Network Engineering for Audio Engineers - Part 4: QoS and IP/Ethernet addresses

2.5 QUALITY OF SERVICE
In the context of IP networks, the phrase "quality of service" (QoS) has a specific meaning, describing the quality of the network with regard to the following:

• Bandwidth
• Dropped packets
• Delay
• Jitter

These are all particularly important for audio applications. Although Web surfing, file transfers, email, and the like are all tolerant to big QoS impairments, audio requires a constant flow of packets, all of which must arrive on time and in proper order. Buffering can correct for most QoS problems, but low-delay audio requires very short buffers.

On LANs, it's no problem to achieve excellent QoS. On private WANs, it's also not much of a problem. On VPNs, however, QoS starts to be an issue. And on the Internet, it is the overriding concern.

Statistical multiplexing has its downside, and we see it here. Because all links that make up the Internet are shared by an unpredictable number of users, with unregulated demands on bandwidth, a user can never be sure what is available to him or her at a given instant.

Let's examine each of the bulleted points above, in turn.

2.5.1 Bandwidth
There has to be enough bandwidth consistently available on the network to support the desired audio transmission bitrate. Including header overhead, for low-delay uncompressed 24-bit/48-kHz audio, this is around 3 Mbps per audio stream. Compression (audio coding) can take this down to less than 100 kbps, but at the cost of delay and audio quality.

The network has to be able to convey audio streams at the required rate consistently, never falling below the minimum. An average bandwidth guarantee is no use to audio applications that need low delay. That's because short receive buffers cannot ride out bandwidth variations.

2.5.2 Dropped Packets
As you've seen, IP routers are allowed to drop packets as a normal part of their operation. This is caused by link overloading, so it depends on a range of factors that are sometimes correctable by careful network engineering when the network is under your control, but are unpredictable and potentially troublesome when the network is being run by someone else.

Audio uses RTP transmission, so there is no lost packet recovery. Any dropped packets are going to result in audio pops and/or dropouts. There is no concealment mechanism for pulse-code modulation (PCM) coding that would cover missing packets.

Recall that low-delay audio cannot withstand the delay that lost-packet retransmission imposes. And other recovery techniques using FEC (forward error correction) also add delay due to the time interleaving that is required to get any significant benefit.

We revisit this topic in Chapter 7. Some codecs are able to effectively conceal 10 percent or even 20 percent random packet loss. But this is no longer low-delay AoIP, by any means.