8.7 LTE

8.7.1 QoS Architecture

With LTE, the quality of service concept is simpler than that of the 3G system. A default bearer supports IP connectivity for basically any service. When the indicated quality of service need is different than that which the default bearer offers, a dedicated bearer is set up – provided that the system has resources for the new bearer.

A traffic flow template (TFT) defines which user flows can be mapped to which bearers. User flows that have similar requirements, can use the same bearer. User flows with differing requirements, will be mapped to different EPS bearers. Traffic flow template exists in the UE for the uplink, and in PDN GW for the downlink.

In addition to traffic flow template, service data flow (SDF) template is defined. This is related to Policy and Charging Control (PCC) rules. SDF may e.g. provide a finer granularity than TFT. SDF and TFT may also in practice be identical. SDF only exists in the core network.

Traffic mapped to the same EPS bearer receives the same QoS treatment. QoS functions in the LTE system include scheduling and queue management policy, RLC configurations (e.g. RLC mode) and a shaping policy.

With LTE, Quality of service Class Indicator (QCI), as the name implies, is one key indication for the QoS of the bearer. In transport Differentiated services codepoint (DSCPs) are available for carrying this information in the IP packets, based on the QCI.

8.7.2 Packet Flows and Bearers

EPS bearer provides the packet data network (IP network) connectivity to the UE. It is comparable to the PDP context concept from 3G.

End-to-end service consists of the EPS bearer (between UE and the PGW) and of the external bearer (typically to the internet, or e.g. to an enterprise VPN). E-RAB is comparable to the RAB in 3G. An EPS bearer is realized with the radio bearer, S1 bearer, and the S5/S8 bearer. S8 interface is for roaming cases. See Figure 8.17.

The connectivity to the external network (PDN, the internet) consists of transport of traffic flow aggregates. These are defined to be collections of multiple service data flows. Further, service data flow is a set of packet flows matching a service data flow template. Service data flows are bound to the bearer (IP-CAN bearer, IP connectivity access network).

PCC rules are needed when EPS bearers are to be established. PCC rules can be statically configured in the PDN GW. Alternatively they can be provided by PCRF (Policy and Charging Rules Function). Rules may also be modified during the lifetime of the EPS bearer.

PCC rules are used to detect these service data flows. Based on the rules, parameters for charging and policy control (including QoS policy) are determined. PCC rules include the service data flow template, and this template possibly contains multiple service data flow filters. Further, the filters may be separate in both uplink and downlink directions. This is illustrated in Figure 8.18.

Figure 8.18 Service data flow filters and detection [36].

In practice, detection means e.g. matching with the source and destination IP addresses, source and destination port numbers, and the protocol used above IP. IP address information may contain a prefix mask, and instead of a single port number, port ranges may be defined. Additionally, the QoS marking on the IP header may also be used. It is also possible to look further into the packet, and examine the transport and application protocol layers.

Due to the detection (packet inspection) capabilities, it is possible to identify packet flows and then apply a policy for charging and QoS control. The EPS bearer, with the QoS characteristics defined, is then established accordingly. The bearer establishment may be subject to admission control on all nodes (eNodeB, SGW, PGW) and other resources (air interface, transport, processing capacity etc).

A default EPS bearer is established for the lifetime of the PDN connectivity (‘always-on connectivity’), when the UE connects to the PDN network. The default bearer is always a non-GBR bearer, which is set up by the MME. For the default bearer, the QoS parameter values of the default bearer are based on subscription data. MME queries the necessary information from the Home Subscriber Server.

Additional bearers are dedicated bearers that may be set up by either UE initiating the request, or by data arriving from the external network. UE may e.g. initiate a VoIP session that triggers setup of a dedicated EPS bearer. Dedicated bearers can be either GBR or non-GBR bearers.

If the bearer set up is due to the external network, the bearer is established via the PCRF and the PDN GW – also e.g. in the case of a VoIP session.

Gx interface is supported between PCRF and the PDN GW. In cases where the S5/S8 interface is Proxy Mobile IP based, the QoS mapping is done by the SGW, with the Gxc interface.

Each dedicated EPS bearer includes a traffic flow template, for both uplink and downlink directions. Traffic is matched against the template, and then mapped accordingly to an EPS bearer. UE maps the traffic in the uplink direction, and the PCEF (in the PDN GW) maps the traffic in the downlink direction. With EPS bearer modifications, the PDN GW delivers traffic flow related information (e.g. source and destination IP addresses, source and destination port numbers, and protocol used) to the UE, so that the UE can associate the application with the correct EPS bearer.

8.7.3 QoS Parameters

The QoS parameters for the EPS bearer are QCI (QoS Class Identifier), ARP (Allocation and Retention Priority), GBR (Guaranteed Bit Rate) and MBR (Maximum Bit Rate). GBR and MBR are applicable only for GBR-bearers. Per UE (aggregate of EPS bearers), additionally QoS parameters APN-AMBR (Aggregate Maximum Bit Rate) and UE-AMBR are defined.

QCI classes are standardized in 3GPP (TS23.203), and shown in Table 8.7.

Table 8.7 QCIs [36].

For Guaranteed Bit Rate (GBR) services, dedicated resources are allocated typically with Admission Control (AC). Non-GBR services do not have a similar guarantee for resource availability.

Priority level is meant for differentiating between multiple service data flows of a single UE, and also for differentiating service data flow aggregates between UEs. Via the QCI, an SDF aggregate is associated with a packet delay bound, and a priority level. Packet delay bound is primarily intended to be used for differentiating scheduling between SDF aggregates If the packet delay bound cannot be met for all SDF aggregates, priority level is used in guiding which SDF aggregate is scheduled first. Priority level N has preference over priority level N+1.

Packet delay budget is interpreted as the maximum delay bound, with a confidence level of 98%. It is defined as the delay between the UE and the PCEF (Policy and Charging Enforcement Function). PCEF is a function of the PDN Gateway. Packet delay budget is the same in uplink and downlink, for a certain QCI.

Packet Error Loss Rate (PELR) specifies an upper bound for packets that are sent by the link layer (e.g. RLC) but not received at the upper layer (e.g. PDCP) at the receiver end. PELR tells the packet loss rate without congestion. PELR is used in selecting the link layer configuration (e.g. RLC acknowledged mode). For a certain QCI value, PELR is the same in uplink and downlink directions.

The PELR value also assumes that the loss in the mobile backhaul (from the base station to the PDN GW) is negligible. So, for the system to meet the PELR loss bound, mobile backhaul should not have a contribution. In non-congested cases, normal conditions and well-designed links, the packet loss ratio should also in practice be minimal in the backhaul. For the mobile backhaul not to have a contribution, the packet loss ratio caused by the backhaul, could be e.g. one decade less than the PELR loss bound.

To estimate for the backhaul induced PELR, an analysis is needed to estimate the packet loss rate when the physical layer has a certain, residual BER (Bit Error Rate). Layer-2 protocols such as the Ethernet and the PPP, include a checksum, which is calculated over the complete frame. If the checksum fails, the frame is discarded. Thus, bit errors lead to frame discards. So the physical layer Bit Error Rate (BER) value has to be considered to assure that there is no significant contribution from the backhaul.

In the mobile backhaul, congestion in the network (packet drops in the egress queues of the routers and switches) may cause further packet loss. The PELR target may be violated in these cases, if the backhaul is congested, even if the air interface has capacity and good radio conditions. Similarly, during temporary backhaul connectivity losses (e.g. due to failing links, and reconvergence of the network), packets may be lost in the mobile backhaul.

Transport level QoS marking is done based on the QCI value of the EPS bearer. eNodeB is responsible for marking in the uplink direction. SGW and PGW mark the packet in the downlink direction and in the uplink directions, similarly based on the QCI value of the EPS bearer. So, at bearer level, EPS bearer is the level of granularity of QoS, and the QCI value of the EPS bearer is further marked to the transport layer.

The bearer establishment/pre-emption priorities are identified by the ARP parameter. It is also needed in the eNodeB, so that pre-emption at the radio interface can be supported.

An example of ARP usage is given in TS23.401, where voice has a higher ARP than video. In cases of congestion, video service is dropped before the voice service. Similarly, ARP may be used to relieve network load. In the case of high traffic load due to e.g. a natural catastrophe, lower priority ARP bearers may be torn down.

GBR is the expected bit-rate provided by the EPS bearer, while the MBR may limit the bit rate with a shaping function. GBR parameter can be utilized e.g. in admission control.

APN-AMBR and UE-AMBR are both subscription parameters which are stored in the HSS. APN-AMBR limits the bit rate of all Non-GBR bearers within the APN. UE-AMBR is the upper limit for all active APN-AMBRs, up to the value of subscribed UE-AMBR. The parameter is set by the MME. So GBR bearers are not included in the UE-AMBR calculation and potential shaping.

Bit rates for GBR, MBR and AMBR, are calculated as the bit stream at S1 excluding GTP-U and IP headers. The QoS parameters have a component for both uplink and downlink directions.

8.7.4 Admission Control

In LTE, for GBR bearers, dedicated resources are allocated, e.g. with admission control. Admission control needs to take into account all functions, radio resources, hardware (processing) resources and also transport.

eNodeB is responsible for the radio admission control. Before establishment of a new GBR bearer, radio resource availability needs to be ensured, both for UL and for DL. Priority levels and QoS required for the new bearer are considered, as well as those of the already established bearers. If resources are available, the bearer is admitted. Otherwise the bearer set-up request will be rejected, unless priority levels and pre-emption indicators suggest that another bearer should be pre-empted in order to allow the new bearer to be established.

8.7.5 S1 Interface

S1 interface includes a user plane connection between the eNodeB and the SGW, a control (signalling) connection (S1-AP) between the eNodeB and the MME, O&M channel towards network management, and often also synchronization (such as IEEE 1588v2). Additionally, transport control plane, such as IP routing protocols, may be carried.

As opposed to 2G and 3G BTS access interfaces, eNodeB S1 interface does not need to transfer radio layer protocols, as those are all terminated in the eNodeB. The requirements for delay, delay variation and loss originate from the needs of the end user experience rather than from any mobile system architecture constraints.

In the user plane, mandatory E-RAB level QoS parameters are QCI, and Allocation and Retention Priority. GBR QoS information is an optional parameter (not needed for non-GBR bearers), consisting of Guaranteed and Maximum bit rates uplink and downlink. Without the GBR QoS information, GBR bearer set-up fails. For the user plane, it is mandated to support configurable DSCP marking. The input information for the marking are the QCIs and other parameters.

For user plane packet loss, the S1 interface behaves differently than packet data on the Iub (assuming acknowledged RLC mode). Lost packets on Iub are retransmitted at the RLC layer. With LTE, RLC layer terminates at the eNodeB. S1 is comparable to the Iu interface of 3G also in this respect. This means that packet loss on S1 is visible to the application layer, as there is no protocol hiding the loss by retransmissions. If the application uses reliable transmission (e.g. TCP), packet loss on S1 causes TCP layer retransmissions.

8.7.6 S1 Example

In LTE the requirements on the backhaul are driven by the applications. There are no technical requirements on delay and delay variation from a controller, since there simply are no radio controllers in an LTE network.

8.7.6.1 Control Plane (S1AP)

Radio network control plane in the case of LTE and eNodeB is again simpler than in 3G, since now the radio layer protocols are not carried over the S1. Correspondingly, cell common channels also terminate in the eNodeB, and are not a concern in the mobile backhaul.

Radio network signaling, S1-AP, is carried over the S1 logical interface. This is marked with DSCP 34 (AF41).

8.7.6.2 Network Control

Network control (e.g. routing protocols) is marked with a DSCP value of 48 (CS 6). There is no difference between radio technologies concerning the transport network control protocol marking. It is needed as CS6 for the same reasons in all radio networks.

8.7.6.3 User Plane

The different end-user applications are indicated by their Quality of Service Indicator (QCI). 3GPP has defined 9 QCIs in TS23.203. Therefore the QoS mapping for LTE in the user plane is determined mostly by a mapping of QCIs to DSCP.

If DSCPs indicating higher drop precedences are used (such as AF32 as an example), then there are enough DSCPs so that each QCI can be mapped to a unique DSCP value. The amount of QCIs is, however, larger than the amount of PHBs. Therefore one has to expect that several QCIs will be treated similarly in the transport network. In this example only DSCPs corresponding to low drop precedence are considered.

From the user plane traffic, voice and real-time gaming traffic have the most stringent requirements, therefore QCI 1 and QCI 3 are marked with DSCP 46 (EF).

IMS signaling traffic is treated as AF41. So QCI 5 is marked with DSCP 34 (AF41).

Regarding the video services we distinguish whether guaranteed bit rates are provided and mark these services differently. QCIs 2 and 4 use GBR-bearers and are marked with DSCP 26 (AF31). QCIs 6 and 7 use non-GBR bearers and are marked with DSCP 18 (AF21).

The default bearer (QCI 9) is marked with DSCP 0 (BE), allowing any premium data traffic to be marked with DSCP 10 (AF11).

QCI 8 and 9 are used to distinguish between interactive and background data traffic, e.g. web serving and file download. Alternatively they differentiate between user classes (ordinary/premium, etc).

8.7.6.4 O&M

O&M traffic is marked with DSCP 16 (CS2), for the same reasons as in the Iub example.

8.7.6.5 Synchronization (by Packet)

Synchronization for LTE (FDD) follows the reasoning discussed in the Iub example.

Assuming IEEE1588v2, the traffic is marked with DSCP46 (EF).

8.7.6.6 Summary

The DSCP marking is presented as a summary in Table 8.8.

Table 8.8 Example QoS mapping for the S1.

A mapping to four queues is considered, as there may not always be as many queues in the backhaul network as there are DSCPs used.

Q1, a strict priority queue, is used for:

• Network control.
• Voice.
• Real-time gaming.
• Synchronization.

The GBR and non-GBR video services should not be merged together. GBR video services might be subject to call admission control. Typically this would not be the case for the non-GBR services. Treating these services together would make the call admission control pointless.

Q2 is used for:

• Control plane (S1-MME).
• QCI 2 and QCI 4 (GBR-bearers for video).
• QCI-5 (IMS signaling).

Q3 is used for:

• QCI 6 and QCI 7 (non-GBR bearers for video).
• QCI 8 (premium data).
• O&M.

The expected behaviour for the AF PHBs can be implemented using WRR/WFQ schedulers. Although e.g. non-GBR video services have more stringent requirements than premium data, this is not a strict priority relation. Instead each of the traffic types should get a defined share of the bandwidth in cases of congestion.

Finally, Q4 is for

• Default bearer.