Professional Documents
Culture Documents
PII: S1389-1286(16)30365-6
DOI: 10.1016/j.comnet.2016.10.017
Reference: COMPNW 6040
Please cite this article as: Ben-Jye Chang , Gunag-Jie Jhang , Minimizing Contention Collision prob-
ability and Guaranteeing Packet Delay for Cloud Big Data Transmissions in 4G LTE-A Packet Random
Access, Computer Networks (2016), doi: 10.1016/j.comnet.2016.10.017
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
Abstract
IP
For transmitting Explosive Bursts Big Data of Mobile Cloud Computing applications, the 4G
LTE/LTE-A standards are specified to provide extreme high data rate and low access delay for various
CR
real-time-demanded cloud services. In the uplink, data packet transmissions of different classes of traffic
of various UEs need randomly contend for the limited number of preambles through the Uplink RACH
US
channel time slots. Clearly, the extremely explosive data contentions certainly yield serious collisions,
AN
and then significantly increase access delay and packet dropping rate. That is, the quality of service (QoS)
of the delay-sensitive-based real-time traffic and the loss-sensitive-based non-real-time traffic cannot be
guaranteed satisfyingly. For overcoming the critical random access issue in cloud services over 4G
M
LTE-A, 3GPP specifies the Uniform Distribution Backoff Procedure and Access Class Barring (ACB) as
ED
the random access mechanism. The Random Access CHannel (RACH) for random contentions in 3GPP
LTE-A neglects some key factors: 1) different classes of traffic requiring different delay bounds, 2) how
PT
to reducing collision probability, 3) intensive congestion traffic and 4) differentiating the collision
domains. This paper thus proposes an adaptive random contention approach (ARC) that consists of three
CE
phases: 1) Sigmoid-based Access Class Barring algorithm, 2) Dynamic Preamble Selection Range (DPSR)
algorithm, and 3) Dynamic Initial Backoff (DIB) algorithm. The main contribution of ARC is based on
AC
the adaptive Sigmoid feature analysis of Cumulative Distribution Function of Normal Distribution
according to the successful contention probability and the RACH congestion state. Numerical results
demonstrate that the proposed approach outperforms the compared approaches in collision probability,
*
This research was supported in part by the National Science Council of Taiwan, ROC, under Grants under Grants
MOST-105-2221-E-224-031-MY2 and MOST-104-2221-E-224-010
1 B.-J. Chang and G.-J. Jhang are with the department of Computer Science and Information Engineering, National Yunlin University of Science
and Technology, Taiwan, ROC. (Corresponding author: B.-J Chang; e-mail:changb@yuntech.edu.tw; phone: +886-5-5342601*4511; fax:
+886-5-5312170).
1
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
goodput and access delay. Furthermore, the mathematical analytical model for the proposed approach is
analyzed. The analysis result is very close the simulation result. It justifies the correctness and efficiency
Keywords —Big data cloud service, random access channel (RACH), LTE-A, differentiate preamble
T
1. Introduction
IP
1.1. Random Access in 4G LTE/LTE-A
Recently, 3GPP specifies the 4G LTE/LTE-A standards as the wireless mobile access
CR
interface operated between various cloud computing services and different-type UEs. For
instance, Machine Type Communication (MTC) through 4G LTE/LTE-A extensively achieves
US
the Internet of Things (IoTs) [1][2]. Various-type communications in IoTs: 1) Machine to
Machine (M2M), 2) Device to Device (D2D), and 3) Human to Human (H2H) [3][4][5] yield the
AN
explosive bursts of the Big Data transmissions that need the data packet service in 4G
LTE/LTE-A. However, 3GPP adopts the out-of-band random access mechanism for data packets
transmissions, i.e., data packets contend among different User Equipments (UEs) by using the
M
RACH time slots. After successful contention, the data packet can be sent through the shared
channels (SCHs). The random access mechanism in 3GPP neglects the collision states, different
ED
classes of traffic and traffic loading. 3GPP suffers from high collision probability (or low
successful probability) and low data throughput that degrades the required Quality of Service
PT
contention-free, in which the total number of preambles (or preamble logical channels) for
random access is defined as 64 preambles. For the explosive bursts transmissions, this paper
focuses on the contention-based mode and can utilize the number of preambles up to 64 [7][8].
AC
In LTE/LTE-A, the 10 ms radio physical frame consists of 10 subframes, and there are two types
of these subframes: 1) the RACH time slot and 2) the data transmission time slot. The UEs with
ready data packets to be sent should contend the preambles of a RACH time slot by using the
p-persistent algorithm, namely the Access Class Barring (ACB) [1][2][4][9] mechanism. If the
contention is successful, the UE can send the data packet on the specified SCH. Conversely, if
two or more UEs selects the same preamble in the Message 1 (MSG1 for Contention Request),
the eNB cannot identify the requested UEs. As a result, the eNB does not reply these UEs by
2
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
sending MSG2 (Contention Reply). That means these contentions fail, and each collided UE
needs to wait a random backoff time based on the Uniform distribution specified by 3GPP (as
demonstrated in Table II) [7]. Since to reduce collision probability and the waiting time of MTCs
become an important issue, 3GPP specifies the Radio Access Network (RAN) to improve MTCs
[4].
1.2 Related works
The related works of random access in LTE/LTE-A are classified into three types: 1) random
access issue, 2) improvements for random access, and 3) performance analyses of random
T
access.
IP
1.2.1 Random Access Issues
CR
In [5][10], 3GPP proposes MTC to achieve M2M emergency and alarming systems. The
sensed or gathered big data should be transmitted to the cloud server periodically or on demand.
The extremely large amount big data is not allowed to use a dedicated channel in 4G
US
LTE/LTE-A, because the total number of dedicated channel is insufficient for MTC. Clearly, the
wireless radio resource and bandwidth are limited, radio resource allocation in MTC becomes
AN
the critical challenge that should be addressed efficiently.
In LTE/LTE-A, the out-of-band random access is adopted. The UE should succeed the
preamble contention first, and then the UE is allowed to send data packets through the SCH
M
channel. Clearly, the MTC supporting a very large number of UEs or machines easily leads to an
extremely high collision probability and access delay [11][12][13].
ED
Prioritizing with Access Class Barring, 2) Dynamic ACB, 3) Cooperative ACB and 4) Backoff,
as described below.
CE
according to the success probability of RACH. It sets a static pre-backoff time value for the
lowest priority UEs (i.e., M2M) transmitting preambles. They avoid yielding a high congestion
in RACH when the eNB determines the state of the high traffic loading, and then guarantees the
QoS of high-priority UEs. Without considering the RACH states of success, collision and idle,
the non-adaptive prioritizing ACB in LTE/LTE-A physical frame limits the performance.
1.2.2.2. Related works of Dynamic ACB
In [16], the ACB factor is adjusted according to the number of backlogged UEs and the new
request UEs. In [17], the ACB factor is adjusted according to the states of the RACH time slots
3
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
and the collision probability. In [18][19], the ACB factor is adjusted according to the arrival rate
of random access in RACH time slots, and thus can effectively decrease the collision probability.
In [20], the ACB factor is adjusted according to the allocation of resource blocks (RBs) for data
slots and the RACH slots. However, the mechanism only considers the same class of traffic,
rather than considers multiple classes of traffic. [20] cannot apply to 3GPP MTC, because 3GPP
MTC supports several classes of traffic, as depicted in previous subsection.
1.2.2.3. Related works of Cooperative ACB
In [21][22], based on the random access traffic load, the ACB factor is cooperatively
T
managed by several neighbor eNBs. The mechanism can reduce the access delay and the
IP
management overhead of the single eNB case. However, the mechanism neglects the priority of
CR
different classes of traffic, and thus degrades the QoS of requirements.
1.2.2.4. Related works of Backoff
In [23], the backoff time is determined based on the traffic class index. Although the backoff
US
penalty can be formulated according to the traffic class, the static setting of parameter cannot
satisfy the dynamic RACH contentions. In [24], two types of backoff algorithms of LTE/LTE-A
AN
and WiMAX are analyzed, namely Dynamic Window Assignment (DWA). DWA increases the
successful probability, but neglect to compare the backoff performance between the backff
algorithms of the Normal Distribution in LTRE/LTE-A and the Truncated Binary Exponential
M
(TBE) in WiMAX.
In [25], to avoid high collision probability of preamble contentions, a polling mechanism is
ED
applied to the existing backoff mechanism. However, [25] increases the access delay.
1.2.3 Performance Analysis of Random Access
PT
In [26], the RACH in LTE/LTE-A is designed for Poisson-distributed arrivals, but it neglects
that the unexpected burst arrivals may result in high collision in RACH. [26] proposes an
CE
analytical model to investigate the RACH with burst arrivals generated. In [27], for MTC
environment, it adopts the Group Paging mechanism to decrease the overloading problem of
AC
random accesses. [27] proposes an analytical model to investigate the RACH with Group Paging
mechanism. Extremely huge random accesses in MTC result in high congestion in RACH. 3GPP
selects Extended Access Barring (EAB) mechanism to barring low priority UE for decreasing
overload problem. [28] proposes an analytical model to investigate the RACH with EAB
mechanism. [29] considers that H2H and M2M perform random accesses within the same
network or in the heterogeneous networks that coexist with macro cell and small cell. [29]
proposes an analytical model to investigate this MTC environment. [30] proposes an analytical
approximation algorithm to estimate the performance of RACH in a one-shot random access
4
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
with the slotted Aloha mechanism. In [31], the collision probability for machine-type
communications using RACH is analyzed according to 3GPP specification [4].
Above all, the analytical studies propose several mathematical analysis of random access in
RACH. These analysis models can justify the correctness of simulation results. However, the
efficient random access control for the LTE/LTE-A RACH contention is required.
1.3. Critical issues, motivations, and objectives
From the 3GPP standards and related works of RACH random access in 4G LTE/LTE-A,
several critical issues need to be addressed efficiently, including:
T
1) Statically controlling the Access Class Barring factor,
IP
2) No differentiating the collision domains for different classes of traffic, and
CR
3) Using the uniform-based distribution backoff algorithm that is completely independent to
the number of retransmissions of an UE.
This motivates us to propose the Sigmoid-based Access Class Barring algorithm, and then
US
dynamically partitions the preamble collision ranges (i.e., collision domains) for different classes
of traffic. By using these two mechanisms, the collision probability is decreased significantly.
AN
Since in the RACH random access control an efficient backoff mechanism acts as an important
role, we propose the dynamic initial backoff algorithm according to the successful probability of
the RACH channel preamble of the eNB and the collision probability of the UE. Clearly, the
M
objectives or contributions of this paper include to reduce collision probability and access delay,
and then to increase goodput. Additionally, the mathematical analytical model for the proposed
ED
performance metrics are defined in Section II. Section III details the proposed efficient RACH
random access approach. Section IV details mathematical analysis model. Numerical results of
CE
analysis and simulation are provided in Section V. Conclusions and future works are given in
Section VI.
AC
2. Network Model
The section first defines the network model and then defines some useful notations. Finally,
we define some performance metrics of the contention control of LTE/LTE-A RACH random
access procedure for the proposed approach, 3GPP standard, and other related works.
We model a 4G LTE/LTE-A network as a graph, G (UE,eNB,M) that consists of different
classes (priorities) of traffic for UEs , the eNB , and M preambles in a Random Access Channel
time slot of a physical (PHY) frame, as demonstrated in Fig. 1.
5
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
In the network model, the eNB provides the Radio Physical Frame of Orthogonal Frequency
Division Multiple Access (OFDMA) and every PHY frame consists of 10 subframes. The RACH
time slot is assigned in some specified subframes (as shown in the gray subframes in Fig. 1)
according to the Physical Random Access Channel (PRACH) configuration [8]. In Fig. 1, 64
Preambles (i.e., preamble logical channels) exhibits in a RACH time slot and these 64 preambles
can be used for two transmission types: the contention-based transmission and the
contention-free transmission [7].According to the out-of-band mechanism in the LTE/LTE-A
specifications [7],this paper focuses on the contention-based one used for random access
T
initialization before data transmissions. Since several configuration modes exhibit in PRACH
IP
configuration, for instance in the mode of PRACH configure 6, the RACH time slots are
CR
assigned in subframes 1 and 6, respectively [8].
First, we assume that N contention-based preambles exhibits in a RACH time slot are
denoted by M n , where 0 n N [7]. Moreover, in the traffic model, we assume that 4G
US
LTE/LTE-A network provides 4 classes (priorities) of traffic for a UE, denoted by UEidr , where
1 r 4 . Note that r 1 denotes the highest class and r 4 denotes the lowest class [3][6][9].
AN
In the contention-based RACH random access, a UE has to initialize the random access
procedure to contend the limited preambles before data packet transmissions. If two or more UEs
M
randomly contend the same preamble, it leads to a collided contention. Conversely, if only one
UE randomly selects a preamble, it leads to successful contention. Thus, the states of a preamble
ED
include: success, collision, and idle. In RACH random access in LTE/LTE-A, the collided UE
needs to execute the uniform distribution-based backoff algorithm, and then waits for a backoff
PT
delay. When the backoff delay is expired, the UE can contend again. In 3GPP, if a data packet
exceeds the maximum re-transmissions ( PTmax 10 ), it will be dropped.
CE
AC
6
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
Contention-based
random access request Data transmission
Preamble (Logical channel)
status
RACH time slot
Idle Preamble id = 63
Preamble logical channel
1 Preamble id = 62
UEpriority contention domain
Success Preamble id = 61
M Available Preambles
eNB (evolved NodeB)
(Logical Channels) Preamble id = 60
2
UEpriority Contention Domain
Collision Preamble id = 49 UE with traffic priority 1
(i.e., M=54)
T
Radio Physical Frame (10ms)
IP
HDFS Job Tracker
Assign Assign
Split 0
Split 1
Task Tracker
CR
Split 2 Task Tracker
(Reduce) (Reduce)
Split 3
eNB MME/S-GW/P-GW
Split 4 Input Sort / Group
4G LTE/LTE-A Mobile
Cloud Service Domain
Communication Wireless Network
US
Fig. 1. Network model
Several important metrics are defined for evaluating the proposed approach and the compared
AN
studies. The metrics include collision probability, success probability, average access delay, and
dropping probability.
RACH
First, the collision probability, PCollision , is defined as,
M
E[ M Collision ]
PCollision = , (1)
E[ M ]
ED
where E[ M ] denotes the average number of available preambles and E[ M Collision ] denotes the
average number of collided preambles.
PT
r ,id
Second, the success probability, PSuccess , is defined as,
E[S]
r ,id
PSuccess , (2)
CE
E[T ]
where E[T ] denotes the average number of transmissions of UEs and E[ S ] denotes the average
AC
TContention TBackoff
E[TAccess ]= , (3)
E[T ]
where E[T ] denotes the average number of transmissions of UEs, TContention denotes the average
contention delay and TBackoff denotes the average backoff delay. Finally, the dropping
probability, PDropping
r ,id
, is defined as,
7
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
r ,id E[Drop]
PDropping = , (4)
E[T ]
Table I. Notation
Notation Descriptions
ACB(x, , u) The Normal Distribution of Sigmoid function for access class barring factor.
BI max The maximum backoff index, i.e., BI max =15.
T
BI rmax
,id
The maximum backoff index of UE id with class r traffic.
r ,id
BI
IP
The backoff index of UE id with class r traffic.
BOValue The corresponding backoff value.
TAccess The total access delay of the UE.
CR
E[TContention ] The average contention delay of an UE.
r
E [ pACB ] The average ACB factor of UE id with class r traffic.
E[T ] The average number of transmissions of UEs.
E[ S ]
E[C ]
E[Drop]
US
The average number of successful of UEs.
The average number of collided of UEs
The average number of dropped preambles of UEs
AN
E[M ] The average number of the available preambles.
E[ M Idle ] The average number of the idle preambles.
E[ M Success ] The average number of the success preambles.
M
r
M CD The number of the available preamble contention domain class r traffic of UE.
i ,b
M block The number of the available preambles of block b at RACH time slot i .
r ,id
P The success probability of UE id with class r traffic.
PT
Success
r ,id
P Collision The collision probability of UE id with class r traffic.
r ,id
PDropping The dropping probability of UE id with class r traffic.
CE
RACH
P Idle The idle preamble probability of in a RACH time slot.
RACH
P Success The success preamble probability of in a RACH time slot.
AC
RACH
PCollision The collision preamble probability of in a RACH time slot.
PTmax The maximum number of preamble transmissions (i.e., PTmax =10).
r ,id
pACB The ACB factor (i.e., p-persistent value) of UE id with class r traffic.
RA Total_time The time period of Random Access simulation.
RA Cont_slot The number of RACH time slots in RA Total_time .
RA rep Interval between two RACH time slots (i.e., RA rep =5 ms).
The average number of random access requests of UEs in block b in RACH time
RA iblock
,b
slots i .
The traffic class (or priority), where rmin r rmax (i.e., rmin =1, rmax =4.
r rmin denotes the highest class, and rmax denotes the lowest class).
8
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
IP
3. The Proposed ARC LTE-A APPROACH
CR
For guaranteeing QoS and reliable cloud services by using random access of various classes of
traffic in LTE-A, this section proposes the Adaptive Random Contention approach for LTE-A
US
(namely ARC), consisting of three phases as listed below and depicted in detail.
Phase 1: The Sigmoid-based Access Class Barring algorithm,
AN
Phase 2: The Dynamic Preamble Selection Range (DPSR) algorithm, and
Phase 3: The Dynamic Initial Backoff (DIB) algorithm.
M
For the transmissions of extreme cloud bursts, 3GPP LTE-A packet data service specifies
the out-of-band random contention for accessing the Random Access CHannel (RACH). In the
out-of-band random contention, before UE’s data packets can be transmitted, UEs need to
PT
contend the limited RACH preambles among the uplink UEs by using the normal distribution
with p -persistent algorithm, where p is a random value [1][2]. Note that the RACH
CE
preamble can be viewed as the virtual channel for the random accesses. A contention can be
sent when the random selection probability is less than p ; otherwise, the contention should
AC
wait for next contention. However, in 3GPP the p value is unspecified and undefined for
various classes of traffic. Certainly, the 3GPP random access specification does not
differentiate the collision domains among different classes of traffic, and thus easily leads to a
high and uncontrollable collision probability.
In LTE-A random access performance, the traffic with a higher successful probability can
have a higher p value for increasing throughput, and vice versa. The Sigmoid function meets
the performance characteristic of the random access mechanism.
9
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
The Sigmoid function, referred to the special case of the Logistic function, is generated by
a mathematic function with an S curve, as shown in Eq. (5),
ex 1
S ( x) or , (5)
1 e x
1 e x
where e is the Euler’s number and x is the impact factor. In Fig. 2, the dash blue curve
indicates the feature of the Sigmoid function, S ( x ) or the Access Class Barring function,
ACB(x, , ) . Note that the impact factor x in S ( x ) and ACB(x, , ) are the same one. The
value of S exponentially increases, as the factor x increasing up to the medium point, and
T
then the value of S increases logarithmically, as the factor x increasing after the medium
IP
point.
CR
US
AN
M
r ,id
ACB probability, pACB , is formulated as
1 x u
ACB(x,u, ) 1 erf ( ) , (6)
2 2
where erf () denotes the error function, x is the impact factor, denotes the expected value
and denotes the standard deviation. Since Phase 1 aims to adaptively control the contention
probability (i.e., the preamble transmission probability) for different classes of traffic under
10
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
First, for parameter , in CDF of Normal distribution as shown in Eq. (7), the expected
value is the mean value of the Sigmoid function curve. Phase 1 thus considers the probability of
the busy period ( 1 PidleRACH ) of the RACH preambles as the pivot. When the pivot value increases
(i.e., the RACH preambles are busily), the upper S curve of the Sigmoid function is applied, i.e.,
the logarithmically increasing curve is applied for decreasing p value of the p -persistent
T
mechanism. Thus, the parameter is formulated by,
IP
1 Pidle
RACH
- r1 r , (7)
max
CR
RACH
where the RACH preamble idle probability, Pidle , is determined by Eq. (8), r denotes the
traffic class index and rmax denotes the total number of traffic classes.
RACH
Pidle US
E[ M idle ]
E[ M ]
. (8)
AN
In Eq. (8), E[ M idle ] denotes the average number of idle preambles and E[ M ] denotes the total
number of available preambles used for the data packet service.
M
Second, for parameter , in CDF of Normal distribution, the standard deviation affects the
width of the Sigmoid function curve. The Sigmoid function smoothly increases, as increasing.
ED
eu 1
+ r . (9)
1 eu rmax
CE
In Eq. (9), a higher value of smoothly increases the ACB(x, , ) function; but, a lower value
of sharply increases the ACB(x, , ) function.
AC
Third, for parameter x , the most important factor in Phase 1 is the successful contention
r ,id
probability, Psuccess . In p -persistent, the transmission of a contention is with probability p . With a
high probability p , the transmission probability increases, and vice versa. Thus, when the
value p (i.e., the parameter x ) should be increased for yielding a high efficient transmission. As
a result, we formulate the parameter x as,
11
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
x rmax Psuccess
r ,id
- r 1 . (10)
T
RACH r ,id
channel idle probabilities ( Pidle ) under different success probabilities ( Psuccess ), in which
IP
r ,id
pACB is determined by Eq. (2). We assume that LTE-A supports four classes of traffic (i.e.,
CR
rmax 4 ) and r 1 is with the highest priority but r 4 is with the lowest priority. Four
r ,id
characteristics exhibit in Figs. 3 (a)(b). First, all of the pACB of different classes increase as
r ,id
Psuccess
r ,id
factor, pACB
US
increasing. Second, the traffic with the highest priority ( r 1 ) yields the highest ACB
r ,id
. Third, the determined ACB factor, pACB , of each class traffic of 90% channel
AN
idle probability is higher than that of 10% channel idle probability. Finally, each ACB factor,
r ,id
pACB , exhibits the characteristic of the Sigmoid function curve, as shown in Fig. 2. That is, the
M
r ,id
ACB factor, pACB , exponentially increases, as the Psuccess
r ,id
increasing up to the medium value,
r ,id
and then the ACB factor, pACB , increases logarithmically, as the Psuccess
r ,id
increasing after the
ED
medium point. Above supreme features justifies the claim of Phase 1: adaptively determining the
r ,id
ACB factor, pACB , for different classes of traffic under different success probabilities, Psuccess
r ,id
.
PT
CE
AC
12
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
(b) 10% channel idle probability
IP
r ,id
Fig. 3.ACB probability, pACB , of 10% and 90% channel idle probabilities under different success
CR
probabilities
3.2. Phase2. Dynamic Preamble Selection Range(DPSR) algorithm
r ,id
After determining the ACB factor, pACB
US
, for p-persistent in Phase 1, Phase 2 aims to
minimize the collision probability of the contentions RACH preambles. Phase 2 thus proposes a
dynamic preamble selection range (DPSR) algorithm to differentiate the collision domains
AN
among different classes of traffic. By dynamically differentiating the collision domains, the
highest class of traffic significantly reduces collision probability and increases goodput.
M
Furthermore, according to the preamble idle probability, DPSR proposes two modes (the shared
mode and the partition mode) for defining the collision domain for different classes of traffic.
ED
Specifically, DPSR switches to the partition mode under a high PidleRACH for efficiently reducing
the collision probabilities among different classes of traffic. Conversely, DPSR switches to the
PT
shared mode under a low PidleRACH for guaranteeing the highest class traffic. The detail descriptions
of DPSR are depicted below.
CE
Phase 2 first determines the weights, and then determines the dynamical preamble ranges for
different classes of traffic. Based on the expected ACB factor E [ pACB
r
] , the total sum of
expected ACB factors of all classes of traffic (namely ) can be obtained by,
AC
rmax
E[p
r rmin
r
ACB ]. (12)
Since from Fig. 3 we have determined the Sigmoid-based ACB factor, pACB
r ,id
, the probability of
a higher class of traffic is higher than that of a lower one. Thus, the weight of the highest class of
traffic ( r 1 ), weight
r 1
, with the highest pACB
r ,id
should have the largest weight, so we set one to it,
i.e., weight
r 1
1 . Then, the weight of class r traffic can be determined by Eq. (13),
13
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
r 1 r 1
].
weight
r
weight
E [ pACB
(13)
Next, the dynamical preamble range of class r traffic, M CD
r
, can be formulated, as shown
in Eq. (14),
r
M CD M weight
r
, (14)
where M denotes the total number of preambles used for the data packet service.
For instance, assume that we have the average access class barring factor of class r traffic of
T
r
UE, E[ pACB ] , are shown below,
IP
E [ p1ACB ]=0.8,
2
E [ pACB ]=0.6,
CR
3
E [ pACB ]=0.3, and
4
E [ pACB ]=0.1 .
weight
4
0.063 .
ED
Finally, we can get the number of the available preamble contention domain class r traffic of
UE, M CD
r
, can be computed by,
1
M 1 54 ,
PT
M CD
2
M CD M 0.56 30 ,
3
M CD M 0.23 12 , and
CE
4
M CD M 0.063 3 .
AC
contention domains of different classes of traffic for the shared and partition modes. Clearly, in
the random access mechanism in LTE-A, the collision probability increases as the traffic load
increasing or as the preamble idle probability decreasing. Thus, the preamble idle probability,
RACH
Pidle , is adopted as the key factor for the mode selection of contention domain, in which DPSR
adopts the partition mode when the preamble idle probability is high (i.e., the preamble
contention loading is low). By partitioning the collision domains, the collision probabilities of
14
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
different classes of traffic can be reduced significantly. Conversely, for guaranteeing the data
rate for the highest class of traffic, DPSR adopts the shared mode when the preamble idle
probability low (i.e., the preamble contention loading is high). That is, when the randomly
generated value is less than PidleRACH , the partition mode is selected, as shown in Eq. (15),
T
For reducing the collision probabilities among different classes of traffic, the partition mode
r ,id r ,id
IP
is adopted as shown in Fig. 4. The lower bound, M CD,lower , and the upper bound , M CD, upper , of a
CR
r ,id
M CD,lower rand () of [ 0 , ( M M CD
r
-1) ] , (16)
r ,id
M CD,upper r
M CD r ,id
1 M CD,lower , (17)
US
where PidleRACH is the maximum number of preambles and M CDr is the number of the available
r , id
M CD RACH time slot
, upper
M
traffic priority 2
(i.e., M=54)
Contention domain for
traffic priority 3
r , id Contention domain for
M CD , lower traffic priority 4
PT
Subframe (1ms)
CE
Subframe : 0 1 2 3 4 5 6 7 8 9
Radio Physical Frame (10ms)
15
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
r ,id
where M CD,lower is set to 0 and M CD,upper
r ,id
is determined based on the idle probability of the RACH
channel preamble of the eNB ( PidleRACH )及the collision contention probability, Pcollision
r ,id
, of class r
T
RACH time slot
IP
Preamble logical channel
contention domain
Contention domain for
traffic priority 1
CR
M Available Contention domain for
Preamble traffic priority 2
(Logical Channels) Contention domain for
Contention Domain traffic priority 3
r , id
M CD
(i.e., M=54) , upper Contention domain for
traffic priority 4
US r
M CD
Subframe (1ms)
AN
r , id
M CD , lower
Subframe : 0 1 2 3 4 5 6 7 8 9
Radio Physical Frame (10ms)
M
In 3GPP [7], a data packet fails on the preamble contention, i.e., the UE fails to receive any
response from the eNB, and leads to a collision. The collided UE then executes the uniform
PT
distribution backoff mechanism to randomly determine the backoff time as well as to wait for
next contention on the random channel slot. In 3GPP [7], the backoff values of 16 possible
CE
backoff cases. Clearly, the backoff index 0 leads to the minimum backoff time, i.e., 0ms.
Conversely, the backoff index12 leads to the maximum backoff time. The backoff index
selection of 3GPP is independent to collision probability and differs to IEEE 802.11’s Truncated
AC
Binary Exponential (TBE) backoff mechanism. As a result, 3GPP suffers from the independence
of collision probability backoff, and thus easily yields extremely high collision while an UE has
a large number of re-transmissions.
Thus, Phase 3 proposes a Dynamic Initial Backoff (DIB) algorithm first to determine the
maximum backoff index ( BI rmax
,id
) for UEs according to the idle probability of the RACH channel
preamble of the eNB ( PidleRACH ), and then randomly selects the backoff index ( BI r ,id or h r ,id ). The
key idea is for avoiding further collisions the backoff time (i.e., backoff index) of a collided
16
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
traffic increases as the idle probability of the RACH channel preamble decreasing, and vice versa.
The DIB algorithm is depicted in detail below.
First, Phase 3 considers several key parameters: 1) the idle probability of the RACH channel
preamble of the eNB ( PidleRACH ), 2) the successful probability of the RACH channel preamble of the
eNB ( Psuccess ) and 3) the collision probability of UE’s random contention ( Pcollision
RACH r ,id
). Then, the
,id ) for class r traffic of an UE id is formulated by,
maximum backoff index ( BI rmax
BI max (1 Psuccess
RACH
) Pcollision
r ,id RACH
, if Pidle rand () 1,
r ,id
T
BI max (20)
BI max , , otherwise.
IP
where BI max denotes the maximum backoff index defined in 3GPP, i.e., BI max 15 , as shown in
CR
Table II.A high idle probability of preambles represents that the RACH channel contention rate
is low, and thus the backoff delay can be reduced certainly. Thus, in Eq. (20), the maximum
BI max (1 Psuccess ) Pcollision , as PidleRACH increasing.
backoff index ( BI max
US
r ,id ) decreases by using the formula RACH r ,id
RACH
Furthermore, BI rmax
,id increases up to
BImax or 15 , as Pidle decreasing.
AN
Then, Phase 3 randomly selects the backoff index ( BI r ,id or h r ,id ) from 0 to BI rmax
,id
for class r
traffic of an UE id ,
BIr ,id or hr ,id rand () of [0,BIrmax
,id
M
]. (21)
The corresponding backoff value, BOValue , defined by 3GPP, as shown in Table II, can be
ED
obtained by,
BOValue Mapping BIr ,id . (22)
PT
Finally, Phase 3 can justify the claim. That is, Phase 3 yields the smaller backoff as the idle
probability of the eNB increasing and the collision probability of an UE decreasing. On the other
CE
hand, Phase 3 yields the larger backoff as the idle probability of the eNB decreasing and the
collision probability of an UE increasing.
AC
In the proposed approach, some state information are determined at eNB (e.g., the preamble
r ,id
idle probability, dynamic preamble range of class r traffic M CD
r
, the lower bound M CD,lower , and
r ,id
the upper bound M CD,upper ) and some are determined at every UE (e.g., successful probability,
information down to all UEs by the Physical Broadcast CHannel (PBCH), the Physical
Downlink Shared CHannel (PDSCH), or the Physical Downlink Control CHannel (PDCCH).
17
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
First, for the case of using PBCH, the eNB state information can be carried by the Master
Information Block (MIB) that consists of a limited number of the most frequently transmitted
parameters essential for initial access to the cell. Second, for the case of using PDSCH, the eNB
state information can be broadcasted by the PDSCH. Third, for the case of using PDCCH, the
eNB state information can be carried as Downlink Control Information (DCI) to UEs.
T
This section proposes the mathematical analyses for the proposed the Adaptive Random
IP
Contention approach for LTE-A (ARC LTE-A). Extended from [26], the analyses for the
CR
proposed ARC LTE-A include: 1) the collision probability of preambles for RACH contentions
( Pc ), 2) the success probability ( Ps ), and 3) the average access delay ( TAccess ).
In the proposed approach, two modes of collision domains for different classes of traffic are
US
adopted in Phase 2 (DPSR). The lower and upper bounds are dynamically determined for
different UEs, and the dynamic computations of bounds result in difficult analyses and the
AN
computation complexity. Thus, in this section, we focus on the analysis of the shared mode.
Fig. 6 demonstrates the timeslots of the random access channel, in which the grey slot
M
denotes the RACH time slot used for preamble contentions and the blank slot denotes the time
slot for data transmission. We assume that the eNB sends a RACH time slot every RA rep time (in
ED
ms) and the total evaluation time is RATotal_time (in ms). Thus, the total number of slots for
contentions, RACont_slots (in slot), within the total evaluation time can be obtained,
PT
RA Total_time
RA Cont_slots = . (23)
RA rep
CE
AC
18
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
(in ms) can be formulated by Eq. (24),
UEireq
UEireq
IP
RA Cont_slot
M UE i e M M e M
i 1
req
CR
Pc , (24)
RA Cont_slot M
where M denotes the number of preambles for random contentions within a time slot
and UEreq
i
US
denotes the number of UE requests of contentions in time slot i . The number of
UEireq
AN
M
successful preambles can be formulated by UE e i
req
. The number of idle preambles can be
UEireq
M
formulated by M e .
M
4.1.2. Case 2. The case of multiple classes of traffic of the proposed approach (collision
probability)
ED
This paper considers multiple classes of traffic and dynamically differentiates their collision
domains, as detailed in Phase 2 (DPSR) and as demonstrated in Fig. 5, for minimizing collision
PT
r
probability and for increasing goodput. Based on the expected ACB factor E [ pACB ] , the key idea
r
is to determine the dynamic preamble range for each class of traffic, M CD .
CE
In the performance analyses, the number of success and idle preambles of each class of traffic
need to be analyzed preciously, respectively. First, for the analysis of the number of success
AC
preambles, the preamble ranges of different classes of traffic are divided as demonstrated in Fig.
7, and then forms rmax blocks, as shown in the RED DASH rectangles. Each block exhibits
different numbers of classes of traffic, e.g., in block 1, i.e., the bottom block, all classes of traffic
can use the preambles of block 1. On the other hand, in block 4, i.e., the top block, only the
highest class of traffic can use the preambles of block 1. Moreover, for each block, the lower and
upper preamble-bounds can be determined from the dynamic preamble range for each class of
r
traffic, M CD , by Eqs. (25)(26),
19
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
0 , if b=1,
i ,b
M lower i ,b1 (25)
M +1, otherwise.
upper
M1 , if b=4,
i ,b
M upper rmax b 1 (26)
M CD -1, otherwise.
Thus, in Fig. 7, the number of preambles within block b can be formulated by Eq. (27),
i ,b
M block M upper
i ,b
M lower
i ,b
1. (27)
i ,b 1 r 1
i ,b 1
For instance, for block 1, M lower 0 and M upper M CD 1.
T
IP
Mapping Contention Domain
CR
The (b+1)-th block of
i ,b
M block contention domain
US
, upper
i ,b
M block
AN
, lower
request
i ,1 i,2 i ,3 i,4
UE req UE req UE req UE req
Subframe (1ms)
ED
Fig. 7. The diagram of preamble logical channel for different classes of traffic.
PT
i ,b
Thus, based on the analyzed number of preambles of block b (namely M block ), the total
CE
average number of preambles for random contentions under multiple classes of traffic (namely
RA iblock
,b
) for block b can be determined by Eq. (28),
AC
rmax -b+1
UE ireq
,r
RA iblock
,b
r M block ,
i ,b
(28)
r 1 M CD
By applying Eq. (25) to Eq. (28), the collision probability for of the preambles under the multiple
classes of traffic of the proposed ARC LTE-A approach can be determined,
RA Cont_slot rmax RAi ,b
iblock
M ,b
iblock
M ,b
RAi ,b
i 1
b 1
M block RA block e
i ,b i ,b block
M block e
i ,b block
Pc . (29)
RA Cont_slot M
20
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
UE i
req, S [ n]
Ps i 1 n 1
RA Cont_slot , (30)
i
T
UE req
i 1
IP
where UEireq,S [ n] represents the success at the n -th contention (i.e., with n 1 collisions) of the
CR
i -th contention slot and PTmax represents . Note that UEireq,S [ n] can be approximately computed
by Eq. (31),
US
UEireq,S [ n] UEireq [ n] e
UEireq
M
. (31)
AN
4.2.2. Case 2. The case of multiple classes of traffic of the proposed approach (success
probability)
For the case of multiple classes of traffic, the success probability of UEs can be derived, as
M
shown in (32),
RA Cont_slot rmax PTmax
RA i ,b
ED
block, S [n]
Ps i 1 b 1 n 1
RA Cont_slot
, (32)
i 1
UE i
req
PT
where RA iblock,
,b
S [n]
denoting the average number of preambles of successful contentions under
CE
which the average delay time of the success at the n -th contention ( Tn ) needs to be computed.
Tn consists of four delays: 1) the slot time (1ms) of the RACH slot, 2) the average total backoff
time of a collision( Tw ), 3) the preamble state detection time ( Tdetect , i.e., 2ms), and 4) the eNB
21
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
Response time for a Random Access contention( WRAR , i.e., 5ms). Consequently, the average
UE i
req, S [ n] Tn
TAccess i 1 n 1
RA Cont_slot PTmax
, (34)
UE
i 1 n 1
i
req, S [ n]
where
T
Note that the performance analyses of this paper neglect the Hybrid Automatic Repeat request
IP
(HARQ) delay. Moreover, in [26] the fixed constant backoff time of 20ms (i.e., Tw 20ms ) is
CR
applied; however, the backoff time defined in 3GPP is a variable, as shown in Table II.
access delay) US
4.3.2. Case 2. The case of multiple classes of traffic of the proposed approach (average
For the case of multiple classes of traffic, Phase 3 randomly selects the backoff index
AN
( BI r ,id or h r ,id ) from 0 to BIrmax
,id
for class r traffic of an UE id . Since BIr ,id is a variable, it is
different to the fixed backoff time, e.g., Tw 20ms in [26]. Thus, the access delay of the case of
M
multiple classes of traffic needs to sum the backoff time of every collision ( Tw,var ), and then to
ED
determine the average Tw,var of all collisions, namely ( Tw,var ). The access delay of UEs of the
RA Cont_slot PTmax
UE i
req, S [ n] Tn ,var
TAccess i 1 n 1
RA Cont_slot PTmax , (36)
UE
CE
i
req, S [ n]
i 1 n 1
where
AC
In Eq. (37), Tn ,var represents the average delay time of the success at the n -th contention.
Next, the average backoff time of collision, Tw,var , needs to be determined by,
H max
Tw,var ph h RA rep , (38)
h H min
where
22
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
-- ph is the randomly selected probability of the backoff index h (i.e., h BI r ,id or h r ,id ) for
class r traffic of UE id ,
T
The minimum and maximum backoff indexes can be determined by,
IP
T WRAR 1
hmin detect , (39)
CR
RA rep
Tdetect WRAR BO value
hmax . (40)
US
RA rep
Moreover, the time for next contention at the h -th RACH time slot, namely th , can be
ED
formulated by,
th t0 ( h 1) RA rep , (42)
PT
where t0 denotes current time. The randomly selected probability of the backoff index h , ph , is
thus formulated by,
CE
BO value
RA rep , otherwise.
BO value
Consequently, the average backoff time of collision, Tw,var , can be determined as indicated in Eq.
23
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
5. Numerical Results
This section evaluates the proposed ARC approach and several related works, including
3GPP LTE/LTE-A specification (namely 3GPP) [7], Access Class barring factor (namely
p-persistent) [1][2][4][9] and the Prioritized Random Access with Dynamic Access barring
(namely PRADA)[14][15]. For clarity depictions, the functions of the compared approaches are
shown in Table II.
Table II. Approach function of several related works
Compared
T
Functions
approaches
1)p-persistent algorithm (p=1)
IP
3GPP
2)Uniform distribution backoff algorithm
1) pACB 0.9, pACB 0.7, pACB 0.3 and pACB 0.1
1 2 3 4
p-persistent_1
CR
2)Uniform distribution backoff algorithm
1) pACB 1.0, pACB 0.8, pACB 0.5 and pACB 0.2
1 2 3 4
p-persistent_2
2)Uniform distribution backoff algorithm
1) pACB 0.5, pACB 0.5, pACB 0.5 and pACB 0.5
1 2 3 4
p-persistent_3
PRADA
US
2)Uniform distribution backoff algorithm
1) UE4priority is set to a static pre-backoff time value as the RACH
congestion state increasing. The UE4priority has the lowest priority.
AN
Related parameters can be referred from [15].
2)Uniform distribution backoff algorithm
M
The performance metrics under different numbers of UEs are evaluated. The performance
metrics include: collision probability, success probability, average access delay, goodput and
dropping probability.
ED
In the simulation environment, for the random access control of the LTE-A Network, the
modules of the physical frame, RACH random access time slot function, the p-persistent
PT
algorithm and the uniform distribution are developed by C++ programming. The network model
and traffic model are detailed below.
CE
In the network model, for evaluating all compared approaches under more realistic
environment within an eNB, four classes of packet data traffic are considered. The physical
AC
frame consists of 10 subframes, and the available preambles of the subframe selected as the
RACH time slot can be used for random contentions among UEs. The backoff delay is computed
according to each evaluated approach.
In the traffic model, the Human and Machine types of transmissions are considered in the
same 4G LTE/LTE-A network [3][4][5][15]. The human type traffic adopts the Poisson
distribution [32][33][34] on-off packet bursts parameters: 1~5 and 1 .The machine type
traffic used for various priorities of emergency, alarming, and regular reporting adopts the Pareto
24
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
distribution to generate the on-off packet bursts with the parameters: 1) _on 1.0 and _ off 1.9 ,
and 1 for Machine RT (emergency), 2) _on 1.2 and _ off 1.6 , and 1 for Machine RT
(alarming) and 3) _on 1.9 and _ off 1.4 , and 1 for Machine RT (Smart Meter) [34].
Several useful parameters for simulations and analyses are listed in Table III.
T
Tablet Pad Smart Watch
Traffic types RT RT RT NRT
IP
Traffic classes 1 2 3 4
Packet size (in
1k 1k 2k 1k
byte)
CR
UEs percentage
5% 10% 10% 75%
[4][5]
Traffic
Pareto Pareto Poisson Pareto
Distribution
Arrival rate ( )
[32][33][34]
Departure rate ( )
[32][33][34]
--
--
US
--
--
1~5
1
--
--
AN
Alpha on ( _ on )
1.0 1.2 -- 1.9
[32][33][34]
Alpha off
( _ off ) 1.9 1.6 -- 1.4
M
[32][33][34]
Beta ( )
1 1 -- 1
[32][33][34]
ED
The parameters for simulations are shown in Table IV [4]. For each run of the simulation, the
simulation time is 5000 (ms).
PT
25
(95% confidential interval)
Simulation time 5000 ms
Number of eNBs 1
{20,50,100,500,1000,3000,5000,7000,10000,15
AC
Number of UEs
000…,50000,55000,60000 }
Radio physical frame duration 10ms
Subframe duration 1 ms
PRACH Config. index 6
RACH time slots per subframe 5 ms
Number of preambles 54
Max. number of preamble retransmissions 10
Ra-Response window size 5ms
Preamble transmission time 1 ms
Preamble processing time 2 ms
Backoff parameter value As shown in Table II
25
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
In Figs. 8(a)-(c), the mathematical analyses of the probabilities of collision, success, and
access delay are analyzed to validate the correctness of the proposed ARC approach. In Fig.8(a),
the collision probabilities of ARC of the simulation and analysis increase as the number of UEs
increasing. Furthermore, the collision probability of simulation is very close to that of analysis.
In Fig.8(b), the success probabilities of simulation and analysis are evaluated, in which both of
the success probabilities decrease as the number of UEs increasing. Importantly, the success
probability of simulation is very close to that of analysis. Similarly, as shown in Fig.8(c), the
access delays of simulation and analysis increase as the number of UEs increasing. Furthermore,
T
the access delay of simulation is very close to that of analysis. From above verifying the
IP
proposed ARC approach, the supreme results in Fig. 8(a)-(c) can justify the claim of the
CR
proposed ARC approach.
US
AN
M
26
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
(c) Average access delay
IP
Fig. 8. Simulation and Analysis for RACH performance.
In Fig. 9, the average of ACB factor, i.e., the pACB
r ,id
probability or the p value of the
CR
p-persistent, of these four classes of traffic under different numbers of UEs are evaluated, in
which all ACB factors decrease as the number of UEs increasing. In addition, the highest class of
US
traffic (i.e., r=1) yields the highest ACB factor. The reason is that the highest class of traffic
always has the highest ACB factor (i.e., r=1), i.e., it has the largest preamble range of the
AN
contention domain. As result, this can justify the highest class of traffic the proposed approach
can yield the highest successful contention probability. Conversely, the lowest one (i.e., r=4)
yields the lowest ACB factor, because the lowest one (i.e., r=4) always has the shortest preamble
M
range. The result meets the characteristic that the highest class of traffic always has the highest
contention probability for resulting in the highest goodput.
ED
PT
CE
AC
27
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
preambles, because the highest class of traffic always has the largest contention domain.
Conversely, the lowest one (i.e., r=4) yields the lowest available preambles.
60
r=3
30 r=4
T
20
IP
10
CR
0
20 100 1000 5000 10000 20000 30000 40000 50000 60000
US
Fig. 10. Available preambles within the contention domain of different numbers of UEs.
Figs. 11-17 evaluate the simulation results of all compared approaches under different
AN
numbers of UEs ranging from 20 to 60000. In Fig. 11(a), the collision probabilities of all
approaches increase as the number of UEs increasing. The proposed ARC approach yields the
lowest collision probability, but the others yield highest collision probabilities. In Fig. 11(b), the
M
success probabilities of all approaches (except for the proposed ARC approach) first increase up
to 0.4, and then decrease down to 0 when the number of UEs increases to 40000. Conversely, the
ED
proposed ARC approach yields the highest success probability (about 0.18) when the number of
UEs increases up to 60000. The reason is that the ARC approach adaptively tunes up the ACB
PT
factor ( pACB
r ,id
) based on the Sigmoid-based function, the differentiation of collision domains
among different classes of traffic and the dynamic backoff algorithm. Conversely, the other
CE
compared approaches use the static parameters setting for the contention probability and the
backoff algorithms. Without the dynamic parameters setting, they suffer from high collision
AC
probability, low successful probability and idle probability. In Fig. 11 (c), the idle probabilities
of all approaches decrease as the number of UEs increasing, because the contention traffic
increases significantly. The proposed ARC approach yields the highest idle probability.
28
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
IP
(a) Collision probability.
CR
US
AN
M
29
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
Fig. 12 evaluates the goodput of all approaches, in which all goodputs (except for ARC)
increase firstly, and then clearly decrease down to 0 when the number of UEs increases to 40000.
However, ARC yields a competitive goodput when the number of UEs is below 3000, but ARC
outperforms the others when the number of UEs is above 7000. Especially, ARC leads to a
significantly high goodput when the number of UEs is extremely high (e.g., from 10000 to
60000). The reason is that ARC adaptively tunes up the ACB factor ( pACB
r ,id
) according to the
Sigmoid-based function, the differentiation of collision domains among different classes of
traffic and the dynamic backoff algorithm. The collision probability is significantly decreased
T
and the successful probability is increased as the number of UEs increasing up to 6000. When
IP
the collision probability is decreased and the successful probability is increased, the goodput is
CR
thus significantly increased in a large number of UEs. However, the compared approaches adopt
static parameters setting for the contention probability and the backoff algorithms. They suffer
from high collision probability and yields low goodput as the number of UEs increasing.
US
AN
M
ED
PT
Fig. 12. Goodput of all compared approach under different numbers of UEs. (Simulation)
Fig. 13 evaluates the average access delay of all approaches under different numbers of UEs.
CE
The average access delays of all approaches increase as the numbers of UEs increasing. ARC
yields the lowest the average access delay, but the other approaches yield the longest one. The
AC
reason is that ARC can adaptively control the ACB factor and backoff value, and thus
significantly reduces the collision and access delay.
30
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
IP
Fig. 13. Average access delay (in ms) (simulation).
CR
From Figs. 14-17, the performance of the highest and the lowest classes of traffic are
evaluated. Fig. 13 evaluates the collision probabilities of two classes of traffic (i.e., the highest
US
high priority and the lowest priority) under different numbers of UEs. The collision probabilities
of all approaches increase as the number of UEs increasing. For the case of the highest class,
AN
ARC yields the lowest collision probability, but the others yield the highest collision probability.
On the other hand, for the case of the lowest class, ARC yields the worst collision probability.
The reason is that in ARC the highest class of traffic always has the highest priority to contend
M
the limited preambles, but the lowest class of traffic does not.
ED
PT
CE
AC
31
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
IP
(b) Collision probability of the lowest class of traffic
CR
Fig. 14. Collision probabilities of the highest and the lowest classes of traffic. (simulation)
Fig. 15 evaluates the average access delay of two classes of traffic under different numbers of
UEs. The average access delay of all approaches increase as the number of UEs increasing. For
US
the case of the highest class, ARC yields the lowest average access delay, but the others yield the
highest delay. For the case of the lowest class, ARC yields the worst access delay when the
AN
number of UEs is between 200 and 1000.
M
ED
PT
CE
32
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
IP
(b) Average access delay of the lowest class of traffic
CR
Fig. 15. Average access delays of the highest and the lowest classes of traffic. (in ms)
(simulation)
Fig. 16 evaluates the goodput of two classes of traffic under different numbers of UEs. The
US
goodput of all approaches increase and then decrease as the numbers of UEs increasing. For the
case of the highest class, ARC yields the highest goodput, but the others yield the lowest ones.
AN
For the case of the lowest class, p-persistent_3 yields the highest goodput and ARC leads to the
worst one. The reason is that ARC always guarantees the highest class of traffic with the highest
priotity to contend the limited preambles.
M
ED
PT
CE
AC
33
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
T
IP
(b) Goodput of the lowest class of traffic
CR
Fig. 16. Goodputs of the highest and the lowest classes of traffic.
(in kbyte) (simulation)
Fig. 17 evaluates the dropping probabilities of two classes of traffic under different numbers
US
of UEs. The dropping probabilities of all approaches increase as the number of UEs increasing.
For the case of the highest class, ARC significantly yields the lowest dropping probability, but
AN
PRADA and 3GPP yield the worst ones. For the case of the lowest class, ARC leads to the
highest dropping, but the others lead to the best dropping. Similarly, the reason is that the ARC
approach always guarantees the highest class of traffic having the highest priority.
M
Note that in Figs. 11-17 although PRADA [14][15] considers different classes of traffic, it
adopts the static parameters, as shown in Table III [14][15] and below. Especially, the PRA
ED
parameters of , , N reset , Treset , T1 , T2 , Textra are set to 18, 8, 32, 8s, 160ms, 4s+Unit[0,1s], 5s,
for all classes of traffic. Since PRADA and 3GPP adopts the static parameter setting for backoff,
PT
rather than adopts the dynamic or adaptive setting mechanism, PRADA and 3GPP yield very
competitive similar results.
CE
AC
T
IP
(b) Dropping probability of the lowest class of traffic
CR
Fig. 17. Dropping probabilities of the highest and the lowest classes of traffic. (simulation)
US
The transmissions of explosive bursts Big Data of Mobile Cloud Computing services require a
high speed cellular networking, e.g., 4G LTE/LTE-A. In the uplink, LTE-A adopts the
AN
out-of-band RACH random access time slots for UEs to contend with a limited number of
preambles. After successful contention, the data packets thus can be transmitted through the
Shared CHannel. However, LTE-A adopts the normal distribution with p -persistent algorithm
M
as the RACH random access mechanism, but suffers from not differentiating the collision
domains among different classes of traffic and increasing collision probability. Moreover, after
ED
collisions, LTE-A adopts the Uniform distribution backoff algorithm, but suffers from the
backoff time is independent to the collision state. As a result, LTE-A easily yields extremely
PT
high collision and long access time. This paper thus proposes the Adaptive Random Contention
approach for LTE-A (namely ARC LTE-A) that consists of three phases: 1) Sigmoid-based
CE
Access Class Barring algorithm, 2) Dynamic Preamble Selection Range (DPSR) algorithm, and
3) Dynamic Initial Backoff (DIB) algorithm. Numerical results demonstrate that the proposed
AC
ARC LTE-A outperforms the compared approaches in collision probability, goodput and access
delay. Additionally, the mathematical analytical model for ARC LTE-A is analyzed. The
analytical results are very close the simulation ones that can justify the claims.
Many aspects of analyzing the performance of the LTE-A relaying network require further
study. For instance, we are currently modeling an analytical model for the real-time cloud
services over B4G LTE-A multihop relaying networks.
35
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
References
[1] S.-Y. Lien, K.-C. Chen, and Y. Lin, “Toward Ubiquitous Massive Accesses in 3GPP
Machine-to-Machine Communications,” IEEE Communications Magazine, vol. 49, no. 4, pp. 66-74,
Apr. 2011.
[2] P. Jain, P. Hedman, and H. Zisimopoulos, “Machine Type Communications in 3GPP Systems,”
IEEE Communications Magazine, vol. 50, no. 11, pp. 28-35, Nov. 2012.
[3] R. Liu, et al., “M2M-oriented QoS categorization in cellular network,” 2011 7th International
Conference on Wireless Communications, Networking and Mobile Computing (WiCOM), pp. 1-5,
Sep. 2011.
[4] 3GPP TR 37.868 V 11.0.0 (2011-10): RAN Improvements for Machine-type Communications, Oct.
2011.
[5] J.-W. Kim, et al., “M2M Service Platforms: Survey, Issues, and Enabling Technologies,” IEEE
T
Communications Surveys & Tutorials, vol. 16, no. 1, pp. 61-76, Feb. 2014.
IP
[6] 3GPP TS 23.107 V 12.0.0 (2014-09): Quality of Service (QoS) concept and architecture, Sep. 2014.
[7] 3GPP TS 36.321 V12.5.0 (2015-03): Medium Access Control (MAC) Protocol Specification, Mar.
2015.
CR
[8] 3GPP TS 36.211 V12.5.0 (2015-03): Evolved Universal Terrestrial Radio Access (E-UTRA);
Physical channels and modulation, Mar. 2015.
[9] 3GPP TS 22.011 V13.1.0 (2014-09): Service accessibility, Sep. 2014.
[10] T. Taleb and A. Kunz, “Machine Type Communication in 3GPP Networks: Potential, Challenges,
US
and Solutions,” IEEE Communications Magazine, vol. 50, no. 3, pp. 178-184, Mar. 2012.
[11] A. Laya, L. Alonso, and J. Alonso-Zarate, “Is the Random Access Channel of LTE and LTE-A
Suitable for M2M Communication? A Survey of Alternatives,” IEEE Communications Surveys &
AN
Tutorials, vol. 16, no. 1, pp. 4-16, Dec. 2013.
[12] M. Hasan, E. Hossain, and D. Niyato, “Random Access for Machine-to-Machine Communication in
LTE-Advanced Networks: Issues and Approaches,” IEEE Communications Magazine, vol. 51, no. 6,
pp. 86-93, Jun. 2013.
[13] M.-Y. Cheng, G.-Y. Lin, and H.-Y. Wei, “Overload Control for Machine-Type-Communications in
M
LTE-Advanced System,” IEEE Communications Magazine, vol. 50, no. 6, pp. 38-45, Jun. 2012.
[14] T.-M. Lin, C.-H. Lee, J.-P. Cheng, and W.-T. Chen, “PRADA: Prioritized Random Access with
Dynamic Access Barring for MTC in 3GPP LTE-A Networks,” IEEE Transactions on Vehicular
ED
[16] D. Suyang, S.-M. Vahid, and V. W. S. Wong, “Dynamic access class barring for M2M.
communications in LTE networks,” 2013 IEEE Globecom Workshops (GC Workshops), pp.
4747-4752, Dec. 2013.
CE
[17] M.K. Giluka, A. Prasannakumar, N. Rajoria, and B.R. Tamma, “Adaptive RACH congestion
management to support M2M communication in 4G LTE networks,” 2013 IEEE International
Conference on Advanced Networks and Telecommunications Systems (ANTS), pp. 1-6, Dec. 2013.
[18] J. Li, H. Tian, L. Xu, and Y. Huang, “An optimized random access algorithm for MTC users over
AC
wireless networks,” 2013 IEEE 77th Vehicular Technology Conference (VTC Spring), pp. 1-5, Jun.
2013.
[19] Z. Jiang and X. Zhong, “Fast retrial and dynamic access control algorithm for LTE-advanced based
M2M network,” AICT 2012, The Eighth Advanced International Conference on Telecommunications,
pp. 24-28, May 2012.
[20] D.-T. Wiriaatmadja and K.-W. Choi, “Hybrid random access and data transmission protocol for
machine-to-machine communications in cellular networks,” IEEE Transactions on Wireless
Communications, vol. 14, no. 1, pp. 33-46, June 2014.
[21] S.-Y. Lien, T.-H. Liau, C.-Y. Kao, and K.-C. Chen, “Cooperative access class barring for
machine-to-machine communications,” IEEE Transactions on Wireless Communications, vol. 11, no.
1, pp. 27-32, Jan. 2012.
36
ACCEPTED MANUSCRIPT
Accept: Oct. 31, 2016; Revision: Oct. 15, 2016; Submission: May 23, 2016.
[22] Y.-H. Hsu, K. Wang, and Y.-C. Tseng, “Enhance cooperative access class barring and traffic
adaptive radio resource management for M2M communication over LTE-A,” 2013 Asia-Pacific
Signal and Information Processing Association Annual Summit and Conference (APSIPA), pp. 1-6,
Oct. 2013.
[23] X. Jian, Y. Jia, X. Zeng, and J. Yang, “A novel class-dependent back-off scheme for machine type
communication in LTE systems,” 2013 Wireless and Optical Communication Conference (WOCC),
pp. 135-140, May 2013.
[24] J.-B. Seo and V. C. M. Leung, “Design and analysis of backoff algorithms for random access
channels in UMTS-LTE and IEEE 802.16 systems,” IEEE Transactions on Vehicular Technology,
vol. 60, no. 8, pp. 3975-3989, Oct. 2011.
[25] W. Jiang, X. Wang, and T. Deng, “Performance analysis of a pre-backoff based random access
scheme for machine-type communications,” 2014 International Conference on Intelligent Green
T
Building and Smart Grid (IGBSG), pp. 1-4, Apr. 2014.
[26] C.-H. Wei, G. Bianchi, and R.-G. Cheng, “Modeling and analysis of random aAccess channels with
IP
bursty arrivals in OFDMA wireless networks,” IEEE Transactions on Wireless Communications, vol.
14, no. 4, pp. 1940-1953, Dec. 2014.
[27] C.-H. Wei, R.-G. Cheng, and S.-L. Tsao, ”Performance analysis of group paging for machine-type
CR
communications in LTE networks,” IEEE Transactions on Vehicular Technology, vol. 92, no. 7, pp.
3371-3382, Mar. 2013.
[28] R.-G. Cheng, J. Chen, D.-W. Chen, and C.-H. Wei, “Modeling and analysis of an extended access
US
barring scheme for machine-type communications in LTE networks,” IEEE Transactions on
Wireless Communications, vol. 14, no.6, pp. 2956 - 2968, Feb. 2015.
[29] D. Niyato, P. Wang, and D.-I. Kim, “Performance modeling and analysis of heterogeneous machine
type communications,” IEEE Transactions on Wireless Communications, vol. 13, no. 5, pp.
AN
2836-2849, Apr. 2014.
[30] C.-H. Wei, R.-G. Cheng, and S.-L. Tsao, “Modeling and estimation of one-shot random access for
finite-user multichannel slotted ALOHA systems,” IEEE Communications Letters, vol. 16, no. 8, pp.
1196-1199, June 2012.
[31] R.-G. Cheng, C.-H. Wei, S.-L. Tsao, and F.-C. Ren, “RACH collision probability for machine-type
M
communications,” 2012 IEEE Vehicular Technology Conference (VTC Spring), pp. 1-5, May. 2012.
[32] F. Capozzi, G. Piro, L.A. Grieco, G. Boggia, and P. Camarda “Downlink packet scheduling in LTE
cellular networks: key design issues and a survey,” IEEE Communications Surveys & Tutorials, vol.
ED
37