Chapter 11

NR URLLC

Abstract

In this chapter a general overview of NR is given with emphasis on the evolution from LTE. With this as base, the focus is set on the parts of the design of NR introduced to enable Ultra-Reliable Low-Latency Communications, going from the physical layer design up to procedures for idle and connected modes.

Keywords

Third generation partnership project (3GPP); 3GPP RAN; LTE; 4G; 5G; mMTC; Critical MTC; cMTC; URLLC; NR

11.1. Background

In this section a brief overview of 5G and New Radio (NR) is given, its relation to LTE is discussed, and Ultra-Reliable Low-Latency Communication (URLLC) is introduced with related design principles.

11.1.1. 5G system

The 3GPP 5G System (5   GS) is a new generation cellular system that aims to broaden the usage of cellular systems. In addition to voice and Enhanced Mobile Broadband (eMBB), 5   GS is intended to support the use case areas of massive Machine Type Communication (mMTC) and critical Machine Type Communication (cMTC).
The 5   GS is based on the 5G Core (5   GC) network and the NR radio access technology together aimed at meeting the 5G IMT-2020 requirements from ITU [1]. A first non-standalone version (NSA) of NR was standardized in 3GPP during 2017. It connects NR to the 4G Core Network (CN) Evolved Packet Core (EPC) through Dual Connectivity (DC) with LTE, effectively adding an NR cell to the existing LTE setup. The full standalone (SA) version, which connects NR to 5   GC, was finalized during 2018 as a part of 3GPP Release-15.
NR builds on the success of LTE and reuses well established concepts from its predecessor in the design of both the lower and higher layers. To that base, new features and modes of operation are added as motivated by the supported use cases. The main advances relate to the flexibility that is built into NR, increasing the scalability and adaptability compared to earlier systems GSM, WCDMA, and LTE. As an example a wider range of frequency bands spanning all the way up to 52.6   GHz is supported.
This chapter gives attention to the NR support for cMTC services, characterized by stringent requirements on service latency and reliability. In the following we will first give a brief introduction of NR and then continue by looking specifically at its Ultra-Reliable Low-Latency Communication (URLLC) features, designed for providing the needed support for critical services.
The process of standardizing the 5   GS as well as the basics of the NR physical layer (PHY) and higher layers are described in Chapter 2. An evaluation of the NR URLLC performance is presented in Chapter 11. For reference, the LTE version of URLLC is presented and evaluated in Chapter 9 and 10, and a comparison of the two technologies is given in Chapter 16.

11.1.2. URLLC

From the conception of 5G, it is designed to address three use case areas:
  1. • eMBB,
  2. • Massive Machine Type Communication (mMTC), as well as,
  3. • Critical Machine Type Communication (cMTC).
These areas can be said to represent the performance dimensions of spectral efficiency, connection efficiency, and service quality, respectively. The structure of NR is flexible enough to provide support for all these use cases within one integrated system, by allowing shifting of the operation point toward one of the three dimensions. A mix of diverse services can thereby be provided over a single 5G radio interface, and even shared on the same carrier.
Ultra-reliable low-latency communication (URLLC) is the technology created to deliver on the cMTC use cases, which need communication links with very high requirements on quality of service (QoS) and availability. Examples of such use cases include protection services for power substations and factory automation with industry robots, as will be studied in Chapter 12, and also further applications such as high-precision remote control and tactile communication, as for example with the case of remote surgery.
To be able to provide URLLC in NR, the system must be able to deliver data packets with short latencies and with high reliability, and the service must be consistently guaranteed. This is achieved by a combination of techniques specified in the standard, and is secured by careful radio resource management in the deployed network. The toolbox of URLLC features must be flexible and powerful enough to make it possible to reach the requirements of the cMTC services. At the same time, since NR also should provide eMBB and mMTC services and connectivity for new verticals, the specific tools applied need to coexist well with the operation of other tools and features. This trade-off explains the design choices taken for URLLC in NR. Compared to LTE URLLC however, in NR URLLC there are no concerns with backwards compatibility, which makes the design less constrained.

11.1.3. NR as the successor of LTE

NR is the natural successor of LTE, and the two RATs share many features. At the basis, both operate on an orthogonal frequency-division multiplexing (OFDM) radio resource grid, and specify the same kinds of physical data and control channels. But some important differences in design should be highlighted:
  1. • High bands: NR is designed to operate at higher frequencies (up to 52.6   GHz) and with higher carrier bandwidths (up to 400   MHz) compared to LTE, where a carrier is limited to 20   MHz and where the highest supported bands are at 5   GHz. The high frequency bands also facilitate small device form factors, thanks to the smaller antennas, which is important for many industrial applications. A scalable OFDM numerology in NR allows for efficient use of higher frequency bands.
  3. • Lean design: Minimized always-on transmissions enable NR networks to have much improved energy performance through extended micro-sleep periods.
  4. • Flexible design: The configuration and use of the NR time-frequency resources provides a high degree of flexibility, giving room for future enhancements and features.
  5. • Low latency: NR offers flexible scheduling and shorter processing times, which are important tools for optimizing the service latency.
  6. • Beam-centric: NR supports new and highly advanced antenna techniques, facilitating an antenna beam-centric design for enabling support of the new frequency range. Fast switching between beams belonging to different nodes is also supported.
At a glance then, NR offers the basic capabilities of LTE, while also being more forward-compatible by flexible design and enabling higher bands and more advanced transmission methods. The main step taken through NR is to be able to deliver on a wide range of use cases with a single integrated system. Using similar radio configurations we would expect roughly the same performance in LTE and NR in terms of spectral efficiency, but beyond the base NR comes with extra gears for higher data rates, lower latency and more options of operation option. Looking at network power consumption, NR is expected to be much more efficient due to the lean design with fewer obligatory signals always transmitted by the base stations.

11.1.4. Introduction of NR URLLC in existing networks

LTE is by now widely deployed on a large range of important spectrum bands ranging up to 5.9   GHz. NR on the other hand supports two frequency ranges (FR). FR1 corresponds to the existing LTE frequency range, while FR2 covers the range from 24 to 52.6   GHz. Traditionally, the mobile network operators (MNO) would stepwise re-farm parts of their existing spectrum to enable a roll-out of a new generation. However, since NR comes with a new set of bands not in current MNO use, an attractive alternative approach is to deploy NR in new rather than in current LTE bands. Therefore, the first NR deployments are expected to come as add-on higher frequency carriers to existing LTE deployments. The use of the FR2 bands, which support large system bandwidths and low latency, also allows NR to support demanding cMTC services with high traffic, such as factory automation, from day one. In addition, low latency can also be achieved on lower frequency bands thanks to short processing time and flexible scheduling with short transmissions.
Since 3GPP Release-15, the Evolved Packet Core (EPC) can, in addition to supporting LTE (called Option 1), also support NR as connected RAN. In this so-called non-standalone (NSA) NR mode (called Option 3), a carrier from an NR base station, called a gNB, provides service in a secondary cell in addition to the service provided in a primary cell by an LTE master eNB. This functionality is based on Dual Connectivity (DC), which in this case called E-UTRA-NR Dual Connectivity (EN-DC). Combined with the availability of new bands in FR2, EN-DC provides an attractive option for a seamless introduction of NR on top of existing LTE deployments.
With the deployment of 5G core networks (5   GC), NR can instead be connected in the main standalone (SA) mode (called Option 2). LTE bands can then be re-farmed to become NR bands, or alternatively NR can be deployed in the data region of existing LTE carriers, by using the method of LTE-NR dynamic spectrum sharing (DSS). With DSS, NR dynamically uses spectrum from an LTE carrier, for instance to guarantee coverage in a lower band for an NR device. DSS can also be used to ensure sufficient resources for legacy LTE devices. For the sharing to work there is no requirement on upgrading nodes or devices to 5   GC-compliant LTE, making it more universally applicable as a means to migrate to NR.
Besides these main three options other NR-LTE combinations based on 5   GC are also possible. Using DC, SA NR and 5   GC-compliant LTE nodes can be connected, either with NR as master node and LTE as secondary node (called Option 4) in NR-E-UTRA Dual Connectivity (NE-DC), or with LTE as master node and NR as secondary node (called Option 7) in NG-RAN-E-UTRA Dual Connectivity (NGEN-DC).
Fig. 11.1 illustrates the main architectural options 1–3 for NR and LTE. With DC, the main node serving the primary cell supports both control plane signaling and user plane data transmissions. The secondary cell provides added capacity by means of data transmission over the user plane. In the pure SA form, NR is run as a separate system either on its own carriers or dynamically scheduled on LTE carriers using DSS.

11.1.5. Radio access design principles

With URLLC we want to transmit data packets with very high quality of service (QoS) requirements, but we still want to use the same system as is otherwise used for more relaxed requirements. This is possible through careful scheduling, using the knobs and levers connected to a set of mechanism specified in the NR standard. These adjustments make it possible to optimize NR for high QoS for the use cases where this is required. In this chapter we will mainly focus on these optimization features and the adjustments relevant for URLLC.
How do we achieve high reliability in varying radio conditions? Broadly speaking, the way to do it is through diversity: the information is duplicated in multiple copies so that at least one attempted copy can be received, or it is spread out in frequency or code domain to ensure successful decoding. But diversity can mean many things, and typically a combination of the following types of tools would be used for achieving reliability in NR:
  1. • Code diversity. A lower code rate means that more coded bits are used to carry the information, and it is more likely that the received transmission conveys enough redundancy to decode the message. The code diversity can be achieved by use of a lower code rate, or by repeating the message, which effectively reduces the effective code rate.
  2. • Frequency diversity. Spreading the message over a wider band, either contiguously or non-contiguously, increases the possibility that some part of the message is sent in good channel conditions. Applying frequency hopping is another way of achieving the same thing.
  3. • Time diversity. Retransmissions of a message has both the effect of lowering the code rate after combining with previous attempts and in addition using the channel at different times, thereby increasing the chance of being in a better condition, in the case of a sufficiently time-varied channel. Attempts at different occasions can also enable interference diversity since different transmitters may be active at the time of transmission.
  4. • Spatial diversity. The use of multiple transmit and receive antennas or transmission points improves the reception by sampling several spatial channels with low correlation.
image
Fig. 11.1 LTE and NR connectivity options 1–3.
How do we then deliver a packet quickly? To answer this, we need to look at what actually constitutes the latency at RAN level when we transmit a packet. This includes delay for:
  1. • Encoding a packet at the packet transmitter side
  2. • Waiting for a transmission opportunity (alignment)
  3. • Over the air transmission duration
  4. • Decoding the packet at the packet receiver side
  5. • Waiting to transmit feedback, and transmission of feedback
  6. • Decoding the feedback at the packet transmitter side, preparing a retransmission, in case of packet failure
  8. • Waiting for a retransmission opportunity
  9. • Retransmission (transmission duration and decoding).
For uplink transmission with dynamic grant, additional steps for sending scheduling request (SR) and downlink control transmission are required before the uplink data transmission can start.
The delay contributions above are then the knobs to turn in order to reduce the delays. Correspondingly, the tools considered for achieving low latency in NR are:
  1. • Enhanced packet encoding/decoding
  2. • Reduced alignment delay by more transmission opportunities
  3. • Reduced transmission time
  4. • Reduced time between downlink data reception and transmission of feedback
  5. • Reduced time between uplink grant and data transmission.
The additional challenge that comes with URLLC is that we want to achieve both qualities at the same time: a packet must be reliably delivered within a certain time. This requirement has the form of being able to deliver a payload of P bytes within a latency of L milliseconds with a probability or reliability of R. In those terms we can define a cMTC service requirement. Delivering the cMTC service in a real system would then be additionally characterized by a coverage, having the form of fraction or coverage C of devices in a certain population to which the service can be supplied. Ensuring the coverage of a service can be interpreted as guaranteeing a minimum SINR level across a certain deployment scenario, for instance in a factory hall or in a city. The available tools for ensuring coverage relate to the transmitting side's ability to provide a strong signal through beamforming and power control, and the receiving side's ability to filter out the interference.

11.2. Physical Layer
The material in this section is based on the 3GPP Release-15 specifications for the NR physical layer [3,4,5,6].

11.2.1. Frequency bands
image
Fig. 11.2 Carrier FR defined for NR.
Naturally, radio propagation is more challenging at the higher frequencies because of higher attenuation and reduced diffraction. One of the main achievements in NR is to provide the tools to cope with these more challenging conditions, and thereby enable also cMTC services.

11.2.2. Physical layer numerology

11.2.2.1. Flexible numerology

NR uses cyclic prefix-based OFDM modulation in both downlink and uplink. It is also possible to use the digital Fourier transform (DFT)-spread OFDM modulation in uplink, also known as single-carrier frequency-division multiple access, (SC-FDMA) could be introduced which is the only available modulation in LTE uplink. The availability of the two options for uplink in NR enables both the scheduling flexibility of OFDM and the possibility of improved coverage of DFT-OFDM, coming from its reduced peak-to-average power ratio (PAPR).
The OFDM subcarriers, which are grouped in sets of 12 into resource blocks (RBs), are separated by a subcarrier spacing (SCS) in frequency domain. NR can be configured with different SCS, defining different numerologies, while LTE only runs on 15   kHz SCS for data transmission. With a subcarrier spacing of W Hz, the duration of an OFDM symbol (OS), excluding the cyclic prefix (CP), is nominally 1/W s [7]. With the basic SCS of W   =   15   kHz, 14 OS including CP takes 1   ms to transmit. One subcarrier during one OS is called a resource element (RE), as in LTE, and can carry one modulated symbol. In NR, the SCS can be set to N∗W for N = 2 μ , with numerology μ = {0, 1, 2, 3}, meaning that 14∗N OSs are transmitted per millisecond. The shorter symbol durations mean that a packet can be transmitted faster, at the cost of using more bandwidth. In NR, with higher frequencies considered, wider frequency bands also become available, so this tradeoff becomes reasonable and interesting for low-latency applications.
The set of available subcarrier configurations are listed in Table 11.1, together with the applicable FR and the supported maximum system bandwidths.

Table 11.1

NR numerologies for data transmissions.
Numerology μ SCS [kHz] Slot duration [ms] Symbol duration [ms] Normal cyclic prefix [μs] Frequency range Maximum nr. of RBs Maximum bandwidth [MHz]
0 15 1 1/14 4.7 FR1 270 50
1 30 0.5 1/28 2.3 FR1 273 100
2 60 0.25 1/56 1.2 FR1 (Optional) 135 100
FR2 264 200
3 120 0.125 1/112 0.59 FR2 264 400

image

At the highest subcarrier spacing intended for wide areas, 60   kHz, the symbol duration is short enough for the delay spread of the radio channel to become an issue in large-distance scenarios (i.e. the delay spread exceeds the CP which implies that the orthogonality between symbols is degraded and in worst case lost). To reduce the harmful effect of inter-symbol interference the CP duration can be set to a longer value (extended CP) at the cost of using fewer data symbols per second, from 14∗N symbols/ms to 12∗N symbols/ms.

11.2.2.2. Frame structure

In time domain the basic unit is the subframe, being 1   ms long, irrespective of used numerology. Depending on the chosen numerology, which determines the symbol duration, each subframe consist of a certain number of slots, each in turn being a set of 14 (normal-CP) or 12 (extended-CP) OS, as illustrated in Fig. 11.3. The basic transmission duration is a slot, but a transmission can occupy also part of a slot, a so-called non-slot transmission or commonly mini-slot, as illustrated in Fig. 11.4, where the frequency domain and different SCS are illustrated. In downlink, a mini-slot can start in any symbol and be 2, 4, or 7 OS long, whereas in uplink it can start in any symbol and be of any length up to the slot length, i.e. 1 to 14 OS long.
In NR the slot structure in terms of uplink and downlink symbols is flexible and there is therefore no principled difference between FDD and TDD slots, as described further below.

11.2.3. Transmissions schemes

11.2.3.1. Beam-based transmissions

image
Fig. 11.3 Slot structure in NR with normal-CP.
image
Fig. 11.4 Slots and example of mini-slots with different SCS.

11.2.3.2. Bandwidth parts

A device operating on an NR carrier can be configured to use only a part of the available bandwidth, a so-called bandwidth part (BWP). This is introduced to allow the operation of lower-complexity devices with less capable radios, and also to allow for further power saving in the devices by limiting the monitored bandwidth. The BWP configured for a device therefore constitutes the activated bandwidth, which could be distributed on different frequency bands and using different numerologies. In NR Release-15 up to 4 BWP can be configured, but only one BWP can be active at a time. This means for instance that the device cannot simultaneously operate in both FR1 and FR2.

11.2.3.3. Duplex modes

As mentioned above, there is one single frame format in NR supporting both TDD and FDD configurations. The device looks for downlink control messages in configured search spaces (contained in CORESETs, as described below), and uses the slot for downlink reception or uplink transmission as dynamically indicated by the gNB. The slot can be partly used for downlink and partly for uplink, with a gap period around the switching point to avoid cross-link interference. A set of slot formats are defined with a sequence of downlink symbols, a set of flexible symbols, and a set of uplink symbols, see examples in Fig. 11.5. The gap period between downlink and uplink is thus taken from the flexible symbols. With TDD operation in a downlink slot the device expects that downlink and flexible symbols can be indicated for data reception, and in an uplink slot the flexible and uplink symbols can be used for transmission. The slot format can either be configured over RRC, dynamically indicated to each device in the DCI, or indicated to a set of devices in a special downlink control information (DCI) called slot format indicator (SFI). The sequence of slot formats indicated to the device the constitutes the TDD pattern of downlink and uplink slots. The fact that all symbols can be indicated as flexible allows for fully dynamic TDD, where the device is allocated downlink and uplink symbols on a per slot basis.
image
Fig. 11.5 Examples of NR slot formats for FDD and TDD use.
The mixed downlink-uplink slot formats also support so-called self-contained slots, where the Hybrid Automatic Repeat Request (HARQ) feedback is sent from the device toward the end of the downlink slot. This setup can enable very short roundtrip times with a high number of retransmissions possible in short time – a key feature for URLLC services. However, given a necessary gap period and allowing for processing on the device side, the number of symbols that can be used for data may be reduced, as further discussed below.
Comparing TDD with FDD, it is obvious that the alignment delay is necessarily longer when the direction is fixed, since we need to wait for the next downlink or uplink period and not only the next slot or mini-slot as in FDD. This is true even if the gNB can specify the configuration on a slot basis, since the uplink or downlink transmission opportunities are once per slot. Note that due to this long alignment delay we expect no significant latency reduction from using mini-slots in TDD. To reduce the latency further we would need more downlink-uplink switching points per slot, but this is not supported in NR Release-15. But even with these limitations, NR TDD operation constitutes a big improvement from LTE when it comes to latency. In LTE the shortest uplink-downlink switch period is 5   ms, placing a high floor on achievable latency, while in NR Release-15 it can be as low as 1/8   ms (one slot with 120   kHz SCS).
The support of dynamic TDD indication is interesting from a system efficiency perspective. Resources in uplink and downlink can then be tailored to the current need, which is not possible with a fixed slot allocation. However, for URLLC services dynamic TDD does not bring significant latency gains above a fixed allocation since the format is anyway set on slot basis with only one switching point, and more importantly the reliability may be risked since cells may independently change configurations resulting in so called cross-link (uplink-to-downlink and downlink-to-uplink) interference, which can have disastrous effects on link quality. It is therefore expected that dynamic TDD will in practice be most useful in controlled environments with limited number of cells and low interference.
When running URLLC in TDD bands, we can instead select a fixed configuration with as high frequency of switching points as possible, for instance by using a sequence of mixed downlink-uplink slot formats with a high SCS for short latency, as is illustrated in Fig. 11.6.

11.2.3.4. Short transmissions

The slot is the basic scheduling interval in NR, meaning that as a baseline the downlink control channel is sent once per downlink slot. This basic type of scheduling in blocks of slots is called Type A mapping, which in downlink can start in symbol 0–3 and have a duration of 3–14 symbols, as shown in Fig. 11.7, and in uplink start in symbol 0 and have a duration of 4-14 OS. As mentioned above it is also possible to use shorter transmissions, which is what is referred to as non-slots or mini-slots, that can be scheduled with a shorter interval. These shorter transmissions can be scheduled in both downlink and uplink through Type B mapping, starting in any symbol and having a duration in the range 1–14 OS in uplink, but in downlink starting in symbol 0–12 and limited to {2, 4, 7} symbols in the downlink, as was exemplified in Fig. 11.4. Allocations of both types are not allowed to span over a slot border, as is seen in Fig. 11.7. This means that transmissions can be very flexibly scheduled, but not always with OS granularity. Of course, the flexibility comes at a cost; in downlink the devices need to monitor more downlink control occasions in order to see if they are scheduled.
image
Fig. 11.6 Examples of possible low latency TDD configurations. Top: a lower SCS with mixed downlink-uplink slots. Bottom: a higher SCS with alternating downlink and uplink slots.
image
Fig. 11.7 Downlink allocations in time domain.

11.2.3.5. Short processing time

When it comes to timing, NR specifies the minimum scheduled timing that the device should manage for two cases:
The length of these delays depends on a set of scheduling factors:
The factor of device processing capability reflects the processing speed of the device, out of which Capability 2 is targeting low-latency services such as URLLC, whereas Capability 1 is the basic processing speed. The defined expressions for d 1 and d 2 are set as the sum of the minimum processing timing parameters N 1 and N 2 , given for Capability 1 and 2 in Table 11.2, and scheduling dependent delays. It should be noted that for 120   kHz SCS Capability 2 is not defined in Release-15, so Capability 1 values are assumed. The range of values reflects the intention of optimizing the allowed latency based on how computationally difficult the tasks are.
For d 1 the sum (in OS) is computed as:
d 1 = N 1 + d 1,1
image
where

Table 11.2

Minimum processing timing parameters for capability 1 and 2.
SCS Capability 1 Capability 2
N 1 image [OFDM symbols] N 2 image [OFDM symbols] N 1 image [OFDM symbols] N 2 image [OFDM symbols]
1 DMRS >1 DMRS 1 DMRS >1 DMRS
15   kHz 8 13 (14 if in OS #12) 10 3 13 5
30   kHz 10 13 12 4.5 13 5.5
60   kHz 17 20 23 9 (FR1) 20 11 (FR1)
120   kHz 20 24 36 20 (Cap. 1) 24 (Cap. 1) 36 (Cap. 1)

image

d 1,1 = { 7 − i [ Mapping type A , i < 7 ] 3 [ Mapping type B , Capability 1 , allocation 4 OFDM symbols ] 3 + d [ Mapping type B , Capability 1 , allocation 2 OFDM symbols ] d [ Mapping type B , Capability 2 , allocation 2 or 4 OFDM symbols ] 0 [ Otherwise ] .
image
where i is the index of the last symbol in the PDSCH allocation, and d is the number of overlapping PDCCH and PDSCH symbols. As seen in the expressions the placement of downlink and uplink control impacts the processing delay. The intention here is to give the device some extra headroom for handling parallel processing tasks.
For d 2 we have the sum (in OS):
d 2 = N 2 + d 2,1
image
where
d 2,1 = { 0 [ Only DMRS in first PUSCH symbol ] 1 [ Otherwise ] .
image
Here the device is allowed an extra symbol to perform the channel estimation from the DMRS before starting the decoding.
From the expressions above we can calculate the processing delays for some interesting URLLC configurations, with the assumption of Capability 2 processing in the device, see Table 11.3. Here it is assumed that 1 DMRS is configured in the PDSCH, that PDCCH and PDSCH overlap in 1 symbol for 4 and 2 symbol transmissions in downlink, and that DMRS is mixed with data on the first PUSCH symbol. It should be noticed here that processing delays do not decrease with shorter allocations, perhaps contrary to expected.
With this specified minimum timing in mind, the network explicitly indicates the uplink timing to the device in the DCI (see Section 11.3.3.1.2) for either the downlink HARQ feedback or the uplink data.

Table 11.3

Example of processing delays with capability 2 for 15   kHz and 120   kHz SCS.
Allocation d 1 image [OFDM symbols] d 2 image [OFDM symbols]
15   kHz SCS 120   kHz SCS (Cap. 1 values) 15   kHz SCS 120   kHz SCS (Cap. 1 values)
Slot [14 symbols] 3 20 6 37
7 symbols 3 20 6 37
4 symbols 4 21 6 37
2 symbols 4 21 6 37

image

11.2.3.6. Downlink multi-antenna techniques

There are many possibilities to improve the signal quality in the downlink. Perhaps the most obvious way is to equip the device with multiple cross-polarized antennas that receive the same transmission. Coherent processing of the signal gives receiver-side directivity, leading to suppression of noise and interference, and thereby improved quality. It is reasonable to expect that more high-end NR URLLC devices targeting cMTC services have at least 4 receive antennas, i.e. 2 pairs of cross-polarized elements.
Also without equipping the device with more antennas it is possible to utilize powerful beamforming on the gNB side, using the larger arrays. With increasing frequency, the antenna elements become smaller and it is feasible to have reasonably sized arrays with many elements. Since the size of the antenna elements are proportional to the wavelength (typically in the range of λ / 2 image ), a doubling of carrier frequency leads to half the array size, which would allow for twice as many elements if the same array size is kept. At the same time, the maximum directional gain of the antenna can be expected to scale with the number of elements. One can therefore typically enable much higher directional gains at the higher carrier frequencies used in NR.
Since the device-specific DMRS in NR are inserted in the downlink transmission, the beamforming can be made transparent to the device and doesn't need to be indicated. The DMRS are beamformed in the same way as data, and when the device demodulates based on the reference signals the beamforming is automatically handled.
For the antenna array the gNB can select the beam in two basic ways; either based on a precoder matrix (known as digital beamforming) or based on transmission weights (known as analogue beamforming). In the analogue setup the beam is a main lobe formed after digital-analog (D/A) conversion, and is typically chosen by sweeping a set of beams in a grid and letting the device report the received signal strength. This is rather robust and low-overhead option, but has the drawback of being restricted to one beam at a time.
The analogue beamforming with fixed beams is well suited for higher carrier frequencies when the focus is to achieve a high beamforming gain. This is because of the higher number of elements required and the thereby increasing processing complexity. A cell can have multiple associated fixed beams associated to it, each with its own set of synchronization and system information transmitted to the device. The device associates to the best beam during random access (initial beam establishment) and can then be moved between beams within the same cell on the physical layer (beam adjustment) based on measurement reports from signals in the beams. This change does not involve the handover procedure as in a cell change.
Using digital beamforming with a precoder matrix, the device is configured to receive pilot symbols, and try out a set of preconfigured precoders from a codebook. The index of the codebook giving the highest quality is indicated in a precoding matrix indicator precoding (PMI) reported by the device. Using precoders makes it possible for the gNB to use multiple beams to different devices at the same time, and it also enables MIMO transmissions with multiple data layers. With URLLC, however, we do not aim to achieve high spectral efficiency but rather robustness. Still, precoder-based beamforming can be a good choice for URLLC as it can be faster and more precise than grid-of-beams beamforming. The drawback is the extensive need for pilots and measurement reports, as well as the risk of using wrong precoder when conditions change.

11.2.3.7. Uplink multi-antenna techniques

For uplink transmissions a device with multiple antenna elements can use beamformed MIMO methods to improve the channel quality. Both non-codebook based (only in the case of downlink-uplink reciprocity over TDD, giving the full precoder matrix) and codebook-based (using an index corresponding to a precoder matrix) precoding with up to 4 antenna ports is supported. For codebook-based precoding the device needs to have the capability of performing fully coherent transmissions and thus be able to control the relative phase of the antennas. The gNB estimates the channel from sounding reference signal (SRS) (see Section 11.2.5.1.2) transmissions from the device, and then indicates a precoder in the uplink grant. The indication of precoder and the additional SRS transmissions means an increased downlink and uplink overhead. On the other hand, the use of beamforming should lead to improved SINR and thereby lead to higher reliability, but only as long as the right precoder is selected. A high rate of SRS transmissions may be needed to ensure the use of the right beam.
In addition to beamforming, the network can use coordinated multi-point reception to improve uplink quality. This implies that the signal is received in multiple locations (e.g. cells belonging to the same three-sector site) and processed jointly. To enable this, no additional signaling is required.

11.2.4. Downlink physical channels and signals

In downlink the most important aspects for URLLC is to ensure high reliability of the control channel PDCCH and the data channel PDSCH. Since the gNB can immediately react to an incoming data packet addressed to the device, scheduling in downlink is faster than in uplink, and the main concern is to provide scheduling opportunities with short intervals.

11.2.4.1. Synchronization and broadcast signals

Slot Periodic 5   ms–160   ms
Subcarrier spacing {15, 30, 120, 240} kHz
Bandwidth 240 subcarriers
Frequency location Predefined raster locations

To enable low energy consumption through long sleep cycles, NR is defined with one unified broadcast message, the synchronization signal block (SSB). This transmission contains both system information and synchronization signals, both used during initial access. By using a long period and differently beamformed SSB, both good energy performance and good coverage can be achieved.
SSB consists of three channels: the primary synchronization channel (PSS), the secondary synchronization channel (SSS), and the physical broadcast channel (PBCH). PSS and SSS are 127 subcarriers wide and PBCH is 240 subcarriers wide, surrounding SSS in frequency and time domain. By placing the three channels in one limited time-frequency region, as shown in Fig. 11.8, and using a longer period (up to 160   ms) it is possible for the network to reduce its power consumption through micro-sleep. Also, the device power consumption can be reduced by using a limited set of frequency domain locations of the SSB.
image
Fig. 11.8 SSB consisting of PSS, SSS, and PBCH.
From PSS and SSS the device derives the basic timing of the cell as well as the physical cell identity. Over PBCH the device receives the master information block (MIB) containing information for how to read the remaining system information, transmitted in system information block one (SIB1) on the downlink data channel.

11.2.4.2. Reference signals

11.2.4.2.1. DMRS

Slot Any
Subcarrier spacing {15, 30, 60, 120, 240} kHz
Time location {1, 2, 3, 4} symbols per slot
Frequency location On {4, 6} subcarriers per resource block

For both data and control in the downlink, device-specific DMRS are used. The DMRS is, just as in LTE, a predefined sequence known by the device from which it can estimate the channel state. Demodulation of up to 12 different antenna ports can be supported depending on configuration as given in Table 11.4, by applying different orthogonal cover codes in time and frequency over the DMRS resources. The DMRS are inserted within the data or control regions and are only transmitted when the device is receiving information.

Table 11.4

Number of supported antenna ports for different DMRS configurations.
#Ports Single Double
Type 1 4 8
Type 2 6 12
image
Fig. 11.9 Left: DMRS placement for PDSCH slots (Mapping A, Type 1, 2 single DMRS) and mini-slots (Mapping B, Type 1, 1 single DMRS). Right: examples of Type 1 and Type 2 DMRS.
For PDCCH (Section 11.2.4.3) the DMRS is inserted on every fourth subcarrier in the region where PDCCH is sent (see Fig. 11.8). PBCH (Section 11.2.4.1) also incorporates its separate DMRS for demodulation.
Two different time-domain mappings and two different frequency-domain types of DMRS are defined for PDSCH (Section 11.2.4.4). The used mapping and type are indicated in the downlink control DCI message.
Mapping A is intended for slot length transmission. In this case the first DMRS is inserted after the downlink control in the third or fourth symbol of the slot, independent of where the data is located.
Mapping B is intended for mini-slot transmission. Here the first DMRS is inserted in the first symbol of the mini-slot allocation, a configuration known as front-loaded DMRS. In case of overlap with a CORESET, DMRS is moved to the first symbol after CORESET.
On top of this basic mapping, extra DMRS can also be inserted for support of e.g. higher speeds, and double DMRS symbols can be used to support more antenna ports. An illustration of the mapping is shown in Fig. 11.9. Up to 4 single or 2 double DMRS symbols can be configured with both Mapping A and B.
On top of the mapping, there are two DMRS types: Type 1 DMRS is placed on every other subcarrier in the symbol, while Type 2 DMRS is placed on 4 out of 12 subcarriers in a RB over a symbol, also shown in Fig. 11.9. Both types can be sent as single or double DMRS symbols.
11.2.4.2.2. PT-RS
11.2.4.2.3. CSI-RS
11.2.4.2.4. TRS

11.2.4.3. PDCCH

Slot Any
Subcarrier spacing {15, 30, 60, 120} kHz
Time location 1-3 symbol duration, any symbol
Bandwidth 2-96 resource blocks

As in LTE, the downlink control channel in NR is called PDCCH, and carries instructions to the device on how to read and transmit data. The PDCCH in NR can be flexibly configured in time and frequency, using a periodicity of once per slot or more, and can span 1–3 OSs. A device is configured with one or several search spaces, defined within control resource sets (CORESETs), where it blindly detects the presence of DCI, described in Section 11.3.3.1.2, which are coded on control channel elements (CCEs) described below. So PDCCH consists of DCI transmitted in CCEs located in CORESETs. The channel coding used for PDCCH is based on Polar codes. The DCI is encoded with a radio network temporary identifier (RNTI), which is the address of the device in the cell, by scrambling of an appended cyclic redundancy check (CRC).
image
Fig. 11.11 PDCCH with DMRS placement, aggregation levels and REGs.
11.2.4.3.1. CCE
PDCCH can be sent either in a non-interleaved manner, meaning on a contiguous allocation, or interleaved manner, where the REGs are distributed for increased diversity. In the non-interleaved case, the REGs are bundled in CCE sizes of 6 as described above and mapped without gaps in frequency domain, while in the interleaved case the CCEs are spread out in frequency using REG bundles of 2 or 6 REGs for 1–2 symbol PDCCH, or 3 or 6 REGs for 3 symbol PDCCH.
11.2.4.3.2. CORESET
A device can be configured with multiple CORESETs, which constitute configured resources on the time-frequency grid where the device is to perform blind decoding of CCEs to find DCIs addressed to it. The location of the basic set, CORESET-0, is signaled in the master information block on PBCH, as described in Section 11.2.4.1, and additional CORESETs are later configured over Radio Resource Control (RRC). These resources are then the possible scheduling occasions for a device, and to achieve low alignment latency in NR URLLC the CORESETs should be configured with short interval. A CORESET can be 1–3 OS long (3 only when DMRS Type A is used in the fourth symbol of the slot, see Section 11.2.4.2.1), and be placed anywhere in the slot with multiple device-specific locations. CORESETs can overlap with each other but not with SSB.
Short intervals between CORESETs enable low latency, but having many CORESETs configured comes with some obvious costs. First, the device has more locations to monitor which increases power and processing needs. Second, more CORESETs likely means more blind decoding attempts performed by a device, with increased risk of falsely detecting downlink control from noise and thereby missing a real transmission. Third, we either need to reserve some resources for potential downlink control (which is expensive from a resource efficiency perspective), or we may be forced to interrupt a data transmission to send downlink control (which disturbs the data transmission). The latter alternative is supported by the feature of downlink pre-emption discussed in Section 11.2.4.4.3. As for sTTI in LTE, the unused CORESETs configured for a device can be reused for PDSCH transmission by indicating this in the DCI.
For 15/30/60/120   kHz SCS, the maximum number of blind decodes a device can be configured with per slot is 44/36/22/20, and the maximum number of CCEs is 56/56/48/32, respectively, with no restriction related to aggregation level.
A CORESET can be said to represent a possible start of a slot or mini-slot. By allowing CORESETs to be configured in any symbol of the NR slot we enable mini-slots to start anywhere, and the location is a matter of scheduling. Given that the available downlink transmission lengths are 2, 4, 7, or 14 symbols in NR Release-15, the gNB can then configure CORESETs with a matching periodicity to enable many scheduling opportunities, as illustrated in Fig. 11.12 for 4 symbol transmissions.
image
Fig. 11.12 Illustration of CORESET placements for slot and 4-symbol mini-slot transmissions.

11.2.4.4. PDSCH

Slot Any
Subcarrier spacing {15, 30, 60, 120} kHz
Duration 1-14 symbols
Bandwidth Any (see Table 11.1)

PDSCH data is coded with defined low density parity codes (LDPC) according to code rates given in three modulation and coding scheme (MCS) tables. When a device receives data, it first reads a DCI message addressed to it on PDCCH. If the DCI is correctly decoded the device knows which MCS table to use, which table entry to use, and on which resources the data is found, and it can then start decoding. The data location is indicated as a start symbol and a duration of the message. This very flexible setup allows for locating the message practically anywhere, except that a transmission cannot spill over into the next slot.
Smaller downlink transport blocks (<3824 bits) are protected with a CRC of 16 bits and larger blocks with a CRC of 24 bits, appended before scrambling with the RNTI of the device.
By replacing the Turbo code used for data in LTE with LDPC improved performance for higher payloads is achieved. The codes used in NR are constructed with the help of two optimally constructed matrices, called base-graphs, one for smaller packets (up to 3840 bits) and one for larger packets (up to 8448 bits). Larger transport block (TB)s are segmented into code blocks(CBs), each with a 24-bit CRC added, before being encoded with LDPC and thereafter concatenated to PDSCH. The size of a set of CBs, a CB Group (CBG), can be defined over RRC, and retransmission can be handled separately per CBG for improved efficiency.
Similar to LTE, data is rate-matched using a circular buffer and sent with incremental redundancy for increased efficiency. A redundancy version (RV) in the sequence {0,2,3,1} is indicated for each data transmission, and the device can then place them in a receive buffer and efficiently combine the different sets of coded bits from multiple copies of a received transport block (from retransmission).
11.2.4.4.1. MCS table for low code rate
In LTE the lowest code rate in MCS-0 is around 0.094, meaning that the code word is about 10 times longer than the information bits. The extra redundancy of using a longer code gives robustness and leads to a reduced error rate. For NR URLLC the design target in Release-15 was to reach a block level error rate (BLER) of 10 −5   at a certain SINR value, the so-called Q-value (see discussion in Chapter 12), which represented the cell-edge of a realistic deployment scenario. To reach this with one transmission attempt the code rate needs to be very low, meaning that many REs are used to send the message. To cater for different needs three MCS tables have been specified in NR, two for higher rate services (table 1 up to 64QAM and table 2 up to 256QAM) with moderate BLER requirement (typically around 10%) such as eMBB in mind, and one with lower rates and low BLER requirement for URLLC in mind (Table 11.3), shown in Table 11.5. This URLLC MCS Table 11.3 includes modulation up to 64QAM using a few higher rates, and in addition reaches down to a code rate of 0.03 with QPSK modulation, which is a third of the lowest LTE rate.

Table 11.5

Low rate MCS table (corresponding to table 5.1.3.1–3 in Ref. [6]).
MCS index Modulation Target code rate R x [1024] Spectral efficiency
0 QPSK 30 0.0586
1 QPSK 40 0.0781
2 QPSK 50 0.0977
3 QPSK 64 0.1250
4 QPSK 78 0.1523
5 QPSK 99 0.1934
6 QPSK 120 0.2344
7 QPSK 157 0.3066
8 QPSK 193 0.3770
9 QPSK 251 0.4902
10 QPSK 308 0.6016
11 QPSK 379 0.7402
12 QPSK 449 0.8770
13 QPSK 526 1.0273
14 QPSK 602 1.1758
15 16QAM 340 1.3281
16 16QAM 378 1.4766
17 16QAM 434 1.6953
18 16QAM 490 1.9141
19 16QAM 553 2.1602
20 16QAM 616 2.4063
21 64QAM 438 2.5664
22 64QAM 466 2.7305
23 64QAM 517 3.0293
24 64QAM 567 3.3223
25 64QAM 616 3.6094
26 64QAM 666 3.9023
27 64QAM 719 4.2129
28 64QAM 772 4.5234
29–31 Reserved

image

To tell the device which table to use, different RNTI encoding of the DCI is used. Devices are normally configured over RRC to expect DCI addressed to a default high-rate table RNTI, and for URLLC devices an additional configuration of a low rate table RNTI is made. The network can then dynamically switch between the two tables depending on the requested service, and the device will attempt decoding with both RNTIs out of which one is correct.
11.2.4.4.2. Downlink repetitions
Compared to HARQ retransmissions these repetitions are not triggered by feedback and are therefore sometimes called blind, automatic, or HARQ-less repetitions. The fact that they are always transmitted regardless of the decoding success makes repetition much less spectrally efficient compared to HARQ retransmissions. Moreover, the fact that the DCI is only transmitted once is a weak spot: if the device fails to decode the downlink control message the multiple copies of data don't matter. However, by using high enough AL for downlink control this problem can be avoided.
11.2.4.4.3. Downlink pre-emption
Unless we have enough critical traffic it is wasteful resource-wise to dedicate a certain frequency band for URLLC transmissions. This is true also for downlink data, where we would need to avoid scheduling longer transmissions in a part of the carrier to give room for shorter transmissions with short notice. And since the downlink control can be sent just before the downlink data, there is no latency gain from using configured grants (see Section 11.3.3.4), in downlink called semi-persistent scheduling. A new mechanism, downlink pre-emption indication, was therefore introduced in NR to remove the issues connected to direct insertion of short data transmissions within ongoing transmissions. This solves both problems: direct access to downlink resources and using the whole carrier for longer transmissions.
If the whole downlink carrier is already used for slot-based transmissions and the gNB receives a high priority packet in the buffer, it can simply pre-empt the high priority data in the middle of the slot by using the next upcoming configured CORESET. Naturally, this means taking resources from the first transmission which will then likely fail. Moreover, it would not only fail, but when a retransmission is later sent to fix the first packet, the receive buffer would be contaminated with other data and the decoding would likely continue to fail. To fix this, a control message, a pre-emption indication (PI) DCI, is sent to the device that received the interrupted message, see illustration in Fig. 11.14. The PI informs the device that the last attempt was pre-empted in an indicated resource region, and the device can then clean the receive buffer from the corrupted bits, and wait for the next retransmission attempt, which may be complete or only cover the impacted CBGs.
image
Fig. 11.13 Illustration of downlink repetition (K   =   3) of a 7-symbol data allocation.
image
Fig. 11.14 Illustration of downlink pre-emption and pre-emption indicator (PI).
Downlink pre-emption thus offers quick downlink resources with affordable overhead. As long as the URLLC traffic is low compared to other traffic, the pre-emptied packets will not cause a big disturbance on the other traffic.

11.2.5. Uplink physical channels and signals

In the uplink, we face two main challenges with URLLC compared to downlink. First, the scheduling is controlled by the gNB while the transmit buffer status is known by the device, meaning that resources either needs to be requested or pre-allocated with cost on either latency or resource efficiency. Second, devices have a limited range of transmission power, fewer antenna elements and smaller antenna spacing, making the uplink more sensitive to poor radio conditions.

11.2.5.1. Reference signals

11.2.5.1.1. DMRS

Slot Any
Subcarrier spacing {15, 30, 60, 120} kHz
Time location {1, 2, 3, 4} symbols per slot
Frequency location {4, 6} subcarriers per resource block

Since the uplink is also based on OFDM, DMRS is used in the same way as for downlink (see Section 11.2.4.2.1), configurable with the same options of Type 1 and 2 placement in frequency domain and Mapping type A and B in time domain. An example configuration of DMRS resources is shown in Fig. 11.15, and an uplink RB pattern example is given in Fig. 11.16. But in uplink it is also possible to run with DFT-precoded OFDM for improved coverage. In this case the DMRS spans an entire symbol across the uplink frequency allocation, as in the LTE case.
11.2.5.1.2. SRS

11.2.5.2. PRACH

Slot Configurable slot set with 10–160   ms period
Subcarrier spacing {1.25, 5, 15, 30, 60, 120} kHz
Duration {1, 2, 4, 6, 12} symbols plus CP
Bandwidth {139, 839} subcarriers, in blocks of {1, 2, 3, 6, 7}

When the device has acquired the system information it can register to a gNB by initiating random access. The first step in random access is to select and transmit a preamble over the physical random access channel (PRACH), followed by a random access response from the network containing a first RNTI and an uplink grant. PRACH is sent on RACH resources in the uplink, which are derived from the system information, and consists of a Zadoff-Chu sequence-based preamble. There are two main types of preambles defined: four long preamble formats (sequence length 839) with a SCS of 1.25 or 5   kHz and only in FR1, and nine short preamble formats (sequence length 139) for 15 or 30   kHz SCS in FR1 and 60 or 120   kHz SCS in FR2 (same as for data). By the use of different preamble formats different CP and repetition can be applied for improved coverage, so that the total duration can be set between 2-12 OS (short formats) and 1-4.3 ms (long formats). Depending on the combination of PUSCH and PRACH SCS the formats will occupy between 2 and 24 RBs on PUSCH, and for short formats always 12 RB. In frequency domain a RACH resource can consist of up to 7 blocks of consecutive copies of the sequence mapped to integer RBs.

11.2.5.3. PUCCH

Slot Any
Subcarrier spacing {15, 30, 60, 120} kHz
Duration {1, 2, 4–14} symbols
Bandwidth From 1 resource block

Since the uplink data transmission duration can be flexibly set in the wide range 1–14 symbols, we must be capable of adjusting the length of the physical uplink control channel (PUCCH) as well. To handle this, NR allows several formats and durations of PUCCH, which is used to send uplink control information (UCI) when it's not transmitted together with data on PUSCH. Five different PUCCH formats (0–4) are defined, see Table 11.6. Up to 4 PUCCH resource sets, mapping to different payload ranges, can be configured for a device, with each set consisting of at least 4 PUCCH configurations, each specifying a format and a resource. The configuration to be used within the payload-determined set is then indicated in the DCI, as described in Section 11.3.3.1.2. Naturally, for URLLC the shorter lengths are more interesting choices since they can be sent more often and can be decoded faster, thereby reducing latency. However, for power limited devices, a longer transmission duration is key to achieve a good enough SINR.

Table 11.6

NR PUCCH formats.
PUCCH format Length (OFDM symbols) Payload (bits) Bandwidth (RBs) Intra-slot hopping possible Multiplexing capacity (devices) Coding
Short PUCCH 0 1–2 1–2 1 Yes (for 2 symbols) 3–6 Sequence selection
2 1–2 >2 1–16 Yes (for 2 symbols) 1 Reed-Muller (3–11   b), Polar (>11   b)
Long PUCCH 1 4–14 1–2 1 Yes 1–7 Block
3 4–14 >2 1–16 Yes 1 Reed-Muller (3–11   b), Polar (>11   b)
4 4–14 >2 1 Yes 1–4 Reed-Muller (3–11   b), Polar (>11   b)

image

Besides delivering channel quality index (CQI) values to be used for scheduling, PUCCH delivers two critical pieces of information: the SR and the HARQ ACK/NACK (A/N). Both of these messages must be delivered fast and reliably, and in uplink this is especially challenging. This is because devices have a strict transmit power limitation of 0.2   W which could leave them power limited at locations where the downlink would still work fine. A power limited device doesn't benefit from a wider allocation with lower code rate, since the power is then split over more resources. What helps instead to improve the quality is to accumulate more energy per transmitted from extending the transmission duration, i.e. with repetition. But this directly contradicts the low latency target, and we are facing a tough trade-off: how to be fast and get high enough quality.
11.2.5.3.1. Long PUCCH
The three PUCCH formats 1, 3, and 4 are often together referred to as long PUCCH. They can be flexibly configured in the range 4–14 symbols. PUCCH format 1 can carry 1–2 bits information over base-12 sequence BPSK or QPSK symbols sent on every other OS, see Fig. 11.17, and can be multiplexed by using orthogonal sequences. This format has a rather high reliability since the code rate is low. The reliability can also be improved further with intra-slot frequency hopping. The number of UCI data symbols for different length and intra-slot hopping is given in Table 11.7. The number of orthogonal sequences that can be used for multiplexing is equal to the number of UCI data symbols. Resource efficiency and reliability are of course attractive aspects, but since it requires multiple symbols there is a latency drawback both in the form of alignment latency (waiting for the next PUCCH opportunity) and reception latency (waiting to the end to decode the message). Therefore, the long PUCCH format is not well suited for URLLC, except when we can anyway ensure a low latency e.g. through using a high SCS.
image
Fig. 11.17 PUCCH formats 0, 1, and 2.

Table 11.7

Number of UCI data symbols in PUCCH Format 1.
PUCCH length #UCI data symbols
No intra-slot hopping Intra-slot hopping
First part Second part
4 2 1 1
5 2 1 1
6 3 1 2
7 3 1 2
8 4 2 2
9 4 2 2
10 5 2 3
11 5 2 3
12 6 3 3
13 6 3 3
14 7 3 4

image

The other long PUCCH formats 3 and 4 can carry more than 2 bits information and are used for CSI reports and for sending multiple HARQ A/N. These formats can also be configured with frequency hopping.
11.2.5.3.2. Short PUCCH
PUCCH formats 0 and 2 are often referred to as short PUCCH formats. Format 0 is 1–2 OS long and consists of a base-12 sequence sent on one RB that is phase rotated to convey up to 2 bits of information (when QPSK modulation is used), just enough for one A/N and one SR bit. This configuration obviously has low latency but since it uses sequence selection and no channel coding it's challenging to achieve high reliability for low-coverage devices. To improve performance the transmission can be repeated in a row. This increases the total signal power in the receiver, thereby improving the sequence detection, but of course extends the duration and increases latency.
A technique that improves quality beyond plain repetition is frequency hopping. By consecutively transmitting the same message on different frequency resources, the frequency diversity of the channel comes to the help. This way, a two-symbol PUCCH can be constructed from a repeated Format 0, see example in Fig. 11.17. A 1 RB resource is used during two subsequent symbols at both ends of the available band to send PUCCH. If the coherence bandwidth of the channel is small enough the attempts are uncorrelated, and we therefore expect a much improved success probability.
Still, with the short duration of a two-symbol PUCCH it can be hard to achieve high reliability. This is especially critical for SR, which is a device's only way to indicate that it has uplink data without grant. It doesn't matter how good the downlink control and uplink data qualities are, if the gNB doesn't know that it should send an uplink grant, nothing gets done (a solution to this problem is to use configured uplink grants as discussed in Section 11.3.3.4). For SR, a combination of the two methods can be used: two-symbol PUCCH with frequency hopping, and repetition of that sequence. Alternatively, if the latency allows, a longer PUCCH can be used to improve reliability.
Also HARQ A/N is a critical information element. Without it, the gNB does not know that a transmission failed and that it should do a retransmission. Even if there is time to perform multiple retransmissions and the combined reception of those would suffice, the chain is broken if a NACK is not delivered. However, compared to SR the A/N doesn't need quite the same level of reliability. This is because a failed data transmission in downlink would be quite rare to begin with, so a retransmission is only required in these rare cases. Some error is therefore acceptable on the HARQ A/N without jeopardizing the total reliability. However, one should keep in mind that the errors on downlink and uplink are likely to be somewhat correlated due to long-term fading dips, so it would often be in just these cases when the NACK is needed that the PUCCH transmission would fail.
In NR Release-15 there is a limitation of 2 PUCCH and 1 HARQ A/N transmission occasions that can be used by a device in a slot. If a short PUCCH is used, e.g. two-symbol long, it would be possible in theory to transmit 7 times per slot. But with the limitation, only two of these occasions can be used and only one for A/N, which effectively places a restriction on the rate of low latency packets a device can receive in downlink.
PUCCH format 2 is also 1–2 symbols long, but can carry a payload of >2 bits, which is required for CSI reports and multiple HARQ A/N feedback. In the range 3–11 bits a Reed-Muller code is used, and for more than 11 bits, Polar code is used after CRC insertion. A DMRS sequence is mixed in on every third subcarrier. The number of RBs used depends on the total payload of the UCI. An example of this format is shown in Fig. 11.17.

11.2.5.4. PUSCH

Slot Any
Subcarrier spacing {15, 30, 60, 120} kHz
Duration 1-14 symbols
Bandwidth Any (see Table 11.1)

In the same way as PDSCH, the uplink data channel in NR, physical uplink shared channel (PUSCH), is sent with OFDM waveforms, and is coded with LDPC with possible CB segmentation using the same CRC attachment. As an option DFT-spread OFDM is available. This is different from LTE, where only DFT-spread OFDM is used in uplink. The reason for using DFT-OFDM and not plain OFDM is to reduce the so-called cubic metric, which is a measurement of the amount of power back-off that needs to be done in the power amplifiers to handle the signal variation over time. High back-off means that the transmit power is reduced and the device becomes more power limited, and as an effect uplink coverage is reduced. When DFT-OFDM is used for the uplink for low back-off, transmissions need to be contiguous, and cannot be spread out over the band. This means less frequency diversity is enabled, and in addition that PUSCH and PUCCH cannot be transmitted at the same time.
As in PDSCH, a CRC is added to the uplink data TB, and additional CRCs in the case of segmented CBs, which is then each scrambled with a RNTI, and DMRS is sent within the data transmission on PUSCH, either in a front-loaded (Mapping type B) or later position (Mapping type A).
Similarly to PDSCH, the duration of PUSCH can be dynamically indicated. But for PUSCH there is no restriction on start and duration other than that an allocation cannot span across a slot border. Thus, PUSCH can start in symbol 0–13 and have a duration of 1–14 symbols.
The same principles of DCI informing the device of where it should transmit works also for uplink, and also for PUSCH the gNB can indicate that the allocation should be done with K slot repetitions (slot aggregation) for increased reliability, in the same way as was described for PDSCH in Section 11.2.4.4.2. However, no mechanism corresponding to the low latency feature of downlink pre-emption (Section 11.2.4.4.3) is supported for uplink in Release-15.
Uplink control information (UCI) can be sent on PUSCH if the device has UCI at the same time as a PUSCH data transmission. This is to avoid simultaneous PUSCH and PUCCH transmissions which can be challenging for power limited devices. The mapping of UCI into PUSCH follows a predetermined pattern, as in LTE, and the UCI is coded with rates set by the so-called beta-factors that the gNB signal in advance over RRC. The value of the beta factor is indicated with an index, see Table 11.8. For HARQ-ACK the device is configured with three indexes I o f f s e t 1 H A R Q − A C K image for 1–2 bits, I o f f s e t 2 H A R Q − A C K image for 3–11 bits, and I o f f s e t 3 H A R Q − A C K image for >11 bits payload, respectively. For CSI part 1 (CQI, RI, CRI) and CSI part 2 (PMI, and in some cases additional CQI) the device is configured with I o f f s e t 1 C S I − 1 image and I o f f s e t 1 C S I − 2 image for 1–11 bits payload, and I o f f s e t 2 C S I − 1 image and I o f f s e t 2 C S I − 2 image for >11 bits payload, respectively. The uplink scheduling DCI (Format 0–1) can be configured to contain a beta-offset indicator field, and in this case 4 indices are configured for each of the 7 parameters mentioned above. In the DCI the network can then dynamically indicate which of the 4 indices should be used.

Table 11.8

Beta factors and indices.
Index Beta factor for HARQ-ACK Beta factor for CSI-1 and CSI-2
0 1.000 1.125
1 2.000 1.250
2 2.500 1.375
3 3.125 1.625
4 4.000 1.750
5 5.000 2.000
6 6.250 2.250
7 8.000 2.500
8 10.000 2.875
9 12.625 3.125
10 15.875 3.500
11 20.000 4.000
12 31.000 5.000
13 50.000 6.250
14 80.000 8.000
15 126.000 10.000
16 Reserved 12.625
17 Reserved 15.875
18 Reserved 20.000
19–31 Reserved Reserved

11.3. Idle and connected mode procedures

11.3.1. NR protocol stack
In the user plane (UP), data radio bearers can be of three types, Master Cell Group (MCG) bearer (where user plane follows the path of the control plane through the master node), Secondary Cell Group (SCG) bearer (where user plane data is mapped to the secondary node), or split bearer where data is split on PDCP layer and mapped two RLC entities in both master and secondary nodes.
In the control plane (CP), NR defines the protocol Radio RRC, closely related to that of LTE. This has the role of conveying control configurations between gNB and the device and to interface to the Non-Access Stratum in the CN.
Which protocols and procedures are relevant for URLLC? Obviously all are necessary, but two protocols ensure that packets are delivered: RLC retransmission on RLC level keeping track of data delivery, and HARQ on MAC level, triggering PHY retransmission attempts. Out of these, RLC can be excluded from a URLLC discussion since the time scale that RLC retransmission typically operates at is much longer than what is relevant for cMTC services. For ensuring delivery at short time scales we therefore rely on the HARQ protocol, or alternatively no protocol at all (just blind retransmission attempts).
image
Fig. 11.18 Protocol stack in the device and interfaces between layers.
We also rely on uplink scheduling protocols for URLLC. As noted above, scheduling is straight-forward for downlink since the gNB is in control of resources, but for uplink data the device needs to either request resources (with SR) or be pre-allocated resources (configured uplink grants). Both these ways of getting an uplink resource have benefits and costs, as discussed below (Section 11.3.3.4).
In addition to diversity on the PHY and to multiple transmission attempts, we can also utilize the diversity from different links. This higher-layer diversity is enabled by branching a packet at the PDCP layer and then following the two legs through, as described further below (Section 11.3.3.7).

11.3.1.1. RRC state machine

One main step of NR is taken in the RRC state machine, which defines the three states IDLE, INACTIVE, and CONNECTED, shown with corresponding transitions in Fig. 11.19. The RRC states also correspond to different CN states. Here, the INACTIVE state is introduced to reduce the signaling and latency for devices with intermittent traffic, by keeping the device context in the gNB and staying in CN CONNECTED state. At the same time, the INACTIVE state allows the device to save battery by sleeping. This scheme is similar to the RRC suspend/resume mechanism that was introduced in LTE Release-13/14 to reduce signaling when going from IDLE to CONNECTED, but the NR solution avoids IDLE altogether and can thereby achieve even lower latency.
To begin with, we should clarify the focus we have here when it comes to procedures. Whether a device is in CONNECTED or IDLE mode will have significant impact on the service latency. A device first needs to wake up and synchronize to the network before data can be received or transmitted, and this process will inevitable take time since it involves the exchange of several messages. This is true also for devices in INACTIVE mode, but the transition is simpler and faster. Even with the enhancements and speed-ups to the CP signaling introduced in NR, we can simply assume that when we speak of URLLC services we assume a latency that is low enough to exclude devices in IDLE mode. In future releases of NR, it is quite possible that the state transition can become faster or that first data can be transmitted earlier, but in Release-15, the CP signaling latency will be significantly higher compared to the UP (user plane) data latency, as shown in Chapter 12. Therefore, in the following we focus on CONNECTED mode procedures, and only discuss CP procedures in the context of handover.
image
Fig. 11.19 RRC states and transitions.

11.3.2. Idle mode procedures

11.3.2.1. Control plane signaling

In cases when a device changes between different states or cells, or during the first attach to the network, it needs to set up the Control Plane (CP) through signaling before it can start to send or receive data over the User Plane (UP). As this CP signaling will take time, it is not obvious that a device that does not have the UP set up already can be counted on for a cMTC service. It therefore makes sense to say, as we did above, that only devices in the CONNECTED state are applicable for URLLC, and that devices need to be kept in this state and can't be allowed to go to a sleep state. However, this strict view can be relaxed a bit by considering handover and less latency critical services that still require high reliability.
For a device moving between cells, a handover is required to attach to the new cell. The handover means that the CP needs to switch to the new cell with RRC signaling. As already said, the signaling to make the transition happen will induce some latency. But the CP signaling is significantly faster in NR compared to LTE. To a large part this is because the processing is faster leading to shorter delay between messages and thereby faster round-trip. As with data, it's possible to reduce the time further with shorter transmissions and higher SCS, at the cost of using more bandwidth.

11.3.3. Connected mode procedures

11.3.3.1. Dynamic scheduling

The basic form of scheduling a device for downlink and uplink data transmissions is by dynamic scheduling using downlink control indication. While this method is resource efficient (a device is given resources only when needed) it has limitations in terms of reliability (SR must be received for uplink, and downlink control must be decoded) and latency in the case of uplink data (transmitting SR and receiving uplink grant).
11.3.3.1.1. Scheduling timeline
The range of the parameters are RRC configured to indices and the indication is done in the DCI. This setup allows for high flexibility in scheduling.
11.3.3.1.2. DCI
The fallback formats use only basic features and have the same smaller size compared to the full scheduling formats for downlink and uplink, see content in Table 11.10. Of these formats the last four (SFI, PI, two types of UPCI) also have the same size. The DCI is protected with a CRC, which is like in LTE a sequence added before scrambling with the RNTI, the configured device address to which the DCI is directed. The CRC for PDCCH is longer in NR, 24 bits compared to 16 bits in LTE. Out of the 24 bits, 3 bits are used to assist the decoding of the used Polar code. This helps reducing the false detection rate that arises from random false CRC checks, which have the approximate incidence of 2 − N for N bits CRC. For URLLC services false detection can be extra problematic, since it can lead to unavailability and buffer contamination and thereby reduced reliability.

Table 11.9

DCI formats.
Fallback DCI Format 0-0 (uplink)
Format 1-0 (downlink)
Non-fallback DCI Format 0–1 (uplink)
Format 1-1 (downlink)
Slot format indicator (SFI) Format 2-0
Pre-emption indication (PI) Format 2-1
Uplink power control indicator (UPCI) Format 2-2 (for PUSCH/PUCCH)
Format 2–3 (for SRS)

image

Table 11.10

DCI fields and size in bits for data transmissions addressed to C-RNTI.
DCI field Format 0-0 (uplink fallback) Format 0–1 (uplink scheduling) Format 1-0 (downlink fallback) Format 1-1 (downlink scheduling)
Uplink/downlink identifier 1 1 1 1
Frequency domain resource assignment Formula Formula Formula Formula
Time domain resource assignment 4 0–4 4 0–4
Frequency hopping flag 1 0–1 1 1
Modulation and coding scheme 5 5 5 5
New data indicator 1 1 1 1
Redundancy version 2 2 2 2
HARQ process 4 4 4 4
TPC command 2 2 2 2
Uplink/SUL indicator 1 1 – –
Carrier indicator – 0–3 – 0–3
Bandwidth part indicator – 0–2 – 0–2
Downlink assignment index 1 – 1–2 – 0–4
Downlink assignment index 2 – 0–2 – –
SRS resource indicator – Formula – –
Precoding information and number of layers – 0–6 – –
Antenna ports – 2–5 – 4–6
SRS request – 2 – 2–3
CSI request – 0–6 – –
CBG transmission information – 0–8 – 0–8
PTRS-DMRS association – 0–2 – –
Beta offset indicator – 0–2 – –
DMRS sequence initialization – 0–1 – 0–1
Uplink-SCH indicator – 1 – –
VRB-to-RB mapping – – 1 0–1
PUCCH resource indicator – – 3 3
HARQ feedback timing – – 3 0–3
RB bundling size indicator – – – 0–1
Rate matching indicator – – – 0–2
ZP CSI-RS trigger – – – 0–2
Transmission configuration indication – – – 0–3
CBG flushing indicator – – – 0–1

image

11.3.3.2. HARQ

HARQ is the protocol handling the default method for triggering retransmissions in the PHY layer. Compared to automatically repeating a message, as with downlink and uplink repetitions described above, the fact that retransmissions are conditionally triggered in HARQ ensures a much-improved resource efficiency. A retransmission is triggered when an ACK is not received in HARQ, and since we target high reliability, most of the packets will indeed result in an ACK, and a retransmission will therefore only rarely be needed. HARQ can thus be seen as a way to use a higher code rate first and then gradually lowering it to the needed level. Incremental redundancy using a circular buffer is used (as in LTE) for this purpose of combining retransmissions into one longer code. This is indeed very resource efficient, but has a drawback of longer latency and the dependency on receiving the feedback.
In NR the HARQ feedback is transmitted in UCI in the uplink for downlink data, and implicitly through a DCI in the downlink for uplink data. The latter is only true for NACK: an unsuccessful uplink transmission results in an uplink grant DCI for a retransmission, while a received uplink transmission results in no feedback on downlink. In NR, there is no equivalent to the LTE PHICH channel giving A/N feedback for uplink data transmissions. The main reason for this is that synchronous HARQ operation with hard-coded timing is not supported in NR, and both uplink and downlink operate using asynchronous HARQ with indicated timing.
If the TB is segmented into CBs the HARQ feedback and retransmission of the CBGs is handled separately by using multiple indication bits in the UCI and DCI, which can improve resource efficiency since only the erroneous parts of a transmission are resent.
NR HARQ is fully dynamic, meaning that there are no fixed connections between transmission and feedback. For downlink data the UCI where to send HARQ A/N is identified from the DCI, and for uplink data a DCI for retransmission is identified from the HARQ process index and a new data indicator, similar to LTE. There are 16 HARQ processes to use for a device.
With HARQ, we can trigger retransmission to ensure high reliability at low resource cost, but only as long as the chain of feedback is not broken. Therefore, we cannot set an arbitrarily high error rate target (using high code rate) for transmissions and rely on many retransmissions, since at some point the feedback may not be delivered (it can also take a long time). However, the reliability of HARQ NACK doesn't need to be on the same level as that of data since it will be rare that the data transmission fails, requiring a retransmission.
For both reasons of latency and reliability there is therefore a limit on how many HARQ retransmissions we can use for URLLC data. The more we can reliably use, the better for efficiency, which for uplink data appears a solid choice, but for downlink data we then rely on safe delivery of UCI which is not obviously achievable when the UE is power limited. This is discussed further in Chapter 12.

11.3.3.3. SR

When a device has URLLC data in its transmit buffer, it triggers a SR if it doesn't have a valid grant. This is often referred to as dynamic uplink scheduling. If the device only operates using one service, this is rather straight-forward: it will use its configured SR resource on PUCCH and the gNB will then know what resource the device needs. But in the case when a device is operating with multiple services (e.g. eMBB for video and cMTC for position) it needs to separate between them.
An SR is configured for a logical channel which is in turn connected to a radio bearer with a certain QoS requirement defined by the network. By using different SR resources, with up to 8 possible to configure in NR, the device can indicate which type of uplink data resource it needs. For cMTC services, we can then assume that the SR is configured on a short PUCCH with short periodicity to allow for low latency. As discussed in Section 11.2.5.3.2 this fast option has a drawback in the form of coverage limitations. Besides the PHY repetition (with frequency hopping in between) it is also possible to do a higher layer repetition of SR. This is enabled by configuring a delay parameter to 0, which has the effect of repeating the SR until the device receives the requested uplink grant. This is then a way to achieve reliability at the cost of latency (we can assume that there will be no additional resource cost from a device using its allocated PUCCH resources). For dynamic uplink scheduling, this possibility is crucial as already noted: until the gNB has received the SR no uplink data can be transmitted.

11.3.3.4. Uplink configured grant
CG Type 1 only consists of the RRC configuration, meaning that all of the information typically sent in DCI is specified there instead and no additional control information is needed.
image
Fig. 11.21 Configured uplink grant Type 1 and 2, and SR-based access for uplink transmission.
CG Type 2 is similar to semi-persistent scheduling in LTE and consist of a configuration pointing out a certain RNTI address and a repetition pattern. A DCI scrambled with that RNTI is then interpreted by the device as an uplink grant that is reoccurring with the configured periodicity. Compared to Type 1, this type is more flexible since the gNB can at any time update the parameters of the uplink transmission in PDCCH, such as MCS and allocation.
With a CG in place, the device can in principle be given an uplink data transmission opportunity in every symbol, that it can readily use if it has data in its buffer. If it has no uplink data, it doesn't transmit anything on the CG. In practice, the shortest transmission duration would likely be two OS to give room for UCI and DMRS, which means that while avoiding overlapping allocations the device can start its transmission at 7 positions in the uplink slot without any further delay. Compared to dynamic scheduling, this offers a dramatic reduction of latency. The cost, however, is that uplink resources become “locked up” for a device, in a way that dynamic scheduling avoids. Unless the device has periodic data of the same period as the CG (this is indeed a possibility) there will be a waste of unused resources. At least this is the case if we want to uniquely assign uplink resources to ensure good SINR. If devices are allowed to share resources the resource management is much more effective, but we then run into the risk of colliding uplink transmissions. For URLLC data this is not an option, since the reliability would be jeopardized.
After a first transmission has taken place on the uplink CG the data will either be successfully received or not. If the transmission failed (but the presence of DMRS was detected) the gNB will issue a DCI for retransmission. Thus, from the first retransmission onwards dynamic grant scheduling will be used (of course the gNB is free to indicate the CG for retransmission also).
If the device has been indicated overlapping dynamic and configured uplink grants it will drop the CG and use the dynamically indicated one.
11.3.3.4.1. HARQ operation
The NR supported HARQ operation is asynchronous, meaning that the process ID (PID) must be indicated in the DCI for the device to know which buffer to use. After activation, the configured uplink grant resources will be numbered with a PID starting from 0 up to a maximum value configured by RRC, after which the numbering restarts at 0. This way the gNB and the device will both know which transmission a given PID refers to. When a retransmission is triggered the device is thereby dynamically scheduled with DCI even if the configured resource can be indicated for retransmission.
11.3.3.4.2. Repetition
It is possible to configure repetition of a TB over the CG, also in the case of mini-slot resources, just as for dynamic PUSCH allocations as mentioned in Section 11.2.5.4. However, as in the case of downlink repetition (illustrated in Fig. 11.13), the repetition in uplink will only take place on slot level, meaning that the same resource will be used for sending a TB in K subsequent slots. This will lead to increased reliability on shorter time compared to HARQ-based retransmission. For lowest latency it would require repetition on mini-slot resources with a slot, but this is not supported in NR Release-15. Over the K repetitions on the uplink CG a RV pattern is applied according to selecting the (mod(n-1,4)+1)th value of an RV sequence at repetition n. The configurable RV sequences are {0,2,3,1}, {0,3,0,3}, and {0,0,0,0}, and the K-sequence can start at any occasion when the RV from the configured RV sequence is 0. This means that when a varying RV sequence is configured, there will be an additional alignment delay to wait for the start of the repetition sequence.

11.3.3.5. Uplink power control

For determining the power used for PUSCH transmission a formula is used:
p PUSCH = min { p cmax , p 0 + α · P L + 10 log 10 ( 2 μ · M r b ) + Δ T F + δ }
image
where p c m a x image is the maximum allowed power per carrier, p 0 image and α image are configurable parameters, P L image is the pathloss estimate, μ image is the numerology ( S C S = 2 μ · 15 k H z image ), M r b image is the uplink allocation, Δ T F image is calculated from the MCS, and finally δ image is the closed-loop parameter. The open-loop components ( p 0 + α ⋅ P L image ) are configured and handled by the device, while the closed-loop component δ image is indicated in an uplink power control DCI. The PUCCH power control is the same as the PUSCH one, but for PUCCH α image is fixed to 1.

11.3.3.6. CSI measurement and reporting

For the gNB to know which MCS to use for a device in downlink it is critical that it knows the status of the channel conditions. The device measures the channel quality on the pilot sounding signals CSI-RS (described in Section 11.2.4.2.3) and then reports a CQI in the CSI reporting, as part of the UCI together with HARQ A/N and SR. The reporting is based on measurement on configured CSI-RS resources and can in a similar way be triggered periodically (on PUCCH), semi-statically (on configured PUCCH or PUSCH resources) or aperiodically (on PUSCH) from a DCI indication.
The value of the CQI is connected to the MCS in such a way that the expected BLER using a certain MCS, when the corresponding CQI is reported, is at a target level. In LTE this expected BLER level is set at 10%, meaning that this is the expected failure rate when a device is scheduled with a MCS corresponding to the reported CQI. In NR, addressing the needs of URLLC and the fact that there is a separate MCS table intended for high reliability, the device can be configured with two different BLER levels: 10% error mapping to CQI table 1 (up to 64 QAM) or CQI Table 11.2 (up to 256 QAM), and 10 −5 error mapping to CQI Table 11.3, shown in Table 11.11. The configuration is semi-static, meaning that it is made over RRC, and not dynamically indicated. It can therefore be so that the CQI reported and the MCS table used do not match, but it is then up to the gNB to translate between the two sets.

Table 11.11

CQI table for 0.001% error rate (corresponding to Table 5.2.2.1–4 in Ref. [6]).
CQI index Modulation Code rate x 1024 Efficiency
0 out of range
1 QPSK 30 0.0586
2 QPSK 50 0.0977
3 QPSK 78 0.1523
4 QPSK 120 0.2344
5 QPSK 193 0.3770
6 QPSK 308 0.6016
7 QPSK 449 0.8770
8 QPSK 602 1.1758
9 16 QAM 378 1.4766
10 16 QAM 490 1.9141
11 16 QAM 616 2.4063
12 64 QAM 466 2.7305
13 64 QAM 567 3.3223
14 64 QAM 666 3.9023
15 64 QAM 772 4.5234

image

In the case of only one antenna port used and one configured CSI-RS resource, the CSI report consists only of CQI. With more antenna ports and CSI-RS resources, the report can also include the codebook index with precoding matrix indicator (PMI), rank indicator, layer indicator, and also a CSI-RS indicator indicating the preferred beam. The reporting can be done for type I single-panel for simple multi-port antennas, type I multi-panel codebook for composite antennas each consisting of multiple ports, or for type II codebook with higher granularity intended for multi-user MIMO.

11.3.3.7. PDCP duplication

As in LTE, data can be duplicated in the PDCP layer for increased reliability from the redundancy over two transmission paths. The procedure is setup over RRC for a split radio bearer and defines an additional RLC entity connected to PDCP by adding an additional logical channel. In the PDCP layer, the incoming data packet in the form of a PDCP PDU is duplicated and sent to the two RLC entities; the master cell group (MCG) and the secondary cell group (SCG), as illustrated in Fig. 11.22. If the MCG and the SCG in turn belong to the same MAC entity the duplication will be over two different carriers, meaning carrier aggregation (CA), and if they belong to different MAC entities the duplication is done over two different cells, meaning dual connectivity (DC). In the case of CA, logical channel mapping restrictions are used to prevent the packets from being sent on the same carrier (which would not give as good redundancy).
On the receiving side the packets are separated all the way up to the PDCP layer, where a duplicate check is done so that only one PDCP PDU is delivered to the data buffer.
image
Fig. 11.22 PDCP duplication of data with Dual Connectivity.