This application is a 35 U.S.C. §371 national stage application of PCT International Application No. PCT/SE2008/051513, filed on 18 Dec. 2008, the disclosure of which is incorporated by reference herein as if set forth in its entirety. The above-referenced PCT International Application was published in the English language as International Publication No. WO 2010/071519 A1 on Jun. 24, 2010.
The present invention relates to systems and method for filtering a signal. For example, in one aspect, the invention relates to a time-domain finite impulse response (FIR) filter designed from a desired frequency response for the case when it is desired that the application of the FIR filter should add a certain amplification to the average power of the signal it is applied to. In some embodiments, the invention addresses the particular case when the filtering operations are performed in fix-point on a digital signal processor (DSP).
In many applications (see e.g., B Sauert, G Enzner, and P. Vary, “Near end listening enhancement with strict loudspeaker output power constraining”, International Workshop on Acoustic Echo and Noise Control, IWAENC 2006, September 12-14, Paris, France and U.S. Pat. No. 5,943,429) a desired filtering effect is achieved by filtering a signal via a linear filtering operation with a time-domain filter that has been designed to have a desired frequency response. Typically the desired effect varies with time and is designed based on the input signal. In some applications, the desired effect should not alter the average power level of the signal, and, hence, the desired frequency response should be designed so as to give a time-domain filter that does not alter the power level of the signal. Alternatively, a fixed gain of the average power level of the signal may be desired of the filtering operation.
On a DSP, filtering operations with FIR filters are usually performed using fix-point operations on signals represented by a limited number of bits: for audio signals typically 16 bits are used. In order to not lose precision in the output, it is desired that the filter coefficients in the FIR filter are properly scaled: if they are too large the filter output will saturate, and if they are too low some of the bits in the output will not be used and some precision will be lost. Therefore, in some applications it is better to incorporate all filtering effects in the filter coefficients, and not to perform the filtering as a combination of a filtering operation and a scaling operation.
J. Proakis and D. Manolakis, “Digital signal processing”, Prentice Hall and U.S. Pat. No. 7,251,271 disclose methods for designing a time-domain FIR filter with a desired frequency response. In practice, for reasons of computational complexity, the order of the time-domain FIR filter needs often to be short. Furthermore, for the same reason, the desired frequency response is usually only computed for a limited number of frequencies.
The limited length of the FIR filter limits the accuracy by which its frequency response is able to approximate the desired frequency response, particularly when the dynamics of the desired frequency response are large. This may cause the amplification of the power of the signal output from the filtering operation to be different from what is desired, as the FIR filter is not able to fulfill the requirements posed by the desired frequency response that would ensure that the desired amplification of the average power of the input signal would be achieved in the output.
Furthermore, the same thing may happen due to the limited frequency resolution of the desired filter frequency response of the filter, even for the case when the frequency response of the time-domain FIR filter is identical to the desired filter frequency response.
The deceivingly simple way of correcting the power of the output from the filtering operation by a simple scaling of the output to the desired level does not work well using a fixed-point implementation of the filtering operation due to possible truncation and loss of precision in the output prior to the scaling.
A possible solution is to perform the filtering, and then, based on the power of the output, properly scale the coefficients of the FIR filter and perform the filtering a second time to form the final filter output. But an FIR filtering operation is expensive. Thus, from a computational point of view, this solution is not desired.
Another possible solution would be to properly normalize the filter output during the filtering operation. On a DSP this is computationally complex. Moreover, because the FIR filtering operation is a computationally demanding task, it typically needs to be fully optimized. Thus, it is generally not feasible to solve the problem in this manner.
What is desired are systems and methods that reduce the problem of improper amplification of the average power of the output of the filtering operation.
In one aspect, the present invention provides systems and methods that reduce the problem of improper amplification of the average power of the output of the filtering operation by tracking the errors in the amplification of the output power and compensating for this.
In one aspect, the invention provides a method of filtering an input signal x(k) (using an FIR filter or other filter) to produce an output signal y(k) such that the ratio of a power level of the output signal (e.g., an average power level of the output signal) to a power level of the input signal (e.g., an average power level of the input signal) will be equal to or about equal to a desired value γ.
In some embodiments, this method includes the following steps: (1) forming a first corrected frequency response
The method, in some embodiments, further includes the steps of: (4) computing the amplification
The step of computing γdiff may comprise computing γ/
In some embodiments, the method also includes computing a first corrected output signal block comprising M samples by computing
In some embodiments, the step of computing the amplification
In some embodiments, gcorr(1)=(1−λ)gcorr(0)+λγdiffgcorr(0), wherein λ is a forgetting factor.
In some embodiments, the input signal represents a sound wave, wherein the sound wave was caused, at least in part, by a person speaking. In this embodiment, the method may further include providing the output signal y(k), or a signal obtained from y(k), to a transducer configured to convert the provided signal into sound.
In some embodiments, the method is performed by a communication device that receives from another communication device the input signal x(k) or a signal from which x(k) was obtained.
In another aspect, the present invention provides an apparatus for filtering an input signal x(k) to produce an output signal y(k) such that the ratio of a power level of the output signal to a power level of the input signal will be equal to or about equal to a desired value γ. In some embodiments, the apparatus includes: (1) a corrected frequency response foaming module configured to form a first corrected frequency response
The above and other aspects and embodiments are described below with reference to the accompanying drawings.
The accompanying drawings, which are incorporated herein and form part of the specification, illustrate various embodiments of the present invention and, together with the description, further serve to explain the principles of the invention and to enable a person skilled in the pertinent art to make and use the invention. In the drawings, like reference numbers indicate identical or functionally similar elements.
An objective of the invention is to filter an input signal x(t) with an FIR filter h(z), which is designed to have a desired frequency response H(ω), to produce an output signal y(t), such that, among other things, a ratio of a power level of the output signal P(y(t)) to a power level of the input signal P(x(t)) equals or substantially equals a desired ratio γ. Mathematically, y(t)=h(z)x(t) and P(y(t))=γ P(x(t)). In some embodiments
H(ω) may be specified for a discrete number L of frequencies ω, (128 and 256 being typical values for L) and is designed to have a desired effect on x(t). In particular, it is desired that the effect of the filtering operation on the average power of y(t) should be a desired specific level of amplification. Typically, the design of H(ω) is made based on x(t) and its spectral properties, making it theoretically feasible to always achieve the desired amplification of the power of the input signal.
The typical way of forming the coefficients hk of h(z) is using the inverse Fourier transform (IFFT)
N≦L is the length of h(z), and wk are the coefficients of a truncation weighting filter. One particular choice of such a filter is shown in
The averaging of the power levels is typically done over a time window. Thus it is desired that
where γ is the desired amplification of x(t) (γ=1 means that the average power level of the signal is unaltered).
In general, due to the problems of limited length of h(z) and limited resolution of H(ω), the actual amplification
For the sake of simplicity and convenience, the invention will be described for the case where the averaging is performed over M samples, H(ω) will be computed for these M samples, and h(z) will be applied to these M samples. However, the invention is not limited to this particular arrangement.
The embodiment described here reduces the amount by which
Referring now to
Next (step 304) a corrected frequency response
Next (step 306) a corrected filter
Next (step 308) the corrected filter
Next (step 310), the resulting amplification
Next (step 312), the amplification error γdiff is computed. In some embodiments,
Next (step 314), a new correction gain gcorr(n) is computed. In some embodiments, gcorr(n)=(1−λ)gcorr(n−1)+λγdiffgcorr(n−1). The value λ is a predetermined forgetting factor for adapting gcorr(n). After step 314, procedure 300 proceeds to step 316 and may also proceed to step 318. In step 316, n is incremented by 1. After step 316, procedure 300 returns to step 304.
In step 318, a corrected output signal block
Referring now to
As further shown in
Device 400 may also include a computing module 412. Computing module 412 is configured to compute a corrected output signal block based on the output signal block and the current and a previous correction value. Further, device 400 may include (i) a digital/analog converter 414 for converting the output signal block (or corrected output signal block) to an analog output signal and (ii) an output module 416 (e.g., an amplifier) that receives the analog output signal and provides the signal (or an amplified version of the signal) to a transducer 490 (e.g., a loudspeaker 490) for converting the analog output signal to speech.
Device 400 may further include a receiver 408 for receiving from another communication device an input signal x(k) or a signal from which x(k) can be obtained.
The above described modules of apparatus 400 may be implemented in hardware (e.g., using one or more processors including, for example, a microprocessor and/or a digital signal processor) and/or software.
In some embodiments, the apparatus shown in
Referring now to
Software 508 is configured such that when processing system 502 executes software 508, communication device 500 performs steps described above with reference to the flow chart shown in
While various embodiments of the present invention have been described above, it should be understood that they have been presented by way of example only, and not limitation. Thus, the breadth and scope of the present invention should not be limited by any of the above-described exemplary embodiments.
Additionally, while the processes described above and illustrated in the drawings are shown as a sequence of steps, this was done solely for the sake of illustration. Accordingly, it is contemplated that some steps may be added, some steps may be omitted, the order of the steps may be re-arranged, and some steps may be performed in parallel.
| Filing Document | Filing Date | Country | Kind | 371c Date |
|---|---|---|---|---|
| PCT/SE2008/051513 | 12/18/2008 | WO | 00 | 6/16/2011 |
| Publishing Document | Publishing Date | Country | Kind |
|---|---|---|---|
| WO2010/071519 | 6/24/2010 | WO | A |
| Number | Name | Date | Kind |
|---|---|---|---|
| 5506798 | Shimada et al. | Apr 1996 | A |
| 5648987 | Yang et al. | Jul 1997 | A |
| 5943429 | Handel | Aug 1999 | A |
| 7113559 | Baas et al. | Sep 2006 | B2 |
| 20010056450 | Kiriaki | Dec 2001 | A1 |
| 20040145502 | Thomson | Jul 2004 | A1 |
| 20050117489 | Serizawa | Jun 2005 | A1 |
| Number | Date | Country |
|---|---|---|
| 1 947 776 | Jul 2008 | EP |
| 64-061178 | Mar 1989 | JP |
| 03-058510 | Mar 1991 | JP |
| 2008-011376 | Jan 2008 | JP |
| Entry |
|---|
| International Preliminary Report on Patentability issued for International Application No. PCT/SE2008/051513 on Mar. 10, 2011. |
| Written Opinion of the International Searching Authority issued for International Application No. PCT/SE2008/051513 on Sep. 23, 2009. |
| Bastian et al. “Near End Listening Enhancement with Strict Loudspeaker Output Power Constraining” ICASSP 2006, IEEE International Conference on Acoustics, Speech and Signal Processing, May 14-19, 2006, pp. I-493-I-496. |
| Number | Date | Country | |
|---|---|---|---|
| 20110252077 A1 | Oct 2011 | US |