The popularity of the Internet in the world led to an explosion of innovation and application development. This trend will continue (mainly by the consumption demand) and therefore doomed to telecom operators dominated in the modern data communications network infrastructure to support voice services.

Although the data is now more than a voice communication services, service providers, the main revenue still comes from the traditional voice business. To protect this source of revenue, carriers must maintain the current quality of voice services, and create support for a variety of new data services network. Next generation network needs in these new business comes in, to take advantage of these business advantages.

Voice services in wireless, wireline, circuit-switched and packet switched networks, are essential to maintain the voice quality as a factor, is a fundamental requirement.

What is echo?

Echo is the result of sound reflection sound repetition. In the telecommunications network, if the speaker's speech to hear his voice after a delay, then there is the echo problem.

Callers hear the echo, depending on the extent of the circuit in the network delay and echo amplitude (volume). End to end delay (also known as "wait time") is the most important factor, which is produced from one end of the call to the voice sound received by the other side of the experience of time. Delay is an echo reflected from the time it takes, it is about twice the end to end delay.

 If more than 10 ms round trip delay (ms), the human ear will notice the presence of echo, resulting in voice quality.

Echo source

 Exist in the telecommunications network are two types of echo: echo and acoustic echo electrical.

Electric Echo

Electrical echo is due to analog local loop caused by impedance mismatch. For example, if you use a mixture of standard wires, or the existence of unused taps and load coils (used to extend the signal transmission from the device), will produce electrical echo. In the public switched telephone network (PSTN), the electrical echo is mainly caused by the converter.

2-wire converter to tap the local loop into two independent lines. A pair used to send the path, and the other for the receive path. Passed most of the signal converter. However, the 2-wire 4-wire loop and the differences between devices has led to a small portion of the received signal "leakage" to the sending path. Since the remote signal receiving part of the signal returned to the same time, so the speaker to hear the echo of what they said.

Electrical echo is not a local call into question, because the path of the relatively short time delay is not serious.

In today's digital networks, 2-wire 4-wire tap into the position usually analog / digital conversion of the position. Either converter or A / D conversion is the same device or different devices for two, 2-wire conversion -4 lines caused by impedance mismatch, resulting in echo.

Acoustic Echo

Acoustic echo, also known as "multi-path echo", it is by telephone between the speaker and microphone caused by acoustic coupling. In the wireless telephone and wireline telephone, or hands-free speaker phone device will appear in this echo. The problem is caused by low quality handset echo the surrounding environment (for example, cars, hotels or factories) or due to cross talk handset.

Echo propagation time

As previously discussed, when the voice circuit is greater than 20ms round trip delay, it will produce echo problem. In the PSTN, most of the delay is the propagation time from the transmission medium due to the transmission medium is, in the long-distance trunk transmission equipment and transmission devices. This delay is called "latency" or "propagation delay."

Transmission speed of the system is to decide whether to feel the echo of the important factors. This is usually in the local PSTN call can not hear the echo of the reasons, despite the fact that the existence of a local call converters and other factors as a result of electrical echo.

In the case of voice packet network, there are several other important delay (in addition to network delays outside) sources, including: processing delay (for example block, compression / decompression, etc.), switching / routing delay and buffer delay. Because of these additional sources of delay in circuit-switched network may not hear the echo of the equivalent of the packet network will be interesting. Therefore, the circuit-switched network and packet network connection between the need for high-quality echo cancellation.

What is the echo cancellation?

Echo canceller is a kind of signal from the remote device echoed locally, and to eliminate the echo after the detection device. In the circuit-switched long distance network, the echo canceller is located in big cities and long distance central office network connections. The echo canceller to eliminate long-distance network delays caused by the apparent electrical echo.
When the voice call over the acceptable limits of delay (for example in a long distance call), telecom operators will provide two ends of the call echo canceller. Protection for each eliminator remote, to prevent their receiving locally generated and returned through the network of any echo.

Echo Canceller "point" echo source (converter) direction, and away from the echo canceller from the benefit of the end of the work.

Echo canceller consists of three main functional components:
    * Adaptive filter (AF)
    * Non-linear processor (NLP)
    * Tone Detector

Adaptive filter is estimated by an echo, and a subtractor composition. Echo estimator to monitor the receive path and dynamically build a production line for the echo of the mathematical model. This line model and receiving voice flow path convolution. So get an estimate of the echo, and is input to the subtractor. This line is subtracted from the transmission path of the linear part of the echo. When a line through the adaptive filter estimate, the echo canceller on the "convergence" in the echo.

Figure 2 depicts a typical circuit-switched network, the digital echo cancellers. When the voice from the remote through the receive path (from Rin to Rout), the echo cancellers monitor the voice. It uses the information estimated echo from the transmit path (from Sin to Sout) subtracting the estimated value.
Echo echo control need to have on the propagation time (reflecting duration) of the echoes to determine the length of the different capabilities and the ability to deal with. Depends on the length of the tail echo cancellation and echo occurred (called "Echo Point") end to end delay between the signal transmission, and the echo propagation time. Residual echo signal through the adaptive filter is the residual echo.

Calculate the remaining non-linear processor, echo, remove all signals below a certain threshold, and does not echo with the sounds as background noise in the simulation of the original background noise to replace these signals.

Echo canceller also includes a tone detector, which enables the user equipment can use a pitch off the echo canceller. For example, data and fax transmission must be turned off during the echo cancellation function. Familiar to start a connection when the modem is a high tone to tell echo canceller in the traffic flow to stop working.

Problems encountered in echo cancellation

Equipment must be carried out to solve the following problem solid echo cancellation: avoid divergence, handling double talk (double talk), to prevent clipping and to avoid annoying background noise changes.
Avoid divergence. Adaptive filter divergence problem is, if you did not use mathematical algorithms to find the right line model solution, there will be divergent.
In the specific line conditions, some algorithms may diverge so that corrupt the signal, or even add the line echo. Echo canceller can be good in almost all line conditions to avoid divergence.
Processing both at the same time to speak. A typical call is a two-way. Listener by saying "Yes, I understand," "ah, ah, is it?" And other speakers said to agree or disagree. In the actual call, the two sides often interrupted each other while talking, or. This situation is called "double talk."

Handling double talk to make it sound quite natural that a technical problem. Echo canceller must be able to do the following:
* Difference between double talk and background noise. Distinguish the exact needs of high-performance echo cancellation.
* Choose not to update the line model in order to avoid divergence. This requires the use of the last line model (just before the start of double talk in the moment) to approximate the line model during double talk.
* The double talk detection, treatment and return to normal processing mode (ie no double talk) to smooth the transition between the three. Transition is not smooth voice will lead off.

To prevent clipping. In a telephone call, if the voice was mistakenly erased a part (usually the first word or the last syllable), then a clipping. Clipping was due to non-linear processor (NLP) inaccurate result. Specifically, NLP is not due to start at the right time and stop. Usually, NLP can not respond fast enough to enter the voice from the local side. It replaced part of the voice with the background noise, resulting in a call difficult to understand. When the NLP disturb a sentence with the remaining echo of the voice level at the end of decay, it will also produce the same results.

To avoid the annoying background noise changes. Background noise must be properly handled in order to make phone calls sound natural. This means that the echo canceller must work accurately and quickly respond to construct an accurate match is on one end of the implementation of echo cancellation environment background noise. When a person spoke, he heard that the NLP-generated simulation of the background noise. When he stops talking and another person talking, then he heard a voice in the "real" background noise. Any difference between the two would call the attention of dispersion.

Senior Octasic echo processor

Octasic OCT6100 Series Echo Canceller coprocessor to solve all of these technical problems. Specifically, the co-processor uses a unique deterministic algorithm, the algorithm can determine a way to provide reliable and excellent quality.
On the other hand, the current market, most of the echo canceller using the same algorithm (using more than 25 years.) The algorithm is recursive, it will calculate the route the previous model and the current signal input to a continuous update of the line model. In this method, the estimator used in each iteration the mutual relationship between the input signal to update its line model. The process is done with variable gain, mutual relationship with a new signal replaces the old part of the line model. If the gain is the wrong choice, then the echo estimator / subtractor may not eliminate all echo or worse - it may be in line to add the annoying sound. In extreme cases, the algorithm may diverge, serious damage to the echo canceller function and makes the calls impossible.
Octasic's OCT6100 Series has realized the advantages of the method (for example, inexpensive memory and more processing power), the use of which is based on "least squares" (LS) algorithm patented. Least squares algorithm does not select the gain, it all of the given set of input signals using only one solution, the solution will be reduced to the minimum energy echo. This method can ensure a more rapid and more accurate convergence, the more solid achievements of echo cancellation, including the improvement of the double talk and background noise processing treatment.