CN101233560B - Method and device for restoring audio signal - Google Patents

Abstract

A method of restoring a corrupted audio signal comprises the steps of ; inputting the corrupted audio signal in a first channel, inputting one or more further correlated audio signals in one or more further channels, and restoring the corrupted audio signal using a Multi-Channel Autoregressive Model that models the corrupted signal as a linear combination of scaled time shifted portions of the further signal and the corrupted signal. Embodiments are described in which the method is used to improve received audio signals in DAB receivers and mobile telephones.

Description

Recover the method and the device of sound signal

Technical field

The invention relates to the method and the device that recover impaired sound signal, one of them of the several applications that the present invention had is for working as identical plan warp from a digital audio broadcasting (DAB, Digital AudioBroadcast) when a channel and an analog fm (FM)/amplitude modulation (AM) Channel Transmission, improves the reception of DAB the down auxiliary of corresponding simulation FM/AM signal.The present invention further uses and has comprised mobile phone and procotol voice (VOIP, Voice Over Internet Protocol) phone.

Background technology

Most audio recording by at least two independently voice-grade channel formed, the digital audio frequency recording in many modern times even comprised 7.1 independently around sound channel, though the application of industrial audio code, captured the advantage of audio frequency redundant mode as mpeg standard, but comprehensively exploitation is still difficulty.

When original medium are damaged, or in the process of transmission, the data that are included in the channel will be impaired, impaired audio file may include the noise of click (clicks), serge bat sound (pops) or cracker (crackles) etc., but can be repaired via the audio frequency restoration methods.

Many effectively real-time (real time) audio frequency recovery algorithms are based on autoregression (AR, Auto-Regressive) model, the output of the all-pole filter (all-pole filter) that will be excited from white noise (white noise) is modeled as the sound signal at random of permanent (stationary), in known single channel autoregressive model, the output of linear-time invariant filter will be restricted to over output valve and white noise input e _nWeight and

$x_n=\underset{i=1}{\overset{P}{\mathrm{\Σ}}}a\left(i\right)x(n-i)+e_n---\left(1\right)$

Using the single channel autoregressive model to recover impaired media fragment can cause audio frequency in various degree to twist (distortion) usually; particularly if include voice or music decompressed data in the media fragment; in addition; method with the construction of single channel autoregressive model institute needs parameter adjustment; as the exponent number (order) of model and block length etc.; sound signal will be excessively level and smooth compared to typical sound signal after this will cause and rebuild; long nick (long gap) interpolation based on autoregressive model; usually in the middle of near breach, locate to show performance of poor quality; as adopting least square method (least square) simulation error to estimate the result of unknown data; therefore, be only applicable to the relative shortage mouth of interpolation usually herein based on autoregressive interpolating method less than 20ms (music is permanent).

For the autoregressive model of single channel more complete description and analysis, see also the school section book of " DigitalAudio Restoration-a statistical model based approach " by name, the author is Simon J.Godsill and Peter J.W.Rayner, by Springer ﹠amp; Verlag published in 1998.

Summary of the invention

Of the present invention first towards for a kind of method of recovering impaired sound signal is provided, comprised step:

Described impaired sound signal is inputed to first channel;

Import one or more associate audio signal to one or more channels; And

Use the multichannel autoregressive model to recover described impaired sound signal, wherein said multichannel autoregressive model is modeled as described impaired sound signal the linear combination of the time scale offset portion of described one or more associate audio signal and described impaired sound signal.

Via using the present invention, can effectively reduce the problem that is produced when using the single channel autoregressive model to recover impaired audio fragment, the past output of time offset value that can be modeled as past output, other channel of an allocated channel from the output that linear-time invariant filter produced of a particular channel add a white noise constant weight and, so the multichannel autoregressive model can be applicable to the recovery of real-time multi-channel audio.

The multichannel autoregressive model is to be built on the observed reading, because the multi-channel audio medium have comprised various channel with redundant data, therefore in any given period, as if unlikely all data all impaired, proof is come modern freely three-dimensional sound record, this type of pop music particularly is the record that is essentially single-tone with a plurality of independent channels transmission, and it only is that audio frequency with several interchannels carries out phase deviation and strong and weak staggered to create the sensation around audio-source.

The multichannel autoregressive model can use following formula to carry out interpolation:

$x_1\left(n\right)=\underset{i=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">P</figure-callout>}}_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">P</figure-callout>}}_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)$

$+\underset{k=1}{\overset{P_3^1}{\mathrm{\&Sigma;}}}{a}_3^1\left(k\right)x_3(n-k+{\mathrm{\&tau;}}_3^1)+...+e_{n1}---\left(2\right)$

Similarly:

$x_m\left(n\right)=\underset{i=1}{\overset{P_m^m}{\mathrm{\&Sigma;}}}{a}_m^m\left(i\right)x_m(n-i)+\underset{j=1}{\overset{P_1^m}{\mathrm{\&Sigma;}}}{a}_1^m\left(j\right)x_1(n-j+{\mathrm{\&tau;}}_1^m)$

$+\underset{k=1}{\overset{P_2^m}{\mathrm{\&Sigma;}}}{a}_2^m\left(k\right)x_2(n-k+{\mathrm{\&tau;}}_2^m)+...+e_{\mathrm{nm}}---\left(3\right)$

Wherein

x ₁, x ₂..., x _m Be channel 1,2 ..., the output of m

Be channel i, the autoregressive coefficient between j

Be channel i, the exponent number of the autoregressive coefficient between j

Be channel i, time migration constant between j

e _N1, e _N2..., e _NmBe the white noise input

When two channels, the multichannel autoregressive model can use following formula to carry out interpolation:

$x_1\left(n\right)=\underset{i=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">P</figure-callout>}}_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">P</figure-callout>}}_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)+e_{n1}---\left(4\right)$

$x_2\left(n\right)=\underset{i=1}{\overset{{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">P</figure-callout>}}_2^2}{\mathrm{\&Sigma;}}}{a}_2^2\left(i\right)x_2(n-i)+\underset{j=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">P</figure-callout>}}_1^2}{\mathrm{\&Sigma;}}}{a}_1^2\left(j\right)x_1(n-j+{\mathrm{\&tau;}}_1^2)+e_{n2}---\left(5\right)$

Of the present invention second towards for a kind of method of the digital audio broadcasting sound signal of remaking is provided, comprised step:

Receive and decipher the digital audio broadcasting sound signal to produce corresponding audio packet;

Receive analog broadcast signal, described analog broadcast signal and described digital audio broadcasting sound signal are broadcasted simultaneously, and described analog broadcast signal contains identical broadcasting plan with described digital audio broadcasting sound signal;

The described analog broadcast signal of demodulation is with from wherein producing simulated audio signal;

Described simulated audio signal is converted to the digitized simulation sound signal;

Import described digital audio broadcasting audio packet and described digitized simulation audio packet to corresponding buffer storage and compensate mistiming between described digital audio broadcasting audio packet and described digitized simulation audio packet so that suitable delay to be provided;

Detect the digital audio broadcasting grouping of impaired or gaps and omissions; And

Use the multichannel autoregressive model and the digitized simulation voice-grade channel of described digital audio broadcasting to come interpolation loss or impaired digital audio broadcasting grouping to divide into groups to recover described impaired digital audio broadcasting.

When receiving feeble signal, the DAB receiver may suffer the situation of packet loss, instability causes in the audio frequency after the recasting and " click (click) ", " serge bat sound (pops) ", " bust sound (drop-off) " or " quiet (silences) " etc. occur because the essence of digital transmission, tonequality will reduce and become.

The simulation (FM or AM) of different frequency and digital wireless signal are transmitted simultaneously (owing to need time encoding in many now wireless stations; digital signal has short the delay usually), most DAB wireless receiver/tuner all can receive and decipher/demodulation DAB and FM/AM signal.

The demodulation via the while/decoding FM/AM and DAB signal, can use the multichannel autoregressive model of the digitized audio that utilizes DAB grouping and produced from the FM/AM signal to predict and recover the DAB grouping, DAB that so will be more now is wireless only to have better performance in the use characteristic that just is converted to the FM/AM signal when the DAB reception is lower than a given threshold value.

Be interpolation DAB grouping, the multichannel autoregressive model can use following formula:

$x_1\left(n\right)=\underset{i=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">P</figure-callout>}}_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">P</figure-callout>}}_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)+{\mathrm{<figure-callout\; id="1"\; label="e\; n"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">e</figure-callout>}}_n$

X wherein ₁Be described DAB signal and x ₂Be described digitized simulation sound signal.

In one embodiment, when lose and/or impaired DAB grouping than a preset period of time when long, then replace described DAB grouping with described digitized simulation sound signal.

Behind a long duration, when damaged packets is not received yet, then damaged packets can not be used to recover the interpolation of DAB audio packet in an other preset period of time, and described other preset period of time is before the described first not impaired DAB grouping and behind a gap greater than described first preset period of time.

Three orientations of the present invention comprise for a kind of wireless receiver is provided: the digital audio broadcasting code translator; The analog broadcasting receiver, described analog broadcasting receiver comprises the detuner that is used to produce simulated audio signal; The storage of first buffer is used to store continuous decoding digital audio broadcasting audio packet; Analog-digital converter is used for the described simulated audio signal of digitizing; Second buffer stores, and is used to store continuous digital signal samples; Packet detector is used for determining whether the digital audio broadcasting audio packet is lost, impaired or not impaired, and produces the packet loss pointer in view of the above; And digital signal processor, described digital signal processor has the input that is used to receive digital audio broadcasting audio packet, digitized simulation audio frequency and described packet loss pointer; Wherein said digital signal processor is used to carry out the multichannel autoregressive model leads from the data of described digital audio broadcasting audio packet and described digitized simulation audio frequency so that the interpolation of the described impaired digital audio broadcasting audio packet of energy with use.

Be interpolation DAB grouping, digital signal processor uses following formula programming:

$x_1\left(n\right)=\underset{i=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">P</figure-callout>}}_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">P</figure-callout>}}_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)+e_{n1}---\left(4\right)$

X wherein ₁Be described DAB signal and x ₂Be described digitized simulation sound signal.

When the gap between between not impaired DAB grouping surpasses given period, the digitized simulation audio frequency will be used to replace described DAB grouping.

In one second short time interval of contiguous not impaired DAB grouping, use described interpolation DAB grouping.

Of the present invention the 4th towards producing first and second associate audio signal for the radiofrequency signal that provides a kind of mobile phone to be used to receive to carry sound signal with from described sound signal, the demodulation on described radiofrequency signal of described sound signal, described mobile phone more comprises signal processor, described signal processor has and is used to receive described first and first and second input and the output of described second sound signal, treated sound signal derives from described output, and described signal processor is through being provided for carrying out the multichannel autoregressive model to activate the interpolation of impaired sound signal.

So mobile phone has various forms, condition obtains two relevant sound signals to carry out the multichannel autoregressive model for making a processor, therefore use rake formula (RAKE) receiver to produce multiple sound signal to receive wireless signal from multiple delay, so multiple inhibit signal is to produce from the signal reflex between described base station and described mobile phone, other possibility has comprised various reception, use a plurality of antennas that spatially separate at base station or at mobile phone, or on the two, all use a plurality of antennas that spatially separate, a plurality of corresponding radio frequency (RF are provided in mobile phone, Radio frequency) channel, or the single RF channel time-division multitask between various aerial signals.

The further options that are used to produce two relevant sound signals have comprised and have provided support at least two or the mobile phone of a Subscriber Identity Module (SIM, the Subscriber Identity Module) card of above sign (identifies) and use a multilateral accord (multi-party protocol) with in conjunction with described signal.

Of the present invention the 5th towards phone for phone that a kind of procotol voice are provided or a kind of wireless compatible authentication, the phone of the phone of described procotol voice or wireless compatible authentication comprises decoder device and signal processor, described decoder device is used to decipher a plurality of associate audio signal that receive by a plurality of different path, and described signal processor is used to receive described a plurality of associate audio signal and enforcement multichannel autoregressive model through decoding and produces treated audio output signal with the interpolation that activates impaired sound signal.

Description of drawings

Above-mentioned and other feature and the advantage of the present invention will be via following narration, in the mode of concrete case study on implementation, and consider in light of actual conditions following subsidiary graphic and release bright, wherein:

First figure has showed DAB receiver of the present invention in the mode of block diagram;

Second figure has showed under condition of acceptance of poor quality and the DAB grouping that receives, and the corresponding FM/AM signal of institute, and use that the present invention obtained one recover DAB output;

The 3rd figure has showed the synoptic diagram of the case study on implementation of incorporating mobile phone of the present invention in the mode of block diagram;

The 4th figure has showed the synoptic diagram of the mobile phone communication case study on implementation that uses mobile phone of the present invention in the mode of block diagram; And

The 5th figure has showed the synoptic diagram of the VOIP communication case study on implementation of incorporating VOIP phone of the present invention in the mode of block diagram.

Embodiment

First figure has showed a wireless receiver, has comprised a DAB receiver 1 and a simulation FM receiver 2.In operation, the two all transfers to identical broadcasting plan with described DAB receiver 1 and described simulation FM receiver 2, now, in Britain, many radio broadcasting stations are transportation simulator and digital wireless signal (ignoring any time skew that signal Processing postpones generation) simultaneously, and many DAB wireless receiver/tuners have the ability that receives DAB and FM/AM analog broadcasting.At the case study on implementation shown in first figure, one first output of DAB receiver, an i.e. audio data packet, provide to a buffer 3, described buffer 3 is as a first in first out (FIFO, First In First Out) buffer is to store these audio data packet, similarly,, can select the length of buffer 3 and 5 to compensate and postpone and the time migration between DAB and simulating signal that causes by 4 digitizings of an analog-digital converter and provide from the output of described FM tuner by signal Processing to a FIFO buffer 5.

The digital signal processor 6 that exports to of buffer 3 and 5 is provided, described digital signal processor 6 disposes a multichannel autoregressive model and recovers any impaired DAB grouping to use digitaling analoging signal, though the program content of DAB grouping and simulating signal is identical, but when both do not have same characteristic features aspect frequency range and amplitude dynamic range, be still laudable, therefore, relatively suggestion uses described digitaling analoging signal to recover impaired DAB grouping, rather than replaces simulating signal with faint DAB grouping.

Buffer 3 and 5 length are so to make digitized simulation sound signal and impartial DAB packet alignment, and are able to so compensate any time skew between DAB and AM/FM broadcasting, and decoding described DAB audio packet, the described FM/AM signal of demodulation and needed mistiming of the described simulated audio signal of digitizing.

Described DAB receiver 1 produces a packet loss signal-arm, described signal is to be coupled to a digital signal processor (DSP via a circuit 9, Digital Signal Processor) 6, when one or more DAB grouping is lost, will make described DSP 6 carry out described multichannel autoregressive model to use DAB grouping history and the described loss of digitized simulation sound signal interpolation DAB grouping.

Provide a DAB packet loss to be not more than a given period, algorithm then proposed by the invention can in be inserted to about 120msecs, the DAB grouping that can recover to lose is to provide the output of the audio frequency with a little degradation, if the loss of grouping has extended beyond one than long duration, then when using described DAB grouping and FM/AM digitized audio and described multichannel autoregressive model interpolation end points, the digitized simulation audio frequency can be substituted in and describedly divide into groups than the DAB in the center section of long duration.

When grouping can't identification or check code (check sum) when incorrect, the packet loss signal-arm that is produced from the DAB receiver can use the pointer from code translator, because when a grouping is not received or finds when impaired, described code translator will normally produce a zero output, another is replaced by the amplitude of the described digital packet of supervision in the buffer storage, and produces the described packet loss signal-arm that is applied to described DSP 6 via a circuit 10 in view of the above.

Under common situation, described multichannel autoregressive model uses following formula to carry out interpolation:

$x_1\left(n\right)=\underset{i=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">P</figure-callout>}}_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">P</figure-callout>}}_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)$

$+\underset{k=1}{\overset{P_3^1}{\mathrm{\&Sigma;}}}{a}_3^1\left(k\right)x_3(n-k+{\mathrm{\&tau;}}_3^1)+...+e_{n1}---\left(2\right)$

Similarly:

$x_m\left(n\right)=\underset{i=1}{\overset{P_m^m}{\mathrm{\&Sigma;}}}{a}_m^m\left(i\right)x_m(n-i)+\underset{j=1}{\overset{P_1^m}{\mathrm{\&Sigma;}}}{a}_1^m\left(j\right)x_1(n-j+{\mathrm{\&tau;}}_1^m)$

$+\underset{k=1}{\overset{P_2^m}{\mathrm{\&Sigma;}}}{a}_2^m\left(k\right)x_2(n-k+{\mathrm{\&tau;}}_2^m)+...+e_{\mathrm{nm}}---\left(3\right)$

Wherein

x ₁, x ₂..., x _mBe channel 1,2 ..., the output of m

Be several channels i, the autoregressive coefficient between j

Be several channels i, the exponent number of the autoregressive coefficient between j

Be several channels i, time migration constant between j

e _N1, e _N2..., e _NmBe the white noise input

Under the double-channel situation of simplifying, will use following formula:

$x_1\left(n\right)=\underset{i=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">P</figure-callout>}}_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">P</figure-callout>}}_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)+e_{n1}---\left(4\right)$

$x_2\left(n\right)=\underset{i=1}{\overset{{\mathrm{<figure-callout\; id="2"\; label="P"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">P</figure-callout>}}_2^2}{\mathrm{\&Sigma;}}}{a}_2^2\left(i\right)x_2(n-i)+\underset{j=1}{\overset{{\mathrm{<figure-callout\; id="1"\; label="P"\; filenames="S2006800280720E00051.png"\; state="\{\{state\}\}">P</figure-callout>}}_1^2}{\mathrm{\&Sigma;}}}{a}_1^2\left(j\right)x_1(n-j+{\mathrm{\&tau;}}_1^2)+e_{n2}---\left(5\right)$

In the case of majority (

And

), please note that each above-mentioned formula is independently, and only recover a voice-grade channel at every turn, other channel that is used to recover then be assumed to be not impaired, in all impaired case of two channels, above-mentioned formula still can use, but performance can be very near the single channel model.

Work as τ, the time migration constant was greater than 1 o'clock, and with the following data of using from other channel, the automatic renewal of energy has been satisfied in the use of following data, when two channels when being similar, rejuvenation is approximate fare-you-well, on the contrary, and when two channels are dissmilarity, its computing method are to be produced by the single channel autoregressive model, therefore, in the middle of the case of majority, can expect that the double-channel autoregressive model surpasses the single channel autoregressive model.

Please pay special attention to, in order to simplify, only suppose simply greater than

And

A real time delay buffer to replace an importing that postpones, therefore following data point can obtain under real-time state.

Can implement double-channel LS autoregression interpolation a son, particularly the content spy in 86 pages and 87 pages via described single channel LS autoregression interpolation that reaches described in the textbook of revising above-mentioned reference incorporates into reference to (incorporated by reference) at this.

Formula (4) is rewritten as x ₁=Ga+e, wherein N is the total length of audio section; Present x ₁And e is (N-P ₁ ¹* 1) hurdle vector; A is for comprising autoregressive coefficient

And

The hurdle vector.

Via finding x ₁And x ₂Between maximum intercorrelation value and estimate that now G is for comprising x ₁And time skew x ₂A matrix:

Use variance to estimate to find the solution the least square estimation of a:

a ^cov＝(G ^TG) ^-1G ^Tx ₁

Rewrite above-listed formula as follows:

e＝Ax

Wherein present x is for comprising x ₁And x ₂Chain (concatenated) hurdle vector of value, and A is for comprising coefficient

And

Suitable matrix:

The position of losing sample can be specified in the block via a detection converting vector i, according to known and unknown sample x _-(i)And x _(i)Add that arrangement back matrix U and K then can be divided into block and the data sample x of N:

X＝Ux _(i)+Kx _(i)

Wherein, definition A _(i)=AU and A _-(i)=AK, and double-channel LS autoregression interpolation separate for:

$x_{\left(i\right)}^{\mathrm{<figure-callout\; id="2"\; label="LS"\; filenames="S2006800280720E00011.png"\; state="\{\{state\}\}">LS</figure-callout>}2\mathrm{CH}}=-{\left(A_{\left(i\right)}^T{A}_{\left(i\right)}\right)}^{-1}{A}_{\left(i\right)}^T{A}_{-\left(i\right)}{x}_{-\left(i\right)}$

Therefore, in the DAB receiver shown in first figure, when the DAB packet drop, DSP 6 receives digital packet x ₁, digitized audio x ₂An and packet loss pointer i, and use these signals to produce the interpolation audio frequency, if packet loss pointer i has extended beyond a long duration, then by digitized audio that interpolation replaced with no longer valid, no matter nothing else, the advantage of described interpolation are that the initiating terminal in the gap in receives DAB divides into groups and end end place provide seamlessly transit (transition) between DAB and FM/AM sound signal.

It is laudable that buffer 3 and 5 can provide the past of audio samples and future value, but DSP 6 access DAB grouping and digitized simulation audio frequency are so that the length in the gap that the DAB that carries out described interpolation and allow decision to be received divides into groups, just, the number of the continuous blank in buffer (or not marking) position, this will make DSP 6 carry out necessary routine with will not impaired DAB packet delivery described D/A converter 7 extremely, and any follow-up analogue audio frequency treatment circuit, the DAB that is used to use double-channel autoregressive model interpolation to lose divides into groups, or when digitized simulation audio delivery during to described D/A converter 7, the DAB grouping is long-time loses.

Second figure has showed the DAB signal of being deciphered under condition of acceptance of poor quality, from second figure (a), can find out, the short DAB grouping that is loss at interval among the figure, these have caused audio frequency processing at interval usually, as: " click (click) ", " serge bat sound (pops) ", " bust sound (drop-off) " or " quiet (silences) ", for alleviating these effects, in having the multichannel autoregressive model of FM/AM channel, use the digitized FM/AM audio frequency among second figure (b), then receive in mode so if packet drop has surpassed a long duration: the interpolation of the center of this period is with no longer valid, DSP will use the FM/AM audio frequency to replace the DAB signal, in any case, two end points in gap one of them, the recovery of DAB grouping is to be used for guaranteeing seamlessly transitting between DAB and FM/AM sound signal.

Though the present invention be consult and use Yu Yike receive the impartial broadcasting of FM/AM the DAB receiver the enforcement situation and disclose, but the present invention is not limited in above-mentioned application, the present invention can use at any audio frequency with two or more correlated channels and receive, for instance, in the record of the three-dimensional sound, can given sample of signal x via L channel and R channel ₁And x ₂, or around in the audio record, will more channel occur and be used for recovering the audio frequency of impaired sound channel.

Below propose the present invention recover impaired sound signal method further can the applicable non-inventory of enumerating fully:

1. the present invention can use the back-up of inferior identical media, as: old record, cartridge could cassette tape etc. recover impaired Digital Media.

2. the present invention can be used for recovering multilingual Digital Media.If one of them of several media fragments on the DVD medium suffers damage, as: impaired or friction when using in the manufacturing, use the second not impaired channel of different language to can be used to recover first channel.

3. the present invention can use not impaired simulated television/video signal audio fragment to recover an impaired Digital Television/video signal audio fragment.

4. the present invention can be used for recovering the impaired audio file of different compressed formats, often, is to be applied to as assisted network stream audio, VOIP, to have identical content but the audio file of different-format, and the fragment of different-format can be used for recovering another fragment.

5. the present invention can be used for recovering to have the ANTENN AUDIO fragment of network audio fragment, and vice versa, for example, can use the not impaired Webisode of DAB broadcasting to recover impaired radio broadcasting, also can use FM broadcasting to recover to broadcast via the DAB of network.

6. the present invention can use as the compressed version with identical audio content that backs up channel to create an intelligent wireless transmitting system, for example, if transmit two copies with different qualities with identical audio content, then can use lower bandwidth channel (for example 8bit 8kHz) to recover higher bandwidth (for example 16bit 44kHz) channel, vice versa.

7. the present invention can use as the one or more different ANTENN AUDIO transmission standard of backup channel and come mobile phone reception in the recovery room, for example, in an indoor environment, via wireless standard relatively poor or that limited to, as: amplitude modulation (AM), frequency modulation (FM), GPRS (General Packet Radio Service) (GPRS), bluetooth (bluetooth) etc., sound that is transmitted or snatch of music, the general length that can be used for recovering impaired is apart from digital cell element wireless transmission standards, as: amplitude modulation (AM), frequency modulation (FM), global mobile communication (GSM), time division multiple access (TDMA) (TDMA), CDMA (CDMA), GPRS (General Packet Radio Service) (GPRS) and bluetooth (bluetooth) or the like, vice versa.

Hereinafter more detailed description the present invention is used for the further application that the mobile phone sound signal is recovered.

Use is described and/or the method declared in claim 1 to 4 is recovered two or more impaired mobile phone audio-source at this.

Two or more audio-source can be by two or the CDMA (CDMA), the broadband-CDMA (W-CDMA that more separate, 3G), global mobile communication (GSM) or other cell element standard base station, perhaps from two or the derivation of more transmission signals simultaneously, alternatively, described two or more audio-source also may come freely to use the wireless signal of some reflections of the single base station of RAKE receiver.

A RAKE receiver uses several baseband correlators individually to handle several multi-path signal compositions, promptly has identical content but the signal that postponed by interdependent period of path, improves communication reliability and performance in conjunction with the output of these correlators to reach.

In IS-95, the two all uses the RAKE reception technique base station and mobile receiver, each correlator in the RAKE receiver all is called RAKE receiver rake and refers to (finger), the output that this base station incoherence ground refers in conjunction with its RAKE receiver rake, just these outputs are added in the middle of the power, the output that mobile receiver as one man refers in conjunction with its RAKE receiver rake just adds to these outputs in the middle of the voltage.Present mobile receiver has three RAKE receiver rakes to refer to usually, but the base station receiver then has four or five rakes to refer to according to equipment manufacturers, there are two kinds to be mainly used to refer to the method exported at present in conjunction with receiver rake, first method is each output of weight equably, so also be called equal gain combining (equal gain combining), second method is estimated weight for using data, signal to noise ratio (S/N ratio) (the SNR that merges back output with maximization, signal-to-noise ratio), this technology for height ratio in conjunction with (maximal-ration combining), in fact, the performance of these two kinds of combination technologies is normally rough identical.

But adopt the mobile phone of RAKE receiver architecture can obtain identical information is arranged on paper but the mutual to each other a plurality of wireless signals that postpone in correlator output place, nominally demodulation and decipher these wireless signals to be created in corresponding a plurality of voice-grade channels with same audio signal, if there is one or more voice-grade channels impaired, can use aforesaid multichannel autoregressive model and recovered, aforesaid multichannel autoregressive model is modeled as a linear combination of the time scale offset portion of other voice-grade channel and described impairment signal with described impairment signal, and this will produce an improved output audio signal.

Although the output of the described correlator after signal processor configuration RAKE receiver is with receiving demodulation and decoding is to provide multitone channel frequently, and aforesaid multichannel autoregressive model handled be applied to these voice-grade channels outputs and be one and implement easily, but be not to use so RAKE receiver to improve sound signal.

What need is can receive two or more sound signal version and use the multichannel autoregressive model to merge the mobile phone of these two versions.

Other can receive the device of many inhibit signals, as provide a plurality of antennas, each antenna provides a separate entity radio-frequency path or the time-division multitask on the single entities radio-frequency path again, and so the diversity receiver is widely known by the people in wireless communication field.

The 3rd figure has showed case study on implementation according to mobile phone proposed by the invention in the mode of block diagram.Shown in the 3rd figure, mobile phone has one first antenna 30, provide one first received signal to a RF layer 31 from described first antenna 30, also described first signal is offered one the one AF layer 32 at described first signal of described RF layer 31 a places demodulation, one second antenna 33 provides one second received signal to the 2nd RF layer 34, offer one the 2nd AF layer 35 in described the 2nd described secondary signal of RF layer 34 places demodulation and with described secondary signal, again these two first and second received signals are provided to first and second input of a signal processor 36.Signal processor 36 can be a microprocessor or a programmable digital signal processor can carrying out the multichannel autoregressive model, takes from other voice-grade channel through the past sample weights of the sample of time skew with recovering in order to use and loses or impaired audio samples in first sound signal.Please pay special attention to, although the case study on implementation among the 3rd figure has used two channels, also can be more than two channels, as the case of RAKE receiver.

A kind of method that mobile phone with two separate audio channels is provided provides the phone of the SIM card of supporting two or more numbers (sign), and use multilateral accord (multi-partyprotocol) to come binding signal, support the SIM card of several numbers can obtain at present.

The 4th figure has showed so arrangement in the mode of block diagram, shown in the 4th figure, have two signs 41, one mobile phone 40 of 42 and one signal processor 43 transfers to base station 44 and receives from base station 44, another mobile phone 45 transfers to base station 44 and receives from base station 44, described another mobile phone 45 uses multilateral accord that transmission is provided to two paths 41 and 42, base station 44 is from described signal to two sign 41 and 42 of mobile phone 45 transmission, described two signs 41 and 42 receive described signal dividually, and two effect voice-grade channels multichannel autoregressive model with aforementioned use formula 4 and 5 in signal processor 43 is handled, though two voice-grade channels are only arranged, herein if there is more voice-grade channel then can use formula 2 and 3.

Another case study on implementation is to use the mobile phone that can use two (perhaps more) various criterions to receive/transmit, as GSM and CDMA, the output of these GSM and CDMA Channel can be used the multichannel autoregressive model and be merged, and this is that the hypothesis associate audio signal is to use two (perhaps more) standards and via Network Transmission.

The combination of the transmission that other are various all might be the potential case study on implementation of the present invention, its condition is for producing two associate audio signal so that satisfy the needs that the multichannel autoregressive model is carried out, and the method that can use multichannel autoregressive model proposed by the invention is improved the quality of the sound signal that is produced.

The present invention further is being applied as the handover (handover) that is used between faint reception period or mobile phone cell element aspect the mobile phone communications, wherein mobile phone may be connected to two different base stations simultaneously during handover, this can use the time-division multi-tasking or finish via having the radio-frequency channel and the antenna that separate on two entities, one of them of antenna transfers to one first base station and receives from described first base station, and antenna one of other transfers to one second base station and receive from described second base station, when when two base stations are receiving, mobile phone also will receive in the same manner the wireless signal of coding or similar sound signal, therefore when the mobile phone demodulation and when deciphering these sound signals that receive, with these signal application to described signal processor processes, these signals will be handled with aforesaid multichannel autoregressive model in described signal processor, so, two base station places are being arranged, particularly, when from two base stations or the signal weakening of one of them, these two sound signals can be applied to implement the multichannel autoregressive model described signal processor to improve the quality of these signals.

It is aspect VOIP phone or a kind of wireless compatible authentication (WI-FI) phone, to improve audio quality and appreciable reception that of the present invention another further used.Shown in the 5th figure, one the one VOIP (or WI-FI) phone 50 and one the 2nd VOIP (or WI-FI) phone 51 are via 52-1 to the 52-n communication of a plurality of paths, phone 50 has and has plurality of networks port 53-1 to 53-n, and phone 51 has plurality of networks port 54-1 to 54-n, this two group networks port can be set up communication via path 52-1 to 52-n, phone 50 has code translator 55-1 to 55-n, it is to obtain input signal from aforesaid network port 53-1 to 53-n, and to make these conversion of signals be the respective audio signal, similarly phone 51 has code translator 56-1 to 56-n, it is to obtain input signal from aforesaid network port 54-1 to 54-n, and to make these conversion of signals be the respective audio signal, and aforementioned code translator 55-1 to 55-n (with code translator 56-1 to 56-n) can be corresponding code translator or the code translator that is configured to the pipeline shape.

In each phone 50 and 51, embed a proper signal processing configuration 57 and 58 respectively, its sound signal after aforementioned code translator 55-1 to 55-n and 56-1 to 56-n reception decoding, and the method according to one of them described recovery one impaired sound signal of claim 1 to 3 is carried out in design, and producing a sound signal of recovering at separately output terminal, signal Processing configuration 57 and 58 will typically comprise through the suitable digital processing units of design.

Therefore, similar audio-frequency information with the form of audio packet via transmitting between the network path of two between two or the more VOIP phone or more separation, the path of each separation all can see through the VOIP phone and connect the different network ports, different hotspot, different port on identical hotspot/approach point, perhaps different network service providers, these voice-grade channels can include the grouping of different CODEC/ standards discretely, the grouping that adds different white noises, the background noise of packet loss or noise, grouping with different average amplitude, perhaps having different time aims at, skew or the grouping that postpones are deciphered each voice-grade channel discretely and are used the multichannel autoregressive model that reaches declaration as previously mentioned to recover sound signal.

Use can be via replacing and merge two or more conversation improve a VOIP phone to another VOIP phone audio quality according to method of the present invention, when receiving quality when being of poor quality, hanging on placement one additional session under the situation of existing conversation, therefore, all conversations are combined with multichannel autoregressive model recovery algorithms to improve reception.

Port 53-1 to 53-n and 54-1 to 54-n can obtain the output of a unity coder, wherein said unity coder with audio-frequency signal coding to transfer to other phone via different transmission path, alternatively, a plurality of scramblers can be provided, each scrambler is corresponding to each port, in this case, may be according to different transmission standards via the different audio-frequency signal codings that transmission path transmitted, these different transmission paths may separate on space, time and frequency.

Certainly, each phone also will comprise the conventional unit that operation is indispensable, as be used for conversion at signal processor 57 (58), the digital analog converter of note amplifier and the sound signal that micropkonic output produced, also be in order to transmit, a suitable microphone, a digital analog converter and a scrambler, be used for sound signal is applied to the network port, the persond having ordinary knowledge in the technical field of the present invention knows these all parts and interconnecting between them, these are not the part of progressive notion of the present invention, so do not show that in graphic this part and this partly also will no longer be given unnecessary details.

This multichannel autoregressive model recovery algorithms can be applied to the phone of VOIP or WI-FI like this, support some standards, for example, access (UMA)/universal access network (GAN)/GSM/3G/CDMA/ phone is moved in the non-permission of the WI-FI/ of standard more than, suppose to decipher each conversation individually, described algorithm can be via improving audio quality in conjunction with the conversation of while various criterion, described algorithm also can aid under the situation of hanging on conversation and switch in a standard and another standard room, the conversation of the standard in conversation of placement earlier (for example, WI-FI) to the conversation in another standard another conversation (for example, GSM), make via multichannel autoregressive model algorithm this switching can be smoothly after, just conversation of poor quality can be omitted.

In addition, algorithm of the present invention can similar methods, under the situation of not interrupting existing conversation, aids in the handover between a standard cell element benchmark and another standard cell element benchmark.

The present invention also can be applied to existing WI-FI standard and mobile phone standard in many ways or the Three-Way Calling agreement, to assist the multichannel recovery algorithms.

Claims (6)

1. a method of recovering impaired sound signal comprises the following steps:

Described impaired sound signal is inputed to first channel;

Import one or more associate audio signal to one or more channels; And

Use the multichannel autoregressive model to recover described impaired sound signal, wherein said multichannel autoregressive model is modeled as described impaired sound signal the linear combination of time migration part of the convergent-divergent of described one or more associate audio signal and described impaired sound signal, and wherein said multichannel autoregressive model uses following formula to carry out interpolation:

$x_1\left(n\right)=\underset{i=1}{\overset{P_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{P_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1);$

$+\underset{k=1}{\overset{P_3^1}{\mathrm{\&Sigma;}}}{a}_3^1\left(k\right)x_3(n-k+{\mathrm{\&tau;}}_3^1)+...+e_{n1};$

$x_m\left(n\right)=\underset{i=1}{\overset{P_m^m}{\mathrm{\&Sigma;}}}{a}_m^m\left(i\right)x_m(n-i)+\underset{j=1}{\overset{P_1^m}{\mathrm{\&Sigma;}}}{a}_1^m\left(j\right)x_1(n-j+{\mathrm{\&tau;}}_1^m);$

$+\underset{k=1}{\overset{P_2^m}{\mathrm{\&Sigma;}}}{a}_2^m\left(k\right)x_2(n-k+{\mathrm{\&tau;}}_2^m)+...+e_{\mathrm{nm}};$

Wherein:

x ₁, x ₂..., x _mBe channel 1,2 ..., the output of m;

Be channel i, the autoregressive coefficient between j;

Be channel i, the exponent number of the autoregressive coefficient between j;

Be channel i, time migration constant between j;

e _N1, e _N2..., e _NmBe the white noise input.

2. method according to claim 1 wherein has two channels and described multichannel autoregressive model to use following formula to carry out interpolation:

$x_1\left(n\right)=\underset{i=1}{\overset{P_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{P_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)+e_{n1};$

$x_2\left(n\right)=\underset{i=1}{\overset{P_2^2}{\mathrm{\&Sigma;}}}{a}_2^2\left(i\right)x_2(n-i)+\underset{j=1}{\overset{P_1^2}{\mathrm{\&Sigma;}}}{a}_1^2\left(j\right)x_1(n-j+{\mathrm{\&tau;}}_1^2)+e_{n2}.$

3. method of digital audio broadcasting sound signal of remaking comprises step:

Receive and decipher the digital audio broadcasting sound signal to produce the grouping of corresponding digital analogue audio frequency;

Receive analog broadcast signal, described analog broadcast signal and described digital audio broadcasting sound signal are broadcasted simultaneously, and described analog broadcast signal contains identical broadcasting plan with described digital audio broadcasting sound signal;

The described analog broadcast signal of demodulation is with from wherein producing simulated audio signal;

Described simulated audio signal is converted to the digitized simulation sound signal;

Input digit audio broadcasting audio packet and described digitized simulation audio packet to corresponding buffer storage compensates mistiming between described digital audio broadcasting audio packet and described digitized simulation audio packet so that suitable delay to be provided;

Detect the digital audio broadcasting grouping of impaired or gaps and omissions; And

Use the multichannel autoregressive model and the digitized simulation voice-grade channel of described digital audio broadcasting to come interpolation loss or impaired digital audio broadcasting grouping to divide into groups to recover described impaired digital audio broadcasting, wherein said multichannel autoregressive model uses following formula to divide into groups with the described digital audio broadcasting of interpolation:

$x_1\left(n\right)=\underset{i=1}{\overset{P_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{P_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)+e_{n1};$

X wherein ₁Output and x for channel 1 ₂Be the output of channel 2, For the autoregressive coefficient of 1,1 of channel and

Be the autoregressive coefficient of 1,2 of channel,

Be the exponent number of the autoregressive coefficient of 1,1 of channel,

Be the exponent number of the autoregressive coefficient of 1,2 of channel,

Be 1,2 time migration constant of channel, e _N1Be the white noise input.

4. method according to claim 3, wherein when described loss or impaired digital audio broadcasting grouping than a preset period of time for long, replace one with described digitized simulation sound signal and divide into groups than the digital audio broadcasting in the center section of long duration.

5. device that recovers impaired sound signal comprises:

Be used for described impaired sound signal is inputed to the device of first channel;

Be used to import the device of one or more associate audio signal to one or more channels; And

Be used to use the multichannel autoregressive model to recover the device of described impaired sound signal, wherein said multichannel autoregressive model is modeled as described impaired sound signal the linear combination of time migration part of the convergent-divergent of described one or more associate audio signal and described impaired sound signal, and wherein said multichannel autoregressive model uses following formula to carry out interpolation:

$x_1\left(n\right)=\underset{i=1}{\overset{P_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{P_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1);$

$+\underset{k=1}{\overset{P_3^1}{\mathrm{\&Sigma;}}}{a}_3^1\left(k\right)x_3(n-k+{\mathrm{\&tau;}}_3^1)+...+e_{n1};$

$x_m\left(n\right)=\underset{i=1}{\overset{P_m^m}{\mathrm{\&Sigma;}}}{a}_m^m\left(i\right)x_m(n-i)+\underset{j=1}{\overset{P_1^m}{\mathrm{\&Sigma;}}}{a}_1^m\left(j\right)x_1(n-j+{\mathrm{\&tau;}}_1^m);$

$+\underset{k=1}{\overset{P_2^m}{\mathrm{\&Sigma;}}}{a}_2^m\left(k\right)x_2(n-k+{\mathrm{\&tau;}}_2^m)+...+e_{\mathrm{nm}};$

Wherein:

x ₁, x ₂..., x _mBe channel 1,2 ..., the output of m;

Be channel i, the autoregressive coefficient between j;

Be channel i, the exponent number of the autoregressive coefficient between j;

Be channel i, time migration constant between j;

e _N1, e _N2..., e _NmBe the white noise input.

6. device of digital audio broadcasting sound signal of remaking comprises:

Be used to receive and decipher the digital audio broadcasting sound signal to produce the device of corresponding digital analogue audio frequency grouping;

Be used to receive the device of analog broadcast signal, wherein said analog broadcast signal and described digital audio broadcasting sound signal are broadcasted simultaneously, and described analog broadcast signal contains identical broadcasting plan with described digital audio broadcasting sound signal;

Be used for the described analog broadcast signal of demodulation with from wherein producing the device of simulated audio signal;

Be used for described simulated audio signal is converted to the device of digitized simulation sound signal;

Be used for input digit audio broadcasting audio packet and described digitized simulation audio packet device, compensate mistiming between described digital audio broadcasting audio packet and described digitized simulation audio packet so that suitable delay to be provided to corresponding buffer storage;

Be used to detect the device of the digital audio broadcasting grouping of impaired or gaps and omissions; And

Be used to use the multichannel autoregressive model of described digital audio broadcasting and digitized simulation voice-grade channel to come the device that interpolation is lost or impaired digital audio broadcasting is divided into groups, to recover described impaired digital audio broadcasting grouping, wherein said multichannel autoregressive model uses following formula to divide into groups with the described digital audio broadcasting of interpolation:

$x_1\left(n\right)=\underset{i=1}{\overset{P_1^1}{\mathrm{\&Sigma;}}}{a}_1^1\left(i\right)x_1(n-i)+\underset{j=1}{\overset{P_2^1}{\mathrm{\&Sigma;}}}{a}_2^1\left(j\right)x_2(n-j+{\mathrm{\&tau;}}_2^1)+e_{n1};$

X wherein ₁Output and x for channel 1 ₂Be the output of channel 2,

For the autoregressive coefficient of 1,1 of channel and

Be the autoregressive coefficient of 1,2 of channel,

Be the exponent number of the autoregressive coefficient of 1,1 of channel,

Be the exponent number of the autoregressive coefficient of 1,2 of channel,

Be 1,2 time migration constant of channel, e _N1Be the white noise input.

Images