical sound processing.
  • the sound field equations are defined and solved for the wave parameters within a conventional software package on a conventional computer architecture.
  • a second implementation involves the use of a conventional table look up.
  • a table can be pre-computed that contains the optimal solutions with a certain resolution.
  • the table can be computed using any of the solving techniques described. Once the table has been computed, one simply looks the solution up in the table to solve the sound field equations. Adding an iteration or approximation process to the look up process can enhance it by minimizing the size of the storage used for the table.
  • FIG. 10 a block diagram of an embodiment of the core solver 78 of FIG. 9 using a table look up implementation with optional approximation is shown.
  • the measurements are rounded to a predefined precision in rounder 80.
  • the rounded measurements are then mapped to an integer space to yield an address in a map to address mapper 82.
  • the address is used by look up 84 to look up in a pre-computed table 86.
  • the table 86 may be stored on any storage device including RAM, ROM, hard disk, and the like. To save space, the table 86 may contain wave parameters in an encoded form.
  • a map to parameter mapper 88 for mapping back to parameter space may optionally be inserted as shown.
  • an interpolation 90 is optionally done to yield the parameter output.
  • the table 86 may contain parameter derivatives besides parameter values.
  • the approximation/interpolation process can be described with equation (28) below. In 28, the large dot denote inner product, m is the
  • WaveP,arm(m, ⁇ ) WP 0 (m, i) + ⁇ RoundingError s • WP g (m, i)
  • FIG. 11 a block diagram of the output generator 56 of
  • FIG. 6 is shown. Recall that the discussion continues to focus on the wave generation embodiment of the apparent incidence processor 22 of FIG. 2.
  • the output generator 56 include a statistical evaluator 92, a wave generation (WG) gain controller 94, a gain smoother 96, a gain mapper 98, and a signal generator 100.
  • the statistical evaluator 92 is optional and when implemented it analyzes the waves to obtain measures of the running signal and noise powers of the sound field.
  • the WG gain controller 94 the individual waves are analyzed and a gain is attached to each wave. The wave gains are generated so that they attenuate unwanted waves, noise, while preserving utility waves.
  • the raw gain output from the WG gain controller 94 is first smoothed and then mapped to the domain of the wave generation.
  • the purpose of the gain smoother 96 is to prevent abrupt gain changes from occu ⁇ ing.
  • the purpose of the gain mapper 98 is twofold. First, the raw gain may exhibit a frequency/value distribution that would cause time domain aliasing to occur if used in the raw state. Second, the raw gain may be defined in another domain or with a different resolution than needed for the signal generator 100. In this case, the gain mapper 98 maps the gain to the different domain/resolution. In the signal generator.100, the waves are synthesized and weighted according to the mapped gain.
  • FIG. 12 a block diagram of the statistical evaluator 92 of
  • FIG. 11 is shown.
  • Each set of wave parameters are analyzed in one of a plurality of signal or noise analyzers 102.
  • the IsSignal and IsNo ise switch signals are multiplied with the wave powers, that is, the squared wave amplitudes.
  • the wave power are summed over all waves in a signal summer 104 and a noise summer 106.
  • the summed signal and noise powers are low pass filtered in a Na ⁇ owBand filter 108 to yield na ⁇ ow band estimates of the signal and noise powers.
  • the effective integration time of the filter 108 controls the speed of the measurement. It must be set large enough that inaccuracies in the wave parameter estimates are filtered out. The na ⁇ ow band measurement may thus be relatively slow.
  • the WideBandPowers output provides the same measurements as the NarrowBandPowers output with the exception that the measurement has been integrated over wide bands in sum over bands integrators 114 before being low pass filtered in WideBand filter 110. Due to the wide bandwidth the measurement may be performed at a faster rate, that is, a shorter integration time, and with a smaller delay than the na ⁇ ow band measurement. Note that the dynamic characteristics of filters 108 and 110 control the update speed of the power signals. Therefore the filters will in general have different characteristics.
  • the signal or noise analysis 102 is based on the measured direction of sound incidence. If this is within a given tolerance equal to a target direction, then the wave/frequency pair is judged to be signal and otherwise it is judged to be noise.
  • the signal or noise analysis 102 is based on the measured direction of sound incidence.
  • the IsNoise signal is generated with the help of a directivity function as shown, for example, in the polar plots of FIG. 14.
  • the IsNoise signal is taken as unity minus the IsSignal signal.
  • the signal or noise analysis 102 is based on the measured wave damping. If this is greater than a given threshold, then the wave/frequency pair is judged to be signal and otherwise it is judged to be noise.
  • an additional path generating a second
  • NarrowBandPowers signal is provided.
  • the two NarrowBandPowers are generated with two different update rates.
  • the WG gain controller 94 includes a strategy chooser 116, a gain function chooser 118, and a plurality of gain function appliers 120.
  • the strategy chooser 116 an overall strategy is chosen based on the wideband power measurement. Strategies can, for example, be to use omni directional response or to use na ⁇ ow beam directional response, among others. The strategy is controlled in wide bands.
  • the gain function appliers 120 can be thought of as the heart of the processing. They directly control the gain of each wave as a function of some or all of the wave parameters including the direction of the sound incidence, the wave damping, and the frequency and amplitude.
  • the gain function chooser 118 selects the gain function that serves best for the cu ⁇ ent signal to noise ratio in view of the strategy input that has been chosen.
  • the output of the gain function chooser 118 will typically control the width of the main lobe of the directional response.
  • two gain strategies are implemented, that is, both omni directional and directional.
  • the strategy chooser 116 compares the wideband signal power with the wideband noise power.
  • the omni directional strategy is chosen in all na ⁇ ow frequency bands covered by wide bands where signal power is greater than a predefined constant times the noise power. In all other bands, the directional strategy is chosen.
  • the gain function appliers 120 operate by comparing the direction of incidence with a target direction. If the direction of incidence is within a predefined tolerance, the cut-off angle, from the target direction, then the raw gain is set to a predefined maximal gain and otherwise the raw gain is set to a predefined minimal gain. This results in a directivity as shown, for example, in the polar plot of FIG. 14a.
  • the gain function chooser 118 outputs the cut-off angle as a GainSelector signal. The cut-off angle is controlled as a function of the na ⁇ owband band signal to noise ratio.
  • the cut-off angle is controlled so as to produce a wide mainlobe of the beam for poor signal to noise ratios and a na ⁇ ow mainlobe for good signal to noise ratios.
  • the cut-off angle is controlled so as to produce a na ⁇ ow mainlobe of the beam for poor signal to noise ratios and a wide mainlobe for good signal to noise ratios.
  • the gain function appliers 120 operate by table look-up and optional approximation/interpolation in a similar way as the table look up process described with respect to FIG. 10 but with the wave parameters as input and the raw gain as output. Furthermore, the table output is mapped to gain space instead of parameter space.
  • This embodiment can implement an arbitrary directivity. For example, any of the directivities depicted in the polar plots of FIGs. 14a and 14b can be implemented.
  • the GainSelector signal switches between a finite number of different gain functions, for example, those implemented by different tables.
  • An example of a set of gain versus direction functions is shown in the polar plot of FIG. 14b.
  • the gain function is chosen to attenuate waves where the absolute value of the wave damping is greater than a predefined threshold. Thus waves are attenuated that are not far field waves.
  • the gain function is chosen to attenuate waves where the value of the wave damping is lower than a given threshold. Thus far field waves are attenuated.
  • the wave estimation and gain control processes will normally be performed on a block of samples.
  • the duration of the blocks will be so large that it is possibly that the raw gain for a specific frequency band in consecutive blocks will differ significantly.
  • an abrupt gain change may cause unwanted audible effects. Therefore it will generally be desirable to prevent abrupt gain changes. This is the purpose of the gain smoother 96 of FIG. 11.
  • the gain smoother 96 of FIG. 11 copies its input to its output without making any changes. In effect, this eliminates the function of the gain smoother 96.
  • the gain is smoothed in the gain smoother 96 of
  • the smoothed output is the average of the raw gains of the most recent Msmooth blocks.
  • the gain is smoothed with exponential averaging of the raw gains of successive blocks in the gain smoother 96 of FIG. 11.
  • FIG. 15 a block diagram of the gain mapper 98 of FIG.
  • the gain mapper 98 includes a reverse analysis transformer 122, a gain windower 124, and a forward gain transformer 126.
  • the raw gain is first converted from the domain of the wave estimation to the time domain in the reverse analysis transformer 122.
  • the converted raw gain is then shortened by applying a window and optionally padding with zeros in the gain windower 124.
  • the length of the window is chosen so as not to provoke time domain aliasing artifacts when the gain is applied downstream.
  • the windowed filter time (FIR) response is finally converted, by the forward gain transformer 126, to the domain that is to be used by the processing downstream.
  • FIR windowed filter time
  • the wave estimation is performed in the frequency domain.
  • the reverse analysis transformer 122 is thus FFT-based.
  • the wave estimation is performed in time domain filter bands.
  • the reverse analysis transformer 122 is then implemented as a reconstruction filter bank.
  • the wave estimation is performed in the time domain.
  • the reverse analysis transformer 122 is thus omitted.
  • the output of the gain mapper 98 is in the frequency domain.
  • the forward gain transformer 126 is thus FFT-based.
  • the output of the gain mapper 98 is in time domain filter bands.
  • the forward gain transformer 126 is thus implemented as an analysis filter bank.
  • the output of the gain mapper 98 is in the time domain.
  • the forward gain transformer 126 is thus omitted.
  • FIG. 16 a block diagram of the signal generator 100 of
  • the signal generator 100 includes at least one wave generator 128, at least one gain applier 130, a wave summer 132, and a reverse signal transformer 134. Based on the amplitude, phase, and frequency of the wave, the wave is generated with the wave generator 128. The gain is applied by the gain applier 130, the individual waves are summed by the wave summer 132, and the sum of all the waves is converted, by the reverse signal transformer 134, back to the time domain as the output of signal generator 100 and the apparent incidence processor 22 of FIG. 2.
  • the signals are generated in the frequency domain.
  • the wave generator 128 thus merely has to output the complex amplitude, A m or
  • the reverse signal transformer 134 performs an inverse frequency transform.
  • the signals are generated in the time domain with sine wave generators for the wave generators 128.
  • the reverse signal transformer 134 is omitted.
  • the signals are generated in the time domain in na ⁇ ow frequency bands by the wave generators 128. Then the ' gain is applied by multiplying in the bands. Finally, the reverse signal transformer 134 is implemented with a reconstruction filter bank.
  • FIG. 17 a block diagram according to another prefe ⁇ ed embodiment of the present invention of the apparent incidence processor 22 of FIG. 2 is shown.
  • the forward filtering method is shown.
  • the processing runs in three stages with the first two being the same.
  • analysis beamforming 52 is performed on the equalized microphone signals.
  • the parameters of the incoming sound waves are estimated in a wave parameter estimator 54.
  • the wave parameters are used in the forward filter 136 to generate filter coefficients for a filter that is applied to the input signals.
  • FIG. 18 a block diagram of the forward filter 136 of FIG.
  • the forward filter 136 include a statistical evaluator 92, a gain smoother 96, and a gain mapper 98 like those described above with respect to FIG. 11.
  • the forward filter 136 includes a forward beamformer 138, a forward filter (FF) gain controller 140, a signal filter 142, and a beam summer 144.
  • the inputs are beamformed by the forward beamformer 138 to produce a number of forward beam signals that are filtered by the signal filter 142 and summed by the beam summer 144 to form the output.
  • the filter responses, that the forward beams are filtered with, are controlled by the FF gain controller 140 that in turn uses the wave parameters as well as the statistically evaluated signal and noise powers from the statistical evaluator 92 to calculate the filter responses.
  • the forward beamformer 138 is optional and may be deleted leaving the input signals to be directly connected to the signal filter 142. When implemented, the forward beamformer 138 serves to remove noise from the signal thus enhancing the noise reduction performance achieved by the wave parameter controlled gain of the FF gain controller 140.
  • the processing is similar to that of the analysis beamformer 52 of FIG. 6.
  • the forward beamformer 138 is identical to the analysis beamformer 52 of FIG. 7.
  • the fbeam signals are taken as the abeam outputs of the analysis beamformer 52.
  • the forward beamformer 138 is in principle identical to the analysis beamformer 52 of FIG. 6 except that where the analysis beamformer 52 is optimized for frequency selectivity the forward beamformer 138 is optimized for low signal delay.
  • two fbeam signals are generated, that is, an omni directional signal and a na ⁇ ow beam, for example a supercardioid.
  • FIG. 19 a block diagram of the forward beamformer 138 of FIG. 18 is shown.
  • one or more of the fbeam signals may be generated with adaptive beamforming.
  • the adaptive beamforming is achieved by first generating, through a plurality of beamformers 146, a number of fixed beam signals. The first being the target beam, pbeam, and the rest being one or more rear beams, rbeam(q).
  • the rbeam signals are filtered by filters 148 and subtracted (150) from the pbeam to form the beamforming output
  • pbeam is an ordinary beam with full sensitivity at the target direction suppressing other directions to some extent
  • rbeam(q) is a number of different beam signals that all have zero sensitivity towards the target direction.
  • the rbeam signals can thus be subtracted from the pbeam without influencing the signal coming from the target direction.
  • the filter responses used to filter the rbeam signals are adapted by adaptors 152.
  • FIG. 20 a block diagram of the adaptor 152 of FIG. 19 is shown.
  • the fbeam output and rbeam signal are converted to the frequency domain by forward transformers 154 and conelated by a conelator 156.
  • the cross-co ⁇ elation is scaled by sealer 158 with an adaptation speed constant, mu, and is normalized with a lowpass filtered estimate, from a power filter 160, of the power of the rbeam signal.
  • the scaled cross-co ⁇ elation is integrated, by integrator and limiter 162, to yield the adapted filter response in the time domain. Besides being integrated, the filter response needs to be limited to eliminate convergence and computation noise problems.
  • a gain mapping is performed on the adapted response by a gain mapper 164.
  • the fbeam and rbeam signals are implemented in the frequency domain.
  • the forward transforms 154 are thus not implemented.
  • the conelator 156, the power filter 160, and the integrator and limiter 162 are implemented in the time domain.
  • the rbeam and fbeam signals are likewise also implemented as time domain signals and the forward transformers 154 are omitted.
  • the gain mapper 164 merely windows the filter response.
  • the integrator and limiter 162 includes a forgetting factor causing the integrated response to tend towards zero during periods of no signal activity.
  • two microphones are positioned along the target axis.
  • the forward beams include an adaptive bea fbeam(il).
  • pbeamftl is implemented with a beamformer 146 of FIG. 19 generating a supercardioid for the target direction as implemented by the beamformer filters defined in (29) below.
  • a single rear beam is used at the adaptive beamforming, rbeam(l,il). It is a cardioid in the reverse direction of the target direction as described by the component filters of (30) below.
  • (29) and (30) describe two beamformers 146 in the frequency domain.
  • PBEAM (il) CMIC( ⁇ ) * HI + CMIC(2) * H2 exp(-y • ⁇ ⁇ tO)
  • e is constant in the range 0.5 to 1 and d(2) is the microphone spacing.
  • FIG. 19 the pbeam signal for this forward beam is taken as the microphone signal cmic(l) directly without the use of the beamformer 146.
  • FIG. 21 a block diagram of the forward filter gain controller 140 of FIG. 18 is shown.
  • the FF gain controller 140 is similar to the WG gain controller 94 of FIG. 13.
  • the strategy chooser 116 and the gain function chooser 118 are comparable to those in FIG. 13 above. A few differences exist between the gain controllers though as will be described below.
  • the signal filtering in the forward filtering embodiments can be based on already beamformed input signals.
  • the FF gain controller 140 therefore has to compensate for the directivity and near field characteristics of the forward beams.
  • the BeamPar signal carries enough information about the forward beam that a plurality of FF gain function appliers 166 can compute a gain that implements the target directivity as shown, for example, in the polar plots of FIG. 14. For static forward beamformers 138 of FIG. 18, the BeamPar signal is not needed.
  • the beam directivity can be "hard coded" into the individual FF gain function appliers 166.
  • the FF gain function applier 166 includes an amplitude updater 168, a plurality of gain function appliers 170, and a wave gain weighter 172.
  • the wave amplitude is co ⁇ ected to take the characteristics of the forward beamformer into account.
  • the gain function applier 170 implements, for each wave, the directivity, amplitude, damping, and like responses in the same manner as described for the gain function appliers 120 of FIG. 13. Since the forward beam contain all waves but can only be assigned a single gain value, all of the different wave gains must be summarized.
  • the FOR WARDGAIN(i) signal is taken as the maximal of the wave gains GAINRAW(j ,i) .
  • the FORWARD GAIN (i) signal is taken as the
  • the FORWARDGAIN ) signal is the power weighted average of the individual wave gains as defined in (31) below.
  • At least one static forward beam signal and at least one adaptive beam signal are implemented.
  • the FF gain function applier 166 monitors and analyses the BeamPar signal from the adaptive beam over time. When the BeamPar signal is stable, indicating a significant noise signal from a constant direction, then the GainSelector signal is switched to use the adaptive beam mainly to build the output. When the BeamPar signal resembles random noise, then the GainSelector signal is switched to remove the adaptive beam from the output.
  • the signal filter 142 is performed in the time domain with FIR or HR filters, hi another embodiment, the signal filter 142 is performed in the time domain within frequency bands. In yet another embodiment, the signal filter 142 is performed in the frequency domain.
  • FIGs. 23 and 24 show co ⁇ esponding multiple output embodiments of the wave generation method and the forward filtering method, respectively.
  • the first output contains the sum of the near field waves while the second output contains the sum of the far field waves.
  • each output contains the sum of the waves originating from a specific range of directions.
  • Each output contains the sum of the waves originating from a specific range of directions. The wide band power of the waves in the sound field is measured. The individual output generation blocks are controlled in a way such that the first output is always generated using a range of directions centered around the origin of the wave with the largest power.
  • the wave generation method or the forward filtering method. It is also possible to combine the two methods. One simply replaces the forward filter 136 of FIG. 17 with a forward filter/output generator 174 that performs both methods.
  • FIG. 25 a block diagram of a forward filter/output generator 174 is shown.
  • the forward filter/output generator 174 is a combination of the output generator 56 of FIG.
  • the various elements are substantially similar except for a wave generator/forward filter (WGFF) gain controller 176 and an output summer 178.
  • the forward filter/output generator 174 contains two output paths. The outputs from both paths are summed by the output summer 178 to yield the combined output.
  • FIG. 26 a block diagram of the WGFF gain controller 176 of FIG. 25 is shown. As with the forward filter/output generator 174 of FIG. 25, the functioning of the WGFF gain controller 176 follows from the descriptions of the WG gain controller 94 of FIG. 13 and the FF gain controller 140 of FIG. 21, respectively.
  • the WGFF gain controller 176 chooses the gain function so that: at high signal to noise ratios, forward filtering is the primary contributor to. the output; and at low signal to noise ratios, wave generation is the primary contributor to the output.
  • the process of finding gains for the waves of the sound has included two main steps, that is, finding the parameters of the waves and deriving a gain based on the parameters found. Both these main processes can be described by mathematical transforms, as depicted in (32) below, and in many cases they are best implemented using techniques known from mathematical pocket calculators Mathematically the gain may be described as a transform directly of the inputs as described in (33) below.
  • FIG. 27 a block diagram of a single combined mathematical transform processor 180 is shown.
  • the combined processor 180 utilizes both the wave generation method and the forward filtering method to implement the core equation solving and the gain control. This implementation is especially useful with portable devices because the size of the table used for the mathematical transform may be greatly reduced as the gain may be described using fewer bits than needed for the description of the wave parameters.
  • the input signals for core solving, P, Q, QA, QB, and BeamExp, as well as the gain control inputs BeamPar and the statistical power measures are inputs to a table lookup and approximation unit 182 similar to that of FIG. 10.
  • the table lookup directly yields the raw gains as output.
  • the statistical evaluation is also performed with the help of the table lookup and approximation unit 182. It contains a model of the mapping from input values to wave parameters to power values.
  • the BlockNarrowBandPowers and BlockWideBandPowers signals contain the power estimates for the cu ⁇ ent block of samples.
  • the block estimates are low pass filtered with appropriate time constants to yield the na ⁇ ow band and wide band power signals, respectively.
  • the combined processor 180 still needs to solve for and output the wave parameters that are needed in order for the wave generation method to function. In pure forward filtering method embodiments contrarily, there will be no need to output the wave parameters.
  • a single table lookup implements the combined core solving of the sound field equations and the gain control. No statistical evaluation is performed.
  • the inputs to the table lookup are the magnitude and the phase of the ratio Q(2) of the quotient signal obtained when, in the frequency domain, the second microphone signal is divided by the first microphone signal.
  • the phase of Q(2) is quantized to one of thirty two possible phases covering the total complex phase range.
  • the magnitude of Q(2) is quantized to one of 512 possible magnitudes covering the range from 0.01 to 100.
  • the gain is stored as a binary value of either one or zero.
  • the table thus implemented requires 16 Kb of storage capacity.
  • noise-canceling microphones For applications where speech is to be picked up from the wearer of a headset, as in a mobile phone, a hearing protector, or the like, two main directions regarding noise suppression have until now been used. The most effective method has been the use of so-called noise-canceling microphones. These microphones amplify near field signals while attenuating far field signals. Unfortunately, noise-canceling microphones have to be placed no farther than two to three centimeters away from the speech source in order to be effective. This may not always be possible or convenient. Another method has been to use directional microphones pointing towards the mouth of the wearer.
  • a prefe ⁇ ed near field embodiment of the present invention enables signal processing methods with which it is possible to produce sound pick-up with near field characteristics. It is possible to obtain noise reduction better than that possible with noise-canceling microphones. Furthermore, it is possible to maintain the near field characteristic with its noise reducing virtues at a distance further away from the speech source than is possible with conventional noise- canceling microphones.
  • the near field method works by dividing the input signal into a number of frequency bands. In each band, the input signals are analyzed to see whether the activity in that band is due to near field sources or to far field sources. If the activity is from near field sources, then that band is replicated in the output with a high gain and otherwise it is replicated with a low gain.
  • FIG. 28 a block diagram of a near field embodiment of the audio processor 14 of FIG. 1 is shown.
  • the near field processor shown is especially well suited for applications where only sound from sources near to the microphones 12 for FIG. 1 should be amplified. Examples of such applications include mobile phones, headsets, and the like.
  • the near field processor includes an analog beamformer 18, at least one A/D converter 24, and at least one D/A converter 26 that are similar to those of FIGs. 2 and 3 above.
  • the near field processor includes a gain smoother 96, a gain mapper 98, and a filter 142 that are similar to those of FIG. 18.
  • the near field processor includes a microphone equalizer 200, a beamformer 202, and a near field gain controller 204.
  • the microphone signals are converted to digital signals, the microphone sensitivities are equalized, and optional beamforming is performed to yield the bmic signals.
  • the first output signal is taken as the reference input bmic(l) filtered with the filter response h.
  • the near field gain controller 204 derives a gain in frequency bands. This gain directly yields the filter response h when mapped from the domain of the gain control to the domain of the filtering.
  • the near field processor utilizes a gain function that maps the input pressures directly into band gain.
  • the microphone equalizer 200 includes a plurality of forward transformers 32 and a plurality of reverse transformers 36 that are similar to those in FIG. 4.
  • the microphone equalizer 200 includes a plurality of microphone equalization updaters 206.
  • one microphone, mic(l) is chosen as the reference.
  • the signals from the other microphone inputs are filtered so that the equalized microphone signals, cmicfi), all have the same absolute sensitivity to sound pressure levels.
  • the equalization is performed by multiplying with a frequency dependent gain, MicEq(i), in the frequency domain.
  • MicEq(i) can be a static gain, measured and saved, for example, at production test time or MicEq may also be updated dynamically.
  • FIG. 30 a block diagram of the microphone equalization updater 206 of FIG. 29 is shown.
  • the phase of the reference microphone signal is compared with the phase of the normalized signal of the microphone to be equalized.
  • the zero phase condition detector 208 outputs a logic one as its ZeroPhase signal output and otherwise the output will be a logic zero. .
  • the accumulator 210 is divided into static and dynamic parts, where the updates only influence the dynamic parts.
  • the effective equalization response is the product of the static and dynamic parts.
  • the static part of the equalization response is measured with standard measuring techniques once at the time of production test or at some other convenient time and saved.
  • a forgetting factor is included with the dynamic part of the accumulator 210. The forgetting factor causes the dynamic response to converge towards zero when no updates are received.
  • means are provided that can save the accumulated equalization response when the near field audio processor is powered down and read the saved response again when the processor is powered up the next time.
  • the microphones used have the same directivity and frequency response except for the small tolerances that the microphone equalizer 200 of FIG. 28 should be able to compensate for.
  • the direction of sound incidence of a sound wave is perpendicular to an axis connecting the reference microphone with the cu ⁇ ent microphone, then the sound wave must arrive with the same amplitude at both microphones.
  • This perpendicular condition is detected by comparing the phases of the two microphone signals in the zero phase condition detector 208. If the phases differ by less than a certain tolerance, then the ZeroPhase signal is generated as a logic one and otherwise it is generated as a logic zero.
  • the signal inband detector 212 for each frequency band evaluates the absolute value of its input signal in the cu ⁇ ent band and the two nearest neighboring bands. If the cu ⁇ ent band carries the highest absolute value, then the Inband signal for the cu ⁇ ent band is generated as a logic one and otherwise it is generated as a logic zero. [0187] Turning now to FIG. 31, a block diagram of the beamformer 202 of FIG.
  • the beamformer 202 is similar to the beamformer 52 of FIG. 7 and also includes a plurality of filters 214 and a summer 216. Again the beamformer 202 is optional and may be omitted.
  • the aim of the beamforming process is to remove noise from the signal prior to the near field gain and filter processing, thereby enhancing the performance of these portions of the process.
  • the microphone inputs are filtered with separate filters and summed to yield the beam output.
  • M microphones are placed along a common axis.
  • the beams are supercardioids.
  • the beams are figures of eight.
  • the beamforming is performed in the time domain with FIR or IIR filters.
  • the beamforming is performed in the frequency domain.
  • the near field gain controller 204 includes a forward transformer 218, a power filter 220, a phase filter 222, a statistical evaluator 224, and a near field gain function applier 226.
  • the beam signals are split in frequency bands or converted to the frequency domain in the forward transformer 218.
  • the power filter 220 the signal powers are measured with a given time constant.
  • the outputs from the power filter 220, R(i) give the ratio between the power of the cu ⁇ ent microphone signal and the power of the reference microphone signal bmic(l).
  • the phase filter 222 the filtered signal phases are compared.
  • the PHI(i) outputs give the difference between the unwrapped phase of the cu ⁇ ent microphone and the unwrapped phase of the reference microphone bmic(l).
  • the statistical evaluator 224 measures the signal and noise powers of different bandwidths and time constants. In the near field gain function applier 226, the raw channel gains are derived.
  • the gain control processing is performed on blocks of samples. For each block, a single complex signal value per frequency band is computed.
  • the power and phase filters 220, 222 only use the values from the cu ⁇ ent block to compute their respective outputs.
  • the power and phase filters 220, 222 averages the signal powers and phases over consecutive blocks.
  • phase averaging is power weighted.
  • no phase information is utilized.
  • the forward transformer 218 is implemented with a time domain filterbank, no phase information is generated or used, and the signal powers are measured with a finite time constant.
  • the forward transformer 218 is FFT based.
  • FIG. 33 a block diagram of the statistical evaluator 224 of
  • FIG. 32 is shown.
  • a signal or noise analyzer 2208 the power and phase inputs are evaluated.
  • the band carries utility signal or noise information. If the band carries signal information, then the co ⁇ esponding part of the IsSignal signal is set to a logic one and the co ⁇ esponding part of the IsNoise signal is set to a logic zero. If the signal carries noise, then IsSignal is set to a logic zero and IsNoise is set to a logic one.
  • the IsSignal and IsNoise switch signals are multiplied with the Wave powers, that is, the squared wave amplitudes.
  • the weighted signal and noise powers are low pass filtered in Na ⁇ owBand filter 230 to yield na ⁇ ow band estimates of the signal and noise powers.
  • the effective integration time of the filter 230 controls the speed of the measurement. It must be set large enough that inaccuracies in the wave parameter estimates are filtered out. The na ⁇ ow band measurement may thus be relatively slow.
  • the WideBandPowers output provides the same measurements as the NarrowBandPowers output with the exception that the measurement has been integrated over wide bands in sum over bands integrators 240 before being low pass filtered in WideBand filter 232. Due to the wide bandwidth the measurement may be performed at a faster rate, that is, a shorter integration time, and with a smaller delay than the na ⁇ ow band measurement. Note that the dynamic characteristics of filters 230 and 232 control the update speed of the power signals. Therefore the filters will in general have different characteristics.
  • the signal or noise analyzer 228 is based on the R(2) signal. If this signal is less than a predefined threshold, then the signal is judged to be utility signal and otherwise it is judged to be noise.
  • NarrowBandPowers signal.
  • the two NarrowBandPowers are generated at two different update rates.
  • the near field gain function applier 226 of FIG. 32 provides the core functionality of the near field processing method. It maps a set of level ratios and optionally phase and signal power information into a gain. The gain should provide , larger amplification for frequency bands containing mainly near field source material and smaller gain for frequency bands containing mainly far field source material.
  • the near field gain function applier 226 of FIG. 32 controls.the gain in the frequency bands as a function of the ratio of the microphone powers in the bands as shown, for example, in the graph of FIG. 36a.
  • FIG. 34 a block diagram of an embodiment of the near field gain function applier 226 of FIG. 32 is shown.
  • the near field gain function applier 226 includes a threshold comparer 242, a combinatorial unit 244, and a gain mapper 246.
  • the threshold comparer 242 generates logic signals as defined in (34) below.
  • the , combinatorial unit 244 performs Boolean algebra on these logic signals to yield an output logic signal, BINGAIN, that indicates whether the respective frequency band should be assigned a high gain for signal or a low gain for noise.
  • the gain mapper 246 maps the output logic signal to actual gain values according to (35) below.
  • the near field gain function can be written as in (36) below.
  • phase is evaluated as well.
  • the na ⁇ ow band powers are evaluated.
  • the gain function can be described with (38) below.
  • FIG. 35 a block diagram of an embodiment of the near field gain function applier 226 of FIG 32 using a table look up implementation with subsequent approximation/interpolation is shown.
  • the function inputs are rounded to a predefined precision by rounder 248.
  • the rounded inputs are then mapped, by address mapper 250, to an integer space to yield an address.
  • the address is used by look up 252 to look up in a pre-computed table 254.
  • the table 254 may be stored on any storage device including RAM, ROM, hard disk, and the like.
  • the table 254 may contain gain values in an encoded form.
  • a gain mapper 256 for mapping back to gain space may optionally be inserted as shown.
  • an interpolator 258 is optionally provided to yield the raw gain output.
  • the table 254 may contain parameter derivatives in addition to parameter values.
  • the WidebandPowers are monitored. At good signal to noise ratios, all of the signals are passed through without attenuation. At poor signal to noise ratios, a near field characteristic is used.
  • the NarrowBandPowers are monitored.
  • gain functions of different widths are chosen. For example, the gain function of different widths are shown in the graph of FIG. 36b.
  • the wideband signal power is compared with the wideband noise power and two gain strategies are implemented, that is, both omni directional and directional.
  • the omni directional strategy is chosen in all na ⁇ ow frequency bands covered by wide bands where signal power is greater than a predefined constant times the noise power. In all other bands, the directional strategy is chosen.
  • the filtering is performed in the time domain with a FLR or HR filter.
  • the filtering is performed with a FFT based FIR filtering.
  • the filtering is performed with a time domain filterbank.

Landscapes

  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Otolaryngology (AREA)
  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Acoustics & Sound (AREA)
  • Signal Processing (AREA)
  • Circuit For Audible Band Transducer (AREA)

Abstract

L'invention concerne un système de traitement sonore comprenant au moins un microphone, un processeur audio et au moins un dispositif de sortie. Le processeur audio comprend un formeur de faisceaux analogue, un égalisateur de microphone et un processeur d'incidence apparente. L'invention comprend deux modes de réalisation distincts de processeur d'incidence apparente, soit un procédé de génération d'ondes et un procédé de filtrage avant. Dans ces deux modes de réalisation, les même principes sont mis en oeuvre pour estimer les propriétés des ondes individuelles du champ sonore. Le système selon l'invention permet de mettre en oeuvre des réponses de directivité arbitraire avec uniquement un petit nombre de microphones, c'est-à-dire deux ou trois microphones. Le système selon l'invention permet d'obtenir une réduction du bruit améliorée, notamment pour des milieux comprenant plusieurs sources de bruit indépendantes et peut être mis en oeuvre pour des signaux et des bruits à statistiques arbitraires.
EP02767372A 2001-08-10 2002-08-12 Systeme de traitement sonore comprenant un generateur d'ondes a reponses de directivite et de gradient arbitraires Withdrawn EP1415503A2 (fr)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
US92778301A 2001-08-10 2001-08-10
US927783 2001-08-10
PCT/EP2002/009031 WO2003015460A2 (fr) 2001-08-10 2002-08-12 Systeme de traitement sonore comprenant un generateur d'ondes a reponses de directivite et de gradient arbitraires

Publications (1)

Publication Number Publication Date
EP1415503A2 true EP1415503A2 (fr) 2004-05-06

Family

ID=25455249

Family Applications (1)

Application Number Title Priority Date Filing Date
EP02767372A Withdrawn EP1415503A2 (fr) 2001-08-10 2002-08-12 Systeme de traitement sonore comprenant un generateur d'ondes a reponses de directivite et de gradient arbitraires

Country Status (2)

Country Link
EP (1) EP1415503A2 (fr)
WO (1) WO2003015460A2 (fr)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7274794B1 (en) 2001-08-10 2007-09-25 Sonic Innovations, Inc. Sound processing system including forward filter that exhibits arbitrary directivity and gradient response in single wave sound environment
US7127076B2 (en) 2003-03-03 2006-10-24 Phonak Ag Method for manufacturing acoustical devices and for reducing especially wind disturbances
EP2254349A3 (fr) * 2003-03-03 2014-08-13 Phonak AG Procédé pour la fabrication des dispositifs acoustiques et pour la réduction des perturbations dues au vent
US7688985B2 (en) 2004-04-30 2010-03-30 Phonak Ag Automatic microphone matching
EP1489883A3 (fr) * 2004-04-30 2005-06-15 Phonak Ag Adaption automatique des microphones
US9357307B2 (en) 2011-02-10 2016-05-31 Dolby Laboratories Licensing Corporation Multi-channel wind noise suppression system and method
CN112259068B (zh) * 2020-10-21 2023-04-11 上海协格空调工程有限公司 一种主动降噪空调系统及其降噪控制方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6130949A (en) * 1996-09-18 2000-10-10 Nippon Telegraph And Telephone Corporation Method and apparatus for separation of source, program recorded medium therefor, method and apparatus for detection of sound source zone, and program recorded medium therefor
US6317703B1 (en) * 1996-11-12 2001-11-13 International Business Machines Corporation Separation of a mixture of acoustic sources into its components
KR100198289B1 (ko) * 1996-12-27 1999-06-15 구자홍 마이크 시스템의 지향성 제어장치와 제어방법
US6023514A (en) * 1997-12-22 2000-02-08 Strandberg; Malcolm W. P. System and method for factoring a merged wave field into independent components

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See references of WO03015460A3 *

Also Published As

Publication number Publication date
WO2003015460A3 (fr) 2003-11-20
WO2003015460A2 (fr) 2003-02-20

Similar Documents

Publication Publication Date Title
US7274794B1 (en) Sound processing system including forward filter that exhibits arbitrary directivity and gradient response in single wave sound environment
AU2002331235A1 (en) Sound processing system including forward filter that exhibits arbitrary directivity and gradient response in single wave sound environment
US10225674B2 (en) Robust noise cancellation using uncalibrated microphones
US8942387B2 (en) Noise-reducing directional microphone array
EP0720811B1 (fr) Systeme de reduction du bruit dans une prothese auditive stereophonique
US10657981B1 (en) Acoustic echo cancellation with loudspeaker canceling beamformer
EP2848007B1 (fr) Réduction du bruit dans un réseau de microphones directionnelle
US6178248B1 (en) Dual-processing interference cancelling system and method
US20080260175A1 (en) Dual-Microphone Spatial Noise Suppression
JP4973655B2 (ja) 適応アレイ制御装置、方法、プログラム、及びこれを利用した適応アレイ処理装置、方法、プログラム
JP2004537944A (ja) オーバーサンプルされたフィルタバンクを用いる指向性オーディオ信号処理
US10269370B2 (en) Adaptive filter control
WO2009042385A1 (fr) Procédé et appareil pour générer un signal audio à partir de multiples microphones
WO2003015460A2 (fr) Systeme de traitement sonore comprenant un generateur d'ondes a reponses de directivite et de gradient arbitraires
EP1415502A2 (fr) Systeme de traitement de son comprenant un filtre de retroaction faisant preuve d'une directivite arbitraire et d'une reponse aux gradients dans un environnement sonore a ondes multiples
WO2003015459A2 (fr) Systeme de traitement du son presentant une reponse du gradient arbitraire
AU2002331238A1 (en) Sound processing system including wave generator that exhibits arbitrary directivity and gradient response
AU2002331236A1 (en) Sound processing system including forward filter that exhibits arbitrary directivity and gradient response in multiple wave sound environment
AU2002331237A1 (en) Sound processing system that exhibits arbitrary gradient response
Samtani et al. FPGA implementation of adaptive beamforming in hearing aids
Geiser et al. A differential microphone array with input level alignment, directional equalization and fast notch adaptation for handsfree communication
Chintakuntla Keystroke Resilience in Laptop Acoustics Using Wiener Beamformer and Spectral Subtraction

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20040217

AK Designated contracting states

Kind code of ref document: A2

Designated state(s): AT BE BG CH CY CZ DE DK EE ES FI FR GB GR IE IT LI LU MC NL PT SE SK TR

AX Request for extension of the european patent

Extension state: AL LT LV MK RO SI

17Q First examination report despatched

Effective date: 20040908

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN

18D Application deemed to be withdrawn

Effective date: 20070301