I am glad that this question has surfaced.
PGW acts as HA and SGW as FA. In addition to this, LTE supports Proxy-MIPv6(PMIPv6) protocol.
PMIPv6, as opposed to traditional MIP, is a network based mobility management protocol, in which the mobile node doesnt need any change to its TCP/IP stack as its IP remains constant.
From what I remember from cdma2000, traditional MIP requires MIP capability on the UE.
When the PPP connection between the UE and the FA/PDSN is establishing, during the IPCP config phase with no 'IP address' option to the FA/PDSN which is an indication that the UE is capable of MIP. The PDSN assigns a 'local'IP(only recognizable by PDSN), and sends the Agent-Advertisement message to the UE, which triggers MIP RRQ, the FA/PDSN relays the MIP RRQ to the HA, HA responds with MIP RRP(containing Home-IP addr) which is relayed to the UE by the FA/PDSN. The UE stack thus looks like
MIP/UDP/IP/.../MAC/PHY.
Whereas using PMIPv6, there is no need for the UE to have MIP/UDP layer.
There are 2 elements in PMIPv6, viz., LMA(Local Mobility Anchor), Mobility Access Gateway(MAG).
LMA role is played by PGW and MAG by SGW.
if the UE changes its point of attachment, the
MAG on the previous link will detect the UE's detachment from the link. It will signal the LMA and will remove the binding and routing state for that mobile by using PBU msg. The LMA, in turn will identify the corresponding mobility session for which the request was received, and accepts the request after which it waits for a certain amount of time to allow the MAG on the new link to update the binding.
The MAG on the new access link, upon detecting the UE on its access link, will signal the LMA to update the binding state(PBU/PBA). After completion of the signaling, the serving MAG will send the Router Advertisements containing the UE's home network prefix(es), and this will ensure the UE will not detect any change with respect to the layer-3 attachment of its interface.
Ref: http://tools.ietf.org/html/rfc5213