Method and device for filtering during a change in an ARMA filter

Abstract

A method and device are provided for filtering digital audio signals using at least one ARMA filter, particularly during a filter change. The method includes the following steps: a step of receiving a first request to change filtering to or from filtering by a first ARMA filter; and, in response to the first request, a step of gradually switching, at each of a plurality of cascaded first filtering blocks, between digital-signal filtering by a first basic filtering cell and digital-signal filtering by another associated basic filtering cell, the first basic filtering cells of the plurality of first filtering blocks factorizing the first filter.

Claims

1. A process comprising: receiving a digital signal by a filtering device comprising a plurality of first cascaded filtering blocks, each filtering block comprising a first elementary filtering cell and another associated elementary filtering cell, all of said first elementary filtering cells of the plurality of first filtering blocks being configured to factorise a first autoregressive moving average (ARMA) filter; filtering the digital signal by the filtering device to produce a filtered digital signal, wherein filtering comprises: receiving by a processor of the filtering device a first request for a filtering change to or from filtering by the first ARMA filter by the filtering device, controlling at least one control element to progressively switch the filtering device between digital signal filtering by the first elementary filtering cell and digital signal filtering by said other associated elementary filtering cell at a level of each of a plurality of first cascaded filtering blocks in response to receiving the first request.

2. The process as claimed in claim 1, wherein the progressive switch at the level of a first filtering block comprises the combination of a filtered signal by the first elementary filtering cell with a filtered signal by the other associated elementary filtering cell to produce a mixed output signal of a filtering block during the progressive switch.

3. The process as claimed in claim 2, wherein said progressive switch uses fades whereof fade coefficients corresponding are applied to the signals filtered by the first elementary filtering cells and by the other elementary filtering cells.

4. The process as claimed in claim 2, wherein within each first filtering block, said other elementary filtering cell is placed in parallel with said first elementary filtering cell to which a switching coefficient is applied, and said other elementary filtering cell uses an identity filter weighted by a complementary switching coefficient.

5. The process as claimed in claim 4, wherein said plurality of first filtering blocks is placed in series with a plurality of cascaded second filtering blocks, each comprising a second elementary filtering cell originating from factorisation of a second ARMA filter placed in parallel with a cell identity filter and weighted by a switching coefficient, all of said second elementary filtering cells of the plurality of second filtering blocks factorising the second ARMA filter, and at the level of the second filtering blocks the process comprises progressive switching of filtering by the second reverse elementary cells of the progressive filtering switching by the first elementary cells at the level of the first filtering blocks to control progressive switching between filtering by the first ARMA filter and filtering by the second ARMA filter.

6. The process as claimed in claim 2, wherein within said first filtering blocks said other elementary filtering cell is placed in parallel with said first elementary filtering cell and comprises a second elementary filtering cell originating from factorisation of a second ARMA filter.

7. The process as claimed in claim 6, wherein if one of the factorisations of the first and second ARMA filters comprises more elementary filtering cells than the other, a cell filter of identity filter type is associated with each supernumerary elementary filtering cell to form said filtering block.

8. The process as claimed in claim 1, comprising, in response to a second request for filtering reverse change of the first request and received during the progressive switch, an inversion step of said filtering switching from the switching state corresponding to the instant of reception of said second request for change.

9. A filtering device for a digital signal, comprising: a plurality of first cascaded filtering blocks, which receive the digital signal and produce a filtered digital signal, each filtering block comprising a first elementary filtering cell and another associated elementary filtering cell, all of said first elementary filtering cells of the plurality of first filtering blocks being configured to factorise a first autoregressive moving average (ARMA) filter; at least one control element configured to control, at the level of each first filtering block, a filtering change to or from filtering by the first ARMA filter; a processor, which is configured to: receive a request for a filtering change to or from filtering by the first ARMA filter; control the at least one control element to progressively switch the filtering device between digital signal filtering by the first elementary filtering cell and digital signal filtering by said other associated elementary filtering cell at a level of each of the plurality of first cascaded filtering blocks, in response to receiving the request.

10. The device as claimed in claim 9, wherein each said first filtering block comprises an adder for combining a signal filtered by the first elementary filtering cell with a signal filtered by said other associated elementary filtering cell to generate a mixed output signal of a filtering block during progressive switching.

11. The device as claimed in claim 10, wherein the at least one control element comprises progressive attenuation means of the filtered signal by the first elementary filtering cell and of the filtered signal by said associated elementary filtering cell before combination, according to two complementary fade coefficients respectively, one closing fade and the other opening fade.

12. The device as claimed in claim 9, wherein within each first filtering block said other elementary filtering cell is placed in parallel with said first elementary filtering cell to which a switching coefficient is applied and said other elementary filtering cell uses an identity filter weighted by a complementary switching coefficient.

13. The device as claimed in claim 12, wherein said plurality of first filtering blocks is placed in series of a plurality of cascaded second filtering blocks, each comprising a second elementary filtering cell originating from factorisation of a second ARMA filter placed in parallel with a cell identity filter and weighted by a switching coefficient, all of said second elementary filtering cells of the plurality of second filtering blocks factorising the second ARMA filter, and at the level of the second filtering blocks, the device comprises control elements of progressive filtering switching by the second reverse elementary cells of the progressive filtering switching by the first elementary cells at the level of the first filtering blocks to control progressive switching between filtering by the first ARMA filter and filtering by the second ARMA filter.

14. The device as claimed in claim 9, wherein within said first filtering blocks, said other elementary filtering cell is placed in parallel with said first elementary filtering cell and comprises a second elementary filtering cell originating from factorisation of a second ARMA filter.

15. A non-transitory memory device comprising a computer program product stored thereon and readable by a microprocessor of a filtering device, the computer program product comprising instructions that configure the microprocessor to perform a process for filtering a digital signal when this program is loaded and executed by the microprocessor, wherein the process comprises: receiving a digital signal by a filtering device comprising a plurality of first cascaded filtering blocks, each filtering block comprising a first elementary filtering cell and another associated elementary filtering cell, all of said first elementary filtering cells of the plurality of first filtering blocks being configured to factorise a first autoregressive moving average (ARMA) filter; filtering the digital signal by the filtering device to produce a filtered digital signal, wherein filtering comprises: the microprocessor receiving a first request for a filtering change to or from filtering by first ARMA filter by the filtering device, at least one control element progressively switching the filtering device between digital signal filtering by the first elementary filtering cell and digital signal filtering by said other associated elementary filtering cell at a level of each of a plurality of first cascaded filtering blocks, in response to receiving the request.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Other particular features and advantages will emerge from the following description, illustrated by the attached diagrams, in which:

(2) FIG. 1 schematically illustrates a first embodiment of the invention;

(3) FIGS. 2, 2a and 2b illustrate operating examples of fade out carried out in the progressive change of IIR filters according to the invention;

(4) FIG. 3 schematically illustrates a second embodiment of the invention;

(5) FIG. 4 schematically illustrates another embodiment of the filtering device according to the invention; and

(6) FIG. 5 shows a particular material configuration of a device or system capable of executing the process according to the invention.

DETAILED DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS

(7) Examples for the present illustration are two digital linear filters with infinite impulse response or IIR filters or recursive filters, designated h.sub.1 and h.sub.2, both having the same even order, designated L, for their respective parts MA and AR.

(8) The invention naturally also applies to other cases of ARMA and IIR filters.

(9) For example, it applies to non-linear filters (especially Volterra filters), to FIR filters, and also in the event where both IIR filters h.sub.1 and h.sub.2 have non-even and/or different orders, or in the event where for the same IIR filter h.sub.1 or h.sub.2 the parts MA and AR also have non-even and/or different orders. The expert will be able to adjust the following explanations to these different cases by obtaining for example a cell of the first order as a complement to cells of the second order, or by obtaining elementary filtering cells whereof the denominator or the numerator is a constant while the associated numerator or denominator is a polynomial of a degree at least equal to 1. In particular, the above publication The Volterra and Wiener theories of nonlinear Systems provides indications on obtaining factorisation of a non-linear system of elementary cells.

(10) As is known per se, linear IIR filters h.sub.i, (i=1 or 2) are defined by the z-transform

(11) $H_i\left(z\right)=\frac{B_i\left(z\right)}{A_i\left(z\right)}=G.Math.\frac{\underset{l=0}{\overset{\frac{L}{2}-1}{.Math.}}\left(1-{}_{i,l}.Math.z^{-1}\right)\left(1-\stackrel{\_}{{}_{i,l}}.Math.z^{-1}\right)}{\underset{l=0}{\overset{\frac{L}{2}-1}{.Math.}}\left(1-{}_{i,l}.Math.z^{-1}\right)\left(1-\stackrel{\_}{{}_{i,l}}.Math.z^{-1}\right)},$

(12) where H.sub.i(z) is the z-transform of the filter h.sub.i, G is the gain of the filter H.sub.i, and .sub.i,l and .sub.i,l are complex coefficients.

(13) The transfer function H.sub.i(z) can also be factorised in cells of the second order as follows:

(14) $\begin{array}{c}{H}_i\left(z\right)=\frac{B_i\left(z\right)}{A_i\left(z\right)}\\ =\frac{\underset{l=0}{\overset{\frac{L}{2}-1}{.Math.}}\left(b_{i,l,0}+b_{i,l,1}.Math.z^{-1}+b_{i,l,2}.Math.z^{-2}\right)}{\underset{l=0}{\overset{\frac{L}{2}-1}{.Math.}}\left(1-a_{i,l,1}.Math.z^{-1}-a_{i,l,2}.Math.z^{-2}\right)}\\ =\underset{l=0}{\overset{\frac{L}{2}-1}{.Math.}}{h}_i^l\left(z\right),\end{array}$

(15) where a.sub.i,l and b.sub.i,l are real coefficients.

(16) These formulae show how an IIR filter h.sub.i can be factorised in cells of the first order and/or in cells h.sub.i.sup.l (l=0 . . . L/21) of the second order, that is, its transfer function is broken down into a product of polynomial cells of the second degree. The first IIR filter h.sub.1 is factorised in a plurality of a first elementary filtering cells h.sub.1.sup.l and the second IIR filter h.sub.2 is factorised in a plurality of second elementary filtering cells h.sup.l.sub.2.

(17) It is evident here that the above generalisation on any IIR filter of any dimensions is immediate, given that any filter can be factorised as a ratio of products of cells of the first and of the second order with real coefficients (lattice breakdown, into cells of the second order).

(18) According to the invention, any other breakdown with cells of greater orders (than the second order) can also be used, if necessary. However, breakdown into cells of the real first and of the second order will be preferred, since the less complex corresponding calculations contribute to better optimisation of the invention.

(19) Note x(n) the digital audio signal to be filtered (X(z) its z-transform) and y(n) the resulting filtered signal (Y(z) its transform), n being the sample index.

(20) The aim of the present invention is managing the filtering during an IIR filter change: for example, activation of an IIR filter in the previous absence of filtering; switching off an IIR filter for the sake of absence of filtering; shifting from one IIR filter to a new IIR filter (as described hereinbelow) or to a modified version of the same current filter.

(21) Such a change can be requested by an operator via a user interface, or by another system (for example noise reduction software). In these various instances, a command for IIR filter change is received.

(22) Reference is made to the example of deactivation of an IIR filter h.sub.2, current for activation of a second filter. In other terms, this is the filter change h.sub.1 in h.sub.2 at time T:

(23) $y\left(n\right)=\{\begin{array}{cc}x\left(n\right)*h_1,& n<T\\ x\left(n\right)*h_2,& nT\end{array}.$

(24) The other examples derive smoothly from this example: for example, taking identity filters (H(z)=z) for h.sub.1 or h.sub.2 to respectively illustrate simple activation of h.sub.2 or simple switching off of h.sub.1.

(25) FIG. 1 schematically illustrates a first embodiment of the invention for managing this switching at the time T from the filter h.sub.1 to the filter h.sub.2. The elementary filtering cells shown in this figure, and the associated control elements can easily be implemented via software and/or material elements.

(26) This first example corresponds to a filtering device 1 implementing in series the filters h.sub.1 and h.sub.2. As emerges from the figure, each filter h.sub.i constitutes a block which is easy to handle to be connected in series to another block representing another IIR filter. Therefore, it is easy to connect in series or substitute any filter to another filter, for the purpose of executing the present invention.

(27) In this configuration, each filter h.sub.i is constituted by the cascading of elementary filtering h.sub.i.sup.l cells which break it down. l is an index of elementary filtering cells depending on completed breakdown. For example, for breakdown into elementary cells of the first order only, L cells are obtained and the index l varies from 0 to L1. Similarly, for breakdown into elementary cells of the second order only, L2 cells are obtained and the index l varies from 0 to L/21.

(28) Associated with each elementary filtering cell h.sup.1.sub.1 at output is a control element 10.sub.i.sup.l, for example a potentiometer module for applying an attenuation factor or fade coefficient to the signal filtered by h.sub.1.sup.l. The whole of these two elements is called cell filter CF.sup.l.sub.i. It should be noted that in software usage, this control element can consist of a fade coefficient applied directly to the signal leaving the corresponding elementary filtering cell.

(29) As emerges from the following explanations, a cell filter CF.sup.l.sub.i comprises an elementary filtering cell and an associated control element for filtering an input signal. According to different configurations, the control element can be placed after the elementary cell or before the elementary cell.

(30) In the case of placement after the elementary cell (in FIGS. 1 and 3 for example), this control element cannot be simple instantaneous switching of interrupter (or switch) type but offers progressive transition (fade type, especially).

(31) In the case of placement before the elementary cell (as in FIG. 4, for example), this control element can prove to be a simple interrupter. In fact, the impulse response time of the elementary cells (these cells have a memory) ensures non-zero contribution of these disconnected cells at the level of corresponding adders, and therefore progressive switching according to the invention at the level of each filtering block. With a control element of progressive type (fade coefficient for example), this contribution due to memory is also modified by the fade coefficient applied by the control element.

(32) In the example of FIG. 1, the control element uses a fade in/fade out function during switching between the two filters h.sub.1 and h.sub.2.

(33) The control elements 10.sup.l.sub.1 controlling the first elementary filtering cells h.sub.1.sup.l are preferably identical, simultaneously applying the same closing fade coefficient C.sub.f during switching to the IIR filter h.sub.2 (for example progressively from 100% to 0%).

(34) The control elements 10.sup.l.sub.2 controlling the second elementary filtering cells h.sup.l.sub.2 are preferably identical, simultaneously applying the same opening fade coefficient C.sub.o during the same switching (for example progressively from 0% to 100%). In particular, the opening fade coefficient C.sub.o can be complementary to the closing fade coefficient C.sub.f, that is, for example C.sub.o+C.sub.f=100%. It should be noted that the ratio C.sub.o/C.sub.f is designated a mixing coefficient.

(35) Also, each cell filter CF.sup.l.sub.i (that is, an elementary filtering cell and its associated control element 10.sup.l.sub.i is placed in parallel with a cell identity filter CF.sup.l.sub.PTi composed by an elementary filtering cell of identity PTi type (that is, letting everything pass through, represented symbolically by a dotted line in the figure) and an associated control element 10.sup.l.sub.PTi of the same type as the control elements 10.sup.l.sub.i. This placing in parallel consists of applying the same input signal to the two elementary cells and acquiring an output signal adding up the respective output signals of each cell filter CF.sup.l.sub.i and CF.sup.l.sub.PTi.

(36) The two cell filters CF.sup.l.sub.i, CF.sup.l.sub.PTi, therefore receive the same input signal {tilde over (s)}.sub.i.sup.l=1(n) and respectively generate a filtered output signal to an adder ADD (or mixer). The latter adds the two acquired filtered signals to generate a filtered output block signal. During progressive switching, this output block signal mixes the two filtered signals, the intensity of the contribution of each elementary filtering cell h.sup.l.sub.i, PTi being a function of the fade coefficient applied by the associated control element 10.sup.l.sub.i, 10.sup.l.sub.PTi.

(37) The filtering block B.sup.l.sub.i is called the unit constituted by the two filter cells CF.sup.l.sub.i, CF.sup.l and by the adder ADD. A filtering block therefore receives an input signal to be filtered and generates a filtered output signal which is the input signal filtered by one of the two elementary filtering cells H.sup.l.sub.i or PTi in a period of stabilised operation (permanent state) or which is, during progressive switching (transitory state), a combination (mixing) of the input signal filtered respectively by each of the two elementary filtering cells. According to the positioning of the control element in the filter cells, the degree of contribution of the filtering from each of the elementary cells is a function of the applied switching coefficients and/or of the memory (impulse response) of these elementary cells.

(38) In detail, the control element 10.sup.l.sub.PTi is complementary to the control element 10.sup.l.sub.i: the control elements 10.sup.l.sub.PT1 implement the opening fade coefficient C.sub.o, while the control elements 10.sup.l.sub.PT2 implement the closing fade coefficient C.sub.f. In other terms, each branch PT1 behaves relative to the fade coefficient as a cell filter CF.sup.l.sub.2 of the second IIR filter (vice and versa for PT2 and the filter cells CF.sup.l.sub.1.

(39) In the case of breakdown of the IIR filters into elementary cells of the second order, there is s.sub.i.sup.l(n) with

(40) $\left(i\left\{1;2\right\},l\left\{0,\mathrm{.Math.},\frac{L}{2}-1\right\}\right)$
the output (filtered signal) of each elementary filtering cell h.sup.l.sub.1.sup.l and s.sub.i.sup.l(n) the output of the filtering block B.sup.l.sub.i that is, the result of output mixing of the signal passing through the cell filter CF.sup.l.sub.i with the signal passing through the cell filter CF.sup.l.sub.PTi, (identity) in parallel.

(41) These outputs are shown as follows for the filter h.sub.1:

(42) $s_1^l\left(n\right)=\underset{k=0}{\overset{2}{.Math.}}{b}_{1,l,k}.Math.{\tilde{s}}_1^{l-1}\left(n-k\right)+\underset{k=1}{\overset{2}{.Math.}}{a}_{1,l,k}.Math.s_1^l\left(n-k\right)$${\tilde{s}}_1^l\left(n\right)=\{\begin{array}{cc}{s}_1^l\left(n\right)={\tilde{s}}_1^{l-1}\left(n\right)*h_1^l& n<T\\ C_f\left(n-T\right).Math.s_1^l+\left(1-C_f\left(n-T\right)\right).Math.{\tilde{s}}_1^{l-1}\left(n\right)& Tn<T+N\\ {\tilde{s}}_1^{l-1}\left(n\right)& nT+N\end{array}$

(43) with the convention {tilde over (s)}.sub.1.sup.1(n) (input signal) and where the fade coefficient C.sub.f(n) is a function of fade out generally decreasing and monotone (examples shown in FIG. 2) and C.sub.o=1C.sub.f in this example.

(44) N designates the switching duration between the two IIR filters. It can be determined empirically or by experimentation and is selected less than the initialisation durations of the techniques of the prior art. N can be selected as equal to a LONG value such as defined previously but calculated for one of said elementary filtering cells, for example the biggest length covering 80% of the energy of an impulse response (infinite) among all the elementary filtering cells.

(45) Correspondingly, the outputs relating to the filter h.sub.2 are shown as follows, for C.sub.o=1C.sub.f:

(46) $s_2^l\left(n\right)=\underset{k=0}{\overset{2}{.Math.}}{b}_{2,l,k}.Math.{\tilde{s}}_2^{l-1}\left(n-k\right)+\underset{k=1}{\overset{2}{.Math.}}{a}_{2,l,k}.Math.s_2^l\left(n-k\right)$${\tilde{s}}_2^l\left(n\right)=\{\begin{array}{cc}{\tilde{s}}_2^{l-1}\left(n\right)& n<T\\ \left(1-C_f\left(n-T\right)\right).Math.s_2^l+C_f\left(n-T\right).Math.{\tilde{s}}_2^{l-1}\left(n\right)& Tn<T+N\\ {\tilde{s}}_2^l\left(n\right)={\tilde{s}}_2^{l-1}\left(n\right)*h_2^l& nT+N\end{array}$

(47) with the input signal convention

(48) ${\tilde{s}}_2^{-1}\left(n\right)={\tilde{s}}_1^{\frac{L}{2}-1}\left(n\right).$

(49) It should be noted that if in this example the same fade coefficient C.sub.f (via C.sub.o=1C.sub.f) is applied to h.sub.2 as for the filter h.sub.1, the latter can be replaced by a different fade coefficient C.sub.f corresponding to a function similar to C.sub.f but the temporal support and the form whereof can be different.

(50) Also, the invention also applies if the second IIR filter h.sub.2 is placed in front of the first filter h.sub.1 (contrary to the figure).

(51) In steady state before the instant T, the input signal x(n) is filtered only by the IIR filter h.sub.1:
y(n)={tilde over (s)}.sub.s.sup.L/21(n)={tilde over (s)}.sub.1.sup.L/21(n)={tilde over (s)}.sub.1.sup.L/22(n)*h.sub.1.sup.L/21={tilde over (s)}.sub.1.sup.1(n)*h.sub.1.sup.9* . . . * h.sub.1.sup.L/21=x(n)*h.sub.1.

(52) This corresponds to a configuration of the device 1 wherein the coefficient C.sub.f=100% and the coefficient C.sub.o=0%. In fact, the signal x(n) is successively filtered by each of the first elementary filtering cells h.sup.l.sub.i then passes through the second part of the device via the identity branches PT2. The contributions of the branches PT1 and h.sup.l.sub.2 are not considered due to the coefficient C.sub.o=0%.

(53) At the instant T, switching between the two IIR filters h.sub.1 and h.sub.2 begins for a period N.

(54) In the present embodiment of the invention, the principle is to progressively switch off the first IIR filter h.sub.1 by progressively reducing the coefficient C.sub.f (and by increasing C.sub.o complementarily) so that the input signal passes progressively through the identity branches PT1. This progressive reduction is conducted via fade out.

(55) Similarly, the second IIR filter h.sub.2 is progressively lit by increasing C.sub.o complementarily to C.sub.f (this is for example fade in), progressively diminishing the signal passing through the branches PT2.

(56) In other terms, during the N switching samples the digital signal is filtered by the two IIR filters h.sub.1 and h.sub.2, their respective contributions varying according to C.sub.f.

(57) For each pair (H.sup.l.sub.1, H.sup.l.sub.2), l, switching at the level of the elementary cells consists of progressively decreasing the contribution of the elementary filtering cell coming from the first IIR filter h.sub.1 for the sake of contribution of the other elementary cell of the pair (coming from the IIR filter h.sub.2).

(58) In each filtering block B.sup.l.sub.i this switching consists of progressively decreasing the contribution of the first elementary filtering cells h.sup.l.sub.1 (respectively cell identity filters CF.sup.l.sub.PT2) to the benefit of the contribution of associated cell identity filters CF.sup.l.sub.PT1 (respectively the second elementary filtering cells h.sup.l.sub.2).

(59) According to the invention, this switching control is undertaken at the level of each filtering block (and of each elementary filtering cell) by means of appropriate means controlling the coefficients C.sub.o and C.sub.f applied by the control elements 10.sup.l.sub.i and 10.sup.l.sub.PTi. Therefore during these N samples, as a filtered output signal for each filtering block B.sup.l.sub.i, for example the barycentre of the output signal of the preceding block (via the connection PTi) is calculated with the result of this signal filtered by the elementary filtering cell h.sup.l.sub.i respectively affected by a fade coefficient (C.sub.f for the elementary cells of h.sub.1, and C.sub.o=f(C.sub.f) for example C.sub.o=1C.sub.f for the elementary cells of h.sub.2) and of the complementary coefficient. These coefficients favour signals filtered by h.sub.1, at the start of switching (C.sub.f close to 1), and favour signals filtered by h.sub.2 on completion of switching (C.sub.f close to 0).

(60) On completion of switching (nT+N), the digital signal x(n) is filtered by the IIR filter h.sub.2 only. In fact, C.sub.f=0 and therefore:

(61) $\begin{array}{c}y\left(n\right)={\tilde{s}}_2^{L/2-1}\left(n\right)\\ =s_2^{L/2-1}\left(n\right)\\ ={\tilde{s}}_2^{L/2-2}\left(n\right)*h_2^{L/2-1}\\ ={\tilde{s}}_2^{-1}\left(n\right)*h_2^0*\mathrm{.Math.}*h_2^{L/2-1}\\ ={\tilde{s}}_1^{L/2-1}\left(n\right)*h_2\\ ={\tilde{s}}_1^{-1}\left(n\right)*h_2\\ =x\left(n\right)*h_2\end{array}$

(62) The first IIR filter h.sub.1 is no longer being used. It can be deactivated such that only the second IIR filter h.sub.2 is retained to continue processing. Fresh switching of the latter to another filter can be done by applying the ideas of the invention again with the breakdown of this other filter in series (in this case the coefficient C.sub.o applied to the elementary cells h.sup.l.sub.2 becomes a closing fade coefficient C.sub.f).

(63) It is not rare, during switching from one IIR filter to another, for a new request for IIR filter change to arrive while switching is incomplete. The new request arrives for example at the time T.sub.1 with T<T.sub.1<T+N.

(64) This can be a request for switching back to the first IIR filter h.sub.1.

(65) In this case, it can be provided that in response the filtering switching is reversed from the switching state at the time when this new request is received.

(66) For example, if at this instant T.sub.1 the coefficients C.sub.f(T.sub.1) are applied to the filter h.sub.1 and C.sub.o(T.sub.1) to the filter h.sub.2, reversing switching consists of applying a new closing fade coefficient C.sub.f to the filter h.sub.2 and a new opening fade coefficient C.sub.o to the filter h.sub.1, substantially taking the form of FIG. 2 (over a period N which can be equal or different to N), but with values of equal origin, respectively, at C.sub.o(T.sub.1) and C.sub.f(T.sub.1) (FIG. 2a).

(67) Reprising these values at origin ensures the absence of discontinuity in the filtered signal, and therefore of audible artefacts.

(68) In a variant reducing return time to the filtering state by h.sub.1 the profile of the applied fade coefficients can be retraced in the reverse direction. In the case for example of a symmetrical profile, this can correspond to being placed instantaneously at the time T.sub.1 and to applying the end of the coefficient profile of FIG. 2 and reversing C.sub.f (applied to 10.sup.l.sub.1 and 10.sup.l.sub.PT2) and C.sub.o (applied to 10.sup.l.sub.1 and 10.sup.l.sub.PT2), with C.sub.f(T.sub.1)=C.sub.o(T.sub.1) and C.sub.o(T.sub.1)=C.sub.f(T.sub.1) to avoid any discontinuity (FIG. 2b). In particular with C.sub.o=1C.sub.f, T.sub.1=N=T.sub.1T), and the transition duration for returning to filtering by h.sub.1 is N=T.sub.1T.

(69) It can also be a request for selecting a new (third) IIR filter h.sub.3. Breakdown of the latter similar to those for the filters h.sub.1 and h.sub.2 is carried out to acquire third elementary filtering cells

(70) 0$h_3^l\text{:}{H}_3\left(z\right)=\underset{l=0}{\overset{\frac{L}{2}-1}{.Math.}}{h}_3^l\left(z\right).$

(71) In this case, it can be provided in response to the new request that fresh filtering switching is carried out from the switching state at the time when this new request is received. For this, the mixing state of the instant T.sub.1 is fixed (that is, the coefficients C.sub.f(T.sub.1) and C.sub.o(T.sub.1) and all of the first and second IIR filters h.sub.1 and h.sub.2 are considered as a single filter (designated h.sub.12) having this fixed state, in series of which the new filter h.sub.3 is placed (for example within filtering blocks placed in series of the filtering blocks comprising h.sub.12).

(72) The preceding ideas are applied to the filters h.sub.12 (starting filter) and h.sub.3 (end filter), especially the presence of associated cell identity filters in parallel of each elementary cell of h.sub.12 and h.sub.3, as well as new closing fade coefficients C.sub.f (for h.sub.12) and opening C.sub.o (equal to 1C.sub.f, for h.sub.3).

(73) It should be noted that to simplify execution, the coefficient C.sub.o is applied to each elementary filtering cell h.sup.l.sub.3 (via control elements 10.sub.3.sup.l) and the coefficient C.sub.f at fixed values C.sub.f(T.sub.1) and C.sub.o(T.sub.1) is applied to the control elements 10.sup.l.sub.i, already existing for filters h.sub.1 and h.sub.2.

(74) FIG. 3 schematically illustrates a second embodiment of the invention for managing this switching at the time T of the filter h.sub.1 to the filter h.sub.2. The elementary filtering cells shown in this figure, as well as the associated control elements, can easily be utilised via software and/or material elements.

(75) This second example corresponds to a filtering device 1 wherein the filters h.sub.1 and h.sub.2 are directly instantiated in parallel. Each elementary filtering cell h.sup.l.sub.1 of the first IIR filter h.sub.1 is associated with an elementary filtering cell h.sub.2 of the second IIR filter h.sub.2 within filtering blocks B.sup.l. In this way using the elementary identity cells PTi is avoided.

(76) Each pair (h.sup.l.sub.1, h.sup.l.sub.2) accompanied by the respective control elements has the same properties as the pair of filtering blocks (B.sup.l.sub.1, B.sup.l.sub.2) of FIG. 1. Accordingly, switching between the two IIR filters is done in the same way.

(77) In this configuration, L/2 pairs corresponding to L/2 filtering blocks B.sup.l are cascaded.

(78) The output {tilde over (s)}.sup.l (n) of a filtering block B.sup.l is shown as follows, for C.sub.o=1=C.sub.f:

(79) ${\tilde{s}}^l\left(n\right)=\{\begin{array}{cc}{s}_1^l\left(n\right)& n<T\\ C_f\left(n-T\right).Math.s_1^l+\left(1-C_f\left(n-T\right)\right).Math.s_2^l\left(n\right)& Tn<T+N\\ s_1^l\left(n\right)& nT+N\end{array}\mathrm{with}{s}_i^l\left(n\right)=\underset{k=0}{\overset{2}{.Math.}}{b}_{i,l,k}.Math.{\tilde{s}}^{l-1}\left(n-k\right)+\underset{k=1}{\overset{2}{.Math.}}{a}_{i,l,k}.Math.s^1\left(n-k\right)$
at the level of each elementary filtering cell h.sub.i.sup.l and the convention {tilde over (s)}.sup.1(n)=x(n).

(80) The ideas provided previously in connection with the first example are also applicable to this second example of a filtering device 1. This is the case in particular for managing a new request for IIR filter change, for the form of functions C.sub.f and C.sub.o, or again the order of the elementary filtering cells.

(81) It should be noted that in the event where the two IIR filters h.sub.1 and h.sub.2 have parts AR and/or MA of different orders, the number of elementary filtering cells can be different to each other. In this case, one or more elementary cells of identity type (PTi) is created which is associated in parallel to each supernumerary elementary filtering cell of the IIR filter having a larger number of elementary cells h.sup.l.sub.1 to form a filtering block B.sup.l. Therefore, each elementary cell of the two IIR filters forms part of a filtering block.

(82) FIG. 4 schematically illustrates a degraded example of execution of the filtering device according to the invention.

(83) In this example, the control elements 10.sup.l.sub.i and 10.sup.l.sub.PTi are positioned within the filter cells before the elementary filtering cells h.sup.l.sub.i and PTi. FIG. 4 corresponds to the particular case where these control elements employ instantaneous transition (C.sub.f(n<T)=1; C.sub.f(nT)=0), represented here by a single interrupter 10.sup.l.

(84) In this case progressiveness of the switching results from the impulse response duration of the elementary filtering cells, according to which the latter continue to generate an output signal (switching off signal) even in the absence of input signal. In fact, the adder ADD combines this switching off signal with the signal filtered by the other elementary active cell, ensuring progressive switching.

(85) It is however noted that the duration of progressive switching is less well controlled here than in the case of FIGS. 1 and 3, where the profile of the coefficient C.sub.f directly controls the switching duration N.

(86) In reference again to FIG. 4, the control elements 10.sup.l.sub.i are single switches or interrupters which do not apply progressive switching. But they produce initialisation durations of the internal variables which are considerably reduced due to use of elementary filtering cells h.sup.l.sub.i according to the invention at the level of which switching is controlled (command 0/1 shown in the figure for controlling simultaneously all the interrupters 10.sup.l).

(87) This degraded embodiment can perform just as well in the series version of FIG. 1 as in the parallel version of FIG. 3.

(88) The invention such as described hereinabove therefore reduces latency in an IIR filter change even more significantly since these IIR filters are broken down into elementary filtering cells having low orders.

(89) Some tests have shown notable efficacy of the invention with respect to known techniques. For example, applying the invention to control switching between the IIR filters of order 4, factorised into two elementary cells of order 2, the inventors reduced by half the switching time relative to the hybrid method mentioned previously, with constant artefact energy (audio click).

(90) The present invention has various applications in the field of digital signal processing, and especially audio signals, since the use of IIR filters varying over time is very widespread.

(91) By way of illustration, this is the case of linear prediction coders (LPC-speech coder). In this case, the filter change LPC from one frame to the other is generally done by interpolation of coefficients. Using the present invention improves filtering stability during switching by having minimal latency and complexity.

(92) This is also the case for acoustic shock suppressor modules. Any telephone terminal plans to integrate limiters on the loudspeaker signal the task of which is to protect the user from potential acoustic shocks (strong signals, Larsen signals, etc.). For example, these limiters are found in self-contained units for the use of representatives, in VoIP clients, in mobile terminals.

(93) In the case of Larsen, the frequential content generally varies rapidly over time. If the frequencies which compose the Larsen over time are supposedly known, using the invention quasi instantaneously adapts the IIR filters to these frequencies and filters the Larsen effect.

(94) Finally, it is also the case of noise reduction systems. Such systems are used in any terminal fitted with a sound pickup, especially hands-free types.

(95) In this case, using noise reduction filters would employ filters which vary over time to adapt to the spectral content of word and/or noise. Applying the ideas of the invention to these systems especially improves the latter with respect to latency.

(96) FIG. 5 schematically shows a device or system 50 for executing the invention, especially equipment fitted with IIR filters and on which the filter switching is carried out.

(97) The system 50 comprises a communications bus 51 to which are attached a central processing unit or microprocessor 52, live memory 53, non-volatile memory 54, a display device 55 for displaying user interfaces, a pointing device 56 and optionally other peripherals 57 (communications interface, diskette or disc reader, etc.).

(98) The non-volatile memory 54 comprises the programs, execution of which executes the process according to the invention, and for example software definitions of IIR filters, optionally already broken down into elementary filtering cells.

(99) During execution of programs, the executable code of these programs is loaded in live memory 53, RAM type, and executed by the microprocessor 52. This execution allows instantiation of the IIR filters and control of their activation and their switching off, that is, also of the switching from one IIR filter to the other, as shown in FIGS. 1, 3 and 4 for example.

(100) The display device 55, such as a screen, enables display of graphic user interfaces for example allowing a user to generate activation, switching off or switching IIR filter commands.

(101) The pointing device 56 can be integrated into the display device, especially when it is a touch screen, or remote, for example a mouse, a touch pad or a graphic tablet, to allow the user to send these commands.

(102) The device described here and the central unit 52 in particular are likely to employ all or part of the processing described in connection with FIGS. 1 to 4 for executing the processes of the present invention and constituting the devices and systems of the present invention.

(103) The preceding examples are only embodiments of the invention which is not limited thereto.

(104) For example, the invention can employ only the block h.sub.1 of FIG. 1 (left part), corresponding to the case of FIG. 3 wherein the filter h.sub.2 is an identity filter (the second elementary filtering cells are also identity cells). Switching according to the invention switches off the filter h.sub.1 for the sake of absence of filtering, without artefact or reducing it significantly.

(105) Similarly, using only the block h.sub.2 of FIG. 1 (right part), switching according to the invention activates the filter h.sub.2 whereas initially no filtering of the digital signal was performed, without artefact or reducing it significantly.

(106) An embodiment of the present invention eliminates at least one of the disadvantages discussed in the Background section.