VoIP: The Voice Quality

Once a portion of a voice conversation is encoded and put into an IP packet, there isn’t much that can degrade it - unless the packet doesn’t make it to where it’s going or gets delayed appreciably en route. Either of these can occur when there’s congestion on the IP network
Once a portion of a voice conversation is encoded and put into an IP packet, there isn’t much that can degrade it - unless the packet doesn’t make it to where it’s going or gets delayed appreciably en route. Either of these can occur when there’s congestion on the IP network.

Virtually all VoIP products use the User Datagram Protocol (UDP) and the Real-Time Protocol (RTP), over IP. This means that voice-containing packets that are lost aren’t retransmitted, whereas most IP "data" packets use the Transmission Control Protocol (TCP), which detects and arranges for retransmission of lost packets.

Still, the various algorithms and protocols used for VoIP react differently to delays and dropped packets. Some VoIP decoding methods will drop any voice packet (a packet containing a voice sample) that is out of sequence or more than, say, 200 ms old. Others have user-settable queues - called "jitter buffers" - that determine how many voice packets/samples will be assembled and how long they’ll be held before being dropped.

Once packets are dropped, different systems have different ways of compensating for the "hole" that is created in the voice stream. Some decoding algorithms will interpolate and create samples to fill these holes; others do nothing, and the missing packets produce interruptions and noise at the receiver’s end. We have found that products vary considerably in the voice quality they deliver under adverse network conditions (typified by heavy congestion).

Since it’s not practical for everyone to build their own test lab and put all the potential products through their paces over both well-behaved and congested IP networks, here are two helpful rules of thumb regarding the characteristics of VOIP products:

• Gateway products that require the least amount of IP network bandwidth per active voice conversation tend to survive better - that is, their voice quality degrades less - as network conditions get worse.

• VoIP products that use smaller IP packets to carry their voice samples survive much better as network congestion gets worse. The smallest voice-containing packet sizes are in the 70-80-byte range; the biggest are 250-300 bytes per voice-containing packet.

Users should be able to query vendors about both of these metrics, and then do their own side-by-side comparison. In comparing voice-over-IP equipment, remember that some (but not all) products offer access to a large number of operational parameters you can tweak to fine-tune voice quality.

In the PSTN, optimum volume levels are well understood. There are generally accepted norms for the relative strength of a voice signal referenced to power (usually measured in dBm, or decibels referenced to a milliwatt), and referenced to background noise (dBrn, or decibels referenced to noise).

However, standards for mapping these norms onto voice-over-IP communications haven’t yet been fully worked out. As a result, some products deliver a voice signal of more-than-sufficient strength (amplitude) but, due to a particular PC’s microphone and speaker, it may simply be too much volume. In this case, the voice quality may sound terrible, because it is set just a little too loud