13.2.1 Initial Access in General

Initial access comprises a set of functions across multiple layers of the air interface in the radio access network (RAN) protocol stack and, to some extent, the CN/RAN interface as in the case of paging and state transition. In LTE, these functions are:

1. Acquisition and cell search, allowing a UE to synchronize to a cell. This includes the following steps: (a) acquisition of frequency and synchronization to a cell, (b) acquisition of frame timing of the cell, (c) determination of the physical layer cell identity of the cell. This is performed using the Primary and Secondary Synchronization Signals (PSS and SSS, respectively). The PSS allows the UE to identify the cell ID and the generic timing information, enabling it to identify the position of the SSS, and the SSS provides the frame timing and the cell identity group to which the cell ID belongs.
2. System information distribution, enabling the UE to acquire the cell system information to be able to access the network. The system information includes the downlink and uplink cell bandwidths and configurations, etc. This information is transferred using the Master Information Block (MIB) and the Secondary Information Blocks (SIBs). The MIB is needed by the UE to be able to read the SIB information. The SIB includes permissions and configurations of the UE, such as cell selection and reselection, random access parameterization, access barring information provision, warning messages, etc.
3. Random access, allowing the UE to connect to the network for initial access, handover purposes, scheduling requests, UE positioning, etc. This is done using a shared channel, namely the aforementioned Random Access CHannel (RACH). The UEs may access the RACH either on a contention basis, or contention free, depending on the purpose; the former approach being mainly used for initial access and RRC connection establishment, and the latter mainly for handover, uplink resynchronization, and positioning.
4. Paging is used for network‐initiated connection setup when the UE is in RRC Idle state. The UE is notified for downlink data using the paging channel in LTE.

While the initial access in 5G will in principle consist of exactly the same steps and functions, some of the listed functions should be significantly enhanced for 5G. In the following subsections, improvements are hence discussed that increase the ability of the system to handle more demanding service requests or an increased number of access requests.

13.2.2 System Information and 5G RAN Lean Design

The LTE system information (SI) comprises information such as access information for the UE, node specific information, system wide information, public warning system (PWS) information, etc. One of the drawbacks with LTE was the rather low possibility for the cell to enter a so‐called micro sleep, also referred to as cell discontinued transmission (DTX). There are several reasons for this related to the system information distribution such as:

• Reference symbols are transmitted even if there is no data;
• System information is transmitted periodically regardless of the load in the system;
• The Physical Downlink Control CHannel (PDCCH) is transmitted across the full system bandwidth, i.e., the same number of PDCCH symbols are used for all resource blocks (RBs);
• For synchronization signals, LTE uses a fixed periodicity of 5 ms, which decreases BS sleep efficiency at low load.

Since system information is always transmitted, it is difficult to make the LTE system more energy efficient at low load. If there is no, or very limited, possibility for BSs for micro sleep, the total power consumption of the network is likely to increase when the BS density is increased [1]. In addition to the economic and environmental reasons to minimize energy consumption, there are also many reasons related to engineering. Today, the main obstacle for miniaturization of radio BSs is heat. Energy consumption directly drives product weight and volume through heat dissipation. Reducing energy consumption will have nonlinear impacts not only on weight and volume, but also on what a BS is, how it is built, what it can do, where it can be deployed, what energy sources it utilizes, etc.

How can 5G be made more energy efficient? Below we go through how the system information can be designed to be more lean in 5G, diving in more detail into the bullets listed before:

Reference Signals

The reference signals (RS) necessary for channel estimation should in 5G typically only be transmitted in the same sub‐frame, over the same bandwidth, and in the same beam, i.e., based on the same beamforming as the corresponding data. This is different compared to LTE, where also the cell‐specific reference signals (CRS) in previous sub‐frames can be used to aid channel estimation, as illustrated in Figure 13‐2, where the red or dark slots are RS. The exact procedure for NR is currently under discussion in 3GPP [2][3].

Synchronization Signals

LTE uses a periodicity of 5 ms for synchronization signals. However, if the periods between the synchronization signals can be increased, the BS sleep efficiency can be increased [5]. The reason is that it takes some time to deactivate and reactivate certain components. For longer sleep durations, more components can be put to sleep, and the sleep power usage becomes lower. The current agreement in 3GPP [6] enables different periodicities of the synchronization channels (5, 10, 20, 40, 80 or 160 ms), depending on, e.g., traffic load. See Sections 11.6 and 13.2.3 for more information on this topic.

Figure 13‐3 shows the relative power consumption per cell for 5G compared to LTE for a range of cell radii for two fixed user densities of 5 or 50 users/km². When the network is densified, the cell radius shrinks. For small cell radii, there is hence a higher probability to have no active users in the cell, which may enable the cell to enter the cell DTX. The major difference between 5G and LTE is the ability to actually utilize the cell DTX when there is no active traffic. In Figure 13‐3, it is assumed that NR has 4 and 6 times higher probability to enter cell DTX if the cell is empty, as detailed in [7]. The figure shows that if 5G is designed so that it allows better cell DTX sleep probabilities than LTE, the power consumption can be decreased substantially.

13.2.3 Configurable Downlink Synchronization for Unified Beam Operation

As mentioned in Section 13.2.1, downlink synchronization is the first task of the initial access for UEs to acquire initial timing and frequency synchronization with the network, and to detect the physical cell ID (PCI), which is used for the reception of most other channels and signals. One of the major targets of the 5G radio system is to support significantly higher carrier frequencies than LTE, e.g., up to 100 GHz. Moreover, the 5G system is envisioned to support different deployment scenarios including heterogeneous networks with ultra‐densely deployed small cells operating in high frequency bands, e.g., millimeter‐wave (mmWave) bands. Additionally, 5G can operate in standalone and non‐standalone mode, as introduced in Section 1.2. While the synchronization signaling requirements for all these scenarios are very different, it is highly desired to design a unified synchronization signal transmission framework to support all these envisioned scenarios. It should also be noted that a 5G radio system endeavours to be highly configurable so that the system can be flexibly configured according to its operation scenario.

As described in Section 13.2.2, a lean system design to achieve much better forward‐compatibility has been considered as one of the main design targets of the 5G radio system. One part of achieving this is to minimize the “always‐on” signals, such as cell specific‐reference signals in LTE. As a result, the synchronization signals (SS) together with MIB channels are the only remaining “always‐on” signals transmitted periodically in the 5G radio system [8]. It is therefore very important to keep the SS transmission frequency on a proper level so that on one hand, UEs can acquire initial synchronization very quickly, and on the other hand, the overhead of SS transmission is kept to a minimum level for the sake of network energy efficiency. To accomplish this, a configurable SS transmission framework has been developed and standardized in 3GPP [9].

As illustrated in Figure 13‐4, in the configurable SS transmission scheme, a series of synchronization signal burst sets are transmitted, where each set contains one or several synchronization signal blocks (SSB). Each SSB consists of 4 OFDM symbols, one PSS, one SSS and two PBCH OFDM symbols carrying MIB. All OFDM symbols in the SSB are transmitted from the same antenna port, i.e., by using the same beam direction. This means that the beam index is implicitly signaled by the time location of the SSBs. To ensure timely acquisition of initial timing and frequency synchronization, regardless of periodicity of synchronization burst set, a certain number of SSBs shall be transmitted within a time window of 5 ms. Since each SSB is transmitted in a particular beam direction, the number of SSBs in 5 ms essentially determines the number of beams used for SS transmission in a cell. Consequently, configurable SS transmission incorporates single‐beam and multi‐beam transmission into a unified solution. Specifically, when single beam operation is applied, there is only one SSB transmitted in a 5 ms time window like in LTE. To realize multi‐beam operation, there are more than one SSB to be transmitted in 5 ms. Due to different beamforming gain requirements in different frequency bands, the maximum number of SSBs in a 5 ms time window is defined as follows: 4 for frequency bands below 3 GHz, 8 for frequency bands above 3 GHz and below 6 GHz, and 64 for frequency bands from 6 GHz to 52.6 GHz [10]. These values set the maximum number of beams that can be used for SS transmission. When more beams (i.e., spatial repetitions) are used, beam width becomes narrower and the beamforming gain increases, there will be less opportunities to put the BS into (micro‐) sleep. As a result, such configurable SS transmission provides a great optimization space for the trade‐off among the different aspects mentioned above.

13.2.4 Digital Beamforming in the Initial Access Phase

The previous sections discussed system information distribution and the synchronization signals transmitted from the BSs for enabling the UE to synchronize to the respective cell and beam. This task is very complicated in cases of highly directional links, since in addition to the time and frequency domains, the 5G network elements (i.e., BS, UE, or relays) have to discover each other in an additional dimension, namely that of space. For instance, each receiver must determine the angles of arrival for the incoming signals before a data link is established. Depending on the beamforming architecture, scanning all possible angles of arrival may entail high delays in the CP. These delays, as we show below, may be well above what is acceptable in the current 4G LTE systems, and will definitely not meet the stringent delay requirements of the 5G system. Note that at the early stages of deployment, the mmWave coverage is projected to be “spotty”, and handovers between small mmWave cells and LTE macro cells will be quite frequent, leading to even higher delays. Therefore, a fast discovery procedure to find the point of attachment in the angular space is needed.

Assuming that mmWave initial access will follow the same five‐step process of LTE, we can identify one major difference: at the end of the synchronization phase, the UE learns where the BS is, and at the end of the random access phase, the BS learns where the UE is.

The main beamforming architectures are analog, digital and hybrid beamforming, as detailed in Section 11.5. The analog architecture relies on inexpensive phase‐shifters in radio frequency (RF) to create and steer a beam towards a desired direction. The digital architecture relies on digital samples obtained by the analog‐to‐digital converters (ADCs) attached to each antenna element of a multi‐antenna array. With analog beamforming, each receiver can steer the beam in only one direction at a time. Therefore, the receiver will have to direct the beam in all the available angles one by one. With digital beamforming on the other hand, thanks to all the available digital samples, the receiver can simultaneously steer beams into as many directions as the number of its antenna elements. This however occurs at the expense of increased power consumption, due to the usage of a higher number of ADCs than in the analog beamforming case. Hence, angular scanning with digital beamforming becomes M times faster, where M is the size of the angular domain the receiver needs to scan. Another factor affecting network discovery is the mode of signal transmission, which is here assumed to be based on single‐stream analog beamforming or omnidirectional analog transmission. This analysis captures the most extreme cases, since multi‐stream transmission is expected to perform somewhat in between omnidirectional and single‐stream analog BF [11]. The total size of the angular space is a function of the signal transmission mode and the beamforming architecture at the receiver: it is the product of the size of the angular space at the receiver and at the transmitter. For example, in the synchronization phase, if both the BS and the UE employ analog directional transmission and beamforming, and have array sizes of 64 and 16 respectively, 1024 angle combinations have to be scanned one by one. Note that if a UE is at the edge of the cell, it may further need to scan the angular space more than once to accumulate enough energy to detect the synchronization signal in a low signal‐to‐noise ratio (SNR) regime.

In Figure 13‐5, we focus on the synchronization phase and assume a transmission scheme where at each synchronization slot the BS randomly picks a direction and transmits the synchronization signal in analog directional way or omnidirectionally (where the transmission angle is fixed). In this example, digital receiver beamforming exhibits undisputable superiority. It can be detected even for SNRs below ‐17 dB. This SNR value was chosen as a threshold, since at this SNR a mmWave system operating at 1 GHz bandwidth can offer data rates of 10 Mbps. Next, we move to comparing five discovery design options where signal transmission is performed sequentially rather than randomly [12]. These options are differentiated and named based on the synchronization signal transmission mode (analog directional or analog omni‐directional), synchronization signal reception architecture (analog directional, analog omni‐directional or digital directional), and random access reception mode (analog directional, analog omni‐directional or digital directional). Note that since at the end of the synchronization phase the BS’s location is known to the UE, except in one case, the random access preamble is always assumed to be transmitted in analog directional, so that all the directivity gain of the antenna array is exploited. The nomenclature of these different schemes and options is explained in Table 13‐2.

The comparison is made in terms of delay and overhead for both edge users, defined as the 1^st percentile of SNR, and also high‐SNR users. The overhead is defined as the ratio of the synchronization signals duration over the transmission period. For example, if the signal duration is 100 µs, and it is transmitted every 5 ms, then the overhead is 2%. Naturally, as the overhead is increased, i.e., as more resources are used for these signals, the delay falls accordingly. Figure 13‐6 shows the delay for the synchronization and random access vs. the overhead. As can be seen, digital beamforming dramatically reduces the synchronization and random access times.

Nevertheless, the issue of energy consumption is an important one. To offset the high power consumption of digital beamforming, one could employ ADCs of low quantization, say 2‐3 bits, as discussed also in Section 11.5.3. Since the power consumption of ADCs scales exponentially with the bit resolution, employing low quantization ADCs will bring the power consumption of digital beamforming to the same level as its analog counterpart, as detailed in Section 16.2.3.2. Note that the effect of reducing the quantization resolution is negligible in low SNR regimes of the edge user.

13.2.5 Beam Finding for Low‐Latency Initial Access

The previous subsection has analyzed the use of digital beamforming (BF) in the initial access phase so as to reduce the synchronization delay. This is particularly important in scenarios with mobility, where the high channel dynamics would necessitate fast mechanisms to find alternative communication links. In this section, we study the performance of large‐but‐finite mmWave networks using codebook‐based analog beamforming. We consider an efficient genetic algorithm (GA) based approach for initial access beamforming. With the proposed algorithm, the appropriate beamforming matrix is selected from a set of predefined matrices such that the network end‐to‐end (E2E) throughput is optimized. The considered system model is depicted on the left side of Figure 13‐7, where a base station with M antennas serves N single‐antenna users. Specifically, the E2E throughput at the end of the K‐th iteration of the genetic algorithm is defined as

(13‐1)

(13‐2)

where is the (i, j)‐th element of the matrix with H denoting the channel matrix and V^K being the selected precoding scheme in the K‐th iteration of the algorithm. Also, P is the transmit power, B is the system bandwidth and N₀ is the power spectral density of the noise. Finally, α represents the relative delay cost for running each iteration of the algorithm. As seen in the above equations, there is a trade‐off between the running delay of the algorithm and improving the beamforming efficiency. As the number of iterations K increases, more accurate beams are selected by the algorithm, and the users’ signal‐to‐interference‐and‐noise ratio (SINR) increases. On the other hand, with α ≠ 0, the delay cost of the algorithm reduces the E2E throughput. Thus, as seen in the following, there is a finite optimal number of iterations maximizing the E2E throughput in delay‐constrained applications.

The details of the proposed scheme can be found in [13]. Shortly, the algorithm is based on the following procedure. The algorithm starts by getting L possible beam selection sets randomly, and each of them means a certain beam formed by transmit antennas, i.e., a submatrix of the codebook. During each iteration, the best selection result is determined, named as the Queen, based on the objective metrics. For instance, the beamforming matrix with the highest E2E throughput is chosen if the above expression for R(K) is considered as the objective function. Next, the Queen is kept and S < L matrices are generated around the Queen. This can be done by making small changes to the Queen such as changing a number of columns in the Queen matrix. Finally, during each iteration L − S − 1 beamforming matrices are selected randomly to avoid local minima. After N_it iterations, considered by the algorithm designer, the Queen is returned as the beam selection rule in the considered time slot.

The proposed scheme is generic in the sense that it can be implemented in the cases with different channel models, beamforming methods as well as optimization metrics. Also, it can reach (almost) the same throughput as in the exhaustive search based approach with significantly less implementation complexity. For instance, on the right side of Figure 13‐7, examples are shown of the GA performance in different iterations in the cases with (α = 0.001) and without (α = 0) costs of running the algorithm, assuming M = 32 transmit antennas, N = 8 single‐antenna users, precoding codebook size 128, SNR = 10 dB, and Rayleigh fading channels. In the figure, the relative achievable throughput compared to the maximum throughput achieved by exhaustive search is plotted. We observe that very few iterations are required to reach the maximum throughput, if the running delay of the algorithm is taken into account. That is, considering the cost of running the algorithm, the maximum throughput is obtained by finding a suboptimal beamforming matrix and leaving the rest of the time slot for data transmission. On the other hand, as the number of iterations increases, the cost of running the algorithm reduces the end‐to‐end throughput converging to zero at K = 1/α. With no cost for running the algorithm, on the other hand, the system performance improves monotonically with the number of iterations. However, the developed algorithm leads to (almost) the same performance as the exhaustive search based scheme with very limited number of iterations (note that with the parameter settings of Figure 13‐7, an exhaustive search implies testing in the order of 10³⁰ possible beamforming matrices). For example, with the parameter settings of Figure 13‐7 and α = 0, the proposed algorithm reaches more than 95% of the maximum achievable throughput with less than 200 iterations, thus making it particularly suitable for delay‐constrained systems.

13.2.6 Optimized RACH Access Schemes

After receiving system information and synchronizing to the cell, the UE should indicate its desire to receive uplink resources. In LTE, the UE randomly selects one of the random access preambles (64 preambles) configured in the broadcasted system information. The current design may create problems, due to collisions, when large numbers of devices simultaneously attempt to access the system. These potential collisions will lead to additional access delays which may impact services. Up to now, several schemes have been proposed in the literature for handling the random access procedure. These schemes can be classified into two large groups, namely pull‐based and push‐based [14]. In the first set of solutions, the RACH preambles are being split in prioritization groups, and the more delay sensitive devices compete against fewer devices for accessing the system. Additionally, the time for a second attempt to access the system could be fine‐tuned according to the collision rate. In the second set of solutions, a push‐based procedure is used to achieve a small data transmission (SDT) from the MTC devices. However, these schemes are designed mainly for prioritizing access based on the transmission requirements and are not aiming at solving the collision rate problem. Also, these focus mainly on traditional MBB use cases, and not on URLLC or mMTC use cases.

Based on the description above, we conclude that there are two key challenges for machine type communication: the massive access and the latency requirement.

For the random access of a vast amount of devices, a solution based on the grouping of the devices seems to be appropriate. Instead of having all the group members performing random access using one of the 64 preambles when they have to transmit, one could aggregate the transmission requests and only one device (i.e., the group head) could perform the RACH request. This will result in a significant reduction in the collision rate in the RACH. A slotted access scheme, where each device transmits according to its needs will further benefit the system.

The proposed solution is shown in Figure 13‐8:

1. The devices are grouped by the network based on their mobility and their communication characteristics (e.g., data to be transmitted, packet delay requirements).
2. The network schedules the cluster heads’ transmission opportunities based on their transmission requirements. The scheduling information includes how many timeslots each device should attempt to access the network and which preambles should be used.
3. The intra‐cluster communication may take place either via a different interface or via scheduled device‐to‐device (D2D) communication.

In Figure 13‐9, it is shown that using the group‐based system access reduces the collision rate significantly, since fewer devices (only the group heads) compete for RACH preambles. This also has a direct impact on the average initial access delay (related to the whole process including random access, random access response, terminal identification, and contention resolution), since the devices are accessing the system with fewer collisions and thus experience fewer retransmissions. For a low number of devices, the collision rate is small since the preambles are enough for the random access. As the number of devices increases, the collision rate and the CP latency increases for both approaches, but in the case of the group‐based access it is considerably lower.

The previous solution can handle the first of the two the key challenges introduced in this section, but it fails to successfully handle the strict latency requirements. One possible solution could be to reserve a set of dedicated preambles for the use of devices with high priority. This solution, however, is not efficient, since the number of RACH preambles is very small (i.e., 64 preambles) and has to be used both for random access and for handover purposes. What could be done is to:

1. maintain the preamble use from the devices without stringent delay requirements as is, and
2. the delay‐sensitive requests apply a combination of preamble signatures at a given random access time slot.

The approach would enable requests with more restrict delay requirements to have higher priority, since combinations of preambles can always be identified by the receiver. The combination of the preambles may take place either in time or frequency domain, thus providing a large set of potential combinations. The network will inform the UE about its reserved preamble combinations and the respective time and frequency shifts for each service priority level after the initial UE attach.

As depicted in Figure 13‐10 for high priority initial access assessment, the preamble coding outperforms the other two approaches, since the combination of preambles reduces the probability of a collision when accessing the system. Collisions of high‐priority transmissions will occur only if two low‐priority devices select the same preambles as the combination dedicated for a high‐priority device that attempts to access the system at the same time. As can be seen in the figure, the proposed scheme provides ~1000 times lower retransmission probability for high‐priority users compared to LTE without resource split, whereas it provides ~100 times lower retransmission probability compared to LTE with resource split.

13.3.1 Fundamentals of the RRC State Machine for 5G

The RRC state machine for the 5G RAN will consist of three RRC states [3][15]: RRC Idle, RRC Connected and a new inactive state, which in this section is called RRC Connected Inactive.

The RRC state machine is designed by mapping the operational functions of the UE to the radio resource control and management, so that the different functions of RRC and RRM procedures can be supported efficiently. RRC states are managed the entire time from UE power‐on until the UE is powered off. For example, after power‐on, the UE starts to search for a public land mobile network (PLMN). When the network has been found and the UE has registered to the network, the network takes over the state management, and the state management process continues depending on the UE data activity, network coverage and user mobility.

Let's assume that a UE should establish a data connection for example to download a newsletter with attached figures. The UE transitions from RRC Idle to RRC Connected state to download the newsletter. When the newsletter is being read, the traffic activity becomes low between UE and network. If the UE remains in RRC Connected state, there would be unnecessary power consumption at the transceiver and allocation of radio resources, even though power and resources could be saved. The simplest option is to disconnect the UE from the network and go to RRC Idle state to save power. This is a valid approach if the data transmission activity remains low for a relatively long period. However, if the user soon clicks another link on the newsletter, then the UE must again request the whole RRC setup, and the network must re‐create the radio bearers from scratch, which would take a lot of signaling and energy. This type of traffic is best served with an intermediate state between active mode and idle mode, or more precisely, a low‐activity state where the context generated by the time‐consuming functions of the connectivity and security are preserved while the UE can still save power during low‐activity periods.

The design of the RRC state machine and number of states reflects the architecture, use cases and evolution of the technology. RRC in LTE has two states, namely RRC Connected and RRC Idle. RRC Connected is the state for active UEs, which can transmit and receive user plane data and control plane signaling. RRC Idle state is the power saving state for low‐activity UEs.

Similarly, in Universal Mobile Telecommunications System (UMTS), the state machine consists of one Idle mode state and four Connected mode states. The Idle mode state is optimized for low‐power and network resource consumption, thus there is no UE context stored in the UE nor in the network. The Connected mode states are optimized for high UE activity where the normal high‐data‐rate traffic for active UEs and most of voice and data traffic are being transmitted and received. For this reason, the UE’s RAN context is stored in the UE and the network. UMTS Connected mode also contains low‐power states where the UE is still connected to the network and listens to the paging and broadcast channels, while the uplink data transfer is not supported without state transition to the active Connected state.

The state machine design for 5G RAN is challenging due to the high number of 5G use cases, which have a high diversity and even contradictory requirements. The state handling mechanism for 5G needs to consider all the 5G use cases and therefore is a key component for 5G design. The user plane and control plane latency should be low to reflect the 5G use cases, and is required to be significantly reduced compared to existing cellular systems. This enables a good user experience and improves the battery life of UEs, since it enables a fast transition from a power‐efficient state to an active state, which means that the devices can spend more time in the low‐power state.

Figure 13‐11 presents the 5G state machine with the three states and their related state transition procedures.

The RRC Connected state is optimized for high UE activity and has similar characteristics as RRC Connected in LTE, e.g., the UE context is stored in the network and in the UE, UE mobility is network controled, and the UE location is known at cell level. In RRC Idle, the UE context is not stored, and the use of this state may be limited to power‐up and to fault recovery procedures. RRC Connected Inactive is proposed as the low‐activity UE state. The new RRC Connected Inactive state will keep the UE context in the UE and the network to avoid a setup of radio bearers and security, and the connection between RAN and CN is kept active. When transitioning from RRC Connected to RRC Connected Inactive, the configured Access Stratum (AS) state information is retained by the UE and network. When transitioning back to RRC Connected, the AS state can be restored without the entire configuration signaling (e.g., state transition from RRC Idle to RRC Connected). Only the radio resources exclusively assigned to the UE are released and can be used by other UEs when the UE enters RRC Connected Inactive. In RRC Connected Inactive, the UE identifies itself to the RAN by a unique UE ID provided by the BS. The UE can move within the pre‐defined area in the RAN by performing cell reselections and without notifying the network. Considering the large number of use cases, the UE behavior can be configured based on the service requirements of the UE. The configurability of the RRC Connected Inactive state is a key feature to differentiate the UE behaviour for different service requirements, while using only a single but flexible low‐activity state.

The identified characteristics of the RRC state model are shown in Table 13‐3. The mobility and system access procedures of the new state model are configured based on different aspects of use cases, device capability, access latency, power saving, security requirements and privacy.

13.3.2 Mobility Procedures for Connected Inactive

The mobility procedures in RRC Connected Inactive state are related to cell selection and reselection, location tracking, location update, and reaching the UE by paging. Mobility during Connected Inactive state is based on cell selections and reselections performed by the UE, similar to cell (re)selections in LTE RRC Idle state. The 5G CN‐RAN connection, for both control and user plane, remains established for the UE in RRC Connected Inactive state. To avoid frequent path switching on the CN–RAN interface, the cell reselections within the allowed RAN tracking area (RTA) are not visible to the CN. Therefore, the CN–RAN interface path and UE context remains at the last serving BS, i.e. the gNB which suspended the UE's active connection to RRC Connected Inactive for the coming low‐activity period. The last serving gNB can now take the role of a mobility anchor, which allows the CP and UP of the CN‐RAN interface to be kept unmodified towards the CN. The overall mobility procedure for RRC Connected Inactive is presented in Figure 13‐12, including the state transitions between RRC Connected and RRC Connected Inactive [16].

If the UE in Connected Inactive state moves out of the RTA, there is a need for location update, either in the form of a periodical update or a RTA update. The gNB, where the UE is currently located, sends the UE’s current cell or RTA (i.e., location) to the anchor gNB, which can initiate the relocation on the CN‐RAN interface and hand over the anchor gNB role to the current gNB.

The anchor gNB initiates the paging in the RAN when receiving downlink packets. Upon receiving the paging response from the UE, the anchor gNB delivers the UE context data and the buffered DL packets to the current gNB where the UE is located. This then becomes the new serving gNB, re‐configures the UE to RRC Connected state and performs the required path switching procedure.

In the UE‐initiated connection, the currently visited gNB will retrieve the UE context from the anchor gNB based on the UE's reported last serving Cell ID, and the visited gNB will buffer UL packet(s) until it has performed the required path switching. In case the UE has not moved from the anchor gNB, the UE context is instantly available and there is no need to perform any path switching.

In the distributed RTA management, the anchor gNB may detect UE movement out of the current RTA based on its received UE registrations. The anchor gNB should deliver the UE context to the new gNB to let it take the role of anchor gNB, to trigger the RTA update procedure, and to perform CN‐RAN interface path switching.

13.3.3 Configurability of the Connected Inactive State

In LTE, there are two RRC states: RRC Idle and RRC Connected. This reduced the standardization effort and simplified the state machine operation for LTE. RRC Idle is optimized for low UE power consumption. The UE Access Stratum (AS) context is neither stored in the UE nor in the network. The RRC Connected state is optimized for high UE activity where the UE AS context is stored in the UE and the network, and the CN‐RAN interface is configured and active. For power saving in the RRC Connected state, discontinuous reception (DRX) is adopted to enable different levels of UE power saving when there is no continuous data transmission. LTE’s two‐state model works well for example in MBB use cases.

However, the model is shown to be inefficient for handling use cases that involve a frequent transmission of small data packets, such as sensor measurements or smart phone keep‐alive messages. The inefficiency comes from signaling procedures, which are used to transition the UE from RRC Idle to RRC Connected, create the UE AS context and configure the CN‐RAN interface. With a large number of UEs which frequently transmit small data packets, this becomes inefficient for the network. It is also costly to keep the UEs in RRC Connected without active data transmission due to dedicated resources, handovers and measurement reports. The configurable RRC state machine was developed to handle the mobility state of UEs with diverse requirements of the UE services, such as battery life, latency, bandwidth, mobility, etc., during RRC Connected Inactive.

The UE state model should fulfil all the requirements coming from different 5G use cases and deployments. All use cases share some common characteristics and default procedures, such as suspending and resuming the connection, tailoring the UE mobility procedures to reflect the UE speed, security and privacy, data rate, latency and reliability, expected UE battery life, etc. A power‐optimized configuration might be used for a UE with a requirement for a long battery life. On the other hand, a configuration optimized for latency can be used for UEs with applications requiring low latency. The configurability options can be classified into commonly used functions between UE and network [17]:

• Configurable discontinuous reception (DRX);
• LTE‐5G tight integration;
• State transitions;
• Measurement configuration;
• Paging and location tracking;
• Synchronization.

A widely configurable DRX at the UE is needed for different traffic patterns and battery requirements. For example, very small packets may be transmitted infrequently by sensors, and those types of devices also need a long battery life. In addition, there may be use cases requiring quite low control plane latency without stringent power consumption requirements, such as vehicles sending and receiving safety and traffic related information. Therefore, the RRC Connected Inactive supports configurable DRX where some devices will monitor the system control channels frequently, while some may stay in low‐activity state for several hours, but also benefit from quick connection resumption.

In case of 5G stand‐alone deployment or a tight integration of LTE and 5G, the dynamic usage of available resources in available radio access technologies (RATs) will enable a wide range of services with the best coverage. The tight integration of LTE and 5G demands common state handling for multi‐radio UEs, especially when connected to the same CN. Non‐integrated state transitions between LTE and 5G would lead to significant signaling load especially on LTE side, since the state transition between 5G Connected Inactive to LTE Idle would result in a UE context release for the UE and network. In RRC Connected Inactive state, the multi‐radio UE can be configured to camp either on LTE or 5G, preserve the UE context between inter‐RAT cell reselections, and monitor the paging from both RATs. Yet another alternative would be to configure multi‐radio UEs to camp simultaneously on both RATs and possibly try to simultaneously access both RATs. This would be beneficial for use cases with a fast establishment of multi‐connectivity after low‐activity periods, but requires more battery power, since two systems are monitored for paging and location tracking.

State transitions can be configured and optimized for different service characteristics. The applications or services transmitting small packets frequently can be configured with different state transition characteristics compared to applications transmitting and receiving large volumes of data. The UE with small data packets can send it as a multiplexed payload of the RRC connection resume request message, since the security context is available in the network. A UE with large data volumes should do a fast state transition to RRC Connected to benefit from maximum data throughput and low latency.

The measurement configuration reflects the characteristics of the requested service. For example, a static device or sensor can be configured to monitor neighbor cells less frequently compared to moving a UE. A UE supporting ultra‐reliable communication should monitor the neighbor cells and control channels more frequently and possibly from multiple RATs to establish fast the suspended multi‐connectivity at the time of connection resumption. Another example is measurement configuration and beamforming at higher frequencies, where UEs can be configured to monitor dedicated reference signals at specific beams or to measure common signals transmitted in wider beams.

In 5G, the RAN controls the UE paging and location tracking, where the gNB that terminates the CN‐RAN interface operates as the paging initiator and mobility anchor. The RAN‐based paging area may be configured to be small to ensure that the inter‐gNB Xn interfaces are available to all the gNBs within the tracking area. In this case, the RAN‐controlled location tracking and paging is mostly suitable for low‐mobility UEs to avoid frequent RTA updates. For UEs that require high mobility, a larger tracking area is beneficial, and the paging can be initiated from CN. In LTE RRC Idle state, the synchronization between UE and network is not maintained. In RRC Connected Inactive state, the UL synchronization is not needed for all UEs. However, UEs requesting low‐latency system access, such as in the context of industry automation, may maintain the UL synchronization. Those UEs can monitor a dedicated channel and therefore enable seamless resumption of the RRC connection for low‐latency system access.

Figure 13‐13 illustrates a procedure where the UE in RRC Connected state has an active connection towards the network. After detecting the inactivity, the UE requests suspension of the connection. Alternatively, the network may notice the end of incoming data and thus detect the potential inactivity period for the UE. The connection suspension using the RRC Suspend command will include the UE service‐specific characteristics, which will fulfil the low‐activity period requirements in terms of power consumption vs. system access latency and granularity of location tracking.

13.3.4 Paging in Connected Inactive

Locating and tracking the low‐activity UEs in the RAN means that the RAN is having a distributed management function, where each cell belongs to an RAN tracking area (RTA). The cells in an RTA have information available about their adjacent RTAs in order to create and deliver up‐to‐date lists on allowed RTAs to the UEs that are in RRC Connected Inactive state.

In the example in Figure 13‐14, the 5G gNBs serve three cells. For example, gNB1 is serving 5G cells 13, 14 and 22. Each gNB connects to a CP entity which is hosting the mobility management function in the core network by using the RAN‐CN CP interface N2, and to the UP function by using the RAN‐CN UP interface N3, see also Section 6.2. The neighboring gNBs are inter connected in the RAN by using the Xn interface.

When a UE is configured to enter Connected Inactive state, it receives from its serving gNB a list of cells belonging to the allowed RTAs. In Figure 13‐14, the UE gets configured with RTA1 and 2, and the UE can move within cells in these RTAs during Connected Inactive state without informing the network about the performed cell reselections. The UE location is known by the RTA, thus the serving gNB does not need to perform the serving gNB change even if the UE performs cell reselection and informs the network about it. A serving gNB change can be done after the UE goes back to RRC Connected for active data transmission.

Each 5G cell advertises in broadcasted information its Cell ID, and the list of cells belonging to a RTA can be signaled to the UE using dedicated signaling. The allowed RTAs are handled transparently to the UE with assistance of the current gNB, which takes care of reporting the UE’s current location to the UE’s last serving gNB.

In the RRC Connected Inactive state, the CN/RAN connection, i.e. both control and user planes remain established for the UE. If the UE reselects a cell that is not in its RTA list, it initiates a RTA update procedure, as shown in Figure 13‐15. After a RACH procedure, the UE sends a RTA update request message that is integrity‐protected using the AS security context that is stored in the UE and the anchor gNB. The message includes shortMAC‐I for UE authentication, the current UE’s RTA list and the resume ID that contains the address of the anchor gNB. If a gNB other than the anchor gNB received the RTA update request message, it forwards the message to the anchor gNB. The anchor gNB may keep the RAN/CN interface and respond with an RTA update response message (case 1) or initiate the RAN/CN interface relocation to the gNB that received the RTA update request message (case 2). The RTA update response message includes a new RTA list. After receiving the RTA update response message, the UE may send a RTA update complete message as an acknowledgement of the successful procedure.

When mobile terminated (MT) data arrives at the next generation core user plane (NGC‐UP), it forwards the data to the anchor gNB of the UE. The anchor gNB buffers the received MT data and initiates the paging procedure to reach the UE, see Figure 13‐16. The anchor gNB sends the paging message to all gNBs in the RTA list of the UE. The gNBs then can page the UE through the cells that are in its RTA list. Here, the anchor gNB needs to keep a gNB‐gNB Xn interface relationship with all the gNBs in the RTA list of the UE.

Upon the reception of the paging message, the UE proceeds to resume its RRC connection from RRC Connected Inactive to RRC Connected state. The RRC resumption procedure may include a context fetching procedure if the UE was camping in a gNB other than the anchor gNB.

13.3.5 Small Data Transmission in RRC Connected State

The new 5G radio needs to support efficient and low‐latency small packet transmission. This is crucial since many applications, sensors and mMTC devices are expected to generate small packets. If the state transition is needed for every small packet, there will be high protocol overhead due to state transitions. Even if the transition from RRC Connected Inactive to RRC Connected is a light‐weight signaling procedure, there is still some overhead and latency which cannot be avoided. Thus, the transmission of packets directly from RRC Connected Inactive would be beneficial for optimized small packet transmission, since the transceiver in a device is active only for a short period. When a UE wants to transmit UL data, it could use a small packet transmit procedure (SPTP). The SPTP, depending on the procedure, may involve just a single contention‐based transmission (one step) or may be preceded by a contention‐based scheduling request and grant (two step) [19]. The SPTP resembles a standard RACH operation, but allows user data transfer already during scheduling request and scheduling grant. This procedure can support many concurrent and HARQ‐enabled transmissions of variable‐sized packets and with minimal control overhead. If there is a downlink packet for the UE, RAN‐based paging [16][20] can be done, and this in turn triggers the UE to initiate the SPTP.

When the UE has been configured for small data transmission during low‐activity periods, a UE in RRC Connected Inactive state can move without any mobility‐related signaling as long as its SPTP configuration is usable, possibly restricted to a set of cell IDs or tracking area IDs. In Figure 13‐17, the UE transmits its small UL data to any gNB in the RAN‐based tracking area (RTA) in SPmsg3. In case its UL data cannot be fully fitted into the small packet block reserved to it with SPmsg2, the UE can indicate the number of still required small packet blocks or the transport block length. The current gNB will schedule further UL grants for RNTI1x, which allow the UE to transmit the remainder of the UL data packet. After the BSR has been fully consumed, gNB1 stops providing UL grants. The UE is still able to receive DL packets as long as its assigned RNTI1x is valid, but the UE has no configured resources for sending scheduling requests, except those for SPmsg1. Accordingly, a possible response from the UE’s communication peer is forwarded to the UE without prior RAN paging, while the UE again utilizes the SPTP in case further UL data becomes available later on.

In Figure 13‐17, it is assumed that the LTE security architecture is retained, and only the anchor gNB is able to decipher and possibly check the received data packet for integrity. Accordingly, the current gNB forwards the ciphered ULdata1 packet to the anchor gNB and receives the deciphered (and decompressed) ULdata1 packet back in return. The current gNB can take the role of the new anchor gNB and update the downlink path towards the UE and trigger the mobility management procedures.

13.4.1 Introduction

As discussed earlier in Section 2.3, the reliability and interruption delay requirements for some 5G services are significantly more stringent compared to 4G. In addition to this, 5G is expected to operate in a wider range of frequencies (1‐100 GHz) than 4G. This also means that beamforming techniques may be needed to compensate for the higher propagation loss at high frequencies. This section presents several different methods for addressing the 5G active mobility requirements. One of the agreements made so far is that 3GPP “will aim to define HO for NR with an interruption as close to zero as possible while only having single Tx/Rx in the UE, and 0 ms interruption at least for the case that the UE supports simultaneous Tx/Rx with source cell and target cell during HO” [21]. This is a major improvement compared to LTE, which always involves a data interruption during handover. Basically, there are two different alternatives to solve this:

• Multi connectivity with role switch;
• Normal handover with enhancement.

In this section, both alternatives are investigated. Another important new topic will be mobility in the context of beamforming. Due to the requirements of a lean design of the 5G DL reference symbols, the UL measurements can complement the DL measurements for beamforming mobility. One method to perform UL measurements for beamforming mobility is described in Section 13.4.2. The beamforming mobility design should support a fast switching and tracking of the communication beam to combat rapid changes in link quality. Also, the design should be able to exploit the availability of multiple overlapping beams that can be used for the communication with a single UE. One solution to fulfil these requirements is a multi‐connectivity solution called cluster‐based mobility, see Section 13.4.3, which is a set of nodes that the UE can detect and which are prepared in advance for a fast re‐routing of the signaling and user data. Section 13.4.4 discusses ways to minimize signaling for multi‐connectivity.

Enhanced normal handover is treated in Sections 13.4.5 and 13.4.6. More precisely, Section 13.4.5 presents ways to avoid data interruption at a handover, and Section 13.4.6 analyzes ways to improve the performance during high velocity by improving the channel state information.

13.4.2 Mobility Management via UL‐based Measurements

Next generation cellular systems must provide a mechanism by which user equipments (UEs) and mmWave gNBs establish highly‐directional transmission links, typically formed with high‐dimensional phased arrays, to benefit from the resulting beamforming (BF) gain and to compensate for the increased pathloss experienced at high frequencies, as detailed in Section 11.5. In this context, directionality requires a fine alignment of the transmitter and the receiver directional paths, an operation which might dramatically increase the time it takes to access the network. Moreover, the dynamics of the mmWave channel imply that the directional path to any cell can deteriorate rapidly, necessitating the need for intensive tracking of the mobile terminal [22].

Therefore, periodical monitoring of the channel quality between each UE and mmWave gNB pair, in order to perform a variety of control tasks (including handover, path selection, radio link failure detection and recovery, beam adaptation, etc.), is fundamental to provide efficient mobility management schemes. However, while channel tracking and reporting is relatively straightforward in cellular systems at conventional frequencies, the mmWave bands present several significant limitations: (i) the high variability of the channel in each link due to blockage; (ii) the need to track multiple directions for each link; and (iii) reports from the UE back to the cells must be made directional.

To address these challenges, a novel multi‐connectivity UL measurement framework has been proposed, as depicted in Figure 13‐18, that, with the joint effort of the legacy LTE frequencies, enables fast, fair and robust cell selection [23].

Unlike in traditional LTE schemes, the proposed framework is based on the channel quality of uplink rather than DL signals. This eliminates the need for the UE to send measurement reports back to the network and thereby removes a possible point of failure in the control signaling path. Moreover, if digital beamforming or beamforming with multiple analog streams is available at the mmWave cell, the directional scan time can be dramatically reduced when using UL‐based measurements. Finally, mobile terminals are the most energy‐constrained network entities, due to their limited battery capacity, contrary to the BS nodes which are always power‐connected and do not suffer from strict energy requirements. Therefore, a UL measurement framework, in which digital beamforming, being the most power‐consuming beamforming architecture, is used at the gNB side, should be preferred to enable a more efficient mobility management scheme.

In detail, each UE directionally broadcasts a sounding reference signal (SRS) in a time‐varying direction that continuously sweeps the angular space. Each potential serving cell scans all its angular directions and monitors the strength of the received SRS, building a report table (RT) based on the channel quality of each receiving direction, to capture the dynamics of the channel. Once the RT of each mmWave gNB has been filled for each UE, each mmWave cell sends this information to a centralized coordinator, for instance residing in the LTE eNB, which, due to the knowledge gathered on the signal quality in each angular direction for each gNB‐UE pair, obtains complete directional knowledge over the cell it controls. Hence, it is able to match the beams of the transmitters and the receivers to provide maximum performance.

Therefore, the coordinator reports to the UE, on a legacy LTE connection, which mmWave gNB yields the best performance, together with the optimal direction in which the UE should steer its beam to reach the candidate serving mmWave gNB in the optimal way. The choice of using the LTE control link is motivated by the fact that the UE may not be able to receive from the optimal mmWave link if not properly configured and aligned. Moreover, since path switches and cell additions in the mmWave regime are common due to link failures, the control link to the serving mmWave cell may not be available either. Finally, the coordinator notifies the designated mmWave gNB, through a high‐capacity backhaul link, about the optimal direction in which to steer the beam, for serving each UE.

The proposed multi‐connectivity UL measurement framework can be used to address some of the most important 5G control plane challenges that arise when dealing with the mmWave frequency bands.

In particular, robust and efficient handover (performed when the UE moves from the coverage of one cell to the coverage of another cell) and beam adaptation (which refers to the need for a user to periodically adapt its steering direction to realign with its serving gNB) can be improved. Frequent handover, even for fixed UEs, is a potential drawback of mmWave systems due to their vulnerability to random obstacles, which is not the case for lower carrier frequencies. Dense deployments of short‐range gNBs, as foreseen in mmWave cellular networks, may further exacerbate frequent handovers between adjacent gNBs. A loss of beamforming information due to channel change is another reason for handover and re‐association. The presented UL measurement framework can ensure more efficient mobility management operations by exploiting the centralized control of the LTE eNB over the network, to periodically determine the UE’s optimal mmWave gNB (and direction) to associate with, or the new direction through which it should steer the beam, when the user is in connected mode, i.e., when it is already synchronized with both the LTE eNB and the mmWave gNBs. The use of both the sub‐6 GHz and the mmWave control planes is a key functionality for efficient handover management. In fact, especially when considering highly unstable mmWave link conditions or initially scarce mmWave deployments, the LTE connectivity ensures a ready backup in case the mmWave links suffer an outage. Furthermore, the handover and beam adaptation decision is forwarded to the UE through the controller, whose legacy link is much more robust and less volatile than its mmWave counterpart, thereby removing a possible point of failure in the control signaling path [24].

In addition, a multi‐connectivity UL measurement framework allows the final attachment decision to be made by the controller operating at LTE frequencies. Therefore, unlike in traditional attachment policies, the association can be possibly performed by accounting for the instantaneous load conditions of the neighboring cells, to guarantee enough fairness and reliability to the whole cellular network.

Moreover, an UL design offers a significantly reduced access delay when a digital architecture is preferred. The main reason is that, due to the BS’s less demanding space constraints with respect to a mobile terminal, a larger number of antenna elements can usually be packed at the gNB side, resulting in a larger number of directions that can potentially be scanned simultaneously through a digital beamforming scheme.

13.4.3 Cluster‐based Beam Mobility Framework

5G communications can be outage‐prone for higher frequency bands, such as mmWave bands. In these bands, communication relies on strong line‐of‐sight (LOS) or near‐LOS components via beamforming. This may lead to frequent handovers between different beams or cells that support a specific location, as discussed before. In order to support frequent cell switching in an agile way, and in a manner that is transparent to the core network, access point clustering can be instrumental.

A cluster is defined as a group of access points (APs) in the vicinity of a UE, capable of serving that UE. The APs included in the cluster can be configured by the network and subsequently reconfigured when the UE moves. Clustering APs can happen in a multi‐tier format, enabling better exploitation of environment characteristics, for instance being based on the relative height of the APs.

To coordinate the mobility within a cluster, one of the APs can be designated as the cluster head (CH), which is connected to the core network through the N2 and N3 interfaces [25], see also Section 6.2. The CH is also connected to all other APs in the cluster. To enable the CH to coordinate the inter‐AP switching in a fast and efficient manner, it is assumed that a limited number of hops exist between the CH and the APs in the cluster. Depending on the topology of the network, the capacity and latency of the transport links, and the network position and role of the CH may vary.

The functional split for a cluster with sufficiently well‐dimensioned transport links can be made quite similar to the functional split of coordinated multi‐point systems. Here, all the intelligence may be located in a central node (i.e., the CH), which is responsible for all control and user plane protocol handling, including how the mmWave beams are tracked at different APs. The CH decides which APs serve the UE or which APs stay in stand‐by mode.

For non‐ideal transport or wireless (self‐)backhauling cases, see Section 7.4, the CH can handle Packet Data Convergence Protocol (PDCP)‐level functionalities, whereas other cluster APs handle the Radio Link Control (RLC), Media Access Control (MAC) and PHY layers, corresponding to the 3GPP split option 2 [26], as detailed in Section 6.6. This way, it is possible to maintain a single PDCP entity in one gNB, while switching the RLC, MAC and PHY from one transmission and reception point (TRP) to another. This protocol split also allows for arranging UE‐specific BS clusters [27], as opposed to static clusters in the case of a higher centralization of functionality.

Intra‐AP beam switching is triggered by the UE measurement feedback to the AP. In case of inter‐AP beam switching, the measurement report will be forwarded to the CH from the current serving AP. The CH will request the target AP for beam switching. If positive feedback is received from the target AP, the UE will eventually be informed (via the CH and the serving AP) to switch its beam, and will be served via the target AP after the switch.

The high path loss, and high susceptibility to blockage are two main factors affecting the mmWave systems’ coverage, especially outdoors. Densely deployed mmWave small cells with multi‐node coordination seem to be a feasible solution to both these issues. The use of coordinated BSs to enhance the data rate and coverage of the network has been widely studied in the context of 4G LTE/LTE‐A networks. Various techniques, e.g., joint transmission, coordinated beamforming, and cooperative communication, have been considered.

The height of the APs has a significant impact on the probability of LOS links, as shown in Figure 13‐19 a) below, where the LOS link between a low‐rise AP and the UE is blocked, but a LOS link between a high‐rise AP and the UE is still available.

Here, we target UEs requiring immersive 5G experiences and thus demanding extremely high data rates. In addition to the mmWave APs installed to street furniture with relatively low height, a small group of significantly simplified APs with beamforming capability are installed on high buildings to provide a beam resource pool (BRP) as shown in Figure 13‐19 b). The high‐rise APs can be implemented as remote radio heads (RRHs) in order to reduce the complexity and cost. When the LOS links between low‐rise APs and those UEs with extremely high data rate requirements are blocked, the beams in the BRP can be used to establish a LOS connection. Therefore, consistent user experiences without interruption can be supported by this two‐tier deployment. In Table 13‐4, it is shown how many high‐rise APs (30 m high) need to be installed together with each 100 low‐rise APs (3 m high) for a target overall LOS probability of 95%, based on the results in [28]. It can be seen that the LOS probability can be significantly enhanced with only a few high‐rise APs.

The high‐ and low‐rise APs can be connected to switches via high‐rate and low‐latency fibre connections. A CH is connected to all high‐ and low‐rise APs via a switch. The functionalities of the CH can be integrated into low‐rise APs so that each low‐rise AP is connected to the high‐rise APs and acquires beam resources from the BRP when needed.

In both architectures, macro cells operating below 6 GHz might be potentially available and offer additional help, but are not essential to the proposed architecture. Two possible system architectures are illustrated in Figure 13‐20.

The UE context will be saved and updated in both the UE‐associated low‐rise APs and the CH. Based on the UE context, the CH maintains a list of candidate beams from the pool in a dynamic manner. Once a low‐rise AP detects that its associated UE is experiencing a low data rate because of LOS link blockage, it sends the request to the cluster head to check if the available beam list is empty or not. If there are no available high‐rise beams, the low‐rise AP continues the transmission to the UE, but with degraded performance. Otherwise, the CH sends the list to the low‐rise AP.

It should be noted that even if the list is not empty, the CH should only choose from those high‐rise beams where the associated high‐rise RRHs are close enough to the target UE, so that the signal strength can be kept at a high level. In the meantime, the CH also requests the available high‐rise APs in the list to measure the pathloss to the target UE to select the best high‐rise AP. If the best high‐rise AP significantly enhances the SINR of the target UE, the CH controls the switch to connect the low‐rise AP to the selected high‐rise AP.

Meanwhile, the low‐rise APs could still maintain the uplink channel between themselves and the UE, since, for 5G immersive experiences, the very high data rate required is normally only on the DL, and the UL connection could still be sufficient even if it does not experience LOS. Another reason that the original UL connection should be maintained is that the low‐rise APs must monitor the UL channel and measure the signal strength of the received signal. Once it is above a certain threshold, e.g. if LOS is available to the low‐rise APs again, the transmission will go back to the low‐rise APs and the high‐rise APs will be released back to the pool. Note that the DL and UL connections could be asymmetric as described above.

There is no need to do handover between low‐ and high‐rise APs. Local traffic steering can easily be employed in the proposed framework to steer traffic from the low‐rise AP to high‐rise AP once the LOS link of a low‐rise AP is lost. Consequently, the signaling procedure can be significantly reduced.

Another option is to integrate the CH with the low‐rise APs as illustrated in Figure 13‐20, so that each of the low‐rise APs can control the switch to select candidate RRHs or beams from the BRP. The operation procedure is similar to the previous case, but the functionalities of the CH are now performed by the low‐rise APs instead. The benefit of such architecture is that the signaling procedure can be further simplified. However, the signaling simplification comes with a cost: If each low‐rise AP must be able to perform CH functionalities, the complexity of the low‐rise APs is increased, whereas in the architecture with a dedicated CH, only the CH node needs to have the capability to perform the relevant functions.

13.4.4 Partly UE‐autonomous Cell Management for Multi‐Connectivity Cases

One of the challenges for UEs in multi‐node connectivity (or multi‐cell connectivity, carrier aggregation, etc.) is that multiple cell associations need to be managed. A fully network‐controlled connectivity management approach will lead to increases in RRC and backhaul signaling if adopted. For multi‐connectivity scenarios with the primary cell on the macro layer and the secondary cell on the small cell layer, numerous studies have shown that the mentioned signaling overhead is dominated by cell management events for the small cell layer, see e.g. [29][30][31].

It is therefore proposed to adopt a UE‐autonomous cell management approach for the small cell layer for cases with multi‐connectivity [32], while the primary cell management (handover) on the macro layer is still fully network‐controlled. Since primary cell management actions happen less frequently, and typically require interaction with the core network, such actions are assumed to continue to be fully network‐controlled and UE‐assisted [30]. The fundamental principles of the UE‐autonomous secondary cell management proposal can be summarized as follows:

1. The network prepares a set of small cells among which the UE is allowed to perform autonomous secondary cell addition and removal, and cell change;
2. The network signals a list of these prepared small cells to the UE, including the measurement that it shall use for performing autonomous secondary cell addition, removal, and change;
3. Once the UE fulfils the criteria for secondary cell addition, removal, or change for the prepared small cells, it takes the corresponding action without first sending RRM measurements to the network, and waiting for the corresponding RRC signaling messaged in the downlink.

The basic principles of UE‐autonomous secondary cell management actions are further exemplified in Figure 13‐21. Here, the notation of master gNB (MgNB) for the macro layer and secondary gNB (SgNB) for the small cell layer is adopted, similar as in LTE‐Advanced. The procedure for UE‐autonomous SgNB addition simply means that the UE sends a RA message to that cell when it wants to add this link (i.e., without involving RRC signaling). The involved SgNB informs the UE's MgNB of the addition. Similarly, for the SgNB change operation, the UE only sends a RA to the new SgNB (i.e., SgNB2 in Figure 13‐21). The SgNB2 thereafter informs the UEs MgNB and requests the release of the users’ connection to the SgNB1. In principle, the SgNB release operation could also have been conducted with a RA message, but it is most efficiently handled with a scheduled RRC message, as pictured in Figure 13‐21. In addition, signaling is also required for configuring the UEs for autonomous secondary cell management, as well as for preparing the SgNBs, as further detailed in [29].

The performance of UE‐autonomous secondary cell management has been evaluated by means of extensive system level simulations. For the highway scenario studied in [29], it is found that the average number of required RRC messages per UE per second is reduced from 4.9 to 0.35 by using UE‐autonomous secondary cell management. Similarly, the associated Xn signaling is reduced by approximately 50%. See also [33] for results on the data interruption time for multi‐node connectivity.

13.4.5 Enhanced Synchronous Handover without Random Access

The basic mobility functionality for LTE is based on UE‐autonomous cell re‐selections for RRC Idle mode and network‐controlled handovers—with UE‐assistance—for RRC Connected state. Numerous measurements from live LTE networks have confirmed that the mobility performance is generally good, observing close to 100% handover success rates and low percentages of ping‐pongs, i.e. undesirable handovers between pair of cells within a short time [34][35]. However, due to the asynchronous nature of the LTE handover functionality with random access (RA) at every cell change, there is an undesirable temporary data interruption gap at every handover. Field measurements reveal that the data interruption time at each handover ranges from, at best, 20‐30 ms and up to 100 ms (or even more) for some networks [34][35]. To overcome this problem for 5G, it is proposed to adopt a time‐synchronized and RA‐less handover functionality [36]. Similar solutions have also been discussed in 3GPP, see for example [37] and [38], but no decision has been taken at this moment. This is a natural evolution as many other features will require time‐synchronized BSs for the 5G‐era.

Following [39], the basic principle of RRC Connected mode synchronous RA‐less handover between a source and a target cell is illustrated in Figure 13‐22.

Once the source cell receives a measurement report to trigger the handover, the source and target cells exchange information and agree on the time the handover should take place. The subsequent handover command (i.e., the RRC reconfiguration message) from the source cell informs the UE of the exact time of the handover. In its simplest form, the UE continues to receive data from the source cell until the time of the handover, after which it starts to receive data immediately from the target cell. Given that the source cell and target cell are fully time‐synchronized and the UE knows the current value of the timing advance (TA) for the source cell, the UE is capable of measuring the received time‐offset from the two cells and compute the TA for the target cell, as outlined in [36]. The handover command from the source cell may also include an uplink pre‐scheduling command for the UE to immediately transmit in the uplink towards the target at the time of the handover.

The timing of the synchronized RA‐less handover functionality is further illustrated in Figure 13‐23. The simplest realization is shown in Figure 13‐23 a), where the UE stops receiving data from the source cell at the time of the handover (as signaled from the source cell in the handover command), and immediately thereafter starts receiving data from the target cell. For this case, the handover interruption time is reduced to a fraction of a subframe (or TTI), accounting for the received time differences of the signals from the two cells and potential UE processing times for performing the switch from source to target. In a more advanced version of the synchronous RA‐less handover functionality, the UE continues to listen to the source cell for a short time period, while in parallel also receiving data from the target cell. For the latter case, there is no data interruption time on the physical layer between the source and target cell. During the time where the UE receives data from both cells, those data packets could be the same (i.e., duplicated) or different data packets. It should also be noticed that for achieving a virtually zero data interruption time during handovers, there needs to be corresponding network support for fast and efficient data forwarding and flow switching between the two involved cells. In [40], more details related to this are reported, where the UE processing times, BS processing times, and backhaul signaling latencies between the source and target cell are taken into account.

The benefit of the proposed synchronous RA‐less handover functionality is a significant reduction of the data interruption time during handovers, approaching virtually zero. Moreover, the handover execution process is reduced compared to that of LTE, as there is no RA at the target cell at every handover. Measurements from LTE networks shows that the RA procedure typically takes on the order of ~10 ms [37]. The faster handover execution process translates to increased mobility robustness since the system is able to react faster, resulting in even lower handover failure probabilities [36]. Finally, the fact that the handover process no longer requires RA translates to savings in the required RA access resources for the system. For example, the handover rate per UE is found to typically vary from few handovers per minute to handovers every second depending on the network topology and velocity of the device [33][35][41][42].

13.4.6 RAN Design to Support CSI Acquisition for High‐Mobility Users

Channel state information (CSI) at transmitters is fundamental in many advanced transmission schemes. However, feedback delays in FDD, framing delays in TDD and transmission control delays of multiple milliseconds result in severe outdating of this information for terminals at vehicular velocities. Backhaul delays increase the problem when using coordinated multi‐point transmission (CoMP). Channel prediction based on extrapolation of the short‐term fading has proven to be inadequate at vehicular velocities and high carrier frequencies.

For this reason, a new scheme was proposed in [43], which may radically extend the prediction horizon when used on vehicles, and which is based on the usage of an additional antenna, a “predictor antenna”, placed in front of the transmission antennas in the direction of travel. This approach can provide an order‐of‐magnitude improvement in channel prediction performance compared to Kalman or Wiener‐extrapolation of previous measurements based on the channel statistics. The principle can be used to improve downlink transmissions that require CSI at transmitters (CSIT) in FDD as well as TDD systems.

There have been investigations in [43] on how different types of antenna designs on vehicles affect the attainable cross‐correlation between the channel measured by the forward predictor antenna and the channel later experienced by the main antenna, when it has moved to the position previously occupied by the predictor antenna. The attainable precision in the prediction of complex channel gains is directly related to this cross‐correlation. The use of monopole antennas placed on flat and uncluttered vehicle roofs was shown to provide the highest correlations, resulting in average cross‐correlations of 0.97‐0.98 for antenna separations of 0.5‐3 wavelengths, as shown in Figure 13‐24.

Those experimental results were based on measured data obtained in cooperation with TU Dresden, in their testbed using a 20 MHz OFDM signal working at 2.68 GHz, at 50 km/h vehicle velocity. The useful antenna correlation is however reduced due to mutual electromagnetic couplings with very closely spaced antennas, around 0.25 wavelengths apart. In [44], a simple and efficient scheme has been found that counteracts this effect. It is based on using an open‐circuit antenna decoupling method. The effect on the average antenna correlation of using such a pre‐compensator is also shown in Figure 13‐24.

Figure 13‐25 shows the attained normalized mean square prediction error (NMSE) for this case (two monopole antennas on a flat vehicle roof), when predicting the channels for 10 kHz wide OFDM subcarriers at 2.68 GHz, at 45‐50 km/h vehicle velocities. The statistics summarize the variability of the prediction accuracy over 1958 subcarriers, for different settings of the spacing between the prediction antenna and the main antenna. The statistics are collected over all predicted subcarriers and separate measurements.

The benefits of the predictor antenna concept to alleviate beamforming mispointing in massive MIMO backhauling of high speed vehicles, even when local scattering and multipath propagation around the vehicle generate very fast fading, was discussed and demonstrated in [45] and [46]. Additional application areas that would benefit from CSIT are coordinated multi‐point (CoMP) with soft handover and robust backhauling for delay and/or mission critical services to fast moving vehicles.

To summarize, the predictor antenna concept has been shown to be feasible, and to be able to provide accurate channel state information for large prediction horizons in time, corresponding to multiple wavelength distances in space.