There are two main advantages mobile network operators have when it comes to offering voice calls over LTE (and HSPA) compared to Internet based companies: Handover of an ongoing call to a 3G or 2G network, which I think is their biggest voice asset, and being able to ensure quality of service, in other words, the voice service can interact with the transport network and the base station to ensure the IP voice packets get precedence. But how is it done in practice?
The mechanism of choice for this in LTE is called a dedicated bearer. In HSPA, it's known as a secondary PDP context. Dedicated bearers / secondary PDP contexts are established when a service in the network requests a priorization of IP packets belonging to a specific media stream between two IP addresses and TCP/UDP ports. So far so good with the theory, but how could the functionality be used in practice?
The IMS One Voice Profile contains an interesting and quite precise answer for this:
An Unacknowledged Radio Bearer for Voice Packets
In chapter 7.3.1 the spec says that on the radio interface the following bearers are established during a voice call: SRB1 + SRB2 (those are signaling bearers to keep the radio connection alive), 4 Acknowledged Mode Data Radio Bearers (AM DRB) and 1 Unacknowledged Data Radio Bearer (UM DRB).
So what are they for? Chapter 7.3.3 gives further details: Acknowledged and unacknowledged mode refers to the layer 3 RLC protocol that can ensure that data that is somehow lost over the air interface is repeated. Repeating lost data (acknowledged mode) is the default RLC operating mode for user data in HSPA today and I expect that to be the same for LTE as well. For a voice data stream, however, it doesn't make sense to repeat lost data as the repeated voice packet would come too late to be useful. This is why an Unacknowledged Mode Data Radio Bearer (UM DRB) is used.
In other words: The voice service (in the network) sends a request to the transport network during the establishment of the voice call to create a dedicated bearer for IP packets being exchanged between two IP addresses and two UDP ports to be mapped to a radio bearer for which no RLC error correction is used. All other IP packets not matching the IP address and UDP port combination requested above are sent over an AM DRB without guarantees for latency and bandwidth.
Prior to 3GPP Release 8, resource reservation was the job of the mobile device. With LTE and 3GPP Release 8, this functionality has now moved to the network. The IMS One Voice Spec remains a bit sketchy on this particular point. The VoLGA Stage 2 specification, however, shows quite clearly how the dedicated bearer is established from inside the network in Figure 9.8.1.
Note that an extra dedicated bearer established for the voice call does not require an extra IP address for the mobile device. In fact, only a single IP address is used as it's the combination of the IP addresses and UDP ports that distinguishes the packets that go through the UM bearer from those that use an AM bearer. For the the application on top (e.g. an IMS client or a VoLGA client) all of this is transparent as the protocol stack below automatically decides which IP packet should be sent over which bearer.
Packet Loss Rate
To ensure that the packet loss in UM mode stays within reasonable limits, the radio transmission characteristics (power output, modulation, coding…) for the UM bearer is configured to ensure that the packet loss rate does not exceed 1%, a value that the voice codec can still tolerate.
Guaranteed Bit Rate
Chapter 7.3.4 of the One Voice Spec then goes on to add that the UM DRB for voice is configured with a guaranteed bit rate and that the network resources are permanently allocated to the user during the call. One of the possibilities to do that is for the base station to tell the mobile device that it can periodically send and receive data without looking for bandwidth assignments. That guarantees the bandwidth for the call and also saves the overhead of dynamic bandwidth assignments which are not needed as the bandwidth requirement is static.
Header Compression
The main thing that makes voice over IP very inefficient compared to traditional circuit switched transmission is the overhead from the IP headers of each packet. To this end, the spec requires that Robust Header Compression (RoHC) is used between the base station and the mobile device. I am not sure yet whether LTE vendors will use RoHC for all data streams or only for some such as dedicated bearers.
DRX
An important feature of the LTE air interface is discontinuous reception (DRX). It allows the UE to put its transceiver to sleep for the DRX period. This is especially important for voice sessions as the bandwidth required is so small that the time between two IP packets containing voice data is very long. Keeping the receiver constantly switched on would waste a lot of energy in the mobile device. So the spec requires DRX configuration while a voice call is ongoing. To be fair, I expect DRX to be used also for best effort transmissions, so it's not a feature that was only made for voice.
Summary
To the voice service on top of the protocol stack, all of this is pretty much transparent. It just requests QoS to be enabled for a data stream via a network interface and gives the network the required parameters (e.g. IP addresses, TCP/UDP ports, bandwidth requirement, etc.) and the rest is done by the transport network. To that end, network operators one day might even discover it as a service they could sell to companies offering Internet based (voice) services.
Nice write up. Thanks.