IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO.

2, APRIL 2018 905

Protocol Sequences With Carrier Sensing

for Wireless Sensor Networks
Yijin Zhang , Member, IEEE, Ming Zhang, Yuan-Hsun Lo, Member, IEEE, and Wing Shing Wong, Fellow, IEEE

Abstract—Protocol sequences are deterministic binary

sequences of a common period, which enjoy some special
Hamming cross-correlation property by design. In contrast to
random or contention-based medium access control schemes, a
protocol sequence-based scheme can serve to provide at least a
certain number of contention-free packet transmissions within
a bounded delay for each asynchronous user in a feedback-free
multiple access system. However, all protocol sequence-based
schemes in the literature require that all sequence entries are
mapped to slots with the same time duration, which produces a
relatively low channel utilization. To overcome this inefficiency
that is undesirable in delay-constrained wireless sensor networks,
building on the idea of combining sequence-based access and Fig. 1. Example of traditional protocol sequence-based schemes for a three-
carrier sensing, this paper proposes a new protocol sequence- user multiple access. For simplicity, the propagation delays are not depicted
based scheme, called the PS-CS. We derive the theoretical in the illustration.
average throughput, average access delay, worst-case delay, and
average energy consumption of the PS-CS. It is shown that
the PS-CS produces average throughput close to the optimal
capacity of p-persistent carrier sense multiple access (CSMA), control (MAC) scheme for asynchronous users. Simply put,
and enjoys smaller access delay than the optimal p-persistent protocol sequences are a set of deterministic and periodic
CSMA. In addition, we study the energy-delay tradeoff, (0, 1)-binary sequences with a common period that enjoy some
impact of carrier sensing fault and channel error, and how to
modify the PS-CS to support real-time downlink for feedback special properties of Hamming cross-correlations by design.
control. Each user is assigned a protocol sequence that is unique from
those of its neighbors by a preallocation [1]–[16] or a dynamic
Index Terms—Access delay, carrier sensing, protocol
sequences, wireless sensor networks (WSNs). allocation based on the geographic positions [17]–[19]. Then,
each user is allowed to send a packet within the bound-
aries of a time slot if and only if the sequence value at that
I. I NTRODUCTION slot equals one. This manner can be viewed as a determin-
A. Background istic backoff that does not require feedback information and
random number generation. Due to lack of synchronization,
ASSEY and Mathys [1] proposed the idea of
M using protocol sequences to define a medium access
the users may start their protocol sequences at different time
slots. One example for a three-user multiple access is shown
in Fig. 1.
Manuscript received June 18, 2017; revised November 3, 2017 and January
3, 2018; accepted February 8, 2018. Date of publication February 19, 2018; The worst-case mutual interference among asynchronous
date of current version April 10, 2018. This work was supported in part by users in a protocol sequence-based scheme can be determined
the National Natural Science Foundation of China under Grant 61301107 or upper-bounded by Hamming cross-correlation functions of
and Grant 11601454, in part by the Open Research Fund of National
Mobile Communications Research Laboratory, Southeast University under protocol sequences, although lack of synchronization invari-
Grant 2017D09, in part by the Grant from the Innovation and Technology ably leads to random pattern of mutual interference. As such, a
Fund of the Hong Kong Special Administrative Region of China under protocol sequence-based scheme can be designed to guarantee
Project ITS/066/17FP, and in part by the Natural Science Foundation of
Fujian Province of China under Grant 2016J05021. (Corresponding author: that each active user, not requiring any feedback information,
Yijin Zhang.) has at least a certain number of contention-free packet trans-
Y. Zhang is with the School of Electronic and Optical Engineering, Nanjing missions within a bounded delay no matter what the relative
University of Science and Technology, Nanjing 210094, China, and also
with the National Mobile Communications Research Laboratory, Southeast offsets among the asynchronous users are [1]–[17]. By con-
University, Nanjing 210018, China (e-mail: [email protected]). trast, ALOHA or carrier sense multiple access (CSMA) type
M. Zhang is with the School of Electronic and Optical Engineering, Nanjing schemes require feedback from the receiver for contention res-
University of Science and Technology, Nanjing 210094, China (e-mail:
[email protected]). olution, and cannot provide guaranteed performance due to
Y.-H. Lo is with the School of Mathematical Sciences, Xiamen University, their probabilistic nature. Hence, protocol sequences are more
Xiamen 361005, China (e-mail: [email protected]). suitable for wireless sensor networks (WSNs) with limited
W. S. Wong is with the Department of Information Engineering, Chinese
University of Hong Kong, Hong Kong (e-mail: [email protected]). computing capability and energy resource, especially when
Digital Object Identifier 10.1109/JIOT.2018.2807403 feedback is difficult to setup.
2327-4662 c 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
906 IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO. 2, APRIL 2018

B. Motivation Since that CSMA type algorithms have been the founda-
In spite of the aforementioned nice features, we note that tions of MAC technology in wireless networks, and many
all conventional protocol sequence-based schemes in [1]–[17] sophisticated CSMA versions can be closely approximated
require that all sequence entries are mapped to slots with by p-persistent CSMA if the average backoff interval is
the same time duration, as illustrated in Fig. 1, for enabling the same [20], the comparative study against the optimal
Hamming cross-correlation functions to accurately reflect the p-persistent CSMA indicates that the PS-CS is a promising
mutual interference patterns. However, such a correspondence MAC candidate for delay-constrained WSNs.
leads to a relatively low channel utilization. Hence, it has been The rest of this paper is organized as follows. In Section II,
shown in [1], [3], [5], and [12] that a protocol sequence-based we summarize the related work on protocol sequences. In
scheme can only achieve the same maximum average through- Section III, after introducing an application scenario and the
put (averaged over all possible relative offsets among users) as considered system model, we describe the proposed PS-CS
in a slotted ALOHA scheme. Further, there has been a com- scheme. In Section IV, we derive the theoretical performance
mon belief that all previously known protocol sequence-based metrics of the PS-CS scheme. Numerical studies are pro-
schemes cannot outperform CSMA type schemes in terms of vided in Section V. We in Section VI discuss the impact
average throughput and average access delay, since that CSMA of unreliable carrier sensing and imperfect wireless channel,
type schemes utilize the channel more efficiently than slot- and in Section VII provide a modification to support real-
ted ALOHA due to the employment of carrier sensing. Such time downlink for feedback control. We conclude this paper
weakness is undesirable in delay-constrained WSNs, for which in Section VIII.
information must be successfully transmitted within a short
time duration for disaster prevention, and small average delay
II. R ELATED W ORK
or worst-case delay is an essential prerequisite to energy sav-
ing. This observation leads us to explore a new type of protocol There has been a body of work in designing and analyzing
sequence-based scheme that enhances the channel utilization traditional protocol sequence-based schemes under different
by the aid of carrier sensing. targeted objectives [1]–[17]. Let the time offsets among the
users be measured in time slot duration units. Let M be the
number of sequences in a sequence set.
The first known protocol sequence set was constructed in [1]
C. Contribution by a decimation decoding technique for characterizing the
This paper proposes a modified scheme with carrier sens- zero-error capacity of the collision channel without feedback.
ing, referred to here as the PS-CS scheme. The key idea of the Such sequences possess the shift-invariant (SI) property, that
PS-CS scheme is to allow each user to perform carrier sens- is, the Hamming cross-correlation function of an arbitrary
ing when a zero is read from the assigned sequence. It will nonempty subset is independent of cyclic shifts of sequences.
be shown that the PS-CS scheme can significantly improve More general constructions of SI sequences can be found
channel utilization in comparison to conventional protocol in [2]–[4]. It was proved in [5] that SI sequences are the only
sequence-based schemes. It is worth pointing out that although solution to achieve the maximum worst-case throughput, and
the benefit comes at the expense of carrier sensing, the PS-CS it was proved in [11] that SI sequences can be used to ensure
scheme still does not need any feedback from the receiver as constant delivery delay for each user. However, an inherent
in conventional approaches. Moreover, under reliable carrier drawback of SI sequences is that the period grows exponen-
sensing, the PS-CS scheme complies with some basic char- tially in the number of users [3], which may lead to individual
acteristics of traditional protocol sequence-based algorithms, short-term starvation problem.
and hence, is still able to provide guaranteed performance. To In view of this issue, some attempts have been made to seek
our best knowledge, this paper is a first attempt to combine short sequences with acceptable throughput. In general, the
sequence-based MAC and carrier sensing. constructed sequences possess the property that the Hamming
To gain a clear insight into the characteristics of the cross-correlation function of an arbitrary subset with two
proposed PS-CS, closed-form formulas are derived for eval- sequences has some small values no matter how we cyclically
uating average throughput, average delay, the worst-case shift these two sequences. For example, a class of protocol
delay, and average energy consumption. It will be con- sequences building on the idea of prime sequences, called
firmed by numerical studies that, the PS-CS can pro- wobbling sequences, was proposed in [7]. Such sequences
duce average throughput close to the optimal capacity of have a period equal to M 4 , and can produce the worst-case
p-persistent CSMA; moreover, it provides significantly better system throughput 0.25. Another class of protocol sequences
delay performance than both traditional protocol sequence- constructed by means of the Chinese remainder theorem,
based scheme and p-persistent CSMA. In addition, it will be called CRT sequences, was presented in [8]. They enjoy
shown that the PS-CS can have better energy-delay tradeoff the worst-case system throughput 0.25, while the period is
in some scenarios even when energy and delay are considered only O(M 2+ ). In addition, protocol sequences with adjustable
equally important, can produce smaller average delay even sequence weights were constructed in [12] by systematically
when some carrier sensing fault and channel error occur, and adding mark chips into some families of prime sequences, and
can be modified to support real-time transmissions from the generalized prime sequences with a more flexible parameters
receiver for feedback control. setting than CRT sequences were constructed in [17].
ZHANG et al.: PROTOCOL SEQUENCES WITH CARRIER SENSING FOR WSNs 907

The concept of user-irrepressible (UI) sequences was The propagation delay between any of the users or the
proposed in [7] and [10]. Such sequences are able to allow receiver is at most tprop seconds. Each user can perform reli-
each user to send out at least one contention-free packet in able carrier sensing to detect a busy medium, and its detection
each sequence period for all possible integer-valued relative time is tcs (tcs > tprop ) seconds. Assume that the transmission
offsets. The design goal of them is to minimize the sequence time of each packet is fixed to tpkt seconds. The channel is
period that determines the worst-case delay. SI sequences divided into time slots of equal duration tslot (tslot ≥ tcs ) sec-
are the shortest known UI sequences for M ≤ 6, CRTp onds, such that tslot = (tpkt +tprop )/l for a fixed positive integer
sequences [10] are the shortest for all prime M ≥ 7, and CRT- l. In other words, each packet exactly fits in l−η channel slots.
UI [9] sequences are the shortest for all nonprime M ≥ 8. Here, η := tprop /tslot is defined as the ratio of the maximum
CRTp and CRT-UI sequences can be seen as two variations of propagation delay to the time slot.
CRT sequences [8]. The users may start their communication session, i.e.,
Protocol sequences are also closely related with other become active from inactive at different time slots. This behav-
sequences studied in the literature, under the name of ior incurs relative offsets between the users, which can be any
cyclically permutable codes [13]–[15], conflict-avoiding arbitrary values that are integer multiples of a slot duration,
codes [21]–[23], and topological transparent schedul- but fixed throughout a transmission session.
ing [24]–[26]. Other classes of sequence-based MAC schemes Remark 1: For a traditional protocol sequence-based
include the applications of Reed-Solomon codes [27] and scheme, a time slot needs to accommodate the transmission
Gold sequences [28]. time of a packet and the propagation delay, and so the time
To the authors’ best knowledge, there is no known study on duration of a slot is required to be tslot = tpkt + tprop .
combining sequence-based MAC and carrier sensing.
C. Protocol Description
III. P ROPOSED PS-CS S CHEME Before accessing the channel by a PS-CS scheme, each user
In the first part of this section, we introduce an application is assigned a distinct protocol sequence. The operations of
scenario and the corresponding system model, while in the transmissions and carrier sensing can only begin at the begin-
rest of this section we describe the proposed PS-CS scheme. ning of a channel slot. The algorithm behaves in two operating
phases, namely, initial phase and normal phase.
A. Application Scenario 1) Initial Phase: When an inactive user wants to become
active, this user performs carrier sensing at the beginning of
Consider a WSN in which multiple users observe the same
a channel slot. If the channel is found to be busy, the user
area of interest. Each user attempts to transmit the latest gen-
keeps silent until the end of the current slot, and still stays
erated packet to a common receive. The common receiver
at the initial phase at the next slot. If the channel is found to
transmits the aggregated data to an access point (AP) if it
be idle, the user enters into the normal phase at the next slot.
has collected a certain number of packets from different users
Such a setup ensures that this user does not begin a packet
or due to the expiry of a strict delivery deadline.
transmission before the completion of ongoing transmissions.
The PS-CS scheme is used for channel access of the users,
In this phase, the user is forbidden to read sequence values.
while an infrastructure-based scheme is used for transmissions
between the receiver and the AP. Under this scenario, higher 2) Normal Phase: Then this user executes the following
average throughput, smaller average access delay, and smaller steps by sequentially looking up the entries of the assigned
worst-case access delay between the users and receiver would protocol sequence, and repeats the sequence periodically.
• If the current sequence value is one, then the user imme-
enable more observation from more users to be quickly deliv-
ered to the AP. This is desirable for fault-tolerance and timely diately sends out a packet. Carrier sensing is not required
detection purposes. It should be noted that there is no need before transmission.
• If the current sequence value is zero, the user performs
to consider queuing delay, since that only the latest generated
packet is contained in the queue of each user. carrier sensing at the beginning of the current channel
This scenario can be further generalized to a cluster-based slot. This operation is used to detect whether one or
WSN deployed over a larger geographical area in which users more users commence transmissions at the beginning of
in the same cluster observe the same area of interest and the this slot. If the channel is found to be idle, the user is
PS-CS scheme is used for channel access within clusters. required to keep silent for one slot (including the carrier
sensing time); and otherwise the user is required to keep
silent for l slots (including the carrier sensing time). The
B. System Model user is forbidden to transmit at the current slot in this
There are K users communicating with a common receiver, case.
which is within the transmission range of each user. We • After the transmission or silence, the user reads a new
assume that the carrier sensing range is at least twice than sequence value and repeats the above operations if it
the transmission range, as shown by real experiments [30], so still has a packet to send. Otherwise, it ends its current
that there are no hidden terminals. The system is limited by transmission session and becomes inactive.
user interference, and a packet can be correctly received if and The PS-CS algorithm is illustrated schematically in Fig. 2.
only if it does not overlap with another packet on the channel. One sees this algorithm with carrier sensing can be reduced to
908 IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO. 2, APRIL 2018

Fig. 3. Example of the PS-CS access for a three-user multiple access. For
simplicity, the propagation delays are not depicted in the illustration.

one and increase the sensing time of the users that read
zero.
Fig. 2. PS-CS algorithm. Benefiting from these features, the PSCS can provide signif-
icantly better throughput and delay performance than tradi-
tional cases, which will be shown in subsequent sections.

a traditional protocol sequence-based scheme [1]–[17] if the

carrier sensing operations (in the dotted boxes) are removed
IV. P ERFORMANCE A NALYSIS
and l is set to be one.
From the algorithm description, we summarize the fol- As described in Section III-C, in a PS-CS scheme, the
lowing distinctive features of the PS-CS scheme. Some of channel status corresponding to one sequence entry can be
them comply with the basic characteristics of traditional classified as idle or busy (successful transmission or con-
protocol sequence-based algorithms, and some of them ben- tention). The status is completely determined by the values of
efit from carrier sensing. Hence, the PS-CS scheme can corresponding entries of the assigned protocol sequences as in
be viewed as a solution lying between protocol sequences traditional cases. Hence, we can analyze the performance of a
and CSMA. PS-CS scheme by examining the assigned protocol sequence
1) Each active user is allowed to look up a new sequence set with random relative offsets. We start this section with
entry only after the completion of ongoing transmis- some basic concepts of protocol sequences [1]–[17].
sions or after every idle channel slot. This ensures that
each active user begins to look up a new sequence entry
A. Basics of Protocol Sequences
simultaneously.
2) As shown in Fig. 3, one major difference from tradi- Let si := [si (0) si (1) · · · si (L − 1)], be the binary proto-
tional protocol sequence-based schemes is that the time col sequence assigned to user i for i = 1, 2, . . . , K, where L
duration occupied by one sequence entry is not constant, is the common sequence period of all K sequences. We can
but varies over time. It can be one channel slot if the alternatively represent si by the characteristic set Ii in ZL ,
channel status corresponding to this sequence entry is which is defined as the set of all positions where si (n) = 1,
idle, i.e., all users read the sequence value zero; or l for n = 0, 1, . . . , L − 1. Define the Hamming weight of si as
channel slots if the channel status corresponding to this the number of ones in si . The duty factor of si is the fraction
sequence entry is busy, i.e., at least one user reads one. of ones in si and is denoted by fi .
Such a manner can be viewed as a deterministic back- Let τi be the relative offset of si . The cyclic shift of si by
off that depends on the current sequence value and the τi is defined as si(τi ) := [si (τi ) si (1 + τi ) · · · si (L − 1 + τi )],
channel status, but does not require any feedback from where the addition is taken modulo L.
(τ ) (τ )
the receiver. Let N(b1 , . . . , bK |s1 1 , . . . , sK K ) denote the number of
3) Each active user performs carrier sensing only when a integer solutions, n, 0 ≤ n < L, such that sj (n+τj ) = bj for all
zero is read from the sequence. It should be noted that j. In particular, let N(hi = 1|s(τ 1) (τK )
1 , . . . , sK ) denote the number
performing carrier sensing when a one is read is not of integers solutions, n, 0 ≤ n < L, such that si (n+τi ) = 1 and
(τ ) (τ )
needed in our algorithm, since that such an operation sj (n + τj ) = 0 for all j = i; and N(0|s1 1 , . . . , sK K ) denote
has no positive impact on further actions of all users, the number of integers solutions, n, 0 ≤ n < L, such that
but would delay the transmissions of the users that read sj (n + τj ) = 0 for all j.
ZHANG et al.: PROTOCOL SEQUENCES WITH CARRIER SENSING FOR WSNs 909

The Hamming cross-correlation between two sequences si Proof: Suppose that the relative offsets of si is τi , for i =
and sj , denoted by Hi,j (τi , τj ), is defined as 1, 2, . . . , K. We note that (τ1 , τ2 , . . . , τK ) is an element in the

L−1 set (ZL )K , where ZL = {0, 1, . . . , L − 1}. Hence, there are LK
 
Hi,j τi , τj := si (n + τi )sj (n + τj ). possible combinations, and the normalized average throughput
n=0 of user i, Ri , needs to be computed over these LK instances. As
It counts the number of time-overlapping ones between si and the sequence period varies over different given relative offsets,
sj in a sequence period when si and sj have relative offsets τi we obtain Ri as the following:
and τj , respectively. L−1 L−1  
(τ1 ) (τK )
Example 1: We illustrate the aforementioned definitions τ1 =0 · · · τK =0 (l − η)N h i = 1|s 1 , . . . , s K
Ri = L−1 L−1  
and notation by the following three protocol sequences with (τ1 ) (τ )
lL · L − τ1 =0 · · · τK =0 (l − 1)N 0|s1 , . . . , sK K
K
common period 8, Hamming weight 4 and duty factor 1/2:   
s1 = [11110000]; s2 = [11001100]; s3 = [10101010]. (l − η)LN(1|si ) K j=1,j=i N 0|sj
=   
Their characteristic sets are I1 = {0, 1, 2, 3}, I2 = {0, 1, 4, 5}, lL · LK − (l − 1)L K j=1 N 0|sj

and I3 = {0, 2, 4, 6} in Z8 . A shifted version of these three l(l − η)fi K j=1,j=i (1 − fj )
sequences can be found as =  .
(1) (3) (2)
l − (l − 1) K j=1 (1 − fj )
s1 = [11100001], s2 = [01100110], s3 = [10101010].
(1) (3) (2)
The second equality of the above equation is due to the
Then, one can check that N(h2 = 1|s1 , s2 , s3 ) = 2, following result shown in [3]:
(1) (3) (2)
N(0|s1 , s2 , s3 ) = 1 and H1,3 (1, 2) = 2.

L−1 
L−1  
K
In the following performance analysis, as in [1]–[5], (τ ) (τ )
... N b1 , . . . , bK |s1 1 , . . . , sK K = L N(bi |si ).
[7]–[12], [16], [17], [20], and [29], we assume that each user
τ1 =0 τK =0 i=1
always has some data to send. This is a fundamental assump-
tion to investigate the performance limit as the offered load
increases and can be applied to analyze WSNs with burst We see from Theorem 1 that the average throughput of any
traffic. user only depends on the chosen duty factors and the packet
length, but is independent of the Hamming cross-correlation
B. Average Throughput property of the adopted protocol sequence set.
Massey and Mathys in their work [1] derived the average For the symmetric case that each sequence has the same
throughput of a traditional protocol sequence-based scheme. duty factor f in a PS-CS scheme, the normalized average
Since that different sequence entries may be mapped to dif- system throughput can be calculated as
ferent number of channel slots in the PS-CS scheme, the
K(l − η)f (1 − f )K−1
throughput analysis in [1] is unapplicable here. Rsym = . (3)
The normalized throughput of user i in the PS-CS for l − (l − 1)(1 − f )K
a given set of relative offsets {τ1 , . . . , τK }, denoted by Under the objective of average throughput maximization for
Ri (τ1 , τ2 , . . . , τK ), can be defined as the fraction of channel PS-CS, we are interested in finding the optimal symmetric
slots in which only user i is transmitting, that is, duty factor as a function of K and l. For l = 1, obviously
  fopt = 1/K. For l > 1, interested readers can refer to [20]
(τ ) (τ )
(l − η)N hi = 1|s1 1 , . . . , sK K and [29] for a detailed discussion, since the throughput of the
Ri (τ1 , τ2 , . . . , τK ) =  . (1)
(τ ) (τ ) p-persistent CSMA with the fixed packet length is prescribed
lL − (l − 1)N 0|s1 1 , . . . , sK K
by the same mathematical expression in (3), except that the
Note that the denominator in (1) denotes the time duration transmitting probability p replaces the duty factor f there. It
of a sequence period, and varies over different given relative was shown in [29] that the following approximation for the
offsets. By considering all possible combinations of relative optimal p holds for l > 1 and Kp  1:
offsets, we obtain the following average individual through-

put. It should be noted that the throughput analysis of the 1 + 2(l − 1) K−1
K −1
∼
popt = . (4)
p-persistent CSMA [20], [29] shares some similarities with our (K − 1)(l − 1)
analysis, since that the idea of using PS-CS can be described
Therefore, we obtain
as a derandomization of the p-persistent CSMA.

Theorem 1: Consider a PS-CS scheme with a protocol 1 + 2(l − 1) K−1

K −1
sequence set {s1 , s2 , . . . , sK } with duty factors f1 , f2 , . . . , fK , fopt ∼
= (5)
respectively. If the relative offset of each user is uniformly and (K − 1)(l − 1)
independently distributed between 0 and L −1, the normalized for l > 1 and Kp  1.
average throughput of user i is equal to
 C. Average Access Delay
(l − η)fi K j=1,j=i (1 − fj )
 (2) According to the application scenario in Section III-A, we
l − (l − 1) K j=1 (1 − fj ) define the medium access delay of a user, as the waiting time
for i = 1, 2, . . . , K. (in units of tslot ) until this user can transmit a contention-free
910 IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO. 2, APRIL 2018

packet starting at the beginning of a time duration that a ran- that each sequence entry with busy channel status is mapped
domly selected protocol sequence entry is mapped to. From to l channel slots and each sequence entry with idle chan-
the protocol description in Section III-C, we believe that the nel status is mapped to one channel slot, to evaluate
average delay performance of a PS-CS scheme highly depends the expected access delay of user a, we need to investigate
on the sequence structure and Hamming cross-correlation the number of sequence entries user a has looked up until the
property, as in traditional cases [16], [17]. completion of its first contention-free transmission, and the
CRT-UI sequences [9] with the period r(2K − 1) are the number of sequence entries with idle channel status user a
shortest known UI sequences if the sequence set size is a has experienced until the completion of its first contention-free
nonprime not smaller than 8 [10]. Here, r denotes the small- transmission.
est prime not smaller than K. To make the average throughput Depending on the assigned protocol sequence and the rel-
of the PS-CS close to its maximum value, we in the PS-CS ative offset τa∗ , there exist K transmissions at the xk (τa∗ )th
adopt a slightly modified version of CRT-UI sequences, called sequence entry for k = 1, 2, . . . , K, such that
MCRT-UI sequences.      
0 ≤ x1 τa∗ < x2 τa∗ . . . < xK τa∗ ≤ L − 1.
For l = 1, we set q = 2K − 1. For l ≥ 2, we set q to be the
integer closest to (K/[rfopt ]) satisfying the condition that q is By the property that the Hamming cross-correlation between
relatively prime to r and not smaller than 2K −1. The mapping any two MCRT-UI sequences is at most one, we know user
f : Zrq → Zr ⊕ Zq defined by f (a) := (a mod r, a mod q) is a must have one contention-free transmission among these K
a bijection from Zrq to Zr ⊕ Zq as r and q are relatively transmissions. Assume that user a successfully transmits with-
prime, and preserves addition and multiplication by integers. out contention for the first time at the Ya th sequence entry
Let I1 := {(y, 0) ∈ Zr ⊕ Zq : y = 0, 1, . . . , K − 1} and since the 0th entry. The possible values of Ya are xk (τa∗ ),
let Ii := {(iy mod r, y) ∈ Zr ⊕ Zq : y = 0, 1, . . . , K − 1} k = 1, 2, . . . , K. See Fig. 4 for illustration.
for i = 2, . . . , K. We obtain Ii by taking the inverse image As Ya is dependent on the relative offsets of all users, by
f −1 (Ii ), for all i. These K sequences have period L = qr. regarding the relative offsets as uniformly distributed random
The following two key features of CRT-UI sequences hold variables between 0 and L − 1, we can treat Ya as a ran-
for MCRT-UI sequences although the latter may have a larger dom variable. Denote Pr(Ya ≤ xk (τa∗ )|τa = τa∗ ) by ϕk for
sequence period: 1) each sequence has a Hamming weight k = 1, 2, . . . , K. Let Ej be the event that the transmission by
K and 2) the Hamming cross-correlation between any two user a at the xj (τa∗ )th sequence entry is contention-free. We
sequences is at most one for any relative offsets. Consequently, can apply the inclusion-exclusion principle to Ej ’s in order to
MCRT-UI sequences also have the UI property. evaluate ϕk
Example 2: Given K = 6, q = 15, the MCRT-UI con-      
struction produces six sequences of period 105 and Hamming ϕk = Pr Ej1 − Pr Ej1 ∩ Ej2 + · · ·
1≤j1 ≤k 1≤j1 <j2 ≤k
weight 6. The characteristic sets are   
+ (−1) k+1
Pr Ej1 ∩ . . . ∩ Ejk . (6)
I1 = {0, 15, 30, 45, 60, 75} I2 = {0, 16, 32, 48, 64, 80}
1≤j1 <...<jk ≤k
I3 = {0, 19, 31, 50, 62, 93} I4 = {0, 20, 33, 46, 79, 92}
Due to the property that the Hamming cross-correlation
I5 = {0, 10, 27, 44, 71, 88} I6 = {0, 11, 29, 40, 58, 87}. between any two MCRT-UI sequences is at most one, it
In the following, we will derive an approximated formula follows that the probability an arbitrary user b, b = a, will con-
for the average delay of the PS-CS. This paper can be viewed tend with user a at a sequence entry indexed by x1 (τa∗ ), x2 (τa∗ ),
as an extension of [16], which derived the average delay of . . ., or xm (τa∗ ) is given by Pa,b,m = (mK/L). If we apply the
the traditional CRT-UI protocol sequence-based scheme. simplifying approximation that the probabilities Pa,u,m for any
With no loss of generality, we focus on user a, and denote user u, u = a, are independent, then
its access delay by Da . Assume the assigned sequence has  
  mK K−1
the characteristic set Ia = {x1 , x2 , . . . , xK } with x1 < x2 < Pr Ej1 ∩ . . . ∩ Ejm = 1 − . (7)
L
. . . < xK . We define δ0  [x1 − xK mod L], δj  [xj+1 − xj
mod L] for j = 1, . . . , K − 1, and δ  (δ0 , δ1 , . . . , δK−1 ). For By substituting (7) into (6), we obtain
d = 0, 1, . . . , K − 1, we further define k   
k mK K−1

K−1 ϕk = (−1)m+1 1− (8)
m L
Cδ (d)  δj δjd m=1
j=0 for k = 1, 2, . . . , K. In particular, we have ϕK = 1
where  denotes the subtraction modulo K. that conforms with the UI property of MCRT-UI sequences.
Suppose that the relative offset of user a is fixed at τa∗ , i.e., Furthermore, obviously
τa = τa∗ . Without loss of generality, we measure the access    
Pr Ya = xk τa∗ |τa = τa∗ = ϕk − ϕk−1
delay of user a, starting from the beginning of a time duration
that is mapped to by a randomly selected entry in user a’s for k = 1, 2, . . . , K. We adopt the convention that ϕ0  0.
sequence. This selected entry is labeled with index 0. Let Aa be the number of sequence entries with idle chan-
Recall that the channel status corresponding to a sequence nel status that user a has experienced starting from the 0th
entry can be classified as idle or busy in the PS-CS. Since sequence entry to the Ya th sequence entry. Let E[Aa |Ya =
ZHANG et al.: PROTOCOL SEQUENCES WITH CARRIER SENSING FOR WSNs 911

Fig. 4. One scenario for calculating the access delay of user a.

xk (τa∗ ), τa = τa∗ ] be the conditional expected value of Aa . By Summing E[Da |τa ] over all possible τa uniformly dis-
the definition of Ya , we know user a’s transmissions from the tributed between 0 and L − 1, we obtain
0th sequence entry to the (Ya − 1)th sequence entry all suffer
1 
L−1
from collisions if Ya > 0. It is easy to see that some other users E[Da ] = E[Da |τa ]
must be involved in these collisions. This implies that there L
τa =0
is a relative offset restriction on each involved user. However,  

K−1
to simplify our analysis, we do not consider this restriction ≈ (l − (l − 1)ϕ1 )E[Ya ] + ϕ1 (l − 1) KϕK − ϕk
and consider the probability users will keep silent or not at k=1
a given slot to be independent of each other. By summing  
l − (l − 1)ϕ1 Cδ (0) + L 
K−1
over all possible relative offsets and exchanging the order of = + (1 − ϕk )Cδ (k)
summation and multiplication, it follows that: L 2
k=1
 

K−1
   + ϕ1 (l − 1) K − ϕk .
E Aa |Ya = xk τa∗ , τa = τa∗ (9)
k=1

L−1 
L−1  (τa∗ )
L−1 xk
1 The equality in (9) is due to the known result proved in [16]
≈ ··· ···  
LK−1
1 Cδ (0) + L 
K−1
τ1 =0 τi =0,i=a τK =0 n=0
E[Ya ] = + (1 − ϕk )Cδ (k) .

K L 2
× 1 − sa (n + τa∗ ) [1 − si (n + τi )] k=1

i=1,i=a
Therefore, by (9), we can approximate the average access
delay of each user in the PS-CS. The accuracy of this approx-
xk (τa∗ )
1    
L−1 
L−1 
L−1
imation will be examined via numerical studies presented in
= 1 − sa n + τa∗ ··· ··· Section V.
LK−1
n=0 τ1 =0 τi =0,i=a τK =0

K
D. Worst-Case Access Delay
× [1 − si (n + τi )]
We are also interested in the worst-case access delay of the
i=1,i=a
PS-CS scheme. By the UI property of MCRT-UI sequences, we
xk (τa∗ )
1   know there must be one contention-free transmission within
= 1 − sa (n + τa∗ ) (L − K)K−1 one sequence period, and hence Ya ≤ L. Assume that user a
LK−1
n=0
  has experienced λ unsuccessful transmissions before its first
∗
 L − K K−1 contention-free transmission. Since that each of these λ unsuc-
= xk (τa ) + 1 − k
L cessful transmissions must involve at least one other user, and
∗
 each other user transmits at most K times in a sequence period,
= xk (τa ) + 1 − k ϕ1 .
we have

Then the conditional expected value of Da provided that Ya − Aa ≤ λ + 1 + K(K − 1) − λ = K 2 − K + 1.

τa = τa∗ can be computed by By using Da = Ya + (Ya − Aa )(l − 1), we further obtain an
upper bound on the worst-case access delay
    
E Da |τa = τa∗ Da ≤ L + K 2 + K − 1 (l − 1) = qr + K 2 + K − 1 (l − 1).

K
  ∗     (10)
= xk τa + 1 l − E Aa |Ya = xk (τa∗ ), τa = τa∗ (l − 1)
k=1 On the contrary, the p-persistent CSMA cannot guarantee any
 
× Pr Ya = xk (τa∗ )|τa = τa∗ upper bound on the worst-case access delay.
Remark 2: For the traditional CRT-UI protocol sequence-

K
  ∗      
≈ xk τa + 1 l − xk τa∗ + 1 − k ϕ1 (l − 1) based scheme, the worst-case delay is l(2K − 1)r (in the
k=1 units of tslot ). By the proof in the Appendix, we find that the
× (ϕk − ϕk−1 ) PS-CS is able to provide smaller worst-case delay than the
  traditional case for all K ≥ 2 and l ≥ 2, although the CRT-
 
K−1
∗
= (l − (l − 1)ϕ1 )E Ya |τa = τa + ϕ1 (l − 1) K − ϕk . UI sequences may have a shorter period than the MCRT-UI
k=1 sequences employed in the PS-CS.
912 IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO. 2, APRIL 2018

TABLE I TABLE II
q IN THE PS-CS FOR K = 4, 8, . . . , 28 AND l = 5, 10 PARAMETERS U SED FOR S IMULATION

E. Average Energy Consumption and EDP

Comparing the PS-CS with a traditional protocol sequence-
based scheme, an obvious difference in the behavior is that the
PS-CS needs to perform carrier sensing, inevitably leading to
extra energy cost. As WSNs usually suffer energy limitations,
it is important to study such side effect alongside the improve-
ment on the throughput and delay performance. To this end,
we derive Eeng , the average energy consumption of the users,
which is defined as the average energy used for each success-
ful transmission. Then for the PS-CS, as each user transmits
for K times and performs carrier sensing for L−K times within
each sequence period, we have
KPtx tpkt + (L − K)Pcs tcs
Eeng =  K−1 (11)
K 1 − KL
where Ptx and Pcs denote the power expenditures correspond-
ing to transmitting and carrier sensing states, respectively.
Clearly, there is a tradeoff involved in energy and other
application-sensitive QoS parameters. So, in delay-constrained Fig. 5. Average system throughput as a function of K for l = 5, 10.
WSNs, energy saving should be achieved by considering its
negative impact on the delay. To capture such tradeoff by
assuming energy and delay are equally important, we use the Little gaps in some cases are due to the infeasibility of con-
energy-delay product (EDP) metric [31], [32]. For the PS-CS, structing MCRT-UI sequences with the exactly optimal duty
it is defined as factors. We also see a significant improvement over the tra-
  ditional CRT-UI sequence-based scheme, owing to the help
1
K
of carrier sensing. A good agreement between analytical and
EDP = Eeng · E[Di ] . (12)
K numerical results can be found in all the cases.
i=1
Then the average delay performance is shown in Fig. 6.
We observe that, benefiting form the special Hamming-cross
V. N UMERICAL R ESULTS correlation and sequence structure, the PS-CS enjoys the sig-
In this section, through extensive simulations developed in a nificantly smaller average access delay than the other schemes
C simulator, we examine the relevant performance metrics of in all the cases. There is a 23.2%−32.7% improvement over
the PS-CS scheme and check the analysis accuracy. The val- p-persistent CSMA, and a 47.7%−59.5% improvement over
ues of q used in the PS-CS are shown in Table I. Based on the the traditional CRT-UI sequence-based scheme. It is also seen
system model described in Section III-B, we consider a sce- that the simulation results of the PS-CS match our approxi-
nario where multiple users with saturated traffic are randomly mation well in all the cases. This implies that the simplifying
placed 30 m away from the receiver. Specifically, reliable assumption used in deriving (9) is reasonable and accurate for
carrier sensing and ideal wireless channel are assumed. The MCRT-UI sequences to reflect the mutual interference patterns
other simulation parameters are summarized in Table II. For among the users.
each case, we carry out a simulation of 105 runs for various We in Fig. 7 further compare the maximal access delay
uniformly distributed random delay offsets. We consider the obtained from the simulation. This is of more practical interest
optimal p-persistent CSMA with the transmitting probability than the theoretical worst-case delay given in Section IV-D.
popt in (4) and the traditional CRT-UI sequence-based scheme The curves show that the PS-CS has a notably smaller maximal
for comparison purposes. In the former, we assume that each access delay. This phenomenon can be attributed to the help
user employs the same carrier sensing mechanism as in the of carrier sensing that reduces the channel idle time, and the
PS-CS, for a fair comparison. UI property of the MCRT-UI sequences that guarantees the
In Fig. 5, we compare the average system throughput of the upper-bounded delay of the PS-CS.
three schemes. The curves show that the PS-CS produces aver- As expected, Fig. 8 shows that the PS-CS has almost the
age throughput close to the optimal capacity of p-persistent same average energy consumption as the optimal p-persistent
CSMA: at least 97.1% of the maximum throughput for l = 5, CSMA, and has more energy consumption than the traditional
and at least 99.7% of the maximum throughput for l = 10. scheme due to carrier sensing. The gap between the PS-CS
ZHANG et al.: PROTOCOL SEQUENCES WITH CARRIER SENSING FOR WSNs 913

Fig. 6. Average access delay (in units of tslot ) as a function of K for l = 5, 10. Fig. 9. Ratio of the EDPs of the PS-CS and p-persistent CSMA to that of the
traditional CRT-UI sequence-based scheme as a function of K for l = 5, 10
and Pcs = 0.5Ptx .

p-persistent CSMA to that of the traditional CRT-UI sequence-

based scheme. It is observed that the PS-CS has smaller
EDP than the optimal p-persistent CSMA in all the cases.
Further, we find that the ratio for the PS-CS has an upward
trend as K increases, and the PS-CS outperforms the tradi-
tional scheme when K is smaller than a certain turning point.
This is because that, for these two schemes, the difference
in average energy consumption increases more quickly with
K than the difference in average access delay, which can
be drawn from Figs. 6 and 8. Also, it can be seen that large
packet length could mitigate the degradation on this ratio. The
reason is that the energy consumption due to carrier sens-
Fig. 7. Maximal access delay (in units of tslot ) in simulation as a function
of K for l = 5, 10. ing in the PS-CS can be made relatively small compared
with the total energy consumption in the traditional scheme if
(Ptx tpkt /Pcs tcs ) is large. These observations suggest that, when
energy and delay are considered equally important in delay-
constrained WSNs, the PS-CS should be more preferred for
small number of users or large packet length.

VI. I MPACT OF S ENSING FAULT AND C HANNEL E RROR

In this paper so far, we have assumed that carrier sensing
fault due to channel impairments or hidden terminals does not
occur. This assumption ensures that every active user in the
PS-CS scheme reads a new sequence entry simultaneously, and
hence, allows us to analyze relevant performance metrics rely-
ing on the sequence structure. Meanwhile, we have assumed
that random channel error due to wireless fading or noise is
Fig. 8. Average energy consumption (in units of Ptx tslot ) as a function of negligible. This simplification allows us to concentrate on the
K for l = 5, 10 and Pcs = 0.5Ptx . fundamental limitations of MAC protocols. In this section, we
relax these two ideal assumptions, and by simulation examine
how unreliable carrier sensing and imperfect wireless channel
and the traditional scheme becomes more noticeable as K would affect the performance of the PS-CS.
increases, since that the extra average energy consumption in To facilitate this paper, we use the model of independent
the PS-CS has a linear growth in K by (11). It should be noted Bernoulli trials by associating to each unideal event a prob-
that energy saving in the traditional scheme is obtained at the ability. We denote by pf the carrier sensing fault probability,
cost of larger average delay, as depicted in Fig. 6. and denote by pe the channel transmission error probability. It
To compare the energy-delay tradeoff of the three schemes, should be noted, since that carrier sensing is required to report
we in Fig. 9 plot the ratio of the EDPs of the PS-CS and a busy medium only after the detection of a signal with used
914 IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO. 2, APRIL 2018

of the delay performance. However, since that carrier sens-

ing is not used in the traditional protocol sequence-based
scheme, compared with it, the PS-CS is losing the advantage
in terms of average delay when pf increases, and meanwhile
the increasing pf exacerbates the energy-efficiency weakness
of the PS-CS. Not surprisingly, this indicates that more carrier
sensing fault would make the PS-CS to pay more energy cost
to get performance improvement.

VII. M ODIFICATION TO S UPPORT R EAL -T IME

F EEDBACK C ONTROL
The study so far has only considered the transmissions from
the users to the receiver for data collection. However, in some
applications, the receiver also needs to deliver control infor-
Fig. 10. Average access delay as a function of pe for K = 12, l = 10 and
different values of pf .
mation to the users for activity control, once some dynamic
adjustment is trigged by the data collection. In the following,
we present how to modify the PS-CS to support such feedback
control in a real-time manner.
We assume tslot = (tpkt + tprop + tcs )/l ≥ 2tcs for a fixed
positive integer l. So, by η := (tprop + tcs )/tslot , each packet
exactly fits in l − η channel slots.
The modified algorithm at the receiver side is simple. The
receiver keeps listening if it has no control information to send,
but otherwise, it begins its transmission at the beginning of a
channel slot after the completion of the ongoing transmission.
Note that carrier sensing is not required before transmission
for the receiver. At the user side, the working procedure of the
initial phase is the same as in Section III-C, but some changes
are made in the normal phase. In the normal phase, an active
user executes the following steps.
1) This user performs carrier sensing at the beginning of a
Fig. 11. Average energy consumption as a function of pe for K = 12, l = 10, slot. If the channel is found to be busy, the user keeps
Pcs = 0.5Ptx and different values of pf .
listening for l slots (including the carrier sensing time).
Otherwise, it is allowed to read a new sequence value.
2) If the new sequence value is one, the user begins its
modulation and spreading characteristics, we assume unreli- transmission at the same slot. If the new sequence value
able carrier sensing only happens if a user determines that the is zero, the user performs carrier sensing again at the
channel is idle when the channel is actually busy. same slot. Then, if the channel is found to be still
Figs. 10 and 11 show the average delay and average energy idle, the users keeps silent for one slot (including the
consumption as a function of pe for K = 12 and l = 10, detection time of two operations of carrier sensing),
respectively. The results with different values of pf are com- and otherwise, it keeps silent for l slots (including the
pared for each case. We see that for each given pf , the average detection time of two operations of carrier sensing).
delay of the PS-CS increases with pe due to the greater number 3) After the transmission or silence, the user repeats the
of unsuccessful transmissions. For the same reason, the aver- above operations if it still has a packet to send.
age energy consumption increases with pe as well. Meanwhile, The first carrier sensing at a time slot is used to for-
we see that for each given pe , both average delay and aver- bid the users to transmit if there is an ongoing transmission
age energy consumption of the PS-CS increases with pf . This from the receiver, and thus can ensure that all transmissions
is due to that the users look up the sequence entries more from the receiver are contention-free. This is desirable for
quickly, i.e., they become more aggressive in accessing the achieving real-time feedback control. Meanwhile, the first
channel, and more carrier sensing fault leads to more par- and second operations of carrier sensing at the same slot
tially time-overlapping collisions. Furthermore, we find that jointly ensure that all users begin to read new sequence values
the PS-CS still performs the best in terms of the average simultaneously.
delay in all the cases. This is because that although the users Briefly speaking, in the modified PS-CS, the receiver occu-
may not look up new sequence entries simultaneously due pies some channel slots as needed for real-time feedback con-
to sensing fault, the Hamming cross-correlation property of trol, while the users only read sequence values and schedule
MCRT-UI sequences employed in the PS-CS still has benefi- their transmissions in the remaining slots. So, the analysis in
cial impact on the mutual interference to some extent, and Section IV is still applicable to evaluate the performance of the
the sequence structure still contributes to the improvement users in an interval between two consecutive feedback control.
ZHANG et al.: PROTOCOL SEQUENCES WITH CARRIER SENSING FOR WSNs 915

Such an interval is of practical interest since that feedback For l = 2, we have

control is only triggered by the data collected in this interval. K(K − 1)(l − 1)
√ − (K 2 − 1)l − (K − 1)2
l−1  
VIII. C ONCLUSION
< 2.5K(K − 1) − 3K 2 − 2K − 1
In this paper, a new protocol sequence-based scheme that
incorporates carrier sensing, called the PS-CS, has been = −0.5K 2 − 0.5K + 1
proposed. This approach significantly improves the channel < 0.
utilization by reducing channel idle time, compared with the
conventional protocol sequence-based scheme. It has been For l = 3, we have
shown that the PS-CS can produce average throughput close to K(K − 1)(l − 1)
√ − (K 2 − 1)l − (K − 1)2
the optimal capacity of p-persistent CSMA; it also outperforms l−1
the optimal p-persistent CSMA and the traditional protocol < 1.5K(K − 1) − (4K 2 − 2K − 2)
sequence-based scheme in terms of the average delay and the
= −2.5K 2 + 0.5K + 2
worst-case delay. Even when energy and delay are considered
equally important, it has been shown that the PS-CS has bet- < 0.
ter energy-delay tradeoff for small number of users or large Therefore, qr + (K 2 − K + 1)(l − 1) < (2K − 1)rl for any l ≥ 2
packet length. Moreover, it has been shown that the PS-CS and K ≥ 2.
still has smaller average delay than other schemes under some
unreliable carrier sensing and imperfect wireless channel. In ACKNOWLEDGMENT
addition, a modification to support real-time feedback control The authors would like to thank the anonymous reviewers
has been presented. These features demonstrate that the PS- for their insightful comments that improve the presentation of
CS offers a promising MAC alternative for delay-constrained this paper.
WSNs. Our future work includes queuing delay analysis under
some scenarios and extension to the multihop topology.
R EFERENCES
A PPENDIX [1] J. L. Massey and P. Mathys, “The collision channel without feedback,”
IEEE Trans. Inf. Theory, vol. IT-31, no. 2, pp. 192–204, Mar. 1985.
The proof of the inequality qr + (K 2
− K + 1)(l − 1) < [2] V. C. da Rocha, Jr., “Protocol sequences for collision channel with-
(2K − 1)rl. out feedback,” IEE Electron. Lett., vol. 36, no. 24, pp. 2010–2012,
Nov. 2000.
Proof: If q = 2K − 1, with r ≥ K ≥ 2 and l ≥ 2, we have [3] K. W. Shum, C. S. Chen, C. W. Sung, and W. S. Wong, “Shift-invariant
  protocol sequences for the collision channel without feedback,” IEEE
qr + K 2 − K + 1 (l − 1) − (2K − 1)rl Trans. Inf. Theory, vol. 55, no. 7, pp. 3312–3322, Jul. 2009.
  [4] C. S. Chen, W. S. Wong, and Y.-Q. Song, “Constructions of robust pro-
= K 2 − K + 1 − (2K − 1)r (l − 1) tocol sequences for wireless sensor and ad hoc networks,” IEEE Trans.
Veh. Technol., vol. 57, no. 5, pp. 3053–3063, Sep. 2008.
≤ (−K 2 + 1)(l − 1) [5] Y. Zhang, Y.-H. Lo, W. S. Wong, and F. Shu, “Protocol sequences for
the multiple-packet reception channel without feedback,” IEEE Trans.
< 0. Commun., vol. 64, no. 4, pp. 1687–1698, Apr. 2016.
[6] Y. Zhang, K. W. Shum, and W. S. Wong, “On pairwise shift-invariant
If q > 2K − 1, as r is a prime, we have q ≤ (K/rfopt ) + 1. protocol sequences,” IEEE Commun. Lett., vol. 13, no. 6, pp. 453–455,
Then, with r ≥ K ≥ 2 and l ≥ 2, we have Jun. 2009.
  [7] W. S. Wong, “New protocol sequences for random-access channels with-
qr + K 2 − K + 1 (l − 1) − (2K − 1)rl out feedback,” IEEE Trans. Inf. Theory, vol. 53, no. 6, pp. 2060–2071,
   
Jun. 2007.
K [8] K. W. Shum and W. S. Wong, “Construction and applications of CRT
≤ + 1 r + K 2 − K + 1 (l − 1) − (2K − 1)rl sequences,” IEEE Trans. Inf. Theory, vol. 56, no. 11, pp. 5780–5795,
rfopt Nov. 2010.
K   [9] K. W. Shum, W. S. Wong, C. W. Sung, and C. S. Chen, “Design
= + r − (K 2 − 1)l − K 2 − K + 1 and construction of protocol sequences: Shift invariance and user irre-
fopt pressibility,” in Proc. IEEE Int. Symp. Inf. Theory, Seoul, South Korea,
− (r − K)(2K − 1)l Jun. 2009, pp. 1368–1372.
K [10] K. W. Shum, Y. Zhang, and W. S. Wong, “User-irrepressible sequences,”
≤ − (K 2 − 1)l + r − (K 2 − K + 1) in Proc. Conf. Sequences Their Appl., Paris, France, Sep. 2010,
fopt pp. 88–101.
− 2(r − K)(2K − 1) [11] L. Salaün, C. S. Chen, Y. Chen, and W. S. Wong, “Constant delivery
delay protocol sequences for the collision channel without feedback,” in
K
≤ − (K 2 − 1)l − (K − 1)2 Proc. Int. Symp. Wireless Pers. Multimedia Commun., Shenzhen, China,
fopt Nov. 2016, pp. 429–434.
[12] C.-C. Chen et al., “Constructions and throughput analyses of proto-
K(K − 1)(l − 1)
=
− (K 2 − 1)l − (K − 1)2 col sequences with adjustable duty factor for collision channels without
feedback,” IEEE Trans. Commun., vol. 64, no. 11, pp. 4736–4748,
1 + 2(l − 1) K − 1
K−1
Nov. 2016.
K(K − 1)(l − 1) [13] Q. A. Nguyen, L. Györfi, and J. L. Massey, “Constructions of binary
≤ √ − (K 2 − 1)l − (K − 1)2 . constant-weight cyclic codes and cyclically permutable codes,” IEEE
l−1 Trans. Inf. Theory, vol. 38, no. 3, pp. 940–949, May 1992.
[14] L. Györfi and I. Vajda, “Construction of protocol sequences for multiple-
For l ≥ 4, the above expression obviously is smaller than access collision channel without feedback,” IEEE Trans. Inf. Theory,
zero. vol. 39, no. 5, pp. 1762–1765, Sep. 1993.
916 IEEE INTERNET OF THINGS JOURNAL, VOL. 5, NO. 2, APRIL 2018

[15] S. Bitan and T. Etzion, “Constructions for optimal constant weight cycli- Yijin Zhang (M’14) received the B.S. degree
cally permutable codes and difference families,” IEEE Trans. Inf. Theory, in information engineering from the Nanjing
vol. 41, no. 1, pp. 77–87, Jan. 1995. University of Posts and Telecommunications,
[16] Y. Wu, K. W. Shum, Z. Lin, W. S. Wong, and L. Shen, “Protocol Nanjing, China, in 2004, the M.S. degree in
sequences for mobile ad hoc networks,” in Proc. IEEE Int. Conf. information engineering from Southeast University,
Commun. (ICC), Budapest, Hungary, Jun. 2013, pp. 1730–1735. Nanjing, China, in 2007, and the Ph.D. degree
[17] Y. Wu, K. W. Shum, W. S. Wong, and L. Shen, “Safety-message broad- in information engineering from the Chinese
cast in vehicular ad hoc networks based on protocol sequences,” IEEE University of Hong Kong, Hong Kong, in 2010.
Trans. Veh. Technol., vol. 63, no. 3, pp. 1467–1479, Mar. 2014. Since 2011, he has been an Associate Professor
[18] Y. Mao and L. Shen, “A framework for protocol sequence allocation in with the School of Electronic and Optical
vehicular ad hoc networks,” in Proc. IEEE Veh. Technol. Conf., Nanjing, Engineering, Nanjing University of Science and
China, May 2016, pp. 1–5. Technology, Nanjing. His current research interests include sequence design
[19] W. S. Wong, “Transmission sequence design and allocation for wide- and resource allocation in communication networks.
area ad hoc networks,” IEEE Trans. Veh. Technol., vol. 63, no. 2,
pp. 869–878, Feb. 2014.
[20] F. Cali, M. Conti, and E. Gregori, “Dynamic tuning of the IEEE 802.11
Ming Zhang received the B.S. degree in information
protocol to achieve a theoretical throughput limit,” IEEE/ACM Trans.
engineering from the Zhejiang University of Media
Netw., vol. 8, no. 6, pp. 785–799, Dec. 2000.
and Communications, Hangzhou, China, in 2013,
[21] W. Ma, C.-E. Zhao, and D. Shen, “New optimal constructions of
and the M.S. degree in information engineering from
conflict-avoiding codes of odd length and weight 3,” Designs Codes
the Nanjing University of Science and Technology,
Cryptography, vol. 73, no. 3, pp. 791–804, 2014.
Nanjing, China, in 2016.
[22] Y. Lin, M. Mishima, and M. Jimbo, “Optimal equi-difference conflict-
Her current research interest includes channel
avoiding codes of weight four,” Designs Codes Cryptography, vol. 78,
access protocol for wireless networks.
no. 3, pp. 747–776, 2016.
[23] M. Mishima and K. Momihara, “A new series of optimal tight conflict-
avoiding codes of weight 3,” Discr. Math., vol. 340, no. 4, pp. 617–629,
2017.
[24] W. Chu, C. J. Colbourn, and V. R. Syrotiuk, “The effects of synchro-
nization on topology-transparent scheduling,” Wireless Netw., vol. 12,
no. 6, pp. 681–690, Dec. 2006. Yuan-Hsun Lo (M’17) received the B.S., M.S., and
[25] Y. Liu, V. O. K. Li, K.-C. Leung, and L. Zhang, “Topology-transparent Ph.D. degrees in applied mathematics from National
scheduling in mobile ad hoc networks with multiple packet recep- Chiao Tung University, Hsinchu City, Taiwan, in
tion capability,” IEEE Trans. Wireless Commun., vol. 13, no. 11, 2004, 2006, and 2010, respectively.
pp. 5940–5953, Nov. 2014. He is currently an Assistant Professor with
[26] J. Lutz, C. J. Colbourn, and V. R. Syrotiuk, “ATLAS: Adaptive topology- the School of Mathematical Sciences, Xiamen
and load-aware scheduling,” IEEE Trans. Mobile Comput., vol. 13, University, Xiamen, China. His current research
no. 10, pp. 2255–2268, Oct. 2014. interests include combinatorics, graph theory, and
[27] C. H. Rentel and T. Kunz, “Bounds and parameter optimization of combinatorial design theory and their applications.
medium access control coding for wireless ad hoc and sensor networks,”
Ad Hoc Netw., vol. 10, no. 1, pp. 128–143, 2012.
[28] D. Kim, D. J. Esteki, Y.-C. Hu, and P. R. Kumar, “A lightweight deter-
ministic MAC protocol using low cross-correlation sequences,” in Proc.
IEEE Glob. Telecommun. Conf. (GLOBECOM), Kathmandu, Nepal, Wing Shing Wong (M’81–SM’90–F’02) received
Dec. 2011, pp. 1–6. the combined bachelor’s and master’s degrees
[29] R. Bruno, M. Conti, and E. Gregori, “Optimal capacity of p-persistent (summa cum laude) from Yale University,
CSMA protocols,” IEEE Commun. Lett., vol. 7, no. 3, pp. 139–141, New Haven, CT, USA, in 1976, and the M.S. and
Mar. 2003. Ph.D. degrees from Harvard University, Cambridge,
[30] G. Anastasi, E. Borgia, M. Conti, E. Gregori, and A. Passarella, MA, USA, in 1978 and 1980, respectively.
“Understanding the real behavior of Mote and 802.11 ad hoc networks: After researching with AT&T Bell Laboratories
An experimental approach,” Pervasive Mobile Comput., vol. 1, no. 2, for ten years, he joined the Chinese University of
pp. 237–256, 2005. Hong Kong, Hong Kong, in 1992, and is currently
[31] S. Lindsey, C. S. Raghavendra, and K. Sivalingam, “Data gathering a Professor of information engineering. He was
in sensor networks using the energy∗delay metric,” in Proc. IPDPS the Chairman of the Department of Information
Workshop Issues Wireless Netw. Mobile Comput., San Francisco, CA, Engineering from 1995 to 2002 and the Dean of the Graduate School from
USA, 2001, pp. 2001–2008. 2005 to 2014, and served as the Science Advisor with the Innovation and
[32] H. Wu and K. Wolter, “Stochastic analysis of delayed mobile offloading Technology Commission of the HKSAR Government from 2003 to 2005. His
in heterogeneous networks,” IEEE Trans. Mobile Comput., vol. 17, no. 2, current research interests include mobile communication, nonlinear filtering,
pp. 461–474, Feb. 2018, doi: 10.1109/TMC.2017.2711014. and networked control.