First, because of the work of the relationship, the author since 2004 contact Echo cancellation (echo cancellation) technology, and has been engaged in a large communications enterprises with ECHO cancellation technology related work, the echo elimination of this seemingly mysterious, high-end and difficult to understand the field of technology is well known. To understand the ins and outs of echo cancellation technology, we have to mention the theory of digital signal processing as the theoretical basis of modern communication technology. First of all, digital signal processing theory has an important branch, called adaptive signal processing. In classical textbooks, the problem of echo cancellation is always discussed as a classical adaptive signal processing case. Since Echo cancellation is a classic application in textbooks, that is, there is nothing mysterious and fresh in the theoretical sense, then the difficulty of echo cancellation is there. The reason why the companies that provide the echo cancellation technology, whether chips or algorithms, are from abroad. Where is the mystery of echo cancellation technology. Second, the principle of echo cancellation from the source of communication echoes, can be divided into acoustic echo (acoustic echo) and line echo (lines echo), the corresponding echo cancellation technology is called acoustic echo cancellation (acoustic echo cancellation, AEC) and circuit back Acoustic Cancellation (line Echo cancellation, LEC). Acoustic echo is due to the fact that the sound of the speaker is repeatedly fed to the microphone (better understood) in a hands-free or conference application, because the line echo is caused by the 24-line matching coupling of the physical electronic circuit (more difficult to understand). There are two main reasons for the Echo: 1. Acoustic echo generated by spatial acoustic reflection (see below):
The man in the picture spoke, The voice signal (SPEECH1) to the lady's room, and due to the reflection of the space, the Echo Speech1 (ECHO) was returned from the microphone and the woman's voice signal (SPEECH2) was superimposed. At this time the man will hear the woman's voice superimposed their own voice, affecting the normal quality of the call. At this time in the lady's room to apply the Echo cancellation module, can cancel out the echo of the man, let the man only hear the voice of the lady. 2. Due to the line Echo introduced in the 2-4-line conversion (see below):
in ADSL modem and switch on the existence of 2-4-line conversion circuit, because there is a mismatch in the circuit, there will be a part of the signal back to form an echo. If there is no echo cancellation on the switch side, the caller will hear their own voice. Whatever the cause, the same thing is needed for voice communication terminals or voice relay switches: When you send, unwanted echoes are removed from the middle of the voice stream. Imagine how difficult it would be to separate a voice stream that has at least two voices mixed, and then remove one of them. Like a bottle of ink and a bottle of red ink poured together, and then need to extract the red ink, which I am afraid it is impossible. So it is not surprising that echo cancellation is considered a mysterious and incomprehensible technique. To be sure, it is impossible to get rid of an echo if it is only a single piece of voice signal mixed with the echo (even the most advanced blind signal separation technology is not possible). But, in fact, in addition to this mixed signal, we can get the original signal that produces the echo, although different from the echo signal. We look at the following AEC Acoustic echo cancellation block Diagram (reproduced in this picture). Figure acoustic Echo cancellation in a voice communication terminal where we can get two signals: one is a blue and red mixed signal 1, which is The actual need to send the speech and the actual unwanted echo mixed with the voice stream, the other is the dotted line signal 2, that is, the original echo-induced speech. Then everyone would say, oh, the original echo was so simple, just cut the 2 off the dotted line from the mixed signal 1. Notice that the dashed signal 2 and echo Echo are different, and direct subtraction will make the speech unrecognizable. We call the mixed signal 1 The proximal signal ne, the dotted signal 2 is called the remote reference signal Fe, and without the FE signal, echo cancellation is an impossible task, like "Ching". Although the reference signal FE and Echo are not exactly the same, there are differences, but the two are highly correlated, which is why Echo calls Echo. At the very least, the semantics of the echoes are the same as the reference signals, but they are also understood, but if you say a word, you will hear your words back again, which is rather uncomfortable. Since the FE and Echo are highly correlated and echo is Fe-induced, we can represent the echo as a FE mathematical function: Echo=f (FE). Function F is called an echo path. In acoustic echo cancellation, function F represents the physical past that the sound is reflected multiple times on the surface of walls, ceilings, etc.In line echo cancellation, function F represents the 24-line matching coupling process of electronic circuits. Obviously, the next thing we're going to do is solve the function F. Get the function F to get echo from Fe, and then subtract echo from the mixed signal 1 to achieve echo cancellation. Although Echo cancellation is a very complex technique, but we can simply describe this approach: 1, Room A of the audio conferencing system received Room B voice 2, The sound is sampled, this sample is called Echo Cancellation reference 3, then the sound was sent to room a speaker and acoustic echo Canceller 4, Room B's voice and room A's voice were picked up by the microphone of room a 5, the sound was sent to the acoustic echo Canceller, compared to the original sample, the removal of the room B sound solve the echo path function F is more difficult to express the mathematical formula. In view of the difficulty of popular expression of mathematical formula, the author has no difficulty in explaining it. The following section expresses the process of solving function F using the principle of adaptive filter. (The following can be skipped) Adaptive Filter Adaptive filter is an algorithm or device that adjusts the filter coefficients automatically and achieves the best filtering characteristics based on the estimation of the statistical characteristics of input and output signals. The adaptive filter can be continuous domain or discrete domain. The discrete-domain adaptive filter consists of a set of tapped delay lines, variable weighting coefficients and automatic adjustment coefficients. The figure shows a discrete-domain adaptive filter used to simulate the signal flow of an unknown discrete system. An adaptive filter for each of the input signal sequence x (n) values, the mean square error of output signal sequence Y (n) compared with the expected output signal sequence D (n) is minimized by a specific algorithm, i.e., the output signal sequence Y (n) approximates the desired signal sequence d (n).
The coefficients of adaptive filters designed with the minimum mean square error as the criterion can be obtained by the Wiener-Hov equation. B. A method proposed by Videro can solve the adaptive filter coefficients in real time, and the result is close to the approximate solution of the Wiener-Hov equation. This algorithm is called the least mean square algorithm or the abbreviation LMS method. By using steepest descent method, the gradient estimation of mean square error is calculated from the current moment filter coefficient vector iteration to calculate the coefficient vector of the next moment in KS as a negative number, its value determines the convergence of the algorithm, V "ε2 (n)" as the mean square error gradient estimation, Adaptive filter applied to the field of communication automatic equalization, Echo elimination In addition, antenna array beamforming, as well as other related domain signal processing parameters identification, noise cancellation, spectral estimation and so on. For different applications, only the input signal and the expected signal are different, the basic principle is the same. (The above can be skipped) The above passage shows that the Echo path function F, which needs to be solved, is the process of the convergence of an adaptive filter W (n). The input signal x (n) is Fe, and the expected signal is ECHO, and the Adaptive Filter converges W (n) is the echo path function F. After convergence, when the actual echo occurs, we pass FE through the function W (n), we can get a very accurate echo, the mixed signal directly minus Echo, get the actual need to send the voice speech, complete the Echo cancellation task. Noteworthy two points: 1, Adaptive filter Convergence phase, the expected signal is full echo, can not be mixed with speech. Because speech and FE are not related, the convergence process of W (n) is disturbed. In other words, the need for echo cancellation algorithm began to converge very quickly, the best is not too late to speak, you say on the convergence of good; after convergence, if the other person began to speak, that is, the speech mixed, this W (n) coefficient will not change, need to stabilize. 2, Echo path may be changed, once the change, echo cancellation algorithm to be able to determine, Because the adaptive filter learning to start again, that is, W (n) needs a new convergence process to approximate the new echo path function F. Basically, the above two points is a dilemma, one needs to adapt to the convergence of the filter to maintain the stability of the coefficient to ensure that the speech speech interference, another need for adaptive filterKeep the update status at all times to ensure that you can track the echo path of change. In this way, only from the mathematical algorithm level, echo elimination is already very difficult. To put it simply, the design of the echo-Cancellation adaptive filter has two conflicting characteristics, namely, fast convergence and high stability, and how to achieve both of these characteristics is the main design challenge. Through the above analysis, I believe that we have a deep understanding of the principles and techniques of echo cancellation, which is a technology that is easy to understand and difficult to implement.
This article is from the "blue Sea and Silver Sand" blog, please be sure to keep this source http://silversand.blog.51cto.com/820613/166095
