Among the five distinct 5G-system (5GS) deployments scenarios, the telco industry seems headed toward Option-2 and Option-3. When 5G core (5GC) is used, we’ve Option-2, so-call the Standalone (SA). For early commercial deploy with EPC (know as EN-DC, eUTRA-NR Dual Connectivity), we’ve Option-3 -the Non-Standalone (NSA).
If you think for a while, it should ring the bell. After all, there was no such thing as Standalone or Non-Standalone EPC! So the question may arise, why they are in 5GC?
NSA or SA signifies that unlike the previous generations, the 5GC is not back-compatible, i.e. it is not possible to integrate 2G/3G/4G RAN with 5GC. LTE has forward-compatibility. It is possible with a software upgrade to reuse an eNB as an NG-eNB, which will happily go along with 5GC. Still, NG-eNB is a 5G RAN node (though not NR). On the other hand, 2G/3G RAN doesn’t stand a chance.
From the interworking point of view, if one ignores the 3GPP lingo (aka reference points), HSPA Core and EPC call flows are indistinguishable. For example, Gn/Gp interface in HSPA and S11/S5/S8 interfaces in EPC use the same GTP-C protocol for signalling (though version differs). All diameter signalling remains the same for both cores. As a result, manufacturers were able to implement multiple nodes, for example, MME and SGSN into a single box.
In 5GC, both GTP-C and diameter signalling are discarded. An exception may be the support for diameter Rx between PCSCF and PCF. These two primary signalling protocols are replaced by stateless RESTful API over HTTP. With a hyper-scalable protocol like HTTP for signalling, an operator no longer has to worry about capacity planning for diameter TPS or specialized firewall/LI software (for GTP-C and Diameter inspect especially in roaming scenario).
Choosing HTTP as signalling protocol implies that 5GC cannot coexist with EPC. The problem is that changing the core network also requires a change in the PDN anchoring point, causing the change of device IP address and service interruption. So, 3GPP specified interworking between 5GC and EPC. Such interworking requires GTP-C based N26 reference-point between MME and AMF as well as PGW-C and PGW-U collocation with SMF and UPF respectively.
Incompatibility between LTE/UMTS and 5GC is mostly because 5GS handles QoS differently than EPS. In EPS, each bearer can have only one QCI value; whereas in 5GS, a single PDU session can carry multiple 5QI values. This mismatch implies that, despite both S1-U in EPC and N3 in 5GC use GTP-U for user-plane, they are not similar as S12/DT in 3G and S1-U in 4G.
Distinctive QoS handling is 5GS also imply that the VoNR is not the same as VoLTE. Since Packet Core is an underlay for IMS, 5GS should be entirely transparent to IMS Core (unless Rx is replaced with RESTful API). As a result, until the VoNR is realized, an operator needs to rely on EPS-FB (fallback to VoLTE) only. Since no interworking supported between 5GC and CS domain, CSFB is not an option.
From the capacity planning point of view, EPC and 5GC interworking, aka the collocation of PGW with UPF will be challenging. Since PGW and UPF are the PDN anchoring point (thus cannot be changed), PGW in 5GC needs to have at least the same or more capacity compared to the PGW in EPC. Many operators may have legacy hardware-based PGW which cannot be incorporated with UPF.
Apparently, 3GPP knew the limitations of the blank-slate design of 5GS, thus mandated virtualization from the very beginning. With virtualization, decoupling PGW from SGW and scaling up with UPF will be less daunting.