Echo is a troubling problem. Most of us have suffered through some call where we had to try to talk with a lot of echo on the wire. It’s very distracting and makes it hard for most people to think straight and talk at the same time.
VoIP does not create echo, but due to the temporal aspect of echo, VoIP systems can and do increase the amount of echo heard when talking.
In any conversation, a certain amount of your own voice is part of what you hear, whether you are talking live, sitting in your office, or on the phone. This “hearing your own voice” is not echo and is referred to as “sidetone.” It’s a normal aspect of talking and listening.
Your own voice becomes echo when it comes to your ear with a significant delay from the time you spoke – longer than 25 milliseconds. But 25 to 150 ms is a typical delay range for international calls and this is why echo cancellation is necessary. Voice over IP calls have a delay budget also in the range of 150ms.
So, what causes echo?
First, let’s look at what doesn’t cause echo. Because delay is a necessary condition for echo, it is virtually impossible for components that are close to the speaker, e.g. on the speaker’s site, to cause echo. Even if part of the transmitted signal is being reflected back in the return channel, the propagation delays will be so brief that it will never be heard as echo.
Also, there is no way for the digital stream of packets in one direction to “bleed into” the digital stream of packets in the other direction. The same is true for the digital parts of the PSTN TDM network. While the underlying electrical signals carrying the bits are, indeed, analog, the corruption of those signals results in digital noise or other problems, not in echo.
So, strictly speaking, echo is never caused by voice over IP. In fact, what happens is that the longer delays introduced by all voice over IP systems reveal echo that was imperceptible with the shorter delays of the PSTN. By delaying existing echo signals more, they fall outside that 25ms window and become audible to us.
Echo is always analog and always at the far end of a conversation.
(Continued…)
Your own voice gets into the return channel in several different ways, but the simplest example would be because the mouthpiece of the phone at the far end is too close to the earpiece, so that your voice is heard and forwarded on the same return channel as the person at the far end speaks on. This is particularly noticeable when using the onboard speakers and microphone that some laptops ship with.
There are other ways echo can occur. Analog telephone handsets, for example, transmit on a “2-wire” system – which means that they transmit and receive on the same sets of wires. Most of the backbone phone system lines are “4-wire” systems, which transmit and receive on separate sets of wires. At some point, there’s an interface that converts the “4-wire” signal to the “2-wire.” But sometimes some of the energy is reflected back into the network in this process, causing an “impedance mismatch” and creating an echo.
Now that we know what echo is – how do we cancel it out?
The louder the echo, the more distracting it is. Echo cancellation, in effect, consists of attenuating the echo part of the signal so that the echo is not as loud. As some point, the echo is so soft that it disappears, compared to the foreground volume of the call.
Echo Return Loss or ERL (see also Figure 1) is the basic measurement of the echo. It is a measure of the loss of loudness between the original signal and the echo. In other words, an ERL of zero means that the echo is fully as loud as the original signal. This is the worst case. As the delay grows longer, an ERL of up to 55 dB is necessary to render the echo quiet enough to avoid being distracting.
One of our products, NetQoS VoIP Monitor , has a feature called Call Watch that can measure and display ERL and ACOM, which is the measurement of the echo after the Echo Canceller (ECAN) has been applied.
Echo cancellation is ideally done at the far end of the call, though it can be done at any analog boundary in the network. (In one sense, you cannot eliminate an echo originating at the far end if you don’t control the equipment at the far end. However, you can minimize it using echo cancellation.)
One place where we try to minimize or cancel echo in a VoIP system will be at the gateway that connects our site to the PSTN. Even though the echo is originating at the far end, echo cancellation here can work within limits.
The echo cancellation device operates by comparing the signal going into the tail circuit with the signal coming out. Basically, the ECAN remembers the signal pattern of the signal going into the tail circuit and looks in the signal coming back to see if it contains this pattern. If I say, “Is that good?” and you reply, “Fine” the ECAN is remembering “is that good?” and looking for it mixed in with “fine.”
Now, there are two important dimensions of ECAN: echo strength and delay. Basically, the echo must be weak enough that it can be distinguished from regular speech – at least 6dB quieter than the speech it appears with. If the echo is louder that this, it cannot be safely removed. Delay is the other dimension. The ECAN looks for the pattern sent into the tail circuit in the signal coming back. ECAN’s have a time window that bounds their operation. They look for the signal going into the tail circuit to be repeated in the signal coming back within a specific time, typically anywhere from 8 to 64ms.
There is one other condition that can exceed the ability of ECANs operation and that is distortion. If the echo itself is so distorted that it doesn’t truly match the pattern seen as the signal was sent into the tail circuit, the ECAN will not be able to recognize it as echo.
Because the ECAN is actually modeling the response of the tail circuit mathematically, it starts each conversation with no knowledge and has to build its model – meaning the first few seconds of the conversation will have more perceived echo than what will follow.
Echo is a difficult problem and not always one that can be solved readily without access to the end-to-end circuit, but with techniques such as echo cancelling, and having monitored metrics to track echo problems, the nastiest effects of echo can be mitigated.
No comments yet.