CN101370056B - Digital audio automatic gain control method and its system - Google Patents

Abstract

The invention discloses a automatic gain control method for digital audio frequency comprising following steps: (1) computing instant power of input signal, (2) detecting if signal of step 1 is background noise, (3) updating gain coefficient g(n) according to power information of input signal of step 1 and self-adaptive filtering study step update, (4) producing new output signal according to new gain coefficient of step 3, (5) computing instant power of output signal, (6) shaping gain (7) performing saturation control and output digital audio frequency signal. The inventive digital audio frequency automatic gain system comprises: a first power computing module, self-adaptive filter module, a second power computing module and gain shaping module. The inventive method and system can automatically control gain for digital audio frequency signal, enable output signal more balanced and smooth, and can be used in voice communication and recording system.

Description

Digital audio automatic gain control method and system thereof

Technical field

The present invention relates to a kind of automatic gain control, relate to a kind of DAB automatic gain control that voice communication, audio/video conference and digital recording etc. are used that is applied to more specifically.

Background technology

(Automatic Gain Control is to be used for digital audio and video signals is done the automatic dynamic gain controlling AGC), makes the user-defined echo signal level that trends towards of signal level finally to obtain the harmonious voice of balance for DAB automatic gain control.This technology generally is used for voice communication and recording system, and concrete the application comprises audio/video conference, the networking telephone, digital recording etc.Distortion does not to a certain degree appear in the signal after digital audio automatic gain control method can guarantee to gain efficiently, and can not gain to background noise.Existing digital AGC technology generally is through comparing the amplitude equalizing value and the predefined AGC thresholding R of phase of history speech output signal, with the variation of ride gain, if amplitude equalizing value surpasses thresholding R, then increases gain, then reduces gain on the contrary.This type gain method is to become in non-linear, time; Owing to can not combine the variation of current demand signal energy and the variation of historical signal energy to come adjustment gain automatically neatly; Can make that so the bigger signal gain of portion of energy is excessive; Cause overflowing, the less signal of portion of energy but gains inadequately, can destroy the continuity of signal and introduce uncomfortable noise and the signal energy of overflowing is done the peak clipping processing simply.And existing technology is seldom considered the situation that has background noise in the gain control process.

Summary of the invention

The technical problem that the present invention will solve provides a kind of digital audio automatic gain control method, and distortion does not to a certain degree appear in the signal after it can guarantee to gain, and background noise is not gained.For this reason, the present invention also will provide a kind of DAB AGC system.

For solving the problems of the technologies described above, digital audio automatic gain control method of the present invention, it may further comprise the steps:

(1) the instantaneous power P of calculating input signal x (n) _x(n);

(2) whether be background noise according to the signal in the power detection step (1), the instantaneous power P of input signal in the step (1) _x(n) be lower than and preset power P _Min, be background noise, directly carry out saturated control and output; Instantaneous power P _x(n) greater than presetting power P _Min, get into next step;

(3) upgrade according to the power information and the adaptive-filtering study step-length of the input signal of step (1), upgrade gain coefficient g (n);

(4) utilize new gain coefficient in the step (3), produce new output signal y (n)=g (n) x (n);

(5) calculate the instantaneous power P that exports signal y (n) _yAnd utilize the instantaneous power P of estimation (n), _y(n) with the echo signal power P of preassignment _RefAsk irregular energy, and it is fed back to the gain coefficient lastest imformation of step (3) as next signal;

(6) according to the power P of input signal _x(n), the shaping that gains;

(7) carry out saturated control and outputting digital audio signal.

DAB AGC system of the present invention is used for the automatic gain digital audio and video signals, and it comprises one first power computation module, is used to calculate the power of input signal; One adaptive-filtering module is used to receive the signal from said first power computation module and output, upgrades gain coefficient, and produces new output signal according to new gain coefficient; One second power computation module is used to calculate the power from the adaptive-filtering module output signal, and feedback signal to adaptive-filtering module, as the gain coefficient lastest imformation of next signal; One gain Shaping Module is used for the signal from adaptive-filtering module output is carried out the gain coefficient adjustment, with the excessive situation of reduction gain coefficient with strengthen the too small situation of gain coefficient, the signal of output after the shaping that gains.

DAB AGC system of the present invention and method; Through upgrading the gain coefficient of the digital audio and video signals of importing; And produce new output signal according to the gain coefficient of this renewal, to the shaping that gains of new output signal, excessive gain and strengthen too small gain effect weakens; The digital audio and video signals that distortion to a certain degree do not occur and background noise is not gained after output one gain makes voice after the processing reach perfect distance and holds weighing apparatus level with both hands.

Description of drawings

Below in conjunction with accompanying drawing and embodiment the present invention is done further detailed explanation:

Fig. 1 is the sketch map of DAB adaptive gain control system of the present invention;

Fig. 2 is the flow chart of DAB adaptive gain control method of the present invention;

Fig. 3 is the sketch map of an instance input signal of the present invention and output signal;

Fig. 4 changes sketch map for the embodiments of the invention gain coefficient;

Fig. 5 is the sketch map of another instance input signal of the present invention and output signal.

Embodiment

Fig. 1 is the sketch map of DAB AGC system of the present invention, is divided into following functional module:

(1) first power computation module 101.This module is used to calculate the power P of input signal x (n) _x(n), wherein n is the time sequence number.Computing formula can be

P _x(n)＝αP _x(n-1)+(1-α)x ²(n) (1)

Wherein α is a smoothing factor, and α is big more, and then variable power is mild more; α is more little, and then power changes with the rapid variation of signal.For the fluctuation that better reflects signal and be unlikely to violent shake, generally select 0＜α＜＜1.

(2) the adaptive-filtering module 103.This module comprises three sub-module: adaptive-filtering study step-length update module, filter factor update module and new output calculated signals module.Filter factor, promptly the more new formula of gain coefficient g (n) does

g(n)＝g(n-1)(1+μ(n)P _x(n)(P _ref-P _y(n-1))) (3)

Wherein μ (n) is the study step-length of adaptive-filtering, P _RefBe a constant, representative of consumer expected objective signal energy level.Adaptive-filtering study step-length more new formula does

$\mathrm{\μ}\left(n\right)=\frac{{\mathrm{\μ}}_0}{\mathrm{\γ}+P_x\left(n\right)}---\left(4\right)$

μ ₀Be the constant between 0 to 1, μ ₀Value have influence on the convergence rate and the stability of adaptive algorithm, bigger μ ₀Can cause restraining faster, but the less stable after the convergence, less μ ₀Can make that algorithmic statement is slack-off, but the stability after the convergence better, in order to consider convergence of algorithm speed and stability, operated by rotary motion μ simultaneously ₀Between 0 to 0.1.γ approaches 0 the disturbance factor, the constant of scope between 0 to 0.1, and the purpose of γ is in order to prevent that denominator from being 0, so generally get the value below 0.1 in this formula.Utilize new gain coefficient, produce new output signal y (n), computing formula does

$y\left(n\right)=\left\{\begin{array}{cc}g\left(n\right)x\left(n\right),& P_x\left(n\right)\&GreaterEqual;P_{\min }\\ x\left(n\right),& P_x\left(n\right)<P_{\min }\end{array}---\left(5\right)\right.$

(3) second power computation module 102.This module is used to calculate y (n) power P of output signal _y(n), wherein n is the time sequence number.Computing formula can be

P _y(n)＝αP _y(n-1)+(1-α)y ²(n) (2)

0＜α＜＜1 wherein.And result of calculation fed back to the gain coefficient lastest imformation of adaptive-filtering module 103 as next signal.

(4) gain Shaping Module 104.Signal from 103 outputs of adaptive-filtering module is carried out the gain coefficient adjustment, and computing formula does

Y (n)=K ₁Y (n) (+K ₂(P _a-P _x(n))) (work as P _a＜P _x(n)＜P _b) (6)

P wherein _MinBe the minimum signal power threshold values of gain, instantaneous power is lower than P _MinSignal be background noise, do not carry out gain controlling.K ₁And K ₂Be constant coefficient, 0.5≤K ₁≤1,1≤K ₂≤8, K ₁Effect be as required the overall magnitude of signal to be certain adjustment, K ₁Value is 1 o'clock, and signal integral body does not gain, K ₂Effect according to the fluctuation of signal, the adjustment of certain ratio, K are done in the part of signal ₂Value can obtain preferably effect during integer 1 between 8.P _aAnd P _bThe lower limit and the upper limit for the signal energy that needs shaping.The purpose of gain shaping is two kinds of situation when adjusting gain coefficient, to occur in order to overcome adaptive filter algorithm: (1) is when gain coefficient is excessive; The energy rank of input signal is also excessive, and it is excessive at this time may to gain, even signal overflows; At this time need do the gain finishing, the reduction gain; When (2) gain coefficient was too small, the energy rank of input signal was too small, and at this time gain effect is obvious inadequately, needed the finishing coefficient, strengthened gain effect.Adaptive filter algorithm has certain robustness, and gain coefficient (weight) is not to change along with the rapid variation of the waveform of signal fully, and the amplitude of coefficient adjustment has certain delay, therefore, need do corresponding gain finishing according to the fluctuation of signal.

Fig. 2 is the flow chart of the inventive method, and it may further comprise the steps:

Step 401, the instantaneous power P of estimation input signal x (n) _x(n), the power calculation method is taked the method for above-mentioned formula (1).

Step 402, whether the power detection of calculating according to step 401 is background noise, the detection method of background noise takes to set P _MinBeing the minimum signal power threshold values, being worth less than this like current demand signal power, then is the direct execution in step 407 of background noise, otherwise order execution in step 403.

Step 403 is upgraded according to power information of calculating and adaptive-filtering study step-length, upgrades gain coefficient g (n), and more new formula is taked above-mentioned formula (3), and the study step-length of adaptive-filtering is upgraded and taked above-mentioned formula (4).

Step 404 is utilized new gain coefficient, produces new output signal, and computing formula is y (n)=g (n) x (n).

Step 405 is calculated the instantaneous power P that exports signal y (n) _y(n), the power calculation method is taked the method for above-mentioned formula (2).And the instantaneous power P of utilization calculating _y(n) with the echo signal power P of user's preassignment _RefAsk irregular energy, and it is fed back to the gain coefficient lastest imformation of step 403 as next signal.

Step 406 is according to the power P of input signal _x(n), the finishing that gains further improves voice quality, and the computational methods of gain shaping are taked the method for above-mentioned formula (6).

Step 407 is carried out saturated control and output signal y (n), and the purpose of saturated control is to overflow for control signal; If input signal is the 16bitPCM code stream; Then saturated control makes the scope of 16bit output signal be [32768 ,+32767], and the output greater than 32767 is set to 32767; Output less than-32768 is set to-32768, and other output is constant.

When importing new voice signal x (n), repeat above step.

Fig. 3 has provided and has used automatic gain control (AGC) method of the present invention that the voice in one period 3 second are carried out process result figure.Fig. 4 has provided the change curve of AGC to the gain coefficient of this section voice; Gain coefficient (original gain coefficient) that does not pass through shaping and the contrast of passing through the gain coefficient of shaping have been write down; It is thus clear that after the gain trimming, the voice signal of output balance more is harmonious; And when background noise, gain coefficient equals 1.In this instance, the AGC selection of parameter is following: the α in formula (1) and the formula (2) gets 0.05, μ in the formula (4) ₀Get 0.03125 and 0.001 respectively with γ, K in the formula (6) ₁, K ₂, P _aAnd P _bGet 0.5,2,0.001 and 0.1 respectively, P in the formula (5) _MinGet 0.0008, represent the power of background noise, P in the formula (3) _RefGet 0.15, represent echo signal power.

Fig. 5 has provided and has used automatic gain control (AGC) method of the present invention that the voice in 6 seconds of another section are carried out process result figure.In this instance, K in the formula (6) ₁, K ₂Get 0.8 and 4 respectively, the echo signal power P in the formula (3) _RefGet 0.2, other values of parameters are identical with the value of instance among Fig. 3.Can find out that from Fig. 3 and Fig. 5 gain effect is more satisfactory, and the background noise part is not gained.

Claims (5)

1. a digital audio automatic gain control method is used for the automatic gain digital audio and video signals, it is characterized in that it may further comprise the steps:

(1) the instantaneous power P of calculating input signal x (n) _x(n);

(2) whether be background noise according to the signal in the power detection step (1), the instantaneous power P of input signal in the step (1) _x(n) be lower than and preset power P _Min, be background noise, directly carry out saturated control and output; Instantaneous power P _x(n) greater than presetting power P _Min, get into next step;

(3) according to the power information of the input signal of step (1) and the adaptive-filtering step-length information of renewal, upgrade gain coefficient g (n);

A kind of variable step update method based on variable power is taked in the renewal of said step (3) gain coefficient, and its computing formula is g (n)=g (n-1) (1+ μ (n) P _x(n) (P _Ref-P _y(n-1))), wherein μ (n) is the step-length of adaptive-filtering, P _RefBe a constant, representative of consumer expected objective signal energy level; The step-length of adaptive-filtering more new formula does μ ₀Value is between 0 to 1, and γ approaches 0 the disturbance factor, and span is between 0 to 0.1;

(4) utilize new gain coefficient in the step (3), produce new output signal y (n);

(5) calculate the instantaneous power P that exports signal _yAnd feed back to the gain coefficient lastest imformation of step (3) (n), as next signal;

(6) according to the power P of input signal _x(n), the shaping that gains;

The computing formula of said step (6) gain shaping is: y (n)=K ₁Y (n) (1+K ₂(P _a-P _x(n))) (work as P _a＜P _x(n)＜P _b), K wherein ₁And K ₂Be constant coefficient, 0.5≤K ₁≤1,1≤K ₂≤8, P _aAnd P _bThe lower limit and the upper limit for the signal energy that needs shaping;

(7) carry out saturated control and outputting digital audio signal.

2. control method according to claim 1 is characterized in that: the power calculation method in said step (1) and the step (5) is respectively P _x(n)=α P _x(n-1)+(1-α) x ²(n) and P _y(n)=α P _y(n-1)+(1-α) y ²(n), 0＜α＜＜1 wherein.

3. control method according to claim 1 is characterized in that: the computing formula of new output signal y (n) does in the said step (4) $y\left(n\right)=\left\{\begin{array}{cc}g\left(n\right)x\left(n\right),& P_x\left(n\right)\&GreaterEqual;P_{\mathrm{Min}}\\ x\left(n\right),& P_x\left(n\right)<P_{\mathrm{Min}}\end{array}\right..$

4. a DAB AGC system is used for the automatic gain digital audio and video signals, it is characterized in that: it comprises one first power computation module, is used to calculate the instantaneous power P of input signal x (n) _x(n); One adaptive-filtering module is used to receive the signal of first power computation module, adopts and upgrades gain coefficient g (n) based on the variable step update mechanism of input signal instantaneous power, and produce new output signal y (n) according to new gain coefficient; One second power computation module is used to calculate the instantaneous power P from the adaptive-filtering module output signal _y(n), and feedback signal power to adaptive-filtering module, as the gain coefficient lastest imformation of next signal; One gain Shaping Module is used for the signal from adaptive-filtering module output is carried out the gain coefficient adjustment, with the excessive situation of reduction gain coefficient with strengthen the too small situation of gain coefficient, the signal of output after the shaping that gains;

The computing formula of said variable step update mechanism based on the input signal instantaneous power is g (n)=g (n-1) (1+ μ (n) P _x(n) (P _Ref-P _y(n-1))), wherein μ (n) is the step-length of adaptive-filtering, P _RefBe a constant, representative of consumer expected objective signal energy level; The step-length of adaptive-filtering more new formula does μ ₀Value is between 0 to 1, and γ approaches 0 the disturbance factor, and span is between 0 to 0.1;

The computing formula of said gain coefficient adjustment is: y (n)=K ₁Y (n) (1+K ₂(P _a-P _x(n))) (work as P _a＜P _x(n)＜P _b), K wherein ₁And K ₂Be constant coefficient, 0.5≤K ₁≤1,1≤K ₂≤8, P _aAnd P _bThe lower limit and the upper limit for the signal energy that needs shaping.

5. control system according to claim 4 is characterized in that: said adaptive-filtering module comprises adaptive-filtering study step-length update module, is used for upgrading adaptive learning step size mu (n) according to the signal power of first power computation module; The filter factor update module is used for according to adaptive updates step size mu (n) and a last power from the adaptive-filtering module output signal, upgrades gain coefficient g (n); And new output calculated signals module, be used for calculating new output signal y (n) according to new gain coefficient.

Images