CN200983612Y - Multi-channel echo elimination system based on SIMO channel blind recognition - Google Patents

Abstract

The utility model provides a multipath echo suppressing system based on SIMO channel blind identification, comprising echo suppressors, microphones, loudspeakers, wherein the near-end microphone is abutting jointed with the far-end loudspeaker, and an echo suppressor is connected between every microphone and loudspeaker. The echo suppressor comprises a SIMO blind identification modular, a sound-access modular, an intercommunication check modular, an echo channel accessing modular, a convolution modular, and an accumulator, wherein the SIMO blind identification modular, the sound accessing modular, the convolution modular and the accumulator are connected with each other, the output of the sound-access modular is connected with the SIMO blind identification modular via the intercommunication check modular, and also is connected with the convolution modular via the echo channel accessing modular. The SIMO blind identification modular and the sound-access modular are connected with the far-end microphones and the near-end loudspeaker with the number of m through the parallel input terminals IX with the number of m, further the accumulator, the echo channel accessing modular, and the intercommunication check modular are connected with the near-end microphone via an input terminal IY, and are connected with a corresponding far-end loudspeaker via an input terminal OUT. The utility model is low in detuning and little in echo wave interference.

Description

Multi-path echo eliminating system based on the SIMO channel blind discrimination

Technical field

The utility model relates to the ECT echo cancellation technique in the sealing voice environment, specifically is meant the multi-path echo eliminating system based on SIMO (the many outputs of the single input of Single Input Multi Output) channel blind discrimination.

Background technology

Along with the development of the develop rapidly of electronic communication, particularly multimedia communication, hands-free communication device has obtained extensive use in mobile communication in the systems such as vehicle-carrying communication and teleconference.But these systems often are difficult to avoid the puzzlement of echo, and are very serious such as echo problems in the sealing voice environment that is made of loudspeaker, room and microphone.In order to improve the Speech Communication quality, echo eliminator is used to eliminate the coupling between loudspeaker and the microphone, disturbs to reduce echo, makes every effort to set up in actual applications effective hands-free communication systems.On the other hand, for Space that strengthens sound and the comfortableness that improves conversation, multi-path voice communication is necessary.Yet for multi-path voice communication, there are many acoustic path to need identification, need carry out multi-path echo and eliminate.It all is to eliminate at the single channel echo that existing echo is eliminated patented technology many.Compare with the elimination of single channel echo, multi-path echo is eliminated some new problems that face.Owing to need pick out many acoustics channels uniquely, multi-path echo is eliminated more difficult more than single channel problem, and traditional multi-path echo arrester can't be estimated the impulse response of near-end room uniquely.In addition, because the strong correlation between a plurality of input signals, the imbalance of traditional multi-path echo arrester is very big, can not satisfy application requirements well.Owing to these two reasons, the performance of traditional multi-path echo arrester is subjected to very big restriction.

Summary of the invention

The purpose of this utility model is to overcome the shortcoming and defect of above-mentioned prior art, provide that a kind of imbalance is little, echo disturb few, the measured multi-path echo eliminating system of Speech Communication matter based on the SIMO channel blind discrimination.

The purpose of this utility model also is to provide the multi-path echo removing method of above-mentioned multi-path echo eliminating system based on the SIMO channel blind discrimination.

The purpose of this utility model is achieved through the following technical solutions: this multi-path echo eliminating system based on the SIMO channel blind discrimination comprises echo eliminator, microphone, loudspeaker, each near-end microphone connects far-end loudspeaker, otherwise, each far-end microphone also connects near-end loudspeaker, each is to microphone, connect an echo eliminator between the loudspeaker, described echo eliminator comprises SIMO blind discrimination module, the sound estimation module, the intercommunication detection module, the echo channel estimation module, convolution module, adder, described SIMO blind discrimination module, the sound source estimation module, convolution module, adder connects, the output of described sound source estimation module also is connected with SIMO blind discrimination module by the intercommunication detection module, and be connected with convolution module by the echo channel estimation module, described SIMO blind discrimination module, the sound source estimation module is passed through m parallel input IX and m far-end microphone of echo eliminator simultaneously, the near-end loudspeaker connect, described adder, the echo channel estimation module, the intercommunication detection module is connected with a near-end microphone by an input IY of echo eliminator, and the output of adder is connected with corresponding far-end loudspeaker by an output OUT of echo eliminator.

For realizing the utility model better, described SIMO blind discrimination module comprises m initial construction matrix module, m follow-up structural matrix module, total structural matrix module, the update calculation module, g parameter cache module, described the 1st initial construction matrix module is connected with the 1st follow-up structural matrix module, m initial construction matrix module is connected respectively with the 1st～m follow-up structural matrix module, described m follow-up structural matrix module is connected with total structural matrix module, total structural matrix module, the update calculation module, g parameter cache module connects successively, the update calculation module with described intercommunication detection module is connected.

Described g parameter cache module comprises m memory.

Principle of the present utility model is: SIMO channel blind discrimination module application is eliminated in multi-path echo, mainly comprised three processes: the contrary and echo path impulse response of far-end room impulse response is estimated or upgraded in (1) under the situation that does not have both sides' intercommunication; (2) on the contrary basis of known far-end room impulse response, estimate far-end room source signal, in conjunction with the estimation of echo path impulse response, and then obtain the estimation of echo-signal; (3) received signal of near-end room microphone deducts echo-signal and realizes the echo elimination.

Basically, do not having under both sides' intercommunication situation, a plurality of input signals of near-end room all are to be got by the same sound source filtering in far-end room, and this has just constituted a SIMO system, thereby can be by SIMO blind discrimination module implementation procedure (1); The impulse response of estimation from far-end room source signal to the near-end room echo can be solved by the single channel method for echo cancellation, adopts auto-adaptive filtering technique to reach this purpose usually.

The utility model imbalance is little, echo disturbs and lacks, and the Speech Communication quality is good, has effectively avoided problems such as " separating not unique ", " the filter imbalance is big " in traditional multi-path echo elimination.

Description of drawings

Fig. 1 is the structural representation of the utility model based on the multi-path echo eliminating system of SIMO channel blind discrimination;

Fig. 2 is the internal structure schematic diagram of multi-path echo arrester shown in Figure 1;

Fig. 3 is the internal structure schematic diagram of SIMO blind discrimination module shown in Figure 2;

Fig. 4 is the internal structure schematic diagram of g parameter cache module;

Fig. 5 is the multi-path echo removing method workflow diagram of the utility model based on the multi-path echo eliminating system of SIMO channel blind discrimination.

Embodiment

Below in conjunction with embodiment and accompanying drawing, the utility model is described in further detail, but execution mode of the present utility model is not limited thereto.

Embodiment

As shown in Figure 1, multi-path echo eliminating system based on the SIMO channel blind discrimination, comprise echo eliminator, microphone, loudspeaker, each near-end microphone connects far-end loudspeaker, otherwise, each far-end microphone also connects near-end loudspeaker, and each is to connecting an echo eliminator between microphone, the loudspeaker.

As shown in Figure 2, echo eliminator comprises SIMO blind discrimination module, the sound estimation module, the intercommunication detection module, the echo channel estimation module, convolution module, adder, SIMO blind discrimination module, the sound source estimation module, convolution module, adder connects successively, the output of sound source estimation module also is connected with SIMO blind discrimination module by the intercommunication detection module, and be connected with convolution module by the echo channel estimation module, SIMO blind discrimination module, the sound source estimation module is passed through m parallel input IX and m far-end microphone of echo eliminator simultaneously, the near-end loudspeaker connect, adder, the echo channel estimation module, the intercommunication detection module is connected with a near-end microphone by an input IY of echo eliminator, and adder output is connected with corresponding far-end loudspeaker by an output OUT of echo eliminator.

As shown in Figure 2, x ₁(t) ..., x _m(t) be meant the signal of gathering by each microphone of echo eliminator IX input, m road altogether.x ₁(t) ..., x _m(t) input to SIMO blind discrimination module, the output of this module is each required parameter g of FIR filter in the sound source estimation module ₁..., g _mThe sound source estimation module is finished the estimation to sound source, and its output is sent to the echo channel estimation module on the one hand, carries out convolution with the output of echo channel estimation module on the other hand.The result of convolution promptly is the estimation to echo, and the estimation by the y (t) of IY input input deducts echo promptly can reach the purpose of eliminating echo.

The sound source estimation module is finished the calculating of following formula (1):

$\hat{s}\left(t\right)=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\left[\underset{\mathrm{\τ}=0}{\overset{N}{\mathrm{\Σ}}}{g}_i\left(\mathrm{\τ}\right)x_i(t-\mathrm{\τ})\right]---\left(1\right)$

M is meant that the IX port has the input of m road in the formula (1), and the output of SIMO blind discrimination module also is the m road simultaneously.g _iBe τ value of i output of SIMO blind discrimination module (τ), detailed description is arranged in Fig. 4.x _i(t-τ) is road t-τ sample value input of IX input port i.N is the exponent number that obtains according to following formula (2):

Wherein l is that the room is long, and w is that the room is wide, and h is the room height, and v is a sampled voice speed, and c is the velocity of sound, probably is 340 meter per seconds, function

Round in the expression.

The echo channel estimation module is mainly finished the evaluation work of following formula (3):

$\left\{\begin{array}{c}e\left(t\right)=y\left(t\right)-\hat{f}{\left(t\right)}^T\·\hat{s}\left(t\right),\\ \hat{f}(t+1)=\hat{f}\left(t\right)-\mathrm{\μ}\·e\left(t\right)\·\hat{s}\left(t\right),\end{array}\right.---\left(3\right)$

μ is the iteration step length of setting in the formula 3.The output of echo channel estimation module

Its meaning is the characteristic estimating of echo channel, wherein $\hat{f}={({\hat{f}}_0,{\hat{f}}_1,\·\·\·,{\hat{f}}_{2N-1})}^T.$ This module adopts adaptive filter method to estimate and renewal It upgrades iterative formula is above-mentioned formula 3.At last, by formula (4) estimation echo, realize the echo elimination again by formula (5):

$\hat{e}\left(t\right)=\hat{f}*\hat{s}\left(t\right)---\left(4\right)$

$y_{\mathrm{out}}\left(t\right)=y\left(t\right)-\hat{e}\left(t\right)---\left(5\right)$

Y in the formula 5 _Out(t) be the final output of having eliminated echo back echo arrester, symbol in the formula 4 " * " expression convolution.

The intercommunication detection module is to detect whether intercommunication of far-end room and near-end room speaker, and its input is the output of sound source estimation module

With the y (t) of IY input input, output is the result who judges.

As shown in Figure 3, SIMO blind discrimination module comprises m initial construction matrix module, m follow-up structural matrix module, total structural matrix module, the update calculation module, g parameter cache module, the 1st initial construction matrix module is connected with the 1st follow-up structural matrix module, m initial construction matrix module is connected respectively with the 1st～m follow-up structural matrix module, m follow-up structural matrix module is connected with total structural matrix module, total structural matrix module, the update calculation module, g parameter cache module connects successively, the update calculation module with the intercommunication detection module is connected.

SIMO blind discrimination module is from the m road signal of IX input input, and its corresponding initial construction matrix module is all followed on each road, as k road x _k(t), it connects k initial construction matrix module, and this module can be constructed X _k(N) matrix is as formula (6)

Element is as x in the matrix _k(N+2), N+2 sample value representing the input of k road.Wherein T is taken as 3N.X with each initial construction matrix module structure _k(N) send into follow-up matrix construction module, this module structure X ⁱ(N) matrix, follow-up matrix construction module module is finished the work of formula (7), and it exports the X of each road correspondence ⁱ(N) matrix.

At matrix X ⁱ(N) in, every or be 0 matrix perhaps is matrix-block X _k(N), its size is (T-N+1) * (N+1).At last with each X ⁱ(N) be aggregated into total structural matrix module, this module structure matrix X (N), the relation of structure such as formula (8).

X (N) is inputed to update module, and when the detection of intercommunication detection module was output as non-intercommunication, then this module was to the symmetrical matrix [X of [(N+1) m] * [(N+1) m] ^T(N) X (N)] carry out feature decomposition, g is the characteristic vector of corresponding minimal eigenvalue, wherein:

$g={(g_1^T,\·\·\·,g_m^T)}^T$

g _k＝[g _k(N)，…，g _k(0)] ^T，k＝1，…，m

Use g to upgrade g parameter cache module.If the detection of intercommunication detection module output result is intercommunication, does not then carry out aforementioned calculation, and do not upgrade g parameter cache module.

As shown in Figure 4, g parameter cache module comprises m memory, and the employed parameter of the in store sound source estimation module of g parameter cache module when not having both sides' intercommunication, can be provided with as required at set intervals and upgrade once; When if both sides' intercommunication is arranged, then needn't upgrade.

As shown in Figure 5, multi-path echo based on the multi-path echo eliminating system of SIMO channel blind discrimination is eliminated the course of work, comprise the elimination of parameter update and echo, parameter update and echo are eliminated parallel carrying out, parameter update is meant under the situation that does not have both sides' intercommunication, estimating or upgrading the contrary of far-end room impulse response is the g parameter, and echo path impulse response is a self adaptation echo channel characteristic estimating in the echo channel estimation module

If there is intercommunication in both sides, echo is eliminated and is proceeded, but parameter is not upgraded; Echo is eliminated and is meant that at the contrary of known far-end room impulse response be on the basis of g parameter, estimates far-end room source signal

Estimation in conjunction with echo path impulse response And then obtain the estimation of echo-signal

The received signal y of near-end room microphone (t) deducts the estimation of echo-signal

The echo that is eliminated clean signal afterwards, thus reach the purpose that echo is eliminated.

The course of work that echo is eliminated is:

(1) the g parameter cache module in the SIMO blind discrimination module is carried out initialization, and to self adaptation echo channel characteristic estimating in the echo channel estimation module

Carry out initialization;

(2) by g parameter cache module the g parameter is sent into the sound source estimation module, simultaneously far-end room microphone signal x ₁(t) ..., x _m(t) also be transfused to the sound source estimation module, the sound source estimation module is handled the estimated value that draws far-end room sound-source signal

And will Send intercommunication detection module and echo channel estimation module to;

(3) the echo channel estimation module receives After, will

With

Convolution obtains the estimation of echo

(4) estimation of echo

Be sent to adder after the negate, the signal y (t) that the near-end room microphone of input summer receives deducts echo and estimates

Obtain

For going echo clean signal afterwards, realize the echo elimination;

(5) return step (2), proceed echo and eliminate.

The course of work of parameter update is:

(1) intercommunication detection module cycle time (0.3～2 second) detects and judges whether to exist the intercommunication signal, and will detect judged result and be input to SIMO blind discrimination module and echo channel estimation module, whether decision carries out parameter update, if there is both sides' intercommunication, need not carry out parameter update, if there is not both sides' intercommunication, then carry out step (2) and step (3).

(2) SIMO blind discrimination module is to the x of input ₁(t) ..., x _m(t) and y (t) handle, result sends the update calculation module to, to upgrade the g parameter in the g parameter cache module;

(3) obtain new echo channel estimated parameter promptly by adaptive algorithm

And to the former preservation of echo channel estimation module

Upgrade.

The course of work that intercommunication is judged is:

(1) by down trigger cycle time, the intercommunication detection module calculates

And the coefficient correlation between the y (t)

(2) judge coefficient correlation ξ and threshold values c ₁Magnitude relationship, c wherein ₁Be positive number, get c ₁=0.1～0.5, if $\left|\mathrm{\&xi;}(\hat{s}\left(t\right),y\left(t\right))\right|{>\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="Y20062005603500102.png,Y20062005603500112.png"\; state="\{\{state\}\}">c</figure-callout>}}_1,$ Think and do not have both sides' intercommunication; Otherwise think and have both sides' intercommunication.

As mentioned above, just can realize the utility model preferably.

Claims (1)

1, multi-path echo eliminating system based on the SIMO channel blind discrimination, it is characterized in that comprising echo eliminator, microphone, loudspeaker, the near-end microphone, it is right that the far-end loudspeaker connect into one to one, each is to microphone, connect an echo eliminator between the loudspeaker, described echo eliminator comprises SIMO blind discrimination module, the sound estimation module, the intercommunication detection module, the echo channel estimation module, convolution module, adder, described SIMO blind discrimination module, the sound source estimation module, convolution module, adder connects, the output of described sound source estimation module also is connected with SIMO blind discrimination module by the intercommunication detection module, and be connected with convolution module by the echo channel estimation module, described SIMO blind discrimination module, the sound source estimation module is passed through m parallel input IX and m far-end microphone of echo eliminator simultaneously, the near-end loudspeaker connect, described adder, the echo channel estimation module, the intercommunication detection module is connected with a near-end microphone by an input IY of echo eliminator, and the output of adder is connected with corresponding far-end loudspeaker by an output OUT of echo eliminator.

Images