Echo + Delay = Trouble

Echo + Packet (IP) Network Delay = More Trouble

Introducing the Problem

If you have experienced echo during a phone call (who hasn’t?), you know how difficult it is to carry on a conversation while your own echo is there to annoy and confuse you. But the echo by itself is not the problem; it is the fact that the echo is delayed that is the problem.

You can hear yourself speak whether you are on the phone or not. But when you are not on the phone, there is little delay between your mouth and your ear so the sound of your own voice is not problematic. But when this delay is increased, as is the case on some phone calls, the echo becomes objectionable. In fact, the degree to which an echo becomes objectionable is a function of both the amount of delay and the degree of attenuation (or loss) of the echo signal level with respect to the signal level of your original speech. The difference in level between the original speech and the echo is known as the Talker Echo Loudness Rating or TELR.

As exhibited in International Telecommunication Union (ITU) recommendation G.131, table 1 shows the minimum requirements on TELR for various round trip delays.

For example, if 20 milliseconds of round trip delay is present, at least 25 dB of TELR is needed to ensure good voice quality. On the other hand, if 80 milliseconds of round trip delay is present, the required TELR is 38 dB.

The two “equations” at the beginning of the article should make more sense now. A VoIP network introduces more delay than its traditional voice network counterpart. As a result of this additional delay in the VoIP network, it is necessary to cancel echo to a greater degree in order to maintain the same level of voice quality.

The conventional wisdom with respect to echo cancellers is that an echo canceller needs to be placed “near” the source of the echo. There are two sources of echo in telephone networks: electrical echo and acoustic echo. Electrical echo is caused by hybrid circuits that are used to connect four-wire circuits to two-wire circuits. Acoustic echo is caused by coupling between speaker and microphone in both handsets and hands-free devices.

So, if we follow the conventional wisdom, we place the echo canceller near all hybrids and near all handsets and handsfree devices. There is plenty of wisdom in this approach, yet there is still a hole that needs to be filled lest echo find its way through this hole.

If we look back at the history of the telephone system we can easily find this hole. Early telephone circuits had no appreciable round trip delay. As a result, there was no need for echo cancellation. As the telephone network grew and delay crept into the system, echo cancellers became necessary. But with such modest delays, the required echo canceller attenuation was also modest, and this is reflected in some of the earlier echo canceller specifications. Today’s echo canceller specifications have become more stringent due to increased round trip delays, but one might argue that nothing has yet prepared us for the delays that may be incurred in VoIP systems.

This leaves us with a dilemma. Let’s take a simple case. Standard telephones do not have echo cancellers built in. The reason is that there is quite a bit of loss between the speaker output and the microphone input. That loss may be sufficient when the phone is used to make phone calls over the traditional telephone system with its relatively short round trip delays. But what about the situation in which that same phone is used to make a phone call over a VoIP network? The added delay could be the proverbial straw that broke the camel’s back.

Let’s take an example as shown in figure 1. We’ll assume that the phone handset has a speaker to microphone loss of 32 dB. According to table 1, the resulting voice quality for round trip delays as high as 40 milliseconds should be fine. But if a VoIP network is introduced as shown in figure 2, and the delay increases to 80 milliseconds or greater, good voice quality is no longer maintained. The echo attenuation must increase in order to maintain the same voice quality.

This example can be extended to hands-free phones – even those with built-in acoustic echo cancellation. The reason is that these devices may have been designed to cancel echo to a degree sufficient only for shorter round-trip delays.

The dilemma is this. We cannot expect existing phones to be upgraded to include or improve echo cancellation just because the users start making phone calls over a VoIP network. Similarly, we may not be able to expect the traditional telephone carriers to add this added echo cancellation capability to their infrastructure equipment, especially since VoIP carriers are in direct competition with them.

That leaves the responsibility of handling this echo with the VoIP carriers and their equipment suppliers.

Introducing the Solution: The Packet Echo Canceller

Traditional echo cancellation is a complex enough problem to solve. Packet echo has issues takes the game to the next level. The length of this article does not allow us to discuss the whole issue of echo cancellation. Instead, we’ll present the issues that a packet echo canceller must handle above and beyond issues that are already handled by traditional line or acoustic echo canceller.

1. Determining the need for additional echo cancellation: The first thing that we should note is that the majority of the circuits connected via a VoIP network will NOT need additional echo cancellation. So, one job that a packet echo canceller needs to do is to decide whether or not a circuit needs additional echo cancellation.

2. Determining the packet network delay: Traditionally, an echo canceller is located “near” the source of the echo. So, the delay between echo canceller and the source of the echo is small. But when the echo canceller is placed somewhere in the packet network, the delay between the echo canceller and the echo source can no longer be assumed to be small. Not only isn’t it small, but the delay could change due to packet loss and jitter buffer adjustment. The packet echo canceller must be able to determine the packet delay and constantly monitor that delay for changes.

3. Packet Loss: A traditional echo canceller is afforded the luxury of having a relatively clean, distortion-free path between it and the source of the echo. A packet echo canceller must be able to deal with lost packets.

4. Time Varying Conditions: Beyond the packet losses and delay jitter, there could be other time varying conditions caused by other line and acoustic echo cancellers in the network that are “closer” to the echo source than the packet echo canceller. The packet echo canceller has to deal with these time varying conditions.

Summary

When transitioning from the traditional telephone network to a VoIP network, we incur additional round trip delay. If we don’t take steps to improve the round-trip echo attenuation, voice quality will be degraded. By incorporating a packet echo canceller into the VoIP infrastructure, we can maintain good voice quality.

Echo + Packet Network Delay + Packet Echo Canceller = Good Voice Quality

References 1. ITU, G.131 Talker Echo and its Control. International Telecommunication Union, November, 2003.