He's not alone. While Cisco AVVID and other vendors are enthusiastically pushing voice over the network, businesses, by and large, are still resisting it. A recent Sage Research survey reports that only 6% of U.S. organizations have implemented VoIP. Another 24% are considering deployment within the next 12 months. But NONE of the rest have plans to even look at VoIP this year, because they view the technology as risky and unproven.
Both vendors and users have valid arguments for their respective positions. Using an enterprise WAN -- either IP, Frame Relay, ATM or DSL -- to bypass the public switched telephone network (PSTN) for voice calls has advantages, especially for international companies, because of the cost savings they can realize on overseas calls. (Toll cost savings may not be as great in the continental U.S. because big PSTN carriers can compete by dropping their prices for bulk-purchased long distance service.) Two other big reasons for gradually introducing packet voice on converged networks are the lower capital costs of equipment and simplified management. LAN PBXs are about 10% to 25% cheaper than a traditional PBX. And with VoIP, companies don't have expensive move-and-change costs when employees relocate offices. An IP phone simply re-registers the user with a gateway from its new location.
Still, companies have concerns about delay and jitter, which are two major characteristics of packet networks that you don't have to deal with on switched circuits. The good news is that if careful analysis and testing is invested throughout your network, you can minimize these effects.
In this workshop, we'll look at the likely sources of delay and jitter, and describe some of the "tuning" choices you can make to improve quality. We'll also look at a recently introduced hardware-based, end-to-end voice quality testing system, which we tested on a set of new IP telephones. A hardware-based approach provides a better, more objective way for you to confirm how well your design and equipment choices will play out on the real network than using human listening panels to evaluate voice quality.
Examine your delay budget
The delay budget encompasses the total time from speaking into an instrument at one end to hearing the reconstructed sound at the other end. Human factor studies show that most listeners can detect end-to-end delays that amount to more than 100 milliseconds (ms). At around 250 ms, the delay budget is annoying; it's hard for each party to murmur in agreement or interrupt the other. Above 450 ms end-to-end delay, only listeners who are desperate to save money will put up with voice over a LAN or WAN.
Jitter induces the stutter, gargling and intermittent breakup that can occur on a bad connection. Called the "cliff effect," the jitter phenomenon is peculiar to digital networks. It is caused by too many dropped, or bunched-up, packets. Audio sent over a network can sound good if packets arrive with roughly the same pacing and spacing they had when they left the voice encoder (vocoder). But if packets arrive erratically, a decoder may have to give up and attempt to resynchronize. The human ear is very sensitive to short breaks like these, which impair the perceived quality of a conversation.
Many voice gateways and IP telephones incorporate jitter buffers to smooth out packet delivery to decoders. They may improve quality to a point, but if a jitter buffer is set for too large a number of milliseconds, unacceptable delay may be added to the overall delay budget.
The trick is establishing a jitter buffer that's big enough to keep voice from sounding bad, ideally without causing delays above that 100 ms touchpoint. You might be able to achieve this end by reducing, for example, the number of
hops between routers or switches that can introduce buffering delays in the network, or by reducing the size of queues at network access points.
RSS
