Latency analysis of mobile transmission

3.1. Mobile communication targets

In traditional mobile communication networks, the primary key performance indicators of interest have been the achievable data rate and spectral efficiency. In 5G, latency has been added as a further key performance indicator by URLLC. The ambition of 5G URLLC was to provide low-latency communication while providing high reliability for maintaining the latency below a specified latency bound. For example, the objective for the 5G standard is to guarantee that a RAN latency of 1 ms can be achieved with 99.999% probability.¶

A solution for reliable wireless transmission with high spectral efficiency is to apply Hybrid Automatic Repeat Request (HARQ) retransmissions to recover from unsuccessful transmissions. However, HARQ leads to an increase in latency due to multiple transmissions causing a notable disturbance in the packet delay distribution. URLLC has introduced two major sets of tools: (i) reducing the radio transmission structure for lower latencies (e.g., processing delays, channel access delays), and (ii) providing higher robustness in the transmission to achieve the same latency reliability with fewer transmission attempts, at the costs of reduced spectral efficiency due to extremely conservative transmission modes.¶

To give an example, an uplink transmission in a millimeter wave carrier can be made in two different configurations [FGS15]:¶

• Normal 5G New Radio (NR) configuration with up to 3 retransmissions for reliability with packet delay from ~500 us to 2.8 ms, with low resource usage,¶

• 5G URLLC NR configuration with single-transmission reliability with packet delay from ~500 us to 900 us, involving high resource usage.¶

Furthermore, very low latencies enabled by URLLC require a thorough network deployment plan (e.g., location and density of base station antennas) to ensure that the capabilities are available throughout the service area. More relaxed latencies are less sensitive to the radio network design.¶

5G URLLC is the main enabler to support time-critical communication standards that have been defined for fixed networks, like IEEE 802.1 TSN and IETF DetNet.¶

3.2. Transmission Latency Breakdown

Generally, the latency contributions in a 5G network are dominated by the RAN [D6G-D3.1]. The transport network only plays a role if a UPF is far away from the gNB; the amount of packet processing at the UPF (and related processing times) is limited in comparison to RAN.¶

In the 5G RAN the main latency contributors are:¶

1. Time-domain reliability based on HARQ¶

2. Mobility with handover interruptions¶

3. Time-division duplex structure¶

4. Congestion due to resource sharing and queuing¶

HARQ allows for a better utilization of the resources while being robust for a defined loss bound. Retransmissions inherently contribute to the latency of the packet with defined probability of retransmission(s). HARQ should be used as reliability tool, in case that it is permitted by the latency bound; it is a tool that combines high reliability with spectral efficiency (at the cost of increased PDV). Reliability can be achieved without HARQ, by using more robust transmission modes. If a (low) latency bound is provided with 99.999% reliability by a robust single transmission, then the large majority of (i.e., 99.99%) of the packets are over-protected with too high resource allocations in order to ensure that also the worst-case packets mostly achieve the latency bound.¶

Mobility is ensured by handover, where a UE switches connection from one base station to another, which can lead to handover interruption times. There are tools to minimize this impact, e.g., L3 make-before-break handover where the resources are allocated and ready before performing the handover, L1 or L2 mobility with multiple transmission-reception point (multi-TRP), multi-connectivity. These options are dependent on deployment and spectrum.¶

Time Division Duplexing (TDD) pattern is sometimes prescribed by national regulation and subject to harmonization of multiple networks. This can place restrictions on applicable configurations. Each TDD pattern introduces at least PDV at transmission time interval (TTI) level since packets need to wait for their time slots to be transmitted.¶

When the network is undergoing congestion at high loads, the opportunities for transmission are restricted and, consequently, additional delay is experienced by the packet. Possible solutions are to apply prioritization, resource partitioning, admission control, traffic policing, reservations, or preconfigured access. In most cases there are implications for the implementation, as well as utilization inefficiencies.¶

3.3. QoS architecture within the mobile network

The packet delay of individual packet is strongly dependent on how the packet is handled within the mobile network. Different packets are treated differently according to the service requirements they are associated with. This allows to provide latency-optimized treatment for dependable time-critical services by applying the Quality of Service (QoS) mechanisms of the mobile network. The handling of QoS for traffic passing through the 5G network is defined in the 5G QoS framework [M23.501][FGAQoS], as summarized in Figure 2. The end-to-end traffic flows passing through the 5G network, denoted as service data flows, are mapped at the ingress to the 5G system at the UE and UPF to QoS flows via traffic filter rules. The QoS flow is the finest level of granularity for specifying the service specific traffic treatment in the 5G system. Each QoS flow can have different traffic forwarding treatment configured in the network, according to the defined QoS requirements.¶

                 NG-RAN                       5GC
<------------------------------------->.<------------------
                 .                     .
+------------+   .    +------------+   .   +-------------+
|   UE       |   .    |    gNB     |   .   |     UPF     |
|+----+ +----------------------------------------+ +----+|
||TFTs| |              PDU Session               | |SDF ||
|+----+ |   +---------------+ #----------------# | |Fltr||
|       |   | Radio Bearer  | |   GTP-U Tunnel | | +----+|
|     +--------------------------------------------+     |
|     |                  QoS Flow                  |     |
|     +--------------------------------------------+     |
|     +--------------------------------------------+     |
|     |                  QoS Flow                  |     |
|     +--------------------------------------------+     |
|       |   +---------------+ |                | |       |
|       |   +---------------+ |                | |       |
|       |   | Radio Bearer  | |                | |       |
|     +--------------------------------------------+     |
|     |                  QoS Flow                  |     |
|     +--------------------------------------------+     |
|       |   +---------------+ #----------------# |       |
|       +----------------------------------------+       |
|            |   .    |            |   .   |             |
+------------+   .    +------------+   .   +-------------+
                 .                     .
                 Uu                      N3
  QoS rules            QoS Profiles         UL and DL
                                            Packet Detection
                                               Rules

The QoS flow is transported through the 5G core network via a GTP-U tunnel between the UPF and the gNB over a transport network. In large networks, the UPF can be placed flexibly in the network topology; this allows the UPF to be placed close to the device (UE) and its application and thereby enabling the shortest possible transport connection and reducing latency [ETR20]. In local deployments (e.g., industrial scenarios) a UPF is typically very close to the gNB and can be even located in the same rack. In the RAN, the QoS flow is transported via a radio bearer over the radio interface between the user equipment (UE) and the gNB.¶

3.4. Latency contributions in different layers of radio protocols

The Service Data Adaption (SDAP) layer maps the QoS flows to Data Radio Bearers (DRBs) and marks the packets with the QoS flow identifier. DRBs can be configured to be either in acknowledged mode (AM) or unacknowledged mode (UM) (see Figure 4); for an acknowledged mode DRB lossless data forwarding at handover is enabled for the Packet Data Convergence Protocol (PDCP) layer and Radio Link Control (RLC) operates in acknowledged mode. The latency impact of SDAP on data transfer is negligible.¶

             Ethernet or IP Traffic
  <------------------------------------------>
     |         QoS Flow                   |
     |<---------------------------------->|
     |                                    |
+--------+          +--------------+ +---------+
| +----+ |          | +----+-----+ | | +-----+ |
| |SDAP| |          | |SDAP|GTP-U| | | |GTP-U| |
| +----+ Radio Bearer +----+-----+ | | +-----+ |
|    |<----------------->|    |    | |    |    |
| +----+ |          | +----+-----+ | | +-----+ |
| |PDCP| |          | |PDCP|     | | | |     | |
| +----+  RLC Channel +----+     | | | |     | |
|    |<----------------->| |     | | | |     | |
| +----+ |          | +----+     | | | |     | |
| |RLC | |          | |RLC | IP  | | | | IP  | |
| +-- -+  Logical Ch. +----+     | | | |     | |
|    |<----------------->| |     | | | |     | |
| +----+ |          | +----+     | | | |     | |
| |MAC | |          | |MAC |     | | | |     | |
| +-- -+ Transport Ch.+----+     | | | |     | |
|    |<----------------->| |     | | | |     | |
| +----+ |          | +----+     | | | |     | |
| |PHY | |          | |PHY |     | | | |     | |
| +----+ |          | +----+-----+ | | +-----+ |
|    ^-------------------^    ^-----------^
|        |          |              | |         |
+---UE---+          +------gNB-----+ +---UPF---+

SDAP: ServiceData Adaptation Protocol
PDCP: Packet Data Convergence Protocol
RLC: Radio Link Control
MAC: Medium Access Control
PHY: Physical Layer

     QoS    QoS               QoS
    Flow1  Flow2             Flow3
      |      |                 |
+-----|------|-----------------|---------+
|     v      v     SDAP        v         |
|   +----------+          +----------+   |
+---|  DRB #1  |----------|  DRB #2  |---+
+---|    AM    |---+  +---|    UM    |---+
|   +----------+   |  |   +----------+   |
|                  |  |                  |
|    PDCP Entity   |  |  PDCP Entity     |
|                  |  |                  |
+------------------+  +------------------+
     ^                   ^     |
     |                   |     |
     v                   |     v
+------------+ +-----------+ +-----------+
|   RLC AM   | | RLC UM UL | | RLC UM DL |
+------------+ +-----------+ +-----------+
+----------------------------------------+
|                MAC and PHY             |
+----------------------------------------+

....................... Air interface (Uu)

At the next layer, the PDCP (Packet Data Convergence Protocol) layer provides ciphering for encryption of user plane data and optionally also integrity protection and verification via a message authentication code that is calculated for each data Protocol Data Unit (PDU). PDCP assigns a sequence number for each data PDU and forwards it to the underlying RLC layer. PDCP can also perform header compression and decompression over the radio link for the IP headers or Ethernet headers of the end-to-end data flow.¶

For acknowledged mode DRBs a copy of each PDCP PDU is stored in a local buffer. At changes of the RLC entity, due to either handover or (re-)configuration of dual connectivity or carrier aggregation, a lossless continuation of data transfer is ensured by forwarding not-yet-acknowledged PDCP PDUs to the new RLC entity.¶

As the underlying protocol layers can lead to packet re-ordering, the PDCP performs packet re-ordering to ensure in-order transmission of data over the DRB. For this, the receiver holds back the received packets until all earlier packets of the DRB have been received and are delivered first. A reordering timer determines how long packets are held back before delivery. In-order delivery leads to head-of-line blocking, which means that a long packet delay of one PDU (e.g., due to a larger number of retransmissions) affects also earlier packets. The impact of this head-of-line blocking is controlled via the reordering timer, which may reduce head-of-line-induced latencies at an increased risk of sending packets out of order. It is possible to configure the PDCP also for explicit out-of-order delivery, in which case no packet delay propagation within a group of PDUs appears.¶

The PDCP can be configured for Service Data Unit (SDU) discard, which enables to set a maximum lifetime on a packet in the radio transmission. If a configured SDU discard timer expires, the PDCP sender removes the packet from its buffer and requests the lower layer to purge the related data. SDU discard can be considered as a latency-based active queue management scheme.¶

The PDCP allows to aggregate multiple radio links over different frequency carriers, based on the NR functionality of carrier aggregation or dual-connectivity. The PDCP connection uses, in this case, multiple RLC entities; this can be used to aggregate the capacity of multiple radio links for the data radio bearer, but it can also be used to provide redundant transmission. For redundant transmission the PDCP entity duplicates PDCP PDUs and transmits them via multiple links; at the PDCP receiver, duplicates are then filtered out.¶

The PDCP uses one or more RLC channels, via one or more RLC instances. RLC provides reliable data transmission over the radio link via its acknowledged mode (AM); it can also be configured to apply the unacknowledged mode (UM) In AM mode, a selective-repeat ARQ protocol is used, in which correct reception of packets is ensured by detecting packet errors or losses and triggering retransmissions as needed. RLC transmitter and receiver entities maintain a sliding-window buffer, and the receiver entity updates the transmitter entity via status reports about correctly received or missing PDUs. The RLC receiver forwards correctly received PDUs to the PDCP receiving entity, which may comprise packets being delivered out-of-sequence. Reordering for in-sequence delivery is then performed in PDCP. RLC applies segmentation of SDUs towards the Medium Access Control (MAC) layer, so that the MAC protocol can multiplex RLC PDUs into the transport blocks sent by MAC to the physical layer.¶

From a packet delay perspective, minor latency contributions are made by packet processing. The larger possible latency contribution in acknowledged mode comes from the ARQ operation. A packet is maintained in the receiver buffer until it is successfully transmitted. For this, several RLC retransmissions can be used, where the maximum number of retransmissions is configurable. An RLC retransmission takes in the order of some tens of milliseconds, so that it can lead to some increased delay of packets that are not correctly transmitted in the first RLC transmission attempt. The need for RLC retransmission depends strongly on the configuration of the reliability that is configured for the lower MAC/PHY layers. For time critical low latency communication, typically the MAC/PHY is configured very reliably so that RLC retransmissions are not necessary. This trade-off we discuss more.¶

MAC entities are responsible for scheduling the radio resources for all bearers in UEs and gNB in both uplink and downlink directions. The RLC data segments received from multiple logical channels are concatenated along with MAC headers, padded if required, and then encoded to fit inside the scheduled Transport Block (TB) to be transmitted through the radio physical layer [NR-5G]. After the successful reception of the TB, the counterpart MAC entity decodes the TB and demultiplexes to the logical channels. Furthermore, the HARQ process of the MAC layer is responsible for handling most of the radio link errors. HARQ combines ARQ with Forward Error Correction (FEC) to efficiently enhance the reliability of communication in wireless channels. Via fast feedback the receiving MAC provides positive (ACK) or negative acknowledgments (NACK) back to the transmitter about successful TB decoding. One of the key functions of the MAC entity at gNB is to perform radio resource allocation for both Uplink (UL) and Downlink (DL) directions every TTI. The exact resource allocation process, considering factors such as Channel State Indicator (CSI), QoS requirements, and buffer occupancy, is beyond the scope of this document. However, it is important to note that the scheduler plays a crucial role in ensuring that the TB size (TBS) aligns with the chosen Modulation and Coding Scheme (MCS) and the number of Physical Resource Blocks (PRBs) allocated for the transmission. In addition to the above functions, the MAC also manages random access control during the initial access of UEs.¶

3.5. Latency Analysis

The latency analysis focuses on the following areas as contributors to the latency:¶

• Processing delays at gNB and UE¶

• Traffic handling / queuing¶

• Data transmission over the radio interface¶

• Reliability mechanisms (like HARQ)¶

In addition, further delays may be incurred due to mobility of devices or activating devices from power-saving idle states.¶

          Mobile BroadBand            Time-critical Communication

Probability                         Probability
  ^                                   ^          Latency
  |                                   |          Bound
  |                                   |           .
  |      xx                           |   xxxx    .
  |     x  x                          |   x   x   .
  |    x   x                          |  x     x  .
  |   x     x                         |  x     x  .
  |   x      x                        |  x      x .
  |  x       x                        | x       x .
  | x         xx                      | x       x .
  | x           xxxxxx                | x        x.
  +                   xxxxx           | x         x
  +-x----------------------x--->      +-x---------.x----->
                  ^          Latency              .^ Latency
                  |                                |
               Long tail                      Too late or lost