You are on page 1of 5

IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 10, NO.

11, NOVEMBER 2021 2499

Age of Information With Joint Packet Coding in Industrial IoT


Tse-Tin Chan , Member, IEEE, Haoyuan Pan , Member, IEEE, and Jiaxin Liang , Member, IEEE

Abstract—This letter studies the information freshness in an


industrial Internet of Things (IIoT) network, measured by Age of
Information (AoI). We consider a scenario where a sink node col-
lects information update packets from different sensors and then
uploads the collected packets to an edge server. This scenario has
two main requirements: high information freshness (low AoI) and
high reliability (low packet error rate, PER). Since update pack-
ets are usually short in practice, previous works usually packed
and encoded multiple short packets from different sensors into a
long packet to improve PER performances. However, while such
a joint coding approach improves reliability, it generally leads
to longer delay and hence possibly higher AoI. This letter inves-
tigates the AoI performance tradeoff by examining the number Fig. 1. A wireless IIoT network scenario: a sink node gathers the update
of packets to be jointly encoded. We consider two AoI metrics, packets from different sensors and then uploads the collected packets to an
average AoI and bounded AoI. In particular, bounded AoI is the edge server.
threshold below which the instantaneous AoI falls for a given
percentage of the time. Our theoretical analysis and numerical
results show that there exist optimal numbers of jointly coded communications system designs [2]–[15]. Although AoI has
packets that minimize the average AoI and the bounded AoI. attracted considerable interest from researchers, most of the
Specifically, a smaller number of packets is usually sufficient to previous AoI studies focused chiefly on upper-layer designs
achieve both high information freshness and high reliability.
such as queue management schemes and packet scheduling
Index Terms—Age of information, information freshness, joint policies [2]–[6].
coding, multiple access, short packet. In this letter, we study the AoI in an industrial IoT
(IIoT) network scenario where a sink node regularly collects
information update packets from different sensors and then
I. I NTRODUCTION uploads the collected packets to an edge server, as shown
UPPORTING Ultra-Reliable Low-Latency Commun-
S ications (URLLC) is crucial for Internet of Things (IoT)
applications that handle time-sensitive information, and Age
in Fig. 1. In contrast to many AoI studies, we focus on the
physical layer design where the packet error rate (PER) is
inevitable due to unreliable wireless channels. There are two
of Information (AoI) is a key performance metric for quanti- main requirements in this scenario: high information freshness
fying the information freshness in IoT systems [1]. AoI was (low AoI) and high reliability (low PER).
introduced in the seminal works [2], [3], and it is defined as Information update packets in IIoT are generally short, pos-
the elapsed time since the generation of the latest received sibly as short as a few tens of bytes [16]. The PER of short
packet. In contrast to delay, which measures only the time packets depends heavily on the finite blocklength of the packet
required to deliver a packet, AoI quantifies the information in the channel coding. This letter uses the Polyanskiy-Poor-
freshness of the received packet from the destination’s perspec- Verdu (PPV) converse bound [17] to characterize the decoding
tive, as determined by both the generation time of the update performance of short packets over unreliable channels. To
packet at the source and the delay in sending that packet to improve the reliability of short-packet communications, small
the destination. Replacing delay and throughput by AoI as packets from multiple sensors can be concatenated into one
the performance metric may lead to fundamental changes in larger packet and jointly encoded at the sink node [18]. On
the one hand, joint coding with a longer code can improve the
Manuscript received June 17, 2021; revised July 29, 2021; accepted August
13, 2021. Date of publication August 16, 2021; date of current version
PER performance of the packets. On the other hand, joint cod-
November 9, 2021. This work was supported in part by the National Natural ing usually causes longer transmission time due to the delay in
Science Foundation of China under Grant 62001298; in part by the Natural waiting for all packets to be jointly encoded, which may lead
Science Foundation of Guangdong Province under Grant 2021A1515012601; to higher AoI. As far as we know, there has been no research
and in part by the Deep Learning and Cognitive Computing Centre, The Hang
Seng University of Hong Kong. The associate editor coordinating the review of addressing this dilemma. Therefore, this letter resolves this
this article and approving it for publication was N. Zlatanov. (Corresponding dilemma by deriving closed-form AoI performances when dif-
author: Haoyuan Pan.) ferent numbers of packets from sensors are jointly coded, using
Tse-Tin Chan is with the Department of Computing, The Hang Seng the PPV converse bound.
University of Hong Kong, Hong Kong (e-mail: ttchan@hsu.edu.hk).
Haoyuan Pan is with the College of Computer Science and Software We consider two AoI metrics throughout this letter, namely
Engineering, Shenzhen University, Shenzhen 518060, China (e-mail: average AoI and bounded AoI. Average AoI measures the
hypan@szu.edu.cn). instantaneous AoI averaged over time [3], [7]. In addition
Jiaxin Liang is with the Department of Information Engineering,
The Chinese University of Hong Kong, Hong Kong (e-mail:
to average performance, we are also interested in worst-case
jiaxin@ie.cuhk.edu.hk). performance, which is critical for many time-sensitive appli-
Digital Object Identifier 10.1109/LWC.2021.3105304 cations. To this end, bounded AoI corresponds to the threshold
2162-2345 
c 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
2500 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 10, NO. 11, NOVEMBER 2021

below which the instantaneous AoI falls for a given percentage focuses on two AoI metrics, namely average AoI and bounded
of the time [12]. AoI, as follows.
Recent works such as [8]–[11] studied the AoI metric in Average AoI: The average AoI [7] of sensor i, Δi , is the
point-to-point short-packet communications. Unlike [8]–[11], time average of the instantaneous AoI, i.e.,
this letter focuses on a multi-hop multi-node scenario, where 
1 T
sensors share the same wireless channel and send their update Δi = lim Δi (t) dt. (2)
packets to a common edge server via a sink [12], [13]. This let- T →∞ T 0
ter also differs from our previous paper [14] which proposed a Therefore, the average AoI of the system Δ is Δ =
stream-based channel coding scheme to improve the AoI per- 1 N Δ . The lower average AoI of the system Δ indi-
N i=1 i
formances, where the number of packets to be jointly coded is cates that the update packets at the edge server are generally
always two. The main contribution of this letter is to address fresher and can better reflect the states of the sensors.
an open problem at the physical layer: the impact of the Bounded AoI: Many time-sensitive applications want to
number of packets jointly coded on the AoI performances in ensure that the instantaneous AoI of the sensors is usually
multi-hop and multi-node scenarios. The closed-form AoI per- low. Bounded AoI was proposed in [12], and it is the thresh-
formances derived and the numerical results show that there old below which the instantaneous AoI falls for a required
exist optimal numbers of jointly encoded packets to mini- percentage of the time γ.
mize the average AoI and the bounded AoI. In particular, The bounded AoI of the system ΔTHR with confidence
high information freshness and high reliability can usually be (percentage of time) γ is the minimum value that satisfies
achieved by jointly encoding a small number of packets. 
1 T
lim I [Δi (t) ≤ ΔTHR ] dt ≥ γ, ∀i (3)
II. P RELIMINARIES T →∞ T 0

A. General System Description where I [Δi (t) ≤ ΔTHR ] is an indicator function which equals
We consider N independent sensors uploading their short 1 if Δi (t) ≤ ΔTHR and equals 0 otherwise [12]. For a cer-
update packets to an edge server with the aid of a sink node tain percentage of time γ, a smaller bounded AoI value of
in a half-duplex mode, as shown in Fig. 1. The sensors are the system ΔTHR indicates that the system can guarantee the
divided into M N groups for joint packet coding at the sink update packets at the edge server to be fresher than the thresh-
node, and each group has M sensors. We assign unique indices old. According to [12], [13], the upper bound of bounded AoI
N } to each sensor and
i ∈ {1, 2, . . . , N } and g ∈ {1, 2, . . . , M with confidence (percentage of time) γ is estimated by
   
each sensor group respectively. Without loss of generality, we 2
consider sensor i as the k-th sensor in group g. Different groups  Δ2i − Δi
ΔTHR = max + Δi (4)
take turns sending their packets to the sink node. The sink node i∈{1,2,...,N } 1−γ
collects packets from the sensors in a group and then forwards 
the jointly encoded packets to the edge server. where (Δ2i ) = limT →∞ T1 0T (Δi (t))2 dt. The upper
We consider a generic communication schedule that each bound (4) has been shown to characterize the performance
sensor sends its latest information update regularly in each of real bounded AoI well in various settings [12], [13].
cycle. The system is allocated with bandwidth B. The signals
are transmitted over an additive white Gaussian noise (AWGN) C. Packet Error Rates (PER) for Short Packets
channel or a fading channel. For simplicity, we consider sym- As in [10], [13], the PER of finite-length packets in AWGN
metric channels where the received power of the signals from channels, ppacket , can be approximated according to the PPV
the sensors to the sink node is the same. Sensors in the same converse bound [17], i.e.,
group have different ways of sending packets for joint packet ⎛ ⎞
coding. We first consider that they send packets in orthogonal 1
⎜ log2 (1 + P ) − SL ⎟
time slots, i.e., in a round-robin manner. The sink node needs ppacket ≈ Q ⎜

2 ⎟
⎠ (5)
to collect all the packets before sending the jointly encoded 1 1− 1
log2 (e) 2L 2
packet to the edge server. As will be shown, this leads to a (1+P )
fairness issue from the perspective of information freshness.
where S is the number of source bits, L is the blocklength
Thus we consider that they send packets simultaneously using
of the encoded packet, P is the effective signal-to-noise
different subchannels. Further details of the system models are
ratio (SNR) at the receiver, e is the Euler’s number, and
described in Sections III-A and IV-A.  u2
Q(x ) = √1 x∞ e − 2 du is the Q-function. Given a ppacket ,

B. Age of Information (AoI) Metrics we can also estimate the blocklength L by (5). Note that the
AoI analysis in this letter can be simply extended to fading
Instantaneous AoI: At time instant t, the instantaneous channels by replacing the expression for PER in (5) with those
AoI [7] of sensor i, Δi (t), is for fading channels such as in [15]. Without loss of generality,
Δi (t) = t − Ui (t) (1) we consider an AWGN channel in this letter.
An update is successfully decoded at the edge server if there
where Ui (t) is the generation time of the latest update received are no errors in the two hops. We denote the PER of the overall
by the edge server from sensor i. Δi (t) is a continuous-time transmission by p = p1 +p2 −p1 p2 where p1 ∈ [0, 1] and p2 ∈
continuous-value stochastic process which can be used as an [0, 1] are the PERs of the sensor-sink link and the sink-server
intermediary to compute other AoI metrics [7]. This letter link respectively. When the decoding fails, the node directly
CHAN et al.: AGE OF INFORMATION WITH JOINT PACKET CODING IN INDUSTRIAL IoT 2501

Fig. 2. Joint packet coding with round-robin transmission.

Fig. 3. An example of ΔRR i (t) in a joint packet coding with round-


discards the packet without any feedback signal. We define X robin transmission. The updates are successfully decoded at tjRR
,4 , tj +2,4 ,
RR
as the number of packets transmitted by sensor i between two
and tj +3,4 except tj +1,4 .
RR RR
successful packet decoding at the edge server. X is a geometric
random variable with parameter 1 − p, e.g., Pr(X = x ) is the
probability that the update of sensor i fails to be decoded x −1
times until its next successful update. successful update of sensor i as
 tjRR
 RR
2
+X ,4 XTround
III. AO I OF J OINT PACKET C ODING W ITH ROUND - ROBIN ARR
i,w = ΔRR
i (t) dt = RR
XTround T0RR + . (6)
tjRR 2
T RANSMISSION ,4

A. System Model An example of ARR i,w is illustrated in the shaded area of


Following the system description in Section II-A, we first Fig. 3. Durations T1 and T2RR are related to the blocklengths
RR
investigate the AoI performance of joint packet coding with LRR
round-robin transmission (RRT) as depicted in Fig. 2. Using and the allocated bandwidth. We have T1RR = B1 and T2RR =
LRR RR RR
the whole bandwidth of the system B, M sensors in the same 2
B where L1 and L2 are the packet blocklengths of the
group send their packets for joint packet coding in orthog- sensor and the sink node respectively.
onal time-slots with duration T1RR . The sink node, which is RR
The average AoI of sensor i, Δi , can be found by1
also allocated with bandwidth B, jointly encodes the received W   RR
packets and then immediately uploads the encoded packet to RR w =1 ARR
i,w E X 2 Tround
Δi = lim W = T0RR +
the edge server in duration T2RR . Afterward, the sensors in W →∞
w =1
2E [X ]
XTround
RR

other groups send their packets to the edge server likewise   


1 RR RR (1 + p) N RR
with the aid of the sink node. We define a round as the total = (M − k + 1)L1 + L2 + NLRR
1 + L .
B 2(1 − p) M 2
transmission time of all the N sensors and the sink node. The (7)
period of a round is TroundRR = NT RR + N T RR .
1 M 2 RR
As shown in Fig. 2, sensor i generates its packet in round The average AoI of the system Δ can then be obtained as
j at time tjRR,1 and completes the transmission at time tj ,2 =
RR N M N M
Δ = N1 = N1
RR RR RR RR
i=1 Δi g=1 k =1 Δg,k where Δg,k
tj ,1 +T1 . The sink node waits until all packets from the same
RR RR
denotes the average AoI of the k-th sensor in group g.
sensor group are collected. Therefore, the sink node sends the Bounded AoI: We first define Si,wRR for the w-th successful
jointly encoded packet at time tjRR ,3 = tj ,2 + (M − k )T1
RR RR
update of sensor i as
and completes the transmission at time tj ,4 = tj ,3 + T2 .
RR RR RR
 
+X ,4 
tjRR 2  2
RR
XTround
Packet decoding for sensor i occurs at time tjRR ,4 . The duration RR
Si,w = ΔRR
i (t) dt = T0RR + XTround
RR
dt
between tj ,1 and tj ,4 is T0 = tj ,4 − tj ,1 = (M − k +
RR RR RR RR RR
tjRR
,4 0
1)T1RR + T2RR . After transmissions of the sensors in other  3
 2  2 RR
XTround
groups, sensor i generates its packet again in round j +1 at time = T0RR RR
XTround + T0RR RR
XTround + . (8)
3
+1,1 = tj ,1 + Tround . Fig. 3 shows an illustrative example of
tjRR RR RR

the instantaneous AoI of sensor i, ΔRR i (t). When the update


RR
Then we can find (Δ2i ) by
is successfully decoded at the edge server, ΔRR i (t) = T0 . In
RR

the absence of an update, Δi (t) increases linearly with time.


RR W
 RR w =1 Si,w
RR
Therefore, ΔRR Δ2i = lim
i (t) is a sawtooth wave as shown in Fig. 3.
W
W →∞
w =1 XTround
RR
   2
 2 T0RR E X 2 Tround
RR E X3 Tround
RR
B. AoI Analysis = T0RR + + . (9)
E [X ] 3E [X ]
RR
In this subsection, we derive the average AoI Δ and esti-
mate the upper bound of bounded AoI Δ  RR of N-user joint Finally, the bounded AoI Δ  RR with confidence γ can be
THR THR
packet coding with RRT. estimated by substituting the parameters into (4).
Average AoI: Without loss of generality, we assume the
(w − 1)-th and the w-th successful updates of sensor i occur 1 The first, second, and third moments about the origin of X can be found
2
1 , E [X 2 ] = 1+p , and E [X 3 ] = 1+4p+p .
at tjRR RR RR
,4 and tj +X ,4 respectively. We define Ai,w for the w-th
by E [X ] = 1−p
(1−p)2 (1−p)3
2502 IEEE WIRELESS COMMUNICATIONS LETTERS, VOL. 10, NO. 11, NOVEMBER 2021

 tjSM
+X ,3
i,w = t SM
successful update of sensor i as ASM ΔSM
i (t) dt =
j ,3
SM )2
(XTround LSM
SM T SM +
XTround . Moreover, we have T1SM = 1
=
0 2 B
M
MLSM LSM
1
B and T2SM = B where LSM
2
1 and LSM
2 are the packet
blocklengths of the sensors and the sink node, respectively.
Following the same procedure as in Section III, the average
SM
AoI of sensor i, Δi , is
W   SM
Fig. 4. Joint packet coding with simultaneous multichannel transmission. SM w =1 Ai,w
SM
E X 2 Tround
Δi = lim W = T0SM +
W →∞
w =1 XTround
SM 2E [X ]
IV. I MPROVING AO I OF J OINT PACKET C ODING W ITH   
1 (1 + p) N SM
S IMULTANEOUS M ULTICHANNEL T RANSMISSION = MLSM + L SM
+ NL SM
+ L . (10)
B 1 2
2(1 − p  ) 1
M 2
In RRT, we find that some sensors have a higher average
Note that the PER p  is generally different from p in Section III
AoI due to the longer waiting time for joint coding at the sink
given the same received power at the sink node, because
node, e.g., the waiting time of a sensor depends on its position
multiple subchannels lead to a lower bandwidth and a lower
k in the sensor group. To address this issue, we investigate the
noise power. For a symmetric channel, the average AoI of the
AoI performance when multiple sensors send their packets to N
system is Δ = N1
SM SM SM
the sink node simultaneously using different subchannels. i=1 Δi = Δi .
Bounded AoI: As in the previous section, we define
SM for the w-th successful update of sensor i as S SM =
Si,w i,w
A. System Model  tjSM
+X ,3
(Δi (t)) dt = (T0 ) XTround + T0 (XTround )2 +
SM 2 SM 2 SM SM SM
We further describe the joint packet coding with simulta- tj ,3
SM
SM )3
(XTround SM
neous multichannel transmission (SMT), following the system
3 We can find (Δ2i ) by (11) as shown at the bot-
.
description in Section II-A, as illustrated in Fig. 4. In SMT,
tom of the page, and then substitute the parameters into (4) to
M sensors in the same group are allocated with different non-  SM with confidence γ.
B . M sensors estimate the bounded AoI Δ THR
overlapping subchannels with equal bandwidth M
in a group send their packets simultaneously in duration T1SM . V. AO I P ERFORMANCE E VALUATION
The sink node, allocated with bandwidth B, jointly encodes the
received packets and then immediately uploads the encoded In this section, we present the average AoI and the bounded
packet to the edge server in duration T2SM . Then, the sensors AoI performances when different numbers of packets are
in other groups transmit their packets likewise. The period of a jointly coded in round-robin transmission (RRT) and simul-
round, which is the total transmission time of all the N sensors taneous multichannel transmission (SMT). In particular, we
and the sink node, is Tround SM = N (T SM + T SM ). show that the optimal number can be found to achieve both
M 1 2
As shown in Fig. 4, sensor i together with other sensors in high information freshness and high reliability.
the same group generate their packets in round j at time tjSM We consider a system with a total of N = 30 sensors. Each
,1 .
They send their packets simultaneously and complete the trans- information update packet from a sensor has S = 100 bits
(i.e., a short packet). The system bandwidth is normalized to
,2 = tj ,1 +T1 . The sink node forwards the
mission at time tjSM SM SM
1. The PERs of the sensor-sink link and the sink-server link
jointly encoded packet at time tjSM ,2 and completes the trans- are fixed to 0.5 × 10−3 , i.e., the total PER after two hops is
mission at time tjSM
,3 = t SM + T SM . The duration between t SM
j ,2 2 j ,1 around 10−3 . The blocklengths of the individual packets (from
and tjSM is T SM = t SM − t SM = T SM + T SM . After the trans-
,3 0 j ,3 j ,1 1 2 the sensors to the sink node) and the jointly encoded packet
missions of the sensors in other groups, sensor i generates its (from the sink node to the edge server) are computed by the
packet again in round j + 1 at time tjSM +1,1 = tj ,1 + Tround . If
SM SM
PPV converse bound under the above PER constraints.
the update packet is successfully decoded at the edge server, Fig. 5 plots the average AoI versus bounded AoI when the
the instantaneous AoI of sensor i, ΔSM i (t), is reset to T0 ;
SM
update packets from M = (1, 2, 3, 5, 6) sensors are jointly
otherwise Δi (t) increases linearly with time.
SM encoded and the received power at the sink node is (1 dB,
3 dB, 5 dB). To evaluate the bounded AoI, we consider the
confidence (percentage of time) γ = 99% to give an upper
B. AoI Analysis
SM
bound on the bounded AoI according to (4). Moreover, we
Now we derive the average AoI Δ and estimate the upper are interested in the case when the confidence γ varies. Fig. 6
bound of bounded AoI Δ  SM of joint packet coding with SMT. plots the AoI performances when γ = (99.9%, 99%, 95%)
THR
Average AoI: Without loss of generality, we assume the and the received power at the sink node is 3 dB.
(w − 1)-th and the w-th successful updates of sensor i occur As we can observe from these two figures, for both RRT and
at tjSM SM SM
,3 and tj +X ,3 respectively. We define Ai,w for the w-th SMT, the average AoI and the bounded AoI generally decrease

 2  2
 2   MLSM
1 +L2
SM   MLSM
1 +L2
SM
W E X2 N E X3 N2
 2 SM w =1 i,w S SM 1 + L2
MLSM SM B B
Δi = lim W = + + (11)
W →∞ SM
w =1 XTround
B E [X ]M 3E [X ]M 2
CHAN et al.: AGE OF INFORMATION WITH JOINT PACKET CODING IN INDUSTRIAL IoT 2503

VI. C ONCLUSION
We have investigated the information freshness in an IIoT
network scenario, where a sink node regularly collects short
update packets from different sensors and then uploads the col-
lected packets to an edge server. Conventionally, multiple short
packets are usually packed and encoded jointly into a larger
packet to improve reliability. However, a large number of pack-
ets to be jointly encoded leads to longer transmission times
and possibly higher AoI. Based on the PPV converse bound
to estimate the PER of short packets, we investigate the num-
ber of jointly-coded packets to achieve both high reliability
and high information freshness. Our numerical results show
that a smaller number of jointly encoded packets is usually
Fig. 5. AoI performance of joint packet coding with RRT and SMT when sufficient to achieve both requirements. Furthermore, simulta-
the received power at the sink node is (1 dB, 3 dB, 5 dB) and γ = 99%. neous transmission through multiple subchannels reduces both
the average AoI and the bounded AoI substantially.

R EFERENCES
[1] M. A. Abd-Elmagid, N. Pappas, and H. S. Dhillon, “On the role of age
of information in the Internet of Things,” IEEE Commun. Mag., vol. 57,
no. 12, pp. 72–77, Dec. 2019.
[2] S. Kaul, M. Gruteser, V. Rai, and J. Kenney, “Minimizing age of
information in vehicular networks,” in Proc. IEEE Int. Conf. Sens.
Commun. Netw., Salt Lake City, UT, USA, 2011, pp. 350–358.
[3] S. Kaul, R. Yates, and M. Gruteser, “Real-time status: How often should
one update?” in Proc. IEEE INFOCOM, Orlando, FL, USA, Mar. 2012,
pp. 2731–2735.
[4] B. Zhou and W. Saad, “Optimal sampling and updating for mini-
mizing age of information in the Internet of Things,” in Proc. IEEE
GLOBECOM, Abu Dhabi, UAE, Dec. 2018, pp. 1–6.
[5] X. Xie, H. Wang, L. Yu, and M. Weng, “Online algorithms for optimiz-
ing age of information in the IoT systems with multi-slot status delivery,”
IEEE Wireless Commun. Lett., vol. 10, no. 5, pp. 971–975, May 2021.
Fig. 6. AoI performance of joint packet coding with RRT and SMT when [6] Q. Abbas, S. A. Hassan, H. Pervaiz, and Q. Ni, “A Markovian
γ is (99.9%, 99%, 95%) and the received power at the sink node is 3 dB. model for the analysis of age of information in IoT networks,” IEEE
Wireless Commun. Lett., vol. 10, no. 7, pp. 1596–1600, Jul. 2021,
doi: 10.1109/LWC.2021.3075160.
[7] A. Kosta, N. Pappas, and V. Angelakis, “Age of information: A new con-
as the received power at the sink node increases. Furthermore, cept, metric, and tool,” Found. Trends Netw., vol. 12, no. 3, pp. 162–259,
SMT outperforms RRT in both AoI metrics under the same M Nov. 2017.
and received power. The performance improvement of SMT [8] R. Devassy, G. Durisi, G. C. Ferrante, O. Simeone, and E. Uysal-
Biyikoglu, “Delay and peak-age violation probability in short-packet
mainly comes from its higher effective SNR. Also, SMT pre- transmissions,” in Proc. IEEE ISIT, Vail, CO, USA, Jun. 2018,
vents some sensors from waiting for a long time for joint pp. 2471–2475.
[9] R. Devassy, G. Durisi, G. C. Ferrante, O. Simeone, and E. Uysal,
packet coding, which is fair to all the sensors in the same “Reliable transmission of short packets through queues and noisy chan-
group. Fig. 6 shows that the bounded AoI decreases as the nels under latency and peak-age violation guarantees,” IEEE J. Sel. Areas
confidence γ decreases in both RRT and SMT. In particular, Commun., vol. 37, no. 4, pp. 721–734, Apr. 2019.
[10] R. Wang, Y. Gu, H. Chen, Y. Li, and B. Vucetic, “On the age of
the improvement by SMT is more significant when γ is large, information of short-packet communications with packet management,”
i.e., a more stringent requirement for the bounded AoI. in Proc. IEEE GLOBECOM, Waikoloa, HI, USA, Dec. 2019, pp. 1–6.
More importantly, both Figs. 5 and 6 indicate that there [11] B. Yu, Y. Cai, D. Wu, and Z. Xiang, “Average age of information in short
packet based machine type communication,” IEEE Trans. Veh. Technol.,
exist optimal numbers of packets for joint coding to mini- vol. 69, no. 9, pp. 10306–10319, Sep. 2020.
mize the average AoI and the bounded AoI. For example, [12] H. Pan and S. C. Liew, “Information update: TDMA or FDMA?” IEEE
Wireless Commun. Lett., vol. 9, no. 6, pp. 856–860, Jun. 2020.
as shown in Fig. 5, when the received power is as high [13] J. Liang, H. Pan, and S. C. Liew, “Is multichannel access useful for
as 5 dB and γ = 99%, M = 3 leads to a minimum of timely information update?” IEEE Wireless Commun. Lett., vol. 10,
both average AoI and bounded AoI simultaneously for SMT. no. 4, pp. 815–819, Apr. 2021.
[14] H. Pan, S. C. Liew, J. Liang, V. C. M. Leung, and J. Li, “Coding of
When the received power is low, e.g., 1 dB, the optimal aver- multi-source information streams with age of information requirements,”
age AoI is reached when M = 3 and the optimal bounded IEEE J. Sel. Areas Commun., vol. 39, no. 5, pp. 1427–1440, May 2021.
AoI is reached when M = 5. Our findings indicate that [15] Y. Gu, H. Chen, Y. Li, and B. Vucetic, “Ultra-reliable short-packet
communications: Half-duplex or full-duplex relaying?” IEEE Wireless
a small number of packets to be jointly coded is usually Commun. Lett., vol. 7, no. 3, pp. 348–351, Jun. 2018.
sufficient to achieve both high information freshness and [16] P. Schulz et al., “Latency critical IoT applications in 5G: Perspective on
the design of radio interface and network architecture,” IEEE Commun.
high reliability. Even with a high confidence level of γ = Mag., vol. 55, no. 2, pp. 70–78, Feb. 2017.
99.9% as shown in Fig. 6, M = 3 is the optimal choice [17] Y. Polyanskiy, H. V. Poor, and S. Verdu, “Channel coding rate in
because of the tradeoff between blocklength (AoI) and PER the finite blocklength regime,” IEEE Trans. Inf. Theory, vol. 56, no. 5,
pp. 2307–2359, May 2010.
(reliability). This is different from conventional ultra-reliable [18] K. F. Trillingsgaard and P. Popovski, “Downlink transmission of
systems, which only consider PER, and the more packets to short packets: Framing and control information revisited,” IEEE Trans.
be jointly coded, the higher the reliability. Our results provide Commun., vol. 65, no. 5, pp. 2048–2061, May 2017.
[19] G. Durisi, T. Koch, and P. Popovski, “Toward massive, ultrareliable, and
new insights when designing ultra-reliable systems with AoI low-latency wireless communication with short packets,” Proc. IEEE,
requirements. vol. 104, no. 9, pp. 1711–1726, Sep. 2016.

You might also like