c-ran enabled seamless mobility mechanism in autonomous
TRANSCRIPT
URSI GASS 2020, Rome, Italy, 29 August - 5 September 2020
C-RAN Enabled Seamless Mobility Mechanism in Autonomous Driving
Jian Zhang(1), Luhan Wang*(1), Wei Zheng (2) and Xiangming Wen(2)
(1) Beijing Advanced Innovation Center for Future Internet Technology, Beijing, China(2) Beijing Key Laboratory of Network System Architecture and Convergence, Beijing University of Posts and
Telecommunications, Beijing, China
Abstract
Connected autonomous driving has already been a newparadigm of driving for safety, efficiency and mobility.Connected autonomous driving can overcome the percep-tual limitations of single vehicle, which has a high demandfor high speed and seamless network connection. With thesuperiority of C-RAN, autonomous driving network can ef-fectively support a vast number of vehicles communica-tions and flexible multi-RRH cooperative transmission toimprove throughput. However, due to the high mobilityof vehicle, the problem of long latency or even link breakcaused by frequent hard handover still exists. In this con-text, based on cooperative transmission and the idea of mul-tiple connections, we studied seamless mobility mechanismin C-RAN enabled autonomous driving. In this paper, wepropose a user-centric dynamic cooperative cluster updat-ing scheme with the purpose of guaranteeing the seamlessof the vehicle during handover process due to mobility.Simulation results demonstrate that the proposed schemecan effectively solve the interruption latency or link inter-ruption problem caused by handover. And compared withthe non-cooperative case, the proposed scheme is superiorin throughput gain.
1 Introduction
Connected autonomous vehicles can break through the per-ceived limitations of traditional single-vehicle automaticdriving, thereby improving safety, efficiency and comfortof mobility, challenging the traditional human-controlleddriving mode [1]. Connected autonomous driving requiresreal-time uploading of massive data and obtaining drivingdecisions from the network. Therefore, this technologyhas extremely high demand for seamless network connec-tion. Simultaneously, owing to limited computing resourceon the vehicle side, the analysis and processing of sensordata needs to be handled at the the mobile network edgeserver, so connected automatic driving requires higher net-work bandwidth to transmit raw sensor data.
In addition, the high mobility of the vehicles makes thechannel conditions more complex and variable, and will in-evitably lead to frequent handover. The current switchingmode is the hard handover of LTE, which has a obvious user
plane interruption delay. In [2], 3GPP assess the U-planeinterruption time during handover. The generic handoverprocedure assumed in E-UTRAN is shown in figure 1, withassociated delays encountered in the procedure. In the fig-ure four constituents for the U-plane interruption are identi-fied, i.e., (a) radio layer process, (b) UL RRC signalling, (c)DL RRC signalling, and (d) path switch. According to thismodel, the total interruption time of the U-plane in the ULis (a) + (b) + (c), whereas the interruption in the DL is (a)+ (b) or (d), whichever is larger. Therefore, frequent han-dovers inevitably lead to a long handover latency or evenlink break, which does not meet the requirements of net-work seamlessness and is also a urgent problem.
Figure 1. U-Plane interruption involved in theintra-MME/UPE HO procedure in E-UTRAN
2 Related Work
Although there is no direct research on the seamless mo-bility of vehicles in the C-RAN enabled networked au-tonomous driving scenario, with the continuous develop-ment of 5G technology, some possible solutions have beenproposed. In [3], the authors use multiple MAC entities torealize seamless intra/inter-cell handover between physicaltransmission and reception points and reduce the interfer-ence levels across the network in NR. The authors in [4]introduce a logical architecture for network-slicing-based5G systems, and present a scheme for managing mobilitybetween different access networks, as well as a joint powerand subchannel allocation scheme in spectrum-sharing two-tier systems based on network slicing. Moreover, the de-scription of the FRAN (Fog Radio Access Network) archi-tecture and the associated mobility and resource manage-
ment mechanisms are provided to reduce the signaling costand increase the net utility for the RAN in [5]. Inspired bythe above literature, considering feasibility and practicality,we propose a simple and feasible seamless mobility mecha-nism based on the C-RAN enabled networked autonomousdriving scenario.
In this paper, we study the seamless handover problem ofvehicles in C-RAN enabled autonomous driving scenarios,aiming at designing a simple and effective seamless mo-bility mechanism. Actually, referring to the idea of multi-connection, if the vehicle always has at least one RRH withbetter channel conditions serving it, the drastic handoverlatency or even link interruptions will be invisible to vehi-cle user. In other words, we eliminate the U-plane inter-ruption delay of the handover process on the existing LTEhandover scheme.The main contributions of this paper aresummarized as: (1)Combined with the characteristics of thescenario, a simple and practical signal model is proposedto simulate vehicle signal transmission in the case of over-lapping cooperative clusters. (2) A novel user-centric low-complexity jointly RRH dynamic clustering and handoverscheme is proposed to maximize user revenue. Consid-ering the actual deployment, we have adopted a network-user joint selection framework (3) We evaluate and ana-lyze the performance of the proposed approach from theextensive simulations, and the results prove that the pro-posed approach can guarantee the seamless throughput ofthe user during the handover process compared with non-cooperation scheme.
Virtual BS Virtual BS
Load Balancing Load Balancing
BBU Pool
X2 X2
Optical Transmission Network
Beam
Figure 2. Basic network scenario
Regarding the notation, (·)T , (·)H , E[·] and ‖·‖ are trans-pose, Hermitian transpose, expectation and L2 normal formoperator.
3 System Modeling
3.1 Scene model
Fig. 2 depicts the data-control separation cloud radio accessnetwork architecture. We can see that the base band units
(BBU) with powerful signal processing ability are groupedtogether to form a BBU pool, and the radio remote header(RRH) are geographically separated from each other butconnected to the same BBU pool via fiber optic transmis-sion links. Autonomous-driving vehicles periodically per-form link measurement and send a measurement report in-cluding the vehicle position and channel status information(CSI) to the BBU. Then the network side carried out a cor-responding cooperative cluster construction and allocationalgorithm, which maintain an optimal service cooperationcluster for each vehicle user.
3.2 Signal model
Considering a downlink of a C-RAN system as shown infig. 3, where a set L of RRHs denoted as L = {1,2, ..., l}are serving a set V of vehicle user equipments denotedas V = {1,2, ...,v}, which are worked as the distributedMIMO style. Each RRH is equipped with M transmit an-tennas, and each vehicle user equipment has N receive an-tennas. Suppose that each RRH can serve up to µ vehicleuser equipments and for each V-UE, there is a correspond-ing RRH cooperative cluster for its signal jointly transmis-sion. Let Li ∈ LV = {L1,L2, ...,Lv} denote the RRH col-laboration cluster corresponding to V-UE i and Vq denotethe set of V-UEs served by RRH q. Hq
i ∈ CN×M denotethe channel martix from q th RRH to ith V-UE. DefineW q
i ∈CM×N as the beamformer from RRH q to V-UE i andlet Wi = [(W 1
i ),(W2i ), ...,(W
Lii )] ∈CMLi×1 denote the beam-
former collection intended for V-UE i. The received signalat ith V-UE can be expressed as
yi = ∑p∈Li
H pi W p
i si + ∑p∈Li
∑j 6=i
H pi W p
j s j+
∑q∈LKq/∈Li
∑l∈Vq
Hqi W q
l sl +Zi(1)
where si ∈C is the normalized data symbol designated forV=UE i and satisfies E[|si|2] = 1, Zi is the additive whiteGaussian noise(AWGN) vector subject to C N (0,σ2
i IN).Hence,the signal-to-interference-plus-noise ratio(SINR) for
Figure 3. System Model of cooperative C-RAN
V-UE i is
SINRi =
∑p∈Li
|H pi W p
i |2
∑p∈Li
∑j 6=i|H p
i W pj |2 + ∑
q/∈Li
∑l∈Vq
|Hqi W q
l |2 +σ2i
(2)
4 Proposed Seamless Mobility Mechanism
In this section,we focus on solving the problem of transmis-sion delay reduction and seamless handover from the per-spective of maximizing user revenue. To obtain the optimaldynamic RRH cooperative cluster for each vehicle user, wedefine the accession and departure rules. The basic idea isto enable RRH to join or leave a cluster based on the prede-fined preferences so as to maximize the user revenue utility.Specifically, we set the algorithm period to a suitable valuesuch as 20ms. At the beginning of each algorithm cycle,determine whether the user is in a state where a RRH needsto be added. If so, perform the add operation and end thealgorithm cycle; if not, determine whether the user needs todelete the RRH, and if necessary, disconnect the RRH andre-add a new RRH, which is equivalent to handover, andfinally end the algorithm cycle.
We introduce a cooperative gain parameter cg as a referenceto which RRH has the greatest possibility to join the cluster.A parameter α is adopted to limit the size of the cluster.Then the accession and departure rules can be respectivelydefined as
Accession rule: RRH b joins the cooperative cluster of ve-hicle user i,
i f
{size(Li) < α and cg(Li,Li∪{b})> 0b = argmax(cg(Li,Li∪{x}))
Departure rule: RRH b′ leaves the cooperative cluster ofvehicle user i,
i f
{size(Li) = α and SINRi < γ
cg(Li,Li∪{b′})> 0 and b′ = argmax(‖Hxi ‖)
where γ is the predefined threshold of SINR for the vehicleuser i in cooperative state. Cooperative gain parameter cgis defined as
cg(Li,Li∪b) = log(SLNRLi∪b
SLNRLi
) (3)
SLNRLi denote the signal-to-leadkage-plus-noise-ratio ofvehicle user i,which can be calculated by
SLNRLi =
∑p∈Li
|H pi W p
i |2
∑p∈Li
∑k 6=i|H p
k W pi |
2
︸ ︷︷ ︸leakage power of the ith V-UE
+ σ2i (4)
To simplify the calculation, when computing Eq. ??, weassume that maximum ratio transmitter (MRT) [6] is usedto maximize the useful signal power. Assume that the totalpower of the RRH is Pt , the power allocated by the basestation to each user is equal and fixed as Ptr, then the beam-forming vector W p
i can be taken as
W pi =√
ρ×(H p
i )H∥∥H p
i
∥∥∥∥W pi
∥∥2= Ptr
(5)
-150 -100 -50 0 50 100 150 200 250 300 350 400-100
-80
-60
-40
-20
0
20
40
60
80
100
1
2
3
4
5
6
7
8
9
10
11
12
13
14 15
16
17
18
1920
21
22
23
24
2526
27
28
29 30
3132
33
RRHV-UERoadRoad
v
V-UEs move to the right at a constant speed v .
Figure 4. Network model
5 Performance Evaluation
5.1 Simulation setup
Simulations were done using MATLAB 2018b and specificsimulation parameters are in table 1. As shown in Fig. 4,the RRH distribution is simulated by the Poisson point pro-cess on both sides of the road, and the vehicle users arerandomly distributed in the lane area. In addition, the smallscale fading coefficient h is modeled as the Jakes [7] wheredoppler frequency fd is 926 Hz and sampling period Td is1e-5s, which is shown in figure 5.
Table 1. Parameters for Simulation
Parameter Value Parameter Value
fd 926Hz µ 4Td 10−5s v 72km/hα 2 γ 3×108bps
pathloss 15.3+37.6 log10(dis) system bandwidth 3MHzN0 -174dbm/Hz RRH transimit power 43 dbm
5.2 Handover seamlessness analysis and per-formance comparison
In this section, we evaluate the seamlessness of user switch-ing and the superiority of the algorithm by observingchanges in user throughput. For the credibility of the re-sults, We take the statistical average throughput of a userunder different channel conditions simulation. We presentthe throughput of a V-UE as a function of time and compareit with non-cooperation scheme. It should be noted that the
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05time[s]
-20
-10
0
10
Magnitu
de[d
B]
Channel Modeled by Jakes, fd=926Hz, Ts=1e-05s
0 0.5 1 1.5 2 2.5 3Magnitude
0
50
100
150
200
250
300
Occ
asi
ons
Channel Modeled by Jakes, fd=926Hz, Ts=1e-05s
-4 -3 -2 -1 0 1 2 3 4Phase[rad]
0
50
100
150
Occ
asi
ons
Channel Modeled by Jakes, fd=926Hz, Ts=1e-05s
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045delay [s]
0
0.2
0.4
0.6
0.8
1
Corr
ela
tion
Autocorrelation of Channel, fd=926Hz
ClassicalSimulated
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1f/fd
0
100
200
300
400
500
600
Magnitu
de
Doppler Spectrum,fd=926Hz
ClassicalSimulated
Figure 5. Channel model
18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40Time[ms]
0
1
2
3
4
5
Thro
ughp
ut[b
ps]
106
DU-CP(Point)Non-CP(Fit Curve)DU-CP (Fit Curve)Non-CP(Point)
Interrupt Delay
Cluster Size=1
Cluster Size=2
Cluster Size=1 Cluster Size=2
Cluster Size=1
Cluster Size=0
Figure 6. Performance comparison of a handover, thealgorithm cycle is 20 milliseconds, and the U-Plane
interruption is 12 milliseconds.
result is intercepted from 18th millisecond to 40th millisec-ond, the throughput is calculated every millisecond and theparameter α is set to 2. As shown in fig. 6, during the han-dover process, the throughput of DU scheme is greater thanthat of Non-CP scheme and remains at a stable non-zerolevel. Obviously, our proposed scheme is far superior to thenon-cooperative one,it can keep user services uninterruptedduring the handover.
Fig. 7 shows a cumulative distribution function(CDF) plotof the average rate of the system for 3MHz system band-width. It can be observed that the system average userthroughput gain of the DU-CP scheme is much larger thanthat of the Non-CP. Moreover, within the range of gamavalues, as the cluster size gama increases, as more RRHsprovide services to users, the greater the user throughputgain obtained by the algorithm, which is also the typicaladvantage of multipoint coordinated transmission.
6 Acknowledgement
This work is supported by the NSFC project, undergrant 61801047, the Beijing Municipal Education Commis-sion Funding Project "5G Connected Autononmous Driv-
1 2 3 4 5 6 7 8 9Average Throughput[bps] 106
0
0.2
0.4
0.6
0.8
1
CD
F
Non-CPDU,gama=2DU,gama=3DU,gama=4
Figure 7. Average rate of the system at 3MHz bandwidth
ing Technology and Applications", and the 111 projectB17007.
References
[1] E. Yurtsever, J. Lambert, A. Carballo, and K. Takeda,“A Survey of Autonomous Driving: Common Prac-tices and Emerging Technologies,” arXiv e-prints, p.arXiv:1906.05113, Jun 2019.
[2] 3GPP, “Feasibility study for evolved Universal Terres-trial Radio Access (UTRA) and Universal TerrestrialRadio Access Network (UTRAN),” 3rd GenerationPartnership Project (3GPP), Technical report (TR)25.912, 1 2016, version 13.0.0. [Online]. Available:https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=1341
[3] J. Liu, K. Au, A. Maaref, J. Luo, H. Baligh, H. Tong,A. Chassaigne, and J. Lorca, “Initial access, mobility,and user-centric multi-beam operation in 5g new ra-dio,” IEEE Communications Magazine, 2018.
[4] H. Zhang, N. Liu, X. Chu, K. Long, A.-H. Aghvami,and V. C. M. Leung, “Network slicing based 5g andfuture mobile networks: Mobility, resource manage-ment, and challenges,” IEEE Communications Maga-zine, vol. 55, no. 8, pp. 138–145.
[5] H. Zhang, Y. Qiu, X. Chu, K. Long, and V. C. Le-ung, “Fog radio access networks: Mobility manage-ment, interference mitigation, and resource optimiza-tion,” IEEE Wireless Communications, vol. 24, no. 6,pp. 120–127.
[6] T. K. Y. Lo, “Maximum ratio transmission,” Commu-nications IEEE Transactions on, vol. 47, no. 10, pp.1458–1461, 1999.
[7] P. Dent, G. E. Bottomley, and T. Croft, “Jakes fadingmodel revisited,” vol. 29, no. 13, pp. 1162–1163, 1993.