An embodiment of the invention relates to digital audio signal processing techniques used to provide an acoustic transparency function in a pair of headphones.
A typical consumer electronics headset contains a pair of left and right headphones and at least one microphone that are connected either wirelessly or via a cable to receive a playback signal from an electronic audio source, such as a smartphone. The physical features of the headphone are often designed to passively attenuate the ambient or outside sounds that would otherwise be clearly heard by the user or wearer of the headset. Some headphones attenuate the ambient sound significantly, by for example being “closed” against the wearer's head or outer ear, or by being acoustically sealed against the wearer's ear canal; others attenuate only mildly, such as loose fitting in-ear headphones (earbuds.) An electronic, acoustic transparency function may be desirable in some usage scenarios, to reproduce the ambient sound environment through the earpiece speaker drivers of the headphones. This function enables the wearer of the headset to also hear the ambient sound environment more clearly, and preferably in a manner that is as “transparent” as possible, e.g., as if the headset was not being worn.
An embodiment of the invention is an audio system that includes a headset that picks up sound in the ambient environment of the wearer, electronically processes it and then plays it through the earpiece speaker drivers, thereby providing acoustical transparency (also referred to as transparent hearing, or hear through mode.) The wearer's sound experience while wearing the headset may thus be equivalent to what would be experienced without the headset (despite the headset passively attenuating the ambient sound.) The headset has a left exterior microphone array and a right exterior microphone array. Each of the microphone signals, from the left and right arrays, is fed to a respective, digital, acoustic transparency filter. The filtered signals are combined and further digitally processed into a left speaker driver signal and a right speaker driver signal, which are routed to left and right earpiece speaker driver subsystems, respectively, of the headset. A data processor performs an algorithm that computes the transparency filters in such a manner that the filters may reduce the acoustic occlusion due to the earpiece, while also preserving the spatial filtering effect of the wearer's anatomical features (head, pinna, shoulder, etc.) The filters may help preserve the timbre and spatial cues associated with the actual ambient sound. A transparent hearing filter design that, to a certain degree, avoids coloring the speaker driver signal, e.g., reduces resonances at higher frequencies, and avoids altering the spatial imaging is desired. Methods are described for how to create non-adaptive transparent hearing filters that are generalized or robust (e.g., are suitable for a population of users.)
In one embodiment, multiple reference measurements are made in a laboratory setting, on different individuals or on different dummy head recordings, and across different headset re-seatings, in order to generalize the design of the transparency filters. This may result in a filter design that works for a population or majority of users. The filter design may be computed, by a mathematical process of joint optimization, or as a particular, closed form solution. A target head related transfer function (HRTF) or, equivalently, head related impulse response (HRIR), is used in both cases, which may be that of a single individual. Such a transparent hearing filter design may reduce coloring of the speaker driver signals (preserving the timbre of the ambient acoustics), while yielding correct spatial imaging (e.g., the sound of an actual airplane flying above the wearer is captured and electronically processed before being played back through the speaker drivers, in such a way that the wearer feels the sound being produced by the speaker drivers is coming from above the wearer's head rather than being “within the user's head.”) It may reduce acoustic occlusion due to the headphone being worn, while also preserving the spatial filtering effect of the wearer's anatomical features (head, pinna, shoulder, etc.)
In another embodiment of the invention, the design of a transparency filter is customized or personalized to the wearer, based on real-time detection of the wearer's acoustic characteristics, using an audio system that has two adaptive subsystems. A first adaptive subsystem computes an adaptive path estimation filter, whose transfer function estimates a path from an input of an earpiece speaker to an output of an interior microphone of a headset, using a playback signal that is driving the earpiece speaker and using an output signal from the interior microphone. The first adaptive subsystem removes a filtered version of the playback signal, which is filtered by the adaptive path estimation filter, from the output signal of the interior microphone. A second adaptive subsystem (running in parallel with the first subsystem) computes an adaptive output filter. The output filter has an input coupled to receive a reference signal produced by an exterior microphone of the headset, and an output that is driving the earpiece speaker. The output filter is computed using a difference between i) a version of the reference signal that has been filtered by a signal processing control block and ii) the output signal of the interior microphone from which the filtered version of the playback signal has been removed.
In one embodiment, the transparency function made be achieved by a processor programming the signal processing control block, which may be a filter that is to be programmed in accordance with a predetermined set of digital filter coefficients (that define the filter and that may be stored in the audio system), wherein the filter so programmed causes the second adaptive subsystem to produce sound pressure at the interior microphone of the headset that is a delayed and frequency-shaped version of sound pressure at the exterior microphone of the headset; this result may be independent of the playback signal, in that the playback signal may coexist with the transparency function. To better evaluate the transparency function in practice, the playback signal may be muted.
Properly configuring the signal processing control block will cause the second adaptive subsystem to adapt the output filter to meet, at a given time, any one of several different transparency conditions. In one condition, referred to here as full acoustic transparency mode, the output filter is automatically adapted to recreate (through the speaker driver) the ambient acoustic environment that is sensed in the reference signal. In another condition, referred to here as full ANC mode, the output filter is producing an anti-noise signal to cancel any leaked ambient sound, across its entire working bandwidth (e.g., conventional ANC operation.) In yet another condition, referred to as a hybrid ANC-transparency mode, the output filter is producing a signal that is designed to cancel the ambient sound in just a portion of the entire audio band (ANC in a low frequency band) while intentionally allowing the ambient sound to come through clearly in another portion of the entire audio band (e.g., a high frequency band.) Other more complex conditions for the adaptive digital output filter are possible, by the proper spectral shaping of the transfer function of the signal processing control block, including for example a tunable strategy for compensating for hearing resonances that are lost in the occlusion effect (especially due to a closed headphone), or a subjective tuning strategy (e.g., a physical or virtual knob allowing “manual” control by the wearer) that allows the wearer to subjectively set the timbre in the transparency mode.
The above summary does not include an exhaustive list of all aspects of the present invention. It is contemplated that the invention includes all systems and methods that can be practiced from all suitable combinations of the various aspects summarized above, as well as those disclosed in the Detailed Description below and particularly pointed out in the claims filed with the application. Such combinations have particular advantages not specifically recited in the above summary.
The embodiments of the invention are illustrated by way of example and not by way of limitation in the figures of the accompanying drawings in which like references indicate similar elements. It should be noted that references to “an” or “one” embodiment of the invention in this disclosure are not necessarily to the same embodiment, and they mean at least one. Also, in the interest of conciseness and reducing the total number of figures, a given figure may be used to illustrate the features of more than one embodiment of the invention, and not all elements in the figure may be required for a given embodiment.
Several embodiments of the invention with reference to the appended drawings are now explained. Whenever the shapes, relative positions and other aspects of the parts described in the embodiments are not explicitly defined, the scope of the invention is not limited only to the parts shown, which are meant merely for the purpose of illustration. Also, while numerous details are set forth, it is understood that some embodiments of the invention may be practiced without these details. In other instances, well-known circuits, structures, and techniques have not been shown in detail so as not to obscure the understanding of this description.
Each of the headphones also includes an earpiece speaker driver subsystem or earpiece speaker 5, that may have one or more individual speaker drivers that is to receive a respective left or right speaker driver signal and produce sound that is directed into the respective ear of the wearer or dummy head. In one embodiment, the headset includes additional electronics (not shown) such as an audio signal communication interface (e.g., a Bluetooth interface, a wired digital audio interface) that receives a playback audio signal from an external audio processing source device, e.g., a smartphone. This playback audio signal may be digitally combined with the transparency signal produced by the DSP block d[n], before the combination audio signal is fed to a driver input of the earpiece speaker 5. To reduce the possibility of too much latency being introduced between the pickup of ambient sound by the microphones 4 and their reproduction through the earpiece speaker 5, the digital signal processing performed by the transparent hearing filters 6 and the DSP blocks d[n] in
Each of the transparent hearing filters 6 is defined by its impulse response h[n] and is identified by its indices x,y. In the particular example shown in
A process for computing the transparent hearing filters 6 may be described with reference to
$$
$$
$$
$$
In the above Eq. 1, R represents a matrix of known convolution matrices convmtx(r,m,s,i), where each convolution matrix contains the known impulse responses illustrated in
With the above in mind, we return to the process for computing the transparent hearing filters 6, where the matrix R needs to be computed. To do so, a group of reference measurements of reproduced ambient sound are recorded in a laboratory setting. This may be done using a number of dummy head recordings that simulate hearing of different individuals, respectively, or using a number of real-ear measurements taken from a number of individuals, respectively. The reference measurements are made while the headset 2 is operating in measurement mode, in an anechoic chamber or other non-reflective laboratory setting. In the measurement mode, the transparency hearing filters 6 and the DSP blocks d[n] depicted in
In one embodiment, each group of L. K. 2. M reference measurements are repeated for a number of different re-seatings, respectively, of the specimen of the headset 2 (as worn on the dummy head or by the individual.) The re-seatings may be informed based on observations of how headsets in general are worn, by different persons. In that case, the matrix R will contain impulse responses for different re-seatings. In yet another embodiment, each group of L. K. 2. M reference measurements are repeated for several different individuals (e.g., several different dummy heads or several individuals), so that R in that case contains impulse responses not just for the different re-seatings but also for the different individuals. As explained below, this results in a solution for h (the vector of impulse responses of the transparent hearing filters 6) that is quite robust in that the transparent hearing filters 6 are smoother and generalized to the variety of wearing conditions.
The process continues with performing a mathematical process to compute the actual impulse responses of all of the individual transparent hearing filters 6, based on the numerous reference measurements that are reflected in the matrix R and for a target HRIR vector, t. In one embodiment, an optimization algorithm is performed that finds an estimate h_hat (for the vector h) that minimizes the expression
p-norm of(R.h+g−t)
where R is the impulse response matrix, t is a target or desired HRIR vector, and g is an acoustic leakage vector which represents the effect of some ambient sound that has leaked past the headphones and into the ear. In the case where the matrix R includes measured impulse responses for several re-seatings, on the same dummy head, a joint optimization process is performed that results in transparency hearing filters 6 (as defined by the computed estimate h_hat) whose transfer functions exhibit fewer spectral peaks and notches at high frequencies, and are therefore more robust or more generalized for a larger population of wearers.
In another embodiment of the invention, the optimization problem in Eq. 1 is solved while applying an L-infinity constraint to the h vector. See equations below. The peaks in the filter design process are kept below or within prescribed levels. This may be preferable to the use of regularization techniques associated with matrix inversions. As an alternative, an L-2 norm constraint may be applied which would constrain the total energy of each h filter (as compared to constraining just the peaks.)
$$
$$
$$
$$
Some benefits of the L-infinity constraint may include the consolidation of the filter design into a single optimization process, avoiding the use of inflexible regularization parameters, directly correlating to a clear filter characteristic by constraining the gains associated with the designed filters, and faster computation using convex optimization solvers.
In yet another embodiment of the constrained optimization problem, an L-2 norm constraint is applied that prescribes a sensitivity parameter, white noise gain (WNG), to avoid boosting a noise floor. This may be viewed as constraining the sum of energy of filters in each band, as opposed to the peaks in bands of individual filters (for the L-infinity constrained solution), or the energy of the individual filters (for the L-2 constrained solution.)
In yet another embodiment, a closed form solution h_hat can be derived, which is given by
h_hat=(R_transpose.R)inverse.R_transpose.(t−g) (Eq.2)
where again R is the impulse matrix, t is the target HRIR vector, and g is the acoustic leakage vector.
Once h_hat has been computed, which defines all of the transparent hearing filters 6, copies of the computed transparent hearing filters 6 are stored into a number of other specimens of the headset 2, respectively. Each of these specimens of the headset 2 is configured to operate in an acoustic transparency mode of operation in which the stored copy of the transparent hearing filters 6 are used as static or non-adaptive filters, during in-the-field use of the headset 2 (by its purchaser-wearer.) The headset 2 as part of an audio system provides acoustical transparency (transparent hearing, or hear through) to the wearer, such that the wearer's experience of the ambient sound while wearing the headset may be more equivalent to what would be experienced without the headset (despite the headset passively attenuating some of the ambient sound.) The transparency hearing filters 6 as computed above help preserve the timbre and spatial cues of the actual ambient sound environment, and work for a majority of wearers despite being a static or non-adaptive solution.
In accordance with another embodiment of the invention, the transparency hearing filters 6 (TH filters 6), in static or non-adaptive form, may be incorporated into an audio system that also includes an acoustic noise cancellation (ANC) subsystem.
In one embodiment, the transparent hearing filters 6 can be disconnected so as to maximize the acoustic noise cancellation effect, during the phone call. For that embodiment, the audio system may also include a number of sidetone filters 7, and multiplexor circuitry (depicted by the switch symbol in
In the sidetone mode, this allows the near end user to also hear some of her own voice during the phone call (as picked up by the exterior microphones 4.) Note that the uplink communications audio signal, which contains the near end user's voice, may be derived from the outputs of the exterior microphones 4, since these can also pick up the near end user's voice during the call.
In another embodiment, the audio system may further include a compressor 16 that is to receive the gain-adjusted version of the transparency signal (assuming the switch is in the TH filter 6 position), to produce a dynamic range adjusted and gain-adjusted version of the transparency signal. The compressor 16 can reduce dynamic range (compression) of the transparency signal, which may improve hearing protection; alternately, it may increase dynamic range (expansion) during an assisted hearing mode of operation in which the wearer of the headset 2 would like to hear a louder version of the ambient sound. An operating profile or compression/expansion profile of the compressor 16 may be adjustable (e.g. threshold, gain ratio, and attack and release intervals) and this, along with the scalar gain provided by the first gain block 9, may be set by the oversight processor 15, based on the latter's analysis of the ambient sound through the exterior microphones 4, the signal from the interior microphone 3, other sensors (not shown), as well as the desired operating mode of the headset (e.g., full transparency mode, full ANC mode, mixed ANC-transparency mode, and assisted hearing mode.) Such analysis may include any suitable combination of howling detection, wind/scratch detection, microphone occlusion detection, and off-ear detection. Such analysis by the oversight processor 15 may also be used by it to adjust or set the gain of the first gain block 9.
In yet another embodiment, also illustrated in
In one embodiment, the second anti-noise signal is produced at all times during an ANC mode of operation, while the first anti-noise signal is either attenuated or boosted by the second gain block 14 depending on decisions made by the oversight processor 15 (in view of its analysis of the conditions give above.)
The embodiments of the invention described above in connection with
Still referring to
pe=pr.(P+Gr.T.S) (Eq. 3)
The first adaptive subsystem has an adaptive filter SE controller 26 that computes the adaptive path estimation filter 25 (filter SE), based on inputs that include i) the playback signal and ii) the output signal of the interior microphone (shown as the output of the transducer block Ge) from which a filtered version of the playback signal has been removed by a digital differencing unit 23. The playback signal is also driving the earpiece speaker (input to path S.) The playback signal is filtered by the adaptive path estimation filter 25 before being removed from the output of the transducer block Ge. The adaptive filter SE controller 26 may implement any suitable iterative search algorithm to find the solution SE, for its adaptive path estimation filter 25, which minimizes the error signal at the output of the differencing unit 23, e.g., a least mean square (LMS) algorithm.
The audio system also has a second adaptive subsystem that should be designed to compute the adaptive output filter 21 (e.g., implemented as a finite impulse response, FIR, or infinite impulse response, IIR, digital filter) to have a transfer function T that meets the following equation:
T=(1−P)/Gr. S (Eq. 4)
This equation expresses the desired response of T that causes the acoustic pressure pe as sensed by the transducer block Ge to match pr as sensed by the transducer block Gr (transparency or hear through.) The adaptive output filter 21 having the desired response T may be computed by an adaptive output filter controller 27 that finds the adaptive output filter 21 which minimizes an error input being a difference between i) a version of the reference signal that has been filtered by a signal processing control block 29 (having a transfer function D) and ii) the output of the differencing unit 23 (which is the signal of the interior microphone from which the SE filtered version of the playback signal has been removed.) This minimization is performed while the reference input of the adaptive filter controller 27 is a version of the reference signal that has been filtered by a filter SE copy 28 which is a copy of the adaptive path estimation filter 25 (that is being adapted by the controller 26.) Any suitable iterative search algorithm may be used for minimization of the error signal at the output of the differencing unit 24, by the adaptive output filter controller 27, e.g., a least mean square (LMS) algorithm.
The error signal at the output of the differencing unit 24 may be written as:
Pr.Gr.D−pr.Gr.T.S.Ge−pr.P.Ge=>0 (Eq. 5)
Assuming T is realizable, then in the presence of broadband signals, the controller 27 will drive Eq. 5 towards zero and the equation can be re-written as:
T=(D−P.(Ge/Gr))/S.Ge (Eq. 6)
Which is a more generalized version of Eq. 4 as the target transparency of pe/pr has not been defined yet. Substituting Eq. 6 into Eq. 3 yields:
pe/pr=D.Gr/Ge (Eq. 7)
According to Eq. 7, by configuring the signal processing control block 29 (having a transfer function D), and based on the ratio of the transducer block responses, Gr/Ge, it is possible use the two adaptive subsystems working together, to automatically adapt the adaptive output filter 21 (transfer function T) to yield a desired transparency (e.g., full transparency when pe/pr=1.) A processor (not shown) can adjust the signal processing control block 29, which causes a change in the computation of the adaptive output filter 21, which in turn changes acoustic transparency through the path S and at the acoustic summing junction 20 of the headset.
When the signal processing control block 29 is a digital filter (whose transfer function D may be realizable with an FIR filter and one or more IIR filters, for example), the processor can program the digital filter in accordance with a predetermined set of digital filter coefficients that define the filter and that may be stored in the audio system. The digital filter (transfer function D) so programmed causes the second adaptive subsystem (and the controller 27) to compute the adaptive output filter 21 so as to yield acoustic transparency through the path S (earpiece speaker) of the headset.
In one embodiment, the signal processing control block 29 includes a full band or scalar gain block (no frequency dependence), whose gain value is adjustable between a low value (e.g., zero) and a high value (e.g., Ge/Gr) with an intermediate value there between. The low value causes the controller 27 to adapt the adaptive output filter 21 to yield no acoustic transparency, because the controller 27 is now adapting the adaptive output filter 21, effectively as a feed forward ANC subsystem, to produce an anti-noise signal that yields ANC at the interior microphone (or at the acoustic summing junction 20.) When the scalar gain block of the signal processing control block 29 is set to its high value, e.g., Ge/Gr, the controller 27 will adapt the transfer function T so as to yield full acoustic transparency at the acoustic summing junction 20 (pe/pr=1.) Setting the scalar gain block to the intermediate value yields partial acoustic transparency.
By including a linear delay element within the signal processing control block 29, e.g., coupled in series or cascaded with the scalar gain block or with a spectral shaping digital filter, it is possible to improve the causality of the transfer function T in Eq. 5. As an example, a linear delay of leading zeroes in an FIR filter is practical.
The following are examples of how the signal processing control block 29 may be used to achieve various, programmable levels or types of transparency (at the acoustic summing junction 20.)
If the target is to have full transparency, then set filter D in Eq. 7 to equal Ge/Gr with some fixed delay; and the adaptive system will drive pe to equal pr. The value Ge/Gr may be trimmed in factory, and programmed into D. D can be an FIR filter, for when Ge and Gr are only different in magnitude, as can be expected in some products over most audio frequencies of interest. Note here that there is no requirement to have run an ANC system.
If the target is to have zero transparency, then set filter D in Eq. 7 to equal zero; and the adaptive system will drive the acoustic pe (while ignoring the playback signal) towards zero. Note also that in this configuration of filter D the adaptive system is transformed into a feed forward adaptive ANC system.
But if the target is to have partial transparency, set filter D in Eq. 7 to some intermediate value between zero and Ge/Gr, with some fixed delay; and the adaptive system will drive the acoustic summing junction 20 to have pe at a lower level than pr. This may provide more comfortable transparency experiences for users in noisy environments, and will result in some amount of ANC at low frequencies.
In another embodiment, the signal processing control block 29 is a filter D that is to be programmed by a processor (in accordance with a predetermined set of digital filter coefficients that define the filter and that are stored in the system) to have a particular spectral shape, such that the filter D so programmed causes the second adaptive subsystem to yield greater acoustic transparency over a first audio frequency band than over a second audio frequency band. Thus, for instance, if D is a high-pass shelf filter normalized such that the response is Ge/Gr at high frequencies, and low or zero at low frequencies, then a hybrid transparency results: ANC (or zero transparency) will happen at low frequencies, and full transparency will occur at high frequencies. One instance of this is a 2^{nd }order IIR shelving filter, with variable gain, and variable corner frequency. Higher order filters may also be used. By changing the overall gain, the adaptive system may provide partial transparency at high frequencies and ANC at low frequencies.
In another embodiment, where filter D is configured to have a particular spectral shape, if filter D is configured to have two or more peaking filters each with positive and/or negative gains set at higher frequencies, then some compensation can be introduced for user hearing responses that are occluded by the headset that has a closed headphone. For instance a peak at or near 3 kHz may be desirable, to correspond to the pinna ear acoustical resonance.
In yet another embodiment, if filter D is configured to be a low-pass shelf filter then subjective tuning can be performed. In other words, the wearer can manually adjust a virtual or physical tuning knob of the audio system (that includes the headset 2) which changes the characteristics of the low-pass shelf filter (e.g., cutoff frequency, roll off rate), if the full transparency mode is considered to sound too bright by some wearers.
In yet another embodiment, where the filter D is again configured with a different gain at low frequencies than at high frequencies, if the gain this time is set anywhere from 1 to 0 at the low frequencies (for partial or full ANC), and to P.(Ge/Gr) at the higher frequencies such that the filter T becomes adapted to zero, then it may be possible here to have a tunable ANC effect or strength with no undesirable boost.
Considering the seven examples above for tuning the filter D, one realization of the filter D is as the combination of an FIR filter to introduce a time delay to improve the causality of filter T in Eq. 6, in cascade with a number of IIR filters to introduce the variations described in the examples 1) through 7) given above. Other realizations are possible.
In example 4 above, the filter T may be implemented as a single FIR filter that can provide variable ANC at low frequencies, and acoustic transparency at high frequencies, if the filter D is configured as a high-pass shelf filter with normalized gain. Note also that the ANC being provided in this case is feedforward ANC, which uses a reference signal that may be produced by a single exterior microphone (that is in the associated headphone.) Now, in the case of a sealed headphone or sealed in-ear ear bud, the wearer experiences her own speech with an undesirable “boominess”, that is caused by ear occlusion (due to the sealed headphone or in-ear earbud.) In accordance with another embodiment of the invention, the audio system of
Referring to
y=pe.Ge−y.X.SE (Eq. 8)
Then re-arranging Eq. 8 for y, gives
y=pe.Ge/(1+X.SE) (Eq. 9)
Then using the error signal at the output of differencing unit 24, the controller 27 will try to drive this:
pr.Gr.D−pe.Ge/(1+X.SE)=>0 (Eq. 10)
Assuming filter T is realizable, Eq. 10 can be rewritten as
pe/pr=D.(Gr/Ge).(1+X.SE) (Eq. 11)
Now, if the feedback ANC subsystem is disabled, e.g., filter X is set to zero, then Eq. 11 matches Eq. 7, as it should.
Recalling Eq. 3 and rewriting to include the addition of feedback ANC:
pe=pr.[P+Gr.T.S]+y.X.S (Eq. 12)
Substituting for y in Eq. 12 using Eq. 9 gives
pe=pr.[P+Gr.T.S]+X.S.pe.Ge/(1+X.SE) (Eq. 13)
which can be re-written as
pe/pr=[P+Gr.T.S]/[1−(Ge.X.S/(1+X.SE))] (Eq. 14)
If the feedback ANC subsystem is disabled, e.g., filter X is set to zero, then Eq. 14 matches Eq. 3, as expected. If the feedback ANC filter X is set equal to −1/S.Ge, then in Eq. 14 pe/pr will go to zero—which is the effect of ANC, as expected.
Setting Eq. 14 equal to Eq. 11, and re-arranging for T gives
T=(D.(1+X.SE−X.S.Ge)−P.(Ge/Gr))/S.Ge (Eq. 15)
When SE=S.Ge, which is feasible given broadband signals and a sufficient FIR filter length in the filter SE, then T simplifies to Eq. 5. So, the filter T here matches the filter T that is in the architecture without the feedback ANC filter X. This equivalence is due to the function of the digital differencing unit 23 and the subtracted SE-filtered feedback ANC (FB-ANC) signal (from the output of the filter X), which removes the feedback ANC effect from the error signal fed to the adaptive controller 27.
Turning now to
The audio system of
The audio system of
The controller 36 (e.g., an LMS engine that adapts the W filter 38) may be part of a conventional feed-forward ANC subsystem. As in Eq. 3, at the acoustic summing junction 20 (at the wearer's ear), Eq. 1 can be written as
pe=pr.[P+Gr.W.S] (Eq. 16)
Now, in accordance with an embodiment of the invention, the adaptive computation of the filter T (by the T filter controller 37) is configured around the signals created at the outputs of the digital differencing units 34, 24, 33 and the related filters D, F and H. The adaptive system driven by the T filter controller 37 will attempt to drive the output of the differencing unit 33 to zero. By studying the block diagram it can be deduced that
Fb.[pr.Gr.D.Hd−pe.Ge.He+pr.Gr.W.SE.He]−pr.Gr.SE.Fa.Hx.T=>0 (Eq. 17)
Assuming T and W are realizable, then this can be reordered as
Fb.[pr.Gr.D.Hd−pe.Ge.He+pr.Gr.W.SE.He]=pr.Gr.SE.Fa.Hx.T (Eq. 18)
Substituting for pe from Eq. 16 into Eq. 18:
Fb.[pr.Gr.D.Hd−pr.P.Ge.He−pr.Gr.W.S.Ge.He+pr.Gr.W.SE.He]=pr.Gr.SE.Fa.Hx.T (Eq. 19)
Dividing through by pr.Gr, and re-arranging for T gives:
T=(Fb/Fa).[D.Hd/Hx−P.(Ge/Gr).He/Hx−W(S.Ge−SE).He/Hx]/SE (Eq. 20)
If the filter SE can train to S.Ge (feasible if the FIR filter that implements the filter SE has enough taps and the playback signal is broadband and above the noise floor), then Eq. 20 is no longer a function of W, and T can be written as
T=(Fb/Fa).[D.Hd/Hx−P.(Ge/Gr).He/Hx]/SE (Eq. 21)
Eq. 21 shows that T is now a function of SE in the audio system of
Eq. 21 shows the filter pairs Fb/Fa, Hd/Hx and He/Hx now affect the shape of filter T. Using phase matched filters with independent frequency response, these filter pairs bring more flexibility to designing a desired filter T. If each pair is equal, then filter T simplifies to an equivalent formula of Eq. 6, and in that case
T=(D−P.(Ge/Gr))/SE (Eq. 22)
In a live system W will be replaced by T when partial or full transparency is needed, and Eq. 22 and Eq. 16 can be combined as
pe=pr.[P+Gr.S.(Fb/Fa).[D.Hd/Hx−P.(Ge/Gr).He/Hx]/SE] (Eq. 23)
Rearranging for P, and again assuming SE=S.Ge gives:
pe/pr=P[1−(Fb/Fa).(He/Hx)]+(Fb/Fa).(Hd/Hx).Gr.D/Ge (Eq. 24)
If each filter pair of F and H are equal then eq. (24) simplifies to the same as Eq. 7, again demonstrating equivalence of
pe/pr=D.Gr/Ge (Eq. 25)
The flexibility of transparency provided by
In both of the audio systems of
W=−P/Gr S. (Eq. 26)
Meanwhile, the adaptive filter SE controller 26 is acting to model the path S and the transducer block Ge, thus
SE=S.Ge (Eq. 27)
If we now convolve W with SE, the response will be
W.SE=−P.Ge/Gr (Eq. 28)
Looking at just low frequencies, such as below 100-200 Hz, the acoustic path P tends to unity gain, for a headphone that presents some passive attenuation of the ambient sound, e.g., a closed headphone, the Eq. 28 will simplify to —Ge/Gr. This computed estimate can then be used by either of the transparency systems in
While certain embodiments have been described and shown in the accompanying drawings, it is to be understood that such embodiments are merely illustrative of and not restrictive on the broad invention, and that the invention is not limited to the specific constructions and arrangements shown and described, since various other modifications may occur to those of ordinary skill in the art. For example, while the transparent hearing filters 6 should be as fast as possible in order to reduce latency, suggesting that dedicated, hardwired digital filter blocks should be used to implement them, a programmable microprocessor that is fast enough to perform all of the desired digital filter algorithms in parallel may alternatively be used. The description is thus to be regarded as illustrative instead of limiting.
This patent application is a continuation of U.S. patent application Ser. No. 15/273,396 filed on 22 Sep. 2016, which is incorporated herein by reference.
Number | Date | Country | |
---|---|---|---|
Parent | 15273396 | Sep 2016 | US |
Child | 16029314 | US |