PaperlayerView exact revision on arXiv

Source-linked AI summary

Improved Handover Through Dual Connectivity in 5G mmWave Mobile Networks

Michele Polese, Marco Giordani, Marco Mezzavilla, Sundeep Rangan, Michele Zorzi

arXiv:1611.04748v3cs.NIcs.IT

TL;DR

MmWave links offer much greater spectrum but are vulnerable to blockage and intermittent channel quality, making handover and path switching challenging. The paper evaluates a dual-connectivity protocol that simultaneously connects UEs to LTE and mmWave cells, combining uplink measurements with a local coordinator. Compared with conventional handover mechanisms, the study reports improvements in packet loss, control signaling, latency, and throughput stability.

  • Problem

    MmWave blockage and rapid channel changes make conventional handover and path switching difficult, while directional scanning can delay failure detection and rerouting.

  • Method

    The paper evaluates dual connectivity between LTE and mmWave cells using coordinated uplink channel measurements, a local coordinator, and realistic end-to-end protocol simulation.

  • Results

    The proposed framework reduces packet loss, control signaling, and latency while increasing throughput stability compared with conventional handover mechanisms.

  • Takeaways & Limitations

    Dual connectivity provides a more robust mobility-management framework for dynamic mmWave access links, with dynamic TTT benefiting some specific scenarios.

Abstract

from arXiv · show

The millimeter wave (mmWave) bands offer the possibility of orders of magnitude greater throughput for fifth generation (5G) cellular systems. However, since mmWave signals are highly susceptible to blockage, channel quality on any one mmWave link can be extremely intermittent. This paper implements a novel dual connectivity protocol that enables mobile user equipment (UE) devices to maintain physical layer connections to 4G and 5G cells simultaneously. A novel uplink control signaling system combined with a local coordinator enables rapid path switching in the event of failures on any one link. This paper provides the first comprehensive end-to-end evaluation of handover mechanisms in mmWave cellular systems. The simulation framework includes detailed measurement-based channel models to realistically capture spatial dynamics of blocking events, as well as the full details of MAC, RLC and transport protocols. Compared to conventional handover mechanisms, the study reveals significant benefits of the proposed method under several metrics.

I. INTRODUCTION

mmWave mobility is difficult because blockage and directional beam scanning can rapidly interrupt links, while path switching must remain fast. The paper proposes dual connectivity, coordinated measurements, a local coordinator, and faster handover procedures, then evaluates their end-to-end benefits.

  • Motivation: MmWave links can deteriorate rapidly because blockage, mobility, and handset or obstacle movements make channel quality highly intermittent.The human body can cause up to 35 dB of attenuation, while common building materials can completely block mmWave signals.
  • Motivation: Multi-connectivity maintains UE connections to multiple cells, allowing alternate routes when one mmWave link is blocked.The paper specifically considers simultaneous connectivity to one 5G mmWave base station and one legacy LTE eNB.
  • Motivation: MmWave handover requires faster failure detection and rerouting, but directional scanning across narrow beams can substantially delay path switching.These delays conflict with 5G’s goal of ultra-low latency and minimal service unavailability during switching.
  • Contributions: The paper provides a comprehensive end-to-end evaluation of mmWave handover and path switching under realistic dynamic scenarios using an ns-3 framework with measurement-based channel models and complete protocol layers.The framework models spatial channel characteristics, blockage dynamics, MAC-layer HARQ, network signaling, and end-to-end transport.
  • Contributions: The proposed dual-connectivity design tracks channel quality and angular directions across multiple links simultaneously, avoiding directional searches when switching is needed.Its uplink control signaling supports periodic measurement and link tracking for handover and beam adaptation.
  • Contributions: A local coordinator anchors traffic at the PDCP layer and manages path switching near the cells, reducing switching delay compared with distant core-network functions.The design also uses LTE-controlled fast switching and secondary-cell handover, with fallback to LTE when the mmWave link degrades.
  • Findings: The study reports reduced packet loss, control signaling, and latency, together with higher throughput stability, for dual connectivity compared with conventional handover mechanisms.It also finds that dynamic TTT can provide improvements in specific mobility scenarios where state-of-the-art methods fail.

B. Related Work

Prior dual-connectivity and handover research largely targeted conventional sub-6 GHz systems, leaving mmWave directionality and channel variability insufficiently addressed. This paper extends LTE dual connectivity for mmWave control, measurement, integration, and handover.

  • Related Work: LTE-Advanced dual connectivity was proposed for macro- and pico-cell systems, but those designs targeted conventional sub-6 GHz frequencies.They did not address the directionality and variability characteristic of mmWave channels.
  • Related Work: Prior 5G control-channel approaches using bands below 6 GHz improve robustness against blockage and extend coverage, but do not provide mmWave’s high capacities.
  • Related Work: Handover research is mature for traditional sub-6 GHz heterogeneous networks but remains recent for mmWave 5G cellular systems.Existing work includes vertical handover decision algorithms and models for macro-, femto-, and pico-cell handovers.
  • Proposed Framework: The proposed LTE-5G architecture connects the UE simultaneously to LTE and mmWave eNBs, using LTE as a user-plane backup and coordinating measurements through the LTE cell.The mmWave eNB connects to the LTE eNB through X2, while the LTE eNB provides the single core-network connection point.
  • Proposed Framework: Potential serving cells scan angular directions and report received SRS strengths, allowing a coordinator to use directional channel knowledge for serving-cell selection and scheduling.The measurement procedure assumes finite sets of directions at each eNB and UE.

1) First Phase – Uplink Measurements:

The uplink measurement framework collects directional channel-quality information across UEs and mmWave eNBs, then uses a coordinator to select the best link and beam over LTE control signaling.

  • Measurement collection: Each UE transmits uplink sounding reference signals across its available angular directions for channel estimation.mmWave eNBs scan their receive directions, either sequentially with analog beamforming or concurrently with digital beamforming.
  • Measurement collection: Each eNB records the highest SINR for every UE based on the best UE–eNB direction pair.These measurements form the local report tables used for network-wide link selection.
  • Network decision: The coordinator aggregates eNB report tables through X2 into a complete report table and selects the maximum-SINR eNB and beam direction for each UE.The selected UE and eNB directions are used to configure the candidate serving mmWave link.
  • Network decision: The LTE control link carries the coordinator’s selected mmWave eNB and UE steering direction, avoiding reliance on a potentially failed mmWave control link.The design also eliminates UE measurement reports sent back to the network, removing a possible control-signaling failure point.
  • Timing and scalability: The measurement process uses periodic 200 µs sounding with 10 µs signals, corresponding to 5% overhead, and its delay is independent of user count and MAC scheduling.With synchronous sounding, the framework scales well with network density; the reported overall delay is 12.8 ms.
  • Protocol control: The protocol stack gives both RATs complete RRC layers while using LTE RRC for dual-connectivity commands and mmWave RRC for link management and measurement reporting.This separates mmWave-specific RRC design from the LTE stack.

B. User Plane (PDCP Layer Integration)

The proposed architecture merges LTE and mmWave protocol stacks at PDCP, enabling non-co-located deployments and routing traffic between LTE and remote mmWave RLC instances.

  • Integration layer: The PDCP layer is proposed as the dual-connectivity integration layer where LTE and mmWave protocol stacks merge.This avoids requiring synchronization among lower layers and permits a clean-slate mmWave PHY-to-RLC design.
  • Deployment: The design supports non-co-located mmWave and LTE cells, reflecting the denser deployment expected for mmWave eNBs.The coordinator or PDCP layer can be deployed in the core network or in a macro LTE cell.
  • User-plane routing: For each bearer, PDCP in the LTE eNB forwards packets either to the local LTE stack or through X2 to the remote mmWave RLC.The architecture therefore supports traffic delivery across both radio access technologies.
  • RAT selection: The LTE connection is retained as an anchor, while mmWave is selected when at least one mmWave eNB is available and LTE is used when mmWave cells are in outage.This choice reflects the higher theoretical mmWave capacity and the possibility of zero mmWave throughput during outage.

C. Dual Connectivity-aided Network Procedures

Dual-connectivity procedures use an LTE anchor and local coordination to switch rapidly between LTE and mmWave links or between mmWave secondary cells without core-network interaction.

  • Procedure overview: The architecture supports fast switching between LTE and mmWave RATs and Secondary Cell Handover across mmWave eNBs.Standalone hard handover requires complete handover procedures, LTE initial access after mmWave loss, and core-network interaction for intra-mmWave handover.
  • Fast switching: When all mmWave eNBs are unavailable, fast switching sends an RRC Connection Switch command over LTE and updates PDCP state to move traffic to LTE.The mmWave eNB is notified through X2 when buffered content must be forwarded.
  • Secondary Cell Handover: Secondary Cell Handover switches from one mmWave eNB to another without core-network interaction.Measurement assistance can identify the target beam and may allow random access to be skipped when timing is maintained with multiple eNBs.
  • SCH decision rule: The SCH algorithm triggers after a target eNB maintains better SINR than the current eNB for a time-to-trigger interval.A third cell within 3 dB of the target during that interval leaves the handover scheduled.
  • SCH decision rule: Dynamic TTT decreases as the SINR advantage of the target over the current cell increases.The evaluation uses TTT_max = 150 ms, TTT_min = 25 ms, Δ_min = 3 dB, and Δ_max = 8 dB.
  • Evaluation: The evaluation uses ns-3 system-level simulations to compare dual connectivity with standalone hard handover using detailed protocol and channel modeling.The framework includes configurable mmWave TDD/MAC behavior and publicly available source code and scenario scripts.

A. Semi-Statistical Channel Model

The semi-statistical channel model combines measured 28 GHz blockage dynamics with statistical channel parameters to represent time-varying mmWave links. It derives SINR from beamforming gain, pathloss, transmit power, and thermal noise, while simplifying blockage as equal attenuation across paths.

  • Channel construction: The model combines measured local blocking events with statistical mmWave channel parameters to capture realistic channel dynamics.Static parameters are generated from measurement-based models, then experimentally measured blockage dynamics are superimposed when obstacles obstruct UE–eNB paths.
  • SINR computation: SINR depends on beamforming gain, pathloss, transmit power, and thermal noise power between each mmWave eNB and the UE.The pathloss model uses distance d and parameters α and β.
  • Channel dynamics: The channel matrix models spatial clusters, power fractions, angular spreads, and small-scale fading affected by UE motion, orientation, Doppler shift, and position.Large-scale fading parameters are updated every 100 ms to simulate sudden link-quality changes.
  • Model assumption: Local blockage is approximated as equally attenuating all propagation paths, although directional obstacles may affect only a subset of paths.The approximation may be reasonable because most power in a fixed direction comes from paths within a relatively narrow beamwidth.

2) Measurement of Local Blockage:

Local blockage is measured experimentally with moving obstacles and incorporated into the statistical channel model as a time-varying attenuation factor. The resulting SINR trace is sampled for coordinator-based handover decisions.

  • Blockage measurement: Moving obstacles placed between transmitter and receiver produce measured blockage traces used to modulate the statistical channel response.Experiments collect Power Delay Profile samples during walking or running blockage events.
  • Blockage measurement: 35–40 dB of attenuation can occur relative to the LOS baseline SINR during local blocking events.This attenuation factor is applied to the time-varying statistical channel response.
  • SINR modulation: The dynamic wideband SINR uses the statistical SINR in LOS and adds an experimentally measured scaling factor δ in NLOS conditions.δ represents the SINR drop measured across blocking scenarios.
  • Trace sampling: The final semi-statistical trace is generated every 125 µs and downsampled to one sample every D seconds for handover decisions.The coordinator builds a CRT from these samples before applying handover logic.

B. SINR Filtering

The evaluation filters noisy SINR measurements, simulates mobility through randomized urban obstacles, and compares handover configurations under varying reporting delays and traffic rates.

  • SINR filtering: A first-order filter converts noisy measured SINR estimates into a time-averaged trace used for channel tracking.The filter parameter η is chosen to minimize squared estimation error relative to the true SINR.
  • SINR filtering: Figure 4 uses SINR samples collected every D = 1.6 ms, with local blockage and fading producing rapid trace fluctuations.The example transitions from NLOS before t = 19.4 to LOS afterward.
  • Simulation scenario: The reference scenario is a 200 × 115 meter urban grid with four randomly deployed buildings, three mmWave eNBs, one co-located LTE eNB, and one moving UE.The default UE speed is v = 5 m/s over a 100 m path.
  • Simulation methodology: Monte Carlo experiments randomize obstacle deployment, generate a CRT every D seconds, and apply one of the evaluated handover algorithms.The study varies CRT delay, UDP packet interarrival time, and handover configuration.
  • Simulation parameters: The experiments test downlink UDP traffic under an SINR outage threshold of Γout = −5 dB and directional antenna configurations.The mmWave eNBs use 8 × 8 UPAs with 16 angular directions, while the UE uses eight directions.

IV. RESULTS AND DISCUSSION

The evaluation compares dual connectivity with fast switching against standalone hard handover across packet loss, latency, signaling, throughput, and handover behavior. Dual connectivity generally improves resilience, while reporting delay affects losses and latency.

  • Packet loss and handover: Dual connectivity triggers more handover or switch events because faster switching responds more frequently to channel dynamics.Increasing D reduces events, while dynamic TTT produces more events than fixed TTT when TTT < 150 ms.
  • Packet loss and handover: Dual connectivity produces fewer packet losses than standalone hard handover.Hard handover losses arise from RLC segmentation, buffer overflow, and outage periods during LTE random access, whereas faster switching reduces these effects.
  • Packet loss and handover: Increasing CRT delay D increases packet loss ratio because less frequent switching raises RLC buffer occupancy and overflow probability.The reported packet-loss figure uses TUDP = 20 µs; at TUDP = 80 µs, loss was zero across tested realizations.

B. Latency

Dual connectivity reduces latency relative to standalone hard handover because fast switching and secondary-cell handover are quicker, avoiding service interruptions. Latency also depends on channel-tracking delay and buffering conditions.

  • DC reduces latency and avoids service interruptions compared with standalone hard handover, despite producing more handover and switch events.Hard handovers take longer than fast switching and secondary-cell handovers.
  • Latency increases as the channel-reporting delay D increases because the UE remains longer on suboptimal cells, increasing RLC buffering and forwarding time.
  • Dynamic TTT produces nearly 15% more handovers than fixed TTT, yet the two configurations have comparable overall latency because some secondary-cell handovers are more timely.
  • Latency is much lower at TUDP = 80 µs because RLC buffers are usually empty and fewer packets require forwarding during switches and handovers.
  • DC can jointly deliver reduced latency and increased PDCP throughput, so the conventional latency-throughput trade-off does not hold in this evaluation.

D. Variance Ratio

The variance ratio measures normalized PDCP-throughput variability across handover configurations. Dual connectivity consistently produces lower variance than hard handover, while more aggressive tracking can increase absolute variability.

  • Rvar is computed as the PDCP-throughput standard deviation divided by its mean, with higher values indicating greater channel instability.
  • Rvar,HH is higher than Rvar,DC for every tested delay, handover metric, and UDP interarrival time, indicating that LTE stabilizes the DC rate.
  • Rvar increases when channel reports are collected more intensively because more handover and switch events create larger periodic throughput variations.
  • RDC/HH remains below 1 for every parameter combination, although dynamic TTT has higher absolute variance than fixed TTT.

E. RRC Traffic

Dual connectivity lowers RRC control traffic relative to hard handover, although it increases X2 forwarding traffic because PDCP packets are forwarded between LTE and mmWave cells.

  • RRC traffic measures UE–mmWave-eNB control operations and is independent of RLC buffer size and UDP packet interarrival time in the tested setting.
  • Fast switching generates lower RRC traffic than hard handover because DC uses smaller control exchanges for switching.A switch message can contain 1 byte per bearer, whereas a single-bearer RRC reconfiguration requires at least 59 bytes.
  • Dynamic TTT produces higher RRC traffic than fixed TTT because it requires more handovers and switches.
  • At D = 1.6 ms, RRC traffic is lower than at D ∈{12.8, 25.6} ms despite more handovers and switches, because frequent monitoring reduces control-PDU retransmissions.
  • DC increases X2 traffic because LTE-to-mmWave PDCP forwarding is required, whereas hard handover mainly forwards RLC-buffer contents during handovers.

G. Final Comments

Across the evaluated scenarios, dual connectivity improves latency, packet loss, control signaling, throughput stability, and average throughput relative to hard handover. Dynamic TTT can provide additional gains in specific mobility scenarios, while the evaluation uses a simulated semi-statistical channel model.

  • Dual Connectivity vs. Hard Handover: DC reduces latency by up to 50% because fast switching and secondary-cell handover are generally faster than traditional handovers.The number of switching events may nevertheless be higher with DC.
  • Dual Connectivity vs. Hard Handover: DC reduces packet loss, lowers user-plane control signaling, and reduces throughput variance by up to 40% compared with hard handover.
  • Dual Connectivity vs. Hard Handover: Overall inbound traffic costs may be equivalent because DC uses X2 forwarding while hard handover uses the S1 link for mmWave-eNB traffic.
  • UDP interarrival time: Latency is much lower at TUDP = 80 µs because mostly empty RLC buffers reduce forwarding during switching and handover events.
  • Fixed vs. Dynamic TTT: Dynamic TTT never degrades the analyzed metrics and can greatly reduce latency in a corner scenario where the UE turns at a T-junction and loses line of sight to both bottom mmWave eNBs.
  • The study uses a simulated semi-statistical channel model with realistic obstacle deployment because temporally correlated mmWave channel measurements were unavailable for an accurate analytical mobility model.
Loading 1611.04748v3…