WiChronos : Energy-Efficient Modulation for Long-Range,

Large-Scale Wireless Networks

Yaman Sangar Bhuvana Krishnaswamy
University of Wisconsin-Madison University of Wisconsin-Madison
sangar@wisc.edu bhuvana@ece.wisc.edu

ABSTRACT processes in the soil, which can, in turn, be used to improve soil
Wireless communication over long distances has become the bottle- health and crop growth. However, due to the high infrastructure
neck for battery-powered, large-scale deployments. Currently used costs for solar panels and batteries to enable wireless information
low-power protocols such as Zigbee and Bluetooth Low Energy retrieval, monitoring sensor networks are deployed only at lim-
have limited communication range, whereas long-range communi- ited locations and have latencies ranging from one day up to one
cation strategies used in cellular and satellite networks are heavy on month [5, 6]. Such monitoring systems typically deploy sensors in
energy consumption. Methods that use narrow-band communica- large fields and share the following requirements: (1) Long battery
tion such as LoRa, SigFox, and NB-IoT have low spectral efficiency, life, (2) Long-range communication to reduce infrastructure costs,
leading to scalability issues. The goal of this work is to develop a (3) Scalability and co-existence, (4) Low cost, with less stringent
communication framework that can satisfy the following require- demands on data rate, latency, and payload size compared to tradi-
ments: (1) Increased battery life, (2) Longer communication range, tional networks For example, sensors used to measure soil moisture,
(3) Scalability in a wireless network. In this work, we propose, de- the chemical and biological properties of soil typically have a reso-
sign, and prototype WiChronos, a communication paradigm that lution of 8-12 bits [7–9], offering a light payload. Also, the gradual
encodes information in the time interval between two narrowband change in the rate of these parameters results in reduced demands
symbols in order to drastically reduce the energy consumption in for data-rate and latency. However, the monitoring systems are
a wide area network with a large number of senders. We leverage often deployed in harsh, remote environments on a massive scale
the low data-rate and relaxed latency requirements of such applica- and are expected to have battery life in the order of years.
tions to achieve the desired features identified above. Based on our Current energy-efficient protocols such as Zigbee [10], black-
prototype using off-the-shelf components, WiChronos achieves an tooth Low Energy (BLE) [11], and backscattering [12–15] consume
impressive 60% improvement in battery life compared to state-of- less than 50mW power during active transmission, allowing for a
the-art LPWAN technologies at distances of over 800 meters. We long battery life. However, their communication range is on the
also show that more than 1000 WiChronos senders can co-exist with order of 10s of meters, and they require infrastructure support for
less than 5% probability of collisions under low traffic conditions. coverage. On the other hand, approaches for long range such as
cellular and satellite networks are expensive and energy consuming,
ACM Reference Format:
limiting the battery life and scalability of the system. In Table 1,
Yaman Sangar and Bhuvana Krishnaswamy. 2020. WiChronos : Energy-
Efficient Modulation for Long-Range, Large-Scale Wireless Networks. In The we compare the battery-life, communication range, scalability, and
26th Annual International Conference on Mobile Computing and Networking achievable data-rate of existing wireless technologies. Evidently,
(MobiCom ’20), September 21–25, 2020, London, United Kingdom. ACM, New existing strategies for addressing low-power and long-range do not
York, NY, USA, 14 pages. https://doi.org/10.1145/3372224.3380898 co-exist efficiently in a large-scale network. This inefficiency can be
explained by the impact of bandwidth (BW) on energy consump-
1 INTRODUCTION tion, communication range, and network capacity. In an additive
Wireless data delivery is key to real-time data collection and analy- white Gaussian noise channel, the maximum channel data-rate is
sis in the fast-growing areas of smart agriculture [1, 2], livestock given by,  
monitoring [3], and precision farming [4]. Large-scale soil mon- S
C = B log2 1 + , (1)
itoring networks currently use cellular and satellite networks to N0 B
collect real-time data over a long period of time [5]. Real-time where B is the signal BW in Hz, S is the signal power and N 0 is the
data collection from in-situ sensors in fields throughout the crop noise spectral density [16]. A wideband signal can achieve a high
season will provide an understanding of the spatio-temporal dynam- data rate, in-turn reducing the time-on-air (the amount of time a
ics of environmental factors and various chemical and biological channel is occupied). With reduced time-on-air, the active transmit
time of the RF module decreases, in-turn reducing energy consump-
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed tion. However, the signal-to-noise ratio (SNR), NS0 B , decreases with
for profit or commercial advantage and that copies bear this notice and the full citation an increase in BW, since the noise power increases with BW. At
on the first page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, long distances, the received signal power decreases due to path
to post on servers or to redistribute to lists, requires prior specific permission and/or a loss; hence a lower BW can decrease noise power and increase
fee. Request permissions from permissions@acm.org. the SNR. Therefore, a narrowband (NB) signal is more suitable for
MobiCom ’20, September 21–25, 2020, London, United Kingdom
© 2020 Association for Computing Machinery.
propagation over long distances with an acceptable SNR.
ACM ISBN 978-1-4503-7085-1/20/09. . . $15.00 1 1
https://doi.org/10.1145/3372224.3380898 Energy ∝ Time-on-air ∝ , Range ∝ , Scale ∝ B.
B B
MobiCom ’20, September 21–25, 2020, London, United Kingdom Yaman Sangar and Bhuvana Krishnaswamy

Battery
Range Scalable Data-rate
Life
(> 1 km) ( > 1000) (kbps)
(> 1 yr)
WiFi X X X <54k
Cellular X X  <10k
BLE  X X <2k
Zigbee  X X <250
Passive RFID [17]  X X <40
Active RFID [18]  X X <40
AmbientBackscatter [13]  X X <1 Fig. 1: WiChronos illustration
LoRaBackscatter [15]   X <37.5 the impact of BW on the time-on-air, without affecting the commu-
pLoRa [14]   X <.0002 nication range. In the case of traditional modulations such as LoRa
LoRa   X <27 or SigFox, the data value of 678 is mapped to bits, which are then
SigFox   X <.6 mapped to symbols for the corresponding modulation technique
and transmitted over the air and hence the time-on-air is propor-
WiChronos    <1 tional to the payload length. Other time interval based modulations
Table 1: Comparison of current wireless technologies and such as M-PPM transmit multiple pulses over air; the time slots in
WiChronos which the pulses are sent indicate the message. In order to achieve
This conflicting requirement on bandwidth makes it challenging a higher datarate, lower latency, and reduce clock skew errors due
for low-power and long-range solutions to co-exist. to imperfect slot synchronization, the number of slots per frame is
LPWAN (Low Power Wide Area Network) technologies such typically set low. Therefore, the number of pulses is high for longer
as LoRa [19], SigFox [20], NB-IoT [21] have been successful in payloads, in turn increasing the time-on-air.
achieving long range [22, 23] and long battery life by limiting the Towards implementing WiChronos in a wireless network, we
duty-cycle (the fraction of time during which a node is in active identify the challenges in designing an optimal anchor symbol
transmit or receive mode) by reducing the number of messages per in terms of length, modulation parameters, and medium access
day. In other words, existing LPWAN solutions reduce the overall control (MAC) in §4. We prototype WiChronos using inexpensive
energy by limiting the airtime of each device in a day. They do off-the-shelf radio modules and microcontroller units (MCU), using
not address the fundamental tradeoff between low-power and long- software-only changes. Based on our experimental implementation
range. Therefore, as the network scales, their cumulative airtime on MSP 430 [25] and Linx-NT [26], we achieved an impressive
will increase, leading to an increase in the probability of collisions, battery life of over 5 years using a coin cell battery (250mAh) at a
rendering them unsuitable for large-scale deployment. range close to 1 km. Also, we estimate the probability of collision to
In this work, we develop a communication framework that ad- be less than 5% in a 1000 node network under low traffic conditions.
dresses the fundamental trade-off between low power and long To this end, we make the following contributions:
range in large-scale networks that have relaxed data-rate and la- • We propose an energy-efficient modulation technique that en-
tency requirements. We propose WiChronos, a modulation technique codes information in the time interval between two symbols,
that encodes information in the time interval between two narrow- minimizing the number of symbols per message.
band symbols. A WiChronos sender transmits exactly two anchor • We propose a spectrally-efficient physical layer design with UNB
symbols per message, namely preamble and postamble, and the data anchor symbols, improving the receiver sensitivity and range.
modulates the duration between them. We leverage the low data • We implement an ALOHA-based MAC protocol and analyze its
rate and small payload requirements of monitoring systems in agri- robustness to collisions in large-scale deployments, allowing the
culture, remote tracking, and derive benefits from the availability network to scale seamlessly.
of resources at the receiver to make transmitters energy efficient. • We prototype the proposed framework using off-the-shelf, low-
WiChronos achieves energy efficiency by minimizing the num- cost RF modules and MCUs with low-power clocks.
ber of symbols and (hence the time-on-air) per message, and long The remainder of this paper is organized as follows. In §2, we pro-
range by transmitting the anchor symbols over ultra-narrowband vide background on the strategies for low power and long range,
(UNB) [24]. The reduced time-on-air and improved spectral effi- and motivate the need for a unifying algorithm in a large-scale
ciency allow the network to scale better. WiChronos trades off network. In §4, we present the design and features of WiChronos,
data-rate to achieve the above three features. followed by a discussion on the challenges and throughput perfor-
Fig. 1 illustrates WiChronos and contrasts it with existing modu- mance in §5. Experimental evaluation is presented in §6. Related
lations. A WiChronos sender with a sensor data of 678 units trans- work is discussed in §3, followed by discussions and future work in
mits an anchor symbol (preamble), goes to sleep for 678 clock cycles §7. Finally, §8 concludes the paper.
(that corresponds to data), and wakes up to send another anchor
symbol (postamble). The anchor symbols consist of a training se- 2 BACKGROUND AND MOTIVATION
quence of alternating 0s and 1s followed by a unique sender address.
The overall power consumed by a sensor node is dominated by
The sleep time encodes the entire message to be transmitted. The
the communication module [27, 28]. Algorithms, protocols, and
time-on-air of WiChronos is thus independent of the payload length
architectures have been developed to reduce the active transmit
(10 bits in this example) and only a function of the anchor symbols;
time and hence reduce the power consumption in the areas of
by separating payload transmission from time-on-air, we reduce
WiChronos : Energy-Efficient Modulation for Long-Range, Large-Scale Wireless Networks MobiCom ’20, September 21–25, 2020, London, United Kingdom

Wireless Sensor Networks (WSN), AdHoc Networks, and Internet- number of messages increases, the overall energy consumption in-
of-Things. We broadly classify existing low power strategies for creases, making them inefficient for an energy-constrained system.
WSNs into the following five categories [4, 29–31]: (1) Duty Cycling, Also, with increasing scale, collision rate increases, which leads to
(2) Routing, (3) Data reduction, (4) Radio module optimization, a decrease in the network throughput [19–21].
(5) Energy harvesting and backscatter. To summarize, existing energy-efficient and long-range strate-
Smart sleep/wake-up protocols [32, 33], event-triggered wake- gies cannot co-exist in a large-scale network. In this work, we
up [34], and scheduled MAC protocols [35, 36] have been pro- develop a communication framework for real-time monitoring sys-
posed to reduce the active transmit and/or receive time of the tem that is energy-efficient, communicates over long distances, and
radio transceiver. Though duty-cycling-based approaches reduce supports a large number of nodes.
the power consumption of the overall network, they are still limited
by the payload length. Data compression [37], prediction [38, 39], 3 RELATED WORK
and cooperative communication [40, 41] techniques have been pro- Timing interval based communication has been studied in the past.
posed to reduce the amount of data to be transmitted, which in We present an overview of these works below.
turn reduces the energy per message. Network topology aware Information Theoretic Analysis : An information theoretic anal-
strategies have been developed to improve the lifetime of a sensor ysis of encoding information in the queuing time has been studied
network through clustering [42], energy-aware routing [43], data in literature [53–57]. While bounds on channel capacity of single-
gathering, and data forwarding [44]. These approaches decrease queue and multiple-queue servers worst-case timing error distribu-
the overall energy consumption of the network by leveraging the tions have been studied, they do not provide algorithms for energy
hierarchical topology and varying the energy constraints of nodes efficiency and/or long-range communication in a wireless system.
in the network [45, 46]. A new wave of battery-less sensors harvest The maximum achievable data-rate of WiChronos aligns with the
energy from ambient signals and/or dedicated sources [13, 15] and capacity of timing channel derived in [53].
use backscattering to communicate. However, these techniques are Timing Channel: Timing based communication has been studied
feasible only within a short range and require an additional signal as a security risk in covert timing channels [58–60], where infor-
source to piggyback on for long-range communication [14, 15]. mation is encoded covertly in the time interval. The majority of the
Active and passive RFID tags [17, 18] are energy efficient but only research work on timing channels has focused on defense mecha-
work for a short range [47] and/or low traffic. RFIDs also are limited nisms to eliminate timing channels.
by the energy-range-scale tradeoff presented in §1. Pulse Time Modulation: This work is motivated by Pulse Posi-
As discussed in §1, the limited range of existing energy-efficient tion Modulation (PPM) and Communication through Silence [61],
strategies can be attributed to the BW and propagation of radio a challenges paper on timing based modulation in wireless sensor
frequency (RF) signals. The range of an RF signal depends on the networks. [61] does not consider long-range and scalability in wire-
link budget that accounts for the gains and the losses. less networks. It does not consider the challenges in the practical
P Rx = PT x + PG − P L , (2) implementation of timing interval modulation in wireless networks
where P Rx is the received power, PT x is the transmit power, PG is such as anchor symbol length, synchronization, and packet errors.
the antenna gains of the system, and P L is the aggregate loss due to To the best of our knowledge, WiChronos is the first attempt at
filter/cable attenuation, known experimental conditions, and path implementing timing interval modulation technique for long-range,
loss (P P L ). Free space path loss is given by energy efficient communication in wireless networks.
λ2 d PPM and its variants such as Differential PPM [62, 63], M-ary
P P L = −10 log10 = 22dBm + 20 log10 , (3) PPM [64–68] have been developed and widely used in one-to-one
(4πd)2 λ
optical networks with ultra wide bandwidth and narrow pulses.
where d is the distance between the sender and the receiver, and
The following requirements of typical PPM implementations make
λ the wavelength. Radio optimization techniques to improve PT x
them challenging for wireless networks.
such as directional antennas [35], Multiple-Input Multiple-Output,
Bandwidth requirements: PPM and its variants typically use an
and mobility-based solutions such as Mobile Ad-Hoc Networks [48–
ultra-wide bandwidth [64] to get a narrow pulse to achieve rea-
50] are energy consuming and not suitable for low-power commu-
sonable data rates while also saving energy. Ultra wide band will
nication. Given the FCC limit on transmit power, the link budget
result in an increase in energy consumption and noise floor in a
depends on the path loss, which is inversely proportional to the
wireless module, affecting both battery life and range. To overcome
wavelength (i.e., path loss increases with increasing frequency).
this challenge, we use narrow bandwidth and achieve long range.
Existing technologies for long-range such as cellular networks,
Clock Synchronization: In UWB, the pulses are short and there-
NB-IoT, Sigfox, and LoRa operate in the sub-GHz bands to reduce
fore symbol synchronization is required for PPM. DPPM, however,
path loss. Cellular and NB-IoT require high-complexity RF front-end
overcomes this challenge by encoding in the time difference be-
and LTE infrastructure for scheduling and synchronization [51].
tween pulses [62]. In this work, we relax synchronization require-
SigFox and LoRa operate at lower BW to achieve long range,
ments with the help of differential modulation and low data-rates.
leading to an increase in the time-on-air. Similarly, Zigbee 3.0 [52]
PPM implemented in one-to-one optical links does not require a
reduces datarate and increases transmit power to improve range
dedicated MAC protocol. However, in a wireless network, the pulses
inturn increasing time-on-air. To achieve long battery life, LPWAN
collide on air and therefore dedicated addressing and MAC protocol
solutions limit the number of messages per day. Therefore, as the
is required [69–72]. Multiple transmitters, therefore, cannot share
MobiCom ’20, September 21–25, 2020, London, United Kingdom Yaman Sangar and Bhuvana Krishnaswamy

1 1
Active Current Idle Current Sleep Current )%.%'%2"

%(",
).",,/+.- " "%0"
(mA) (mA) (nA)
MSP430 [75] 1 0.7 45
MCU 
6
")-*,
STM32L[76] 0.45 0.3 140 %(",
*
) $*,
Linx-NT [26] 42.5 42.5 1 %)#* 6
7
 7  

Tx
CC1101 [77] 27.4 27.4 200
%(",6%)#*.+5
Linx-NT 22.5 22.5 1



*-.('"
Rx ")!",
)!"1

&
CC1101 15.4 15.4 200 

,"('"
,"('" *-.('"

Table 2: Current consumption of individual modules 

.*+%(",&

.,.%(",&
the channel without incurring packet losses or increasing energy ")! ,1 6
%(",&


consumption [73, 74]. In this work, we remove the need for MAC
by uniquely identifying the transmitter in each pulse.

%)#* 6

,1 3 .+5

'""+
Despite its energy efficiency, PPM is not popular in wireless
%)#* 7

6

owing to the above challenges. We propose WiChronos, a special 
).",,/+.
",0% "
*/.%)" 7 
case of the more general pulse position modulation for wireless
Fig. 2: System implementation flow diagram
communication and design the modulation to cover a long range
and accommodate a large scale. In §6, we will compare the energy V and given δt, the current drawn, which determines the energy
efficiency, range, and scalability of M-PPM with WiChronos. consumption can be classified into six categories as,
(1) Ia−u : Active - MCU (5) Ir −r f : Active listen -
4 WICHRONOS: INTERVAL MODULATION (2) Is−u : Sleep - MCU RF module
In the majority of the digital modulation techniques used in wire- (3) Ii−r f : Idle - RF module (6) It −r f : Active transmit -
less networks, data modulates the characteristics of the carrier (4) Is−r f : Sleep - RF module RF module
signal [78]. For e.g., 2-FSK maps bit 0 and bit 1 to two distinct Table 2 enumerates the above currents for selected OTS hard-
frequencies, transmitting 1 bit per symbol. The total number of ware. The active and sleep currents of MCUs are few mA and μA
symbols per message, therefore, depends on the modulation tech- respectively, as noted in row 2 of Table 2. It −r f depends on the
nique and the message length. Higher-order modulation techniques transmit power of the RF module and the protocols implemented
such as 64-QAM can encode more bits per symbol, but have larger in the transceiver. The active currents of RF module are on the
energy per bit, Eb while lower-order modulations such as 2-FSK order of tens of mA. Therefore, an energy-efficient transmitter must
have lower Eb but an increased number of symbols per message. minimize the amount of time spent in active mode. By encoding
A modulation technique that reduces the number of symbols per information in the time interval between symbols, we reduce the
message without increasing Eb is required to achieve energy efficiency. active transmit time of the RF module and the MCU. The energy
With this insight, we propose WiChronos, a modulation technique consumption is further reduced by operating the MCU and the RF
that encodes information in the time interval between two anchor module in sleep mode during information transfer (clock count).
symbols. Fig. 1 illustrates an example with an integer data 678 from Fig. 2 details our implementation in MSP430 FR2355. On reading
a 10-bit sensor. The transmitter sends a predefined preamble and analog data from the sensor, the transmitter converts it to a digital
waits for 678 clock cycles and then a predefined postamble. The value (D), maps to clock cycles Tinf o (with redundancy to correct
receiver, with apriori knowledge of the anchor symbols of each for timing errors) and sets TimerA to Tinf o . The MCU triggers the
transmitter, identifies the sender from the preamble and triggers RF module to send a preamble and then set the RF to sleep. The MCU
the corresponding timer to start (or stop) on the reception of the itself goes to sleep mode, with TimerA running in the background.
preamble (or postamble). In the absence of any timing errors and After Tinf o cycles, TimerA interrupts the MCU to turn on the RF to
prior knowledge of clock rate, the number of clock cycles counted send postamble. Since Tinf o is the time difference between the sym-
at the receiver is equal to that at the transmitter. bols, the transmitter and receiver clocks need not be synchronized.
The total number of symbols transmitted on-air and the energy It is sufficient to use the same clock rate. Since the RF and MCU
consumed by the RF module depends only on the length of the modules are in sleep mode during data transmission (Tinf o ), the
anchor symbols and is not affected by the length of the sensor overall energy consumption is unaffected by the message length.
output (10-bits in this example). WiChronos hence satisfies the two The energy consumption of data transmission is,
requirements identified earlier for energy-efficient modulation i.e., E = V (Ia−u + It−rf ) 2tanchor + V (Is−u + Is−rf )Tinf o (4)
(1) Minimum number of symbols, (2) Non-increasing energy per bit
where tanchor denotes the duration of anchor symbols. The overall
Eb . In the rest of this section, we discuss the challenges in achieving
energy consumption is therefore determined by the modulation
the promise of WiChronos and propose solutions to address them.
parameters used to transmit the anchor symbols, the anchor symbol
length, and the corresponding current drawn. Optimum anchor
4.1 Energy Efficiency
symbol length is crucial to achieving the promised energy efficiency.
A key insight for the energy-efficient design of WiChronos is that
the RF module is the bottleneck of a sensor node [28]. Energy 4.2 Anchor Symbol Design
consumed by a sensor node is given by V · I · δt, where V is the Predefined preambles and sync words are commonly used in com-
voltage, I the average current drawn for a duration δt. For a constant munication systems to indicate the start of a packet, as well as
assist the receiver with frame (and/or bit) synchronization [78].
WiChronos : Energy-Efficient Modulation for Long-Range, Large-Scale Wireless Networks MobiCom ’20, September 21–25, 2020, London, United Kingdom



1

Probability of Anchor Symbol Error

0.5 False Positive 50 50 

  
Probability of False Positive

Anchor Symbol Error  

0.4 

0 0 
  

0.1 SNR

Power (dBm)
 

SNR (dB)
0.3 Received Power
Noise Floor
-50 -50 
0.2

0.01



  

X
0.1 -100 -100  

0.0 
 

 
-150 -150  
0.001
0 2 4 6 8 10 12 14 16 18
100 1k 10k 100k 1M
Anchor Symbol Length (bits) 
Bandwidth (Hz)
Fig. 3: Anchor symbol length design Fig. 4: Impact of bandwidth on Fig. 5: Collisions in the network using
receiver sensitivity ALOHA
The anchor symbols play the same role; indicate the start of a mes- and AGC can further reduce the settling time and false positives,
sage (to start and stop clock counting) and assist the receiver to leading to shorter anchor symbols.
identify the sender. Since we specifically focus on networks with
relaxed data-rate requirements, the anchor symbols are transmitted 4.3 Long-Range Communication
at the order of few kilobits per second over narrow BW. At such The second challenge identified in this work is achieving long-range
low data-rates, the time and the number of bits required for the Au- within the energy constraints. As discussed in §2, propagation loss is
tomatic Gain Control (AGC) in the RF front end to settle, is reduced. directly proportional to the distance and the frequency of operation.
Further, UNB communication requires much shorter preamble than Following the IEEE 802.11ah standard [81], we operate in the sub-
coded modulations [79]. For example, TI CC1125 [80] uses an AGC GHz spectrum and narrow bandwidth (NB) for long-range. The
design with a 4-bit settling time along with smart carrier sensing range also depends on the receiver sensitivity (the minimum received
to reduce false positives. The lower bound on the anchor symbol power required to demodulate the signal with an acceptable BER),
length is thus determined by the RF front end and the modulation a characteristic of the receiver. For a given propagation channel
parameters. In order to differentiate between the start counting and frequency of operation, the received power from Equation 2
and stop counting triggers, each sender is assigned two unique depends on the transmit power, antenna gains, and the channel.
addresses i.e., each sender has a unique preamble and postamble. SNR depends on the received power and the noise floor, given by,
We define the optimum anchor symbol length to be the shortest Noise floor = −174 + N F + 10 log10 BW . (6)
length that enables reliable decoding at the receiver i.e., it must
Here N F is the noise figure (the ratio of input SNR to output SNR
minimize the probability of error in decoding and has a low false-
at the receiver) and BW is the receiver BW [35]. Fig. 4 shows SNR
positive rate. For a given modulation, SNR, and the corresponding
and N F as a function of BW . For a given received power, the noise
BER pb , the probability of symbol error and false-positive are,
e   ϵ   floor increases with receiver BW, in turn decreasing the SNR and
N k N −k N N the receiver sensitivity. NB and UNB modulations are capable of
Pe = · pb q , P f = (0.5) (5)
k i=0
i long-range due to their improved receiver sensitivity and SNR from
k=0
where e is the maximum number of bit errors that can be tolerated reduced BW. However, decreasing the receiver BW implies reduced
by the receiver for an N-bit anchor symbol. Fig. 3 plots the above transmit BW, leading to an increase in time-on-air and energy
probability of symbol error Pe and probability of false positive P f consumption. Thus, even in low data-rate applications, wireless
as a function of anchor symbol length for e=0. Pe increases with systems operate at higher datarates to save battery life. To overcome
symbol length; therefore, shorter anchors are desirable to reduce the impact of narrow BW, existing LPWAN strategies limit the
symbol error while P f decreases with symbol length. With short number of messages to reduce the total time-on-air [82].
symbols, the probability of a random bit pattern from other devices Due to its reduced time-on-air WiChronos can operate in UNB
or interference being falsely detected as the anchor is high. We without significantly increasing the energy consumption. The increase
identify the smallest anchor symbol length that renders low P f and in time-on-air from UNB transmission of the anchor symbols is
Pe for a given anchor symbol modulation, expected SNR, desired much smaller than that of the entire message (as in other LPWAN
false positive rate, and symbol error rate. In this work, we use techniques). In our evaluation, we modulate the anchor symbols
BFSK to modulate the anchor symbols with an SNR ≥ 10dB, BER using BFSK in the 902-928MHz spectrum with a BW of 100kHz.
−3
of ≤ 10 . We choose 10-bit anchor symbol to achieve P f ≤ 0.01 RF modules such as TI CC1101 [77], CC1120 [83] can allow us
and Pe ≤ 0.001. It must be noted that this shortest anchor symbol to reach 10s of km with much lower BWs. Operating at the 902-
length is required to transmit each pulse in any variant of PPM. 928MHz spectrum also offers the advantage of mitigating multipath
The anchor symbol length calculation is unaffected by the receiver reflections and wideband interference. Devices operating in NB
since we leverage the resource asymmetry in infrastructure mode and UNB are unaffected by multipath as they experience no inter
and let the receiver to always be in listen mode. symbol interference [84, 85]. This is a physical layer phenomenon
The minimum anchor symbol length is determined by the settling and we discuss this in our future work as an area of further analysis.
time of the RF module, the probability of false positives, and the prob-
ability of symbol error. RF modules with advanced carrier sensing
MobiCom ’20, September 21–25, 2020, London, United Kingdom Yaman Sangar and Bhuvana Krishnaswamy

4.4 Medium Access Control for SigFox and LoRa are 46.3% and 12.75% compared to WiChronos
The third challenge identified in this work is co-existence in a large- at 1.2% for a BW of 100kHz and 1.8% for 600Hz.
scale deployment. In spite of the vast research on MAC protocols for With increasing network size, the anchor symbol length will in-
large-scale [33, 86], ALOHA-based algorithms are implemented in crease as log function of network size, since each sender is assigned
LPWANs due to its simplicity. ALOHA allows a sender to transmit a unique preamble and postamble as the local address. For example,
whenever it has data without any coordination between the senders. a 2-byte anchor can address a maximum of 211 senders each with
16
The simplicity of ALOHA makes it vulnerable to collisions in large- a unique 16-bit preamble and postamble ( 22 ) and 4 initial bits for
15
scale and/or heavy-traffic deployments. A packet collision occurs if RF settling ( 224 ). The trade-off between network scale and energy
one sender begins transmission during an on-going transmission efficiency will determine the anchor symbol length in practice.
i.e., twice the data transmission period is vulnerable to collisions.
LPWANs using ALOHA-based MAC protocols suffer from collisions 5 ACCURACY-THROUGHPUT TRADEOFF
in large-scale networks when the channel is used for transmission The promise of long battery life and range in a large-scale is achieved
more than 20% of the time (on average) [78]. at the cost of datarate. WiChronos achieves energy efficiency by
WiChronos (by design) has low time-on-air and leverages the offloading the communication burden to timers at the sender and
simplicity of ALOHA without getting penalized by collisions. Data the receiver and hence the data-rate of a link depends on the
transmission time and time-on-air are different for WiChronos, clock rate at the transceiver. Consider an MCU with a clock rate
unlike existing digital modulations. As shown in Fig. 5,in cur- fc = 32.768 kHz. If increasing clock cycles each represent a value, a
rent systems, the entire transmission is vulnerable to collisions, total of 32768 values can be conveyed in 1 s i.e., 15 bits (loд2 (32768))
whereas, in WiChronos, only the anchor symbols are vulnerable, can be communicated in one second(plus anchor symbol time). To
decreasing the probability of collisions. During the data transmis- generalize, the data-rate of a WiChronos link is given by
sion of one sender (Tinf o ), the other sender can transmit their d · fc
anchor symbols without affecting the on-going transmissions. We Rb = . (8)
Tinf o
derive the probability of collisions given a node is transmitting a
message following the textbook approach used in the analysis of Tinf o , the time to transmit a d-bit message is in the range of [0, 2d −
ALOHA [87]. We assume a Poisson arrival process with a cumu- 1]. For a uniformly distributed data source where the values of Tinf o
lative arrival rate of λ. Given a sender started a transmission, the are equiprobable, the expected value (average) of Tinf o = 2d −1 .
event is a success if no other anchor is sent within 2tanchor . The achievable data-rate is lower than that of LPWAN tech-
Pr(collision) = 1 − Pr(no other event in 2T  ) nologies with comparable range. The numerator of data-rate in
 Equation 8 is a linear function of payload length while the denomi-
= 1 − e −λ 2 T , where T’ = tanchor (7) nator is an exponential function. This is in contrast with existing
Fig. 6 plots the probability of collision for a given node transmit- modulation techniques where the data-rate is a constant that is
ting a 1-byte message every ten minutes as a function of the network independent of the payload length. WiChronos is thus not suitable
size. For SigFox, we assume its default overhead of 11 bytes at a for applications with high data-rate and low latency demands.
datarate of 600bps and the shortest preamble allowed for LoRa with
a spreading factor (SF) of 7, code rate 4/5 and 125kHz BW [88], 5.1 Maximizing Data-rate
operating in implicit header mode without CRC. We turn off CRC We present two strategies to improve our data-rate performance:
and MAC overheads, as well as ACKs in LoRa for a fair comparison. (1) Accurate higher clock rates (2) Efficient Time encoding. As
An arrival rate of 2λ is used in the calculation for WiChronos to shown in Equation 8, the data-rate of WiChronos is directly propor-
account for the two anchor symbols. tional to the clock rate fc at the transmitter. Increasing fc reduces
With an increase in the number of sensors, the probability of the time between the anchor symbols for the same number of clock
collision for SigFox is the highest among compared technologies, cycles and improve datarate. The accuracy of the clock plays a sig-
due to its very low data-rate and very high time-on-air. The above nificant role in improving data-rate. Commercially available ultra
parameters for LoRa are chosen for low-power and do not offer low-power crystal oscillators [89], and MEMS clocks [90] offer high
long range. For longer range, LoRa recommends higher SF, which accuracy, high rate clocks. We will discuss the impact of clock skew
will lead to increased time-on-air, thus increasing the probability on data-rate and accuracy later in the section.
of collision. The rate of increase of probability of collisions with The second strategy for improving data-rate is aimed at minimiz-
network scale is slower for WiChronos due to its reduced air-time. ing the average wait time between anchor symbols using apriori
It can achieve energy efficiency without limiting the number of mes- knowledge of the source distribution, analogous to source coding.
sages and/or the network size.. As shown in Fig. 6, reducing BW The expected (average) time to transmit a d-bit message is,
increases the probability of collision due to increase in symbol du- d −1
2
ration. However, this increase in collision is smaller for WiChronos
E = i Pr(v(i)),
as only the anchor symbol duration contributes to collision as op-
i=0
posed to SigFox or LoRa where the entire payload increases the
where Pr(v(i)) is the probability of sensing a data value v(i). Mini-
collision probability. Therefore, we can leverage NB and UNB to
mizing E will minimize the average time to send a value between 0
improve range without significantly affecting scalability and energy
and 2d − 1, which will maximize the data-rate. This can be reduced
efficiency. In a network of 1000 nodes, the probability of collision
to a continuous Knapsack problem [91], where the data values
WiChronos : Energy-Efficient Modulation for Long-Range, Large-Scale Wireless Networks MobiCom ’20, September 21–25, 2020, London, United Kingdom

1

$ "
&$#  #$
&$#
       
Probability of Collision

"#$$" "#$$"
0.1
!%! !%!
   
   

%""$
 "#$$"
#!  "#$$"

#  μ # #!
0.01    μ
# #
Sigfox, 600Hz 
WiChronos, 600Hz
LoraWAN, 125kHz
WiChronos, 100kHz
0.001
0 200 400 600 800 1000
 
  

Number of Nodes 
Fig. 6: Probability of collisions Fig. 7: Timing components of WiChronos sender
must be assigned a time value such that E is minimized. It has been 5.2.2 Processing timing error . In Fig.7, let tp be the average time to
proven that the optimum solution for continuous knapsack problem send an anchor symbol. tp is composed of the time to perform the
is achieved using greedy algorithm [91] i.e., arrange all possible operations in MCU such as ADC, timer setup, serial communication,
sensor data in decreasing probability of occurrences and map to enter and exit sleep mode, radio wake-up time, modulation, and
increasing time values in [0, 2d − 1]. Mapping the most recurring transmission. Variations in the number of cycles to execute any of
data to the shortest time, we optimize the data-rate performance. these operations can vary tp in the range [−ϵ, +ϵ]. We assume that
the operations in the MCU are deterministic and do not contribute
5.2 Timing Error Correction to ϵ. The radio wake-up time can vary within a bounded range
The maximum achievable data-rate of a WiChronos link relies ( [26].) We correct for this timing error by triggering the data timer
on the accuracy of clock rate, processing time, data transmission, after tp + ϵ. On reading a sensor data of D, the MCU sets a transmit
and the propagation delay. Fig.7 presents a detailed timeline at a counter for tp + ϵ cycles and triggers the radio module to transmit
WiChronos transmitter. The currents and the duration listed are the preamble. At the end of the transmit counter, the MCU and the
specific to the MCU and radio used in our prototype. By default, the radio are set to sleep mode for D cycles. By increasing the transmit
radio and the MCU are in sleep mode. The total time to send data time from tp to tp + ϵ, the transmitter removes the variability in
includes the radio wake-up time, the anchor symbol transmit time, the preamble (and postamble) transmit time. The receiver counts
and the information time. Timing errors in one or more of these the number of cycles between the preamble and postamble and
components will affect the received Tinf o , leading to bit errors. subtracts (tp + ϵ) to obtain Tinf o . We correct for variations in tp
We broadly categorize bit errors in received data into the follow- by increasing the overall time to transmit as, tanchor ≥ tp + ϵ. For
ing categories, (1) errors in the anchor symbol, (2) anchor symbol timing errors bounded by ϵ, the increase in anchor symbol duration
loss, (3) processing and propagation time errors, (4) clock counting from tp to tp + ϵ can achieve 100% error correction.
error. Existing error detection and correction mechanisms are de- 5.2.3 Clock cycle error. Timing errors due to clock skew at the
signed for bit errors and require rethinking for timing errors. We transceivers can lead to incorrect Tinf o and hence bit errors. We
propose a simple error detection mechanism for anchor symbol propose a simple error correction mechanism that spreads out the
loss and correction mechanisms for bounded timing errors. data value (in time) to account for the clock skew. Consider the
5.2.1 Anchor symbol error and loss. Though we choose the receiver 32kHz crystal oscillator used in MSP430 FR2355. Though the quartz
bandwidth and anchor symbol length to minimize BER, there is crystal oscillators have high stability, it has a tolerance of ± 30ppm,
a non-zero probability for anchor symbol error or loss. Existing resulting in an error of ± 0.9Hz in room temperature [89]i.e., the
coding techniques such as Hamming codes will be used to correct measured clock cycle can vary by ≈ 1 clock cycle. To correct for this
for single bit errors in the anchor symbol. Anchor symbol loss is clock counting error, we assign data value D to clock cycles that are
detected using (1) timeout, (2) stateful receiver. We use the prior separated by 3 i.e., the minimum difference between two adjacent
knowledge of the maximum payload length and the processing Tinf o transmitted over air is set to three, to correct for variations
time at the sender to detect anchor loss using timeout. For each in the clock cycles measured. To generalize, when using a clock
sender, the timeout value is set toTO = tanchor +Tinf o−max , where rate with a tolerance of ±δ , the minimum difference between two
tanchor is the sum of maximum processing and transmission time transmitted clock cycles is set to δ . Therefore, a sensor data D is
to send an anchor symbol and Tinf o−max is the maximum clock mapped to δ + D · (2δ + 1). Using the above redundancy in time to
cycles mapped to a data. The receiver detects the loss of an anchor correct for clock cycle errors will reduce the effective data-rate since
symbol whenever the counter exceeds TO. The second detection fewer unique clock cycles represent data. Therefore, accurate(low
mechanism maintains the state of the anchor symbol reception for tolerance) clocks are crucial to the design of WiChronos.
each sender. For every anchor symbol received, the receiver verifies 5.2.4 Propagation error. The time to receive an anchor symbol
the previous symbol received and the timer value is stored only if includes the propagation delay of the RF signal. At distances of a few
a postamble is followed by a preamble. Else, the counter value is kms, propagation delays of RF signals traveling at the speed of light
discarded, reset and marked as an error, since recurring postambles are on the order of μseconds. Therefore, changes in propagation
(or preambles) indicate loss of a preamble(or postamble). delays would also be on the order few μseconds. Error correction
mechanism proposed above for clock skew is modified to correct
MobiCom ’20, September 21–25, 2020, London, United Kingdom Yaman Sangar and Bhuvana Krishnaswamy

with SigFox and M-PPM. Easy-to-program, Arduino based inte-

grated LoRa module [92] (RFM69HCW) does not allow for energy

management and has low battery life. We implemented an energy
management module in SX 1272 [93] operated using MSP430; we





reduce its preamble to 6 bytes, remove Cyclic Redundancy Check
 



   and MAC headers, disable ACKs, and set to transmit-only mode
to minimize its energy consumption and provide fair comparison.




Throughout this section, unless otherwise mentioned, we operate


LoRa and Linx at 915.37 MHz center frequency. LoRa is implemented



at a spreading factor of 7, coding rate 4/5 and BW of 125kHz.
As shown in Fig.8, the sender is powered by a battery and the
Fig. 8: Experiment locations and set up receiver is always-on. Both the transceivers use the external clock in
4
MPPM Variant 1 the MSP and commercially available 1/4-wave whip antennas. Our
MPPM Variant 2
WiChronos implementation in MSP shown in Fig.2, circuit design, and hardware
3
setup to recreate our results will be made publicly available1 .
Energy (mJ)

2 6.2 M-PPM Vs WiChronos

Following our discussion in §3, deploying PPM in a wireless system
1 is challenging. Recently M-ary PPM has been considered for low
data rate IOT networks [94]. WiChronos is a special case of the
0 general PPM systems. In this section, we compare M-ary PPM
n=1 n=2 n=3 n=4 n=5 WiChronos
m=65536 m=512 m=128 m=64 m=32 against WiChronos on its energy efficiency, range, and scalability.
Fig. 9: Energy consumption of M-ary PPM and WiChronos MPPM encodes a d-bit message to one of 2d symbols which are
for propagation errors too. A 50% error in propagation delay at a represented by n pulses in one of the M time slots. For example, in
distance of 10 km can lead to an error of ± 0.8 cycles at a clock Fig.1, a 10-bit message is communicated using n=2 pulses in one of
 
rate of 32 kHz. Apriori estimate of channel and propagation model the M=32 slots [ 32 10
2 > 2 ]. By transmitting these pulses over NB
is used to encode redundancy. For bounded error in propagation or UNB, long range can be achieved [94]. In the rest of this section,
delays, we can achieve 100% error correction by further increasing we assume a 100kHz BW for WiChronos and M-PPM pulses and
the minimum difference between adjacent Tinf o . Combining these therefore both the modulations can achieve the same range.
error detection and correction mechanisms, WiChronos can correct Unlike an optical system, a single pulse needs to be long enough
for all bounded errors and detect symbol losses. for a wireless receiver to decode. Thus, the pulse length calculation
follows that of the anchor symbol. We choose a 1-byte sequence
6 PERFORMANCE EVALUATION of alternating 1s and 0s as a single pulse since that is the smallest
length that can be communicated using Linx NT i.e., the shortest
6.1 Experimental setup
MPPM (and WiChronos) pulse is set to 8 bits. The start of a message
We implement WiChronos using Linx NT [26] as RF module and is denoted by a 1-byte pulse, followed by the sender address, similar
TI MSP430FR2355 [75] for MCU as shown in Fig.8. Linx-NT has to a WiChronos anchor symbol. Shorter pulse lengths using other
extremely low sleep current(1nA), and a transparent mode with- RF modules will benefit M-PPM as well as WiChronos. Our imple-
out protocol overheads. The anchor symbols are transmitted at mentation of WiChronos uses 2-byte anchor symbol, where the first
9.6kBaud rate using BFSK over 100kHz bandwidth (BW) with a byte is used for synchronization and the second byte to identify the
frequency deviation of ± 30kHz [26]. The MCU costs $2.72 and the sender. Once an M-PPM receiver decodes the start of a message, it
radio with antenna is $45.8, bringing the overall cost to be < $50. synchronizes the clock and waits for a pulse.The transmitter is in
Other less expensive transceivers such as TI CC1101 and CC1125 deep sleep mode when it isn’t actively transmitting.
which cost below $5 can also be used to implement WiChronos. We implement the following two variants of MPPM that tradeoff
Linx NT is chosen for its simplicity, allowing majority of the pa- energy and scalability. (1) Each pulse is 1-byte i.e. the shortest pulse.
rameters to be set through hardware external pins. As part of our The short pulses consume lower energy, at the cost of MAC over-
future work, we plan to explore various radio modules. heads. Since the receiver cannot differentiate pulses from multiple
We use a 32.678kHz external crystal oscillator for data trans- senders, a dedicated MAC protocol [73] or coding [65, 94] to coor-
fer (counting) as well as processing (MCU). Since information is dinate transmission is required. (2) Each transmitter has a unique
encoded in the time difference, absolute clock synchronization is pulse. This eliminates MAC overheads but renders each pulse longer,
not needed. During system setup, the receiver stores the clock rate like a WiChronos anchor symbol, trading off energy for scalability.
and anchor symbol of each transmitter. To understand the impact Energy Consumption: Fig. 9 compares the energy consumption
of timing errors, we estimate the maximum possible timing error of WiChronos and the above variants of M-PPM, in transmitting a
analytically, verify it experimentally, and show that it is possible 2-byte payload. The energy consumption of WiChronos depends
to correct all bounded errors (as explained in §5.2). We compare only on the length of the anchor symbols, while that of M-PPM
against LoRa experimentally and present analytical comparison
1 https://github.com/Yaman-Sangar/WiChronos
WiChronos : Energy-Efficient Modulation for Long-Range, Large-Scale Wireless Networks MobiCom ’20, September 21–25, 2020, London, United Kingdom

12 10.5
3.0 False Positive
10.0
Symbol Loss
2.5 WiChronos
7.0 10
False Positives (%)

LoRa, SF = 7

Battery Life (Years)

10

Symbol Loss (%)

6.5

Battery Li al nodes
2.0 LoRa, SF = 10

of individu
8
6.0 SigFox
1.5 6
5.5

fe (Years
8 4
1.0 5.0 2
60
0.5 4.5 0

)
0 50
0
0 40
6 20 30
0

N ota

)
es
et l
0

0
0.0

(T

0
w nu

ag
20

ec tw s)
0.5

or m
0

ns be (

ss
ve n
k b

co me load
uti ee
10

Sc er

me
6
0

al of

0
80
1 Byte 2 Byte 3 Byte

e d

(T ffic
00
0 2 4 6 8 10

a
Tr
10

i
Anchor Symbol Length (Bytes)

ev
Payload (Bytes)

ic
es
)
Fig. 10: Experimental verification of Fig. 11: Battery life of WiChronos Vs Fig. 12: Battery life of WiChronos Vs.
anchor symbol length LoRa and SigFox scale and traffic load
increases linearly with the number of pulses n. For n = 2, the energy symbol length and offers room for a further decrease in symbol
consumption of M-PPM is the same as WiChronos and requires length. In the rest of our evaluation, we use 2-byte anchor symbols
M=512 slots to transmit a 2-byte message. While increasing n in- unless stated otherwise. The receiver determines the preamble using
creases energy consumption, it decreases latency. correlation and starts decoding the address byte. A look-up table
Scalability:An energy efficient M-PPM system (variant 1) would on the receiver is used to identify the transmitter.
choose the shortest pulse, which will require MAC overheads and/or
additional coding techniques to reduce collision rate. While a scal- 6.4 Energy Efficiency
able M-PPM (variant 2) eliminates MAC overheads, it increases We evaluate the energy efficiency of WiChronos with 2-byte anchor
energy per pulse (similar to the anchor symbol of WiChronos), symbol and compare with LoRa and SigFox. The energy manage-
increasing the overall energy consumption and the pulse duration. ment module we implemented results in longer radio wake-up time.
Data Rate: The frame duration of M-PPM is fixed for a given pay- We experimentally measured the average radio wake-up time of
load length i.e., M=32, with 2 pulses can encode 10 bits and always Linx to be 4.2ms and that of LoRa ≤ 500μs. In Fig.11, we compare
consumes 32 slots. WiChronos on the other hand, has variable mes- the life of a 250mAh, 3.3V coin-cell battery for varying payload
sage duration between the symbols that depends on the payload. lengths sent every 10 minutes. WiChronos has an impressive 10.25
WiChronos leverages lower data rate to achieve energy efficiency. years life for any payload length as its energy is only a function of
the anchor symbol length. LoRa with SF=7 has a battery life of ≈
6.3 Anchor symbol length 5 years, but decreases with payload length. LoRa with higher SF
In this section, we evaluate the robustness of our anchor symbol used for longer range drastically reduces battery life (7 months at
length design in §4.2. In our evaluation, we design the length to SF=10) while that of SigFox is the lowest at two months.
achieve an SNR ≥ 10 dB resulting in a BER of 10−3 for BFSK [78]. When transmitting a 2-byte payload, WiChronos reduces energy
At BER = 10−3 , we determine the optimum symbol length to be 10 consumption by 1.7x compared to LoRa-SF7, 11x to LoRa-SF10, and
bits, with a false positive probability P f ≤ 0.01 and symbol error 13.55x to SigFox (BW normalized). The improvement in battery life,
probability Pe ≤ 0.001 from Fig.3. Linx allows integer multiples of which increases with payload length, is due to, 1) reduced air-time
bytes (not bits) and hence we choose a 2-byte anchor symbol. The 2) non-increasing energy-per-bit Eb . Eb for a WiChronos sender
first byte is a sequence of alternating 1s and 0s to indicate the start decreases with increasing payload as the overall energy remains
of message and the second byte to identify the sender uniquely. constant while the number of bits increases. In classical modulation,
In Fig.10, we experimentally determine P f and Pe for varying Eb is constant and hence their energy increases with payload length.
anchor symbol lengths. With increasing anchor symbol length, P f The energy efficiency will be further improved by using modules
decreases while Pe increases. At 2-byte anchor symbol, P f is below with smaller sleep currents and faster wake-up time.
0.01 and Pe below 0.08. The experimental Pe is higher than the The energy efficiency of WiChronos depends only on the length
expected Pe derived in Equation 5 due to intermittent pedestrian of the anchor symbol and the traffic load. Fig.12 shows the impact of
traffic and strong winds. We present the impact of other environ- traffic load and network scale on the battery life of a single node. For
mental conditions on the symbol error rate later in the section. a given traffic load, battery life of a node decreases with increasing
From experiments we infer that the Linx-NT has a higher than network scale, since the anchor symbol length increases with scale.
average carrier sensing time, resulting in higher false positives. RF The anchor symbol length is the sum of minimum decodable pulse
modules with smarter carrier sensing can further reduce P f and (8 bits in this case) and loд2 (N etworkSize) + 1 to uniquely identify
hence the anchor symbol length. CC1125 [80] has a 4-bit settling the senders. Similarly, with increasing traffic load, the overall time-
time which can be reduced further by freezing the AGC setting on-air increases, in turn decreasing the battery life. At higher traffic
for known input levels. In addition to that, the maximum transmit loads where a payload is being sent once every 15 or 30 seconds,
power of Linx-NT is 13.5 dBm while FCC allows up to 30dBm. An the battery life reduces considerably as compared to low traffic
increase in transmit power and decrease in BW can improve SNR, scenarios. As discussed earlier, WiChronos derives energy savings
reducing Pe which will allow us to reduce the anchor symbol length partly from reducing the time-on-air which increases significantly
further. The presented result is thus an upper bound on the anchor when the traffic load and/or the scale increases.
MobiCom ’20, September 21–25, 2020, London, United Kingdom Yaman Sangar and Bhuvana Krishnaswamy
10 1000
1 MHz ± 30 ppm 80
Symbol Error Rate
32kHz ± 30 ppm
Payload Error Rate 70

Average Throughput (bps)

Anchor Symbol Loss (%)

8 100 1 MHz ± 150 ppm
32kHz ± 150 ppm 60
6 10 50
Error (%)

40
4
1
30
2 20
0.1
10
0
0.01 0
10 15 20 25 59F, Light wind 28F, Moderate wind 24F, Strong wind
0 100 200 300 400 500 600
Distance (m) Payload (bits) Temparature (F)/Wind
Fig. 13: WiChronos range Fig. 14: Datarate performance Fig. 15: Impact of environmental
conditions
6.5 Long Range 6.6.1 Processing time error correction. Fig.7 shows the timeline of
The long-range performance of WiChronos depends on the radio WiChronos data transmission, where the duration and currents
module and the modulation parameters used for anchor symbol. are specific to the components used in our implementation. As
In §4.3, we derived the achievable communication range as a func- discussed in §5.2, processing time error is corrected by elongating
tion of receiver BW. The maximum achievable range using Linx the transmission time tp to tp + ϵ. For the system under consid-
is limited to 914 meters, due to its reduced transmit power of 13.5 eration, tp = 196 cycles and the estimated tp + ϵ = 276 cycles.
dBm. We verified the feasibility of long-range in indoor and outdoor At system clock rate of 32.678 kHz, the components of tp with
environments using 1/4-wave whip antenna. Indoor experiments their corresponding average (A) and maximum (M) clock cycles
were performed in a 15 m x 10 m office space and along corridors are, (1) Radio wake-up time + MCU processing (A:137 , M:200 )
up to 75 m long. Outdoor experiments were performed in streets (2) MCU processing (A:8, M:8) (3) 2-byte anchor symbol transmis-
with car traffic, bike paths with pedestrian traffic, and open fields sion time at 9.6kBaud (A:51 ,M:68. We estimate the upper bound of
at varying distances and weather conditions as shown in Fig.8. each operation to determine ϵ, and validated experimentally. On
Fig.13 plots the percentage of symbol and payload errors at reading a sensor data, the MCU sets TimerB to count up to 276
varying distances for a 2-byte payload. The received symbol error cycles during radio wake-up and anchor transmission. The effec-
remained below 10% at 525 m. At distances up to 850 m, WiChronos tive processing and transmission time is therefore a predetermined
implemented on Linx was feasible under line-of-sight conditions. constant (276 cycles) and variations within this upper bound does
Payload error is the error in decoded data (time). On detecting an an- not affect the measured Tinf o at the receiver. The receiver counts
chor symbol loss, the receiver ignores the data. In our experiments, Tinf o and subtracts 276 to decode data, achieving 100% accuracy.
timing errors remained bounded and therefore, payload error is 6.6.2 Clock skew error. Clock skew directly affects the accuracy of
zero, provided the anchor symbols were received correctly. This the data received. We evaluate the performance of the two clocks
shows that under bounded errors, proposed mechanisms detect present in MSP430 1) an external crystal oscillator (ECO) 2) an
anchor symbol loss and correct timing errors with 100% accuracy. internal Digitally Controlled Oscillator (DCO) and analyse their
SNR at the receiver can be improved by increasing the transmit impact on data-rate. Columns 2-4 in Table 3 show the clock rate and
power. RF modules with the maximum allowable Effective Isotropic worst case tolerance of these clocks, along with other commercially
Radiated Power(EIRP) and lower BW can reach distances of over available low-power clocks. Clock tolerance is an indicator of the
10 km [80, 83, 95]. It must be noted that the achievable range of accuracy of the clock. For example, with the 32.678 kHz ECO with
our prototype is constrained by the hardware used here and is a tolerance ± 30 parts per million [89] the measured number of
not an inherent limitation. Linx was chosen for proof-of-concept clock cycles vary only by ± 1 cycle. We experimentally verified
because of the ease of hardware programming it offers, simplifying the tolerance levels of the ECO and the DCO by transmitting a
communication between the MCU and the radio. The sleep current predefined value of Tinf o every minute for over a week. Column 5
of Linx is extremely low which helps keeping the power budget shows the tolerance determined experimentally for the ECO and
down. Other RF modules such as TI CC1101, CC1125, CC1190 with DCO. Note that the theoretical tolerance values are an upper bound.
UNB modulations have been shown to achieve 10s of km, using Therefore, the clock skew correction proposed in §5.2 uses the
software programming. Any RF module with a protocol-free mode tolerance values to determine δ . For the ECO, since δ = 1, a data
can be integrated with an MCU to implement WiChronos. D is mapped to Tinf o = 3D + 1. We correct for clock skew by
introducing redundancy, which reduces the effective number of
6.6 Error correction and data-rate data values D that can be communicated, reducing the data-rate.
We identified four categories of error viz., anchor symbol loss, Higher tolerance requires higher redundancy for correction, neg-
processing time error, clock skew error, and propagation error. The atively affecting data-rate. Fig.14 plots the achievable data-rate as
results presented in the rest of this section incorporate the anchor Clock Tolerance Tolerance Expt. Tolerance
symbol loss detection. For the other three errors, we evaluate the
Rate ppm cycles cycles
data-rate performance of WiChronos, the accuracy of received data
1 MHz [75] ± 10000 ±10000 ± 720
in the presence of timing error, and the impact of the proposed
32.678 kHz [89] ± 30 ±1 ±1
error correction mechanisms on data-rate. We implement the error
correction mechanisms proposed in §5.2 at the transceivers.
1 MHz [90] ± 50 ±5 Not validated
Table 3: Impact of Clock rate and skew on datarate
WiChronos : Energy-Efficient Modulation for Long-Range, Large-Scale Wireless Networks MobiCom ’20, September 21–25, 2020, London, United Kingdom

Fig. 16: Collision performance in a multiple access network Fig. 17: Goodput Vs load for WiChronos & LoRa
categorize the traffic load as high, medium and low when a 2-byte
payload is sent every 15 s, 60 s, and 600 s respectively. As shown in
Fig. 16, this probability in a 100 node network with high traffic is
9.1%, 31.3%, and 95% for WiChronos, LoRa with SF-7, and SF-10 re-
spectively. At low traffic, this decreases to 0.24%, 0.93%, and 7.2% for
WiChronos, LoRa-SF7, and SF-10 respectively. To compare against
SigFox, we use a BW of 600Hz for both WiChronos and SigFox. For
low traffic, collision probability is 1.6% and 60.3% for WiChronos
and SigFox. Note that the collision probability increases with both
network scale and traffic, due to larger volume of data, resulting in
higher vulnerability to collisions. This increase in collisions is lower
for WiChronos than that of LoRa and SigFox due to its reduced
time-on-air. Since the channel is active only for the duration of the
Fig. 18: Goodput Vs load for WiChronos & SigFox anchor symbols, the vulnerable period for collision of WiChronos
a function of payload for different clocks. When using clock skew is smaller for the same bandwidth and payload length.
error correction, the tolerance has a higher impact on the data-rate We also validate the theoretical results with experiments for high-
than the clock frequency. For a given tolerance, the data-rate of traffic, small-scale network (zoomed-in image). The experimental
WiChronos with 32kHz and 1MHz are comparable, due to the re- results reflect the cumulative losses due to collisions, false positives,
dundancy introduced for clock skew correction. Accurate clocks and anchor symbol losses. On an average, the theoretical collision
are key to the success of WiChronos. Low-power, high accuracy rate is 75% of the experimental rate; the theoretical results are
clocks in the order of MHz [90] can be integrated with the MCU to an approximation for a large scale network and we expect the
improve the data-rate performance of WiChronos. Note that the experimental results to match theoretical for larger networks.
data rate decreases with increasing payload length, irrespective of The reduced collision also improves the average network good-
clock tolerance, due to the exponential dependence of information put performance (total number of successful data bits in the network
time on the payload. Prior knowledge of source distribution can in a given duration). We present the average network goodput as a
help with optimum data-time encoding to improve data-rate. function of traffic load for various network scales in Figs 17 and 18,
where each transmitter sends a 2-byte message every traffic-load
6.7 Multiple Access Control seconds. The above results are BW normalized for fairness, with
WiChronos and LoRa using 100kHz and 125kHz respectively in
In Fig. 16, we verify the effectiveness of WiChronos using ALOHA
Fig 17 and 600Hz for WiChronos and SigFox in Fig 18. Due to an
in a large-scale network under varying traffic load. We also experi-
increase in the volume of data, the goodput increases with traf-
mentally validate the collision probability derived in Equation 7 in
fic initially. However, it reaches an optimal point beyond which,
a network of 12 senders in heavy traffic scenario. We program the
collisions dominate and reduce the number of successful transmis-
senders to wake up every 16 seconds to communicate a random
sions, inturn reducing goodput. For a given scale, the goodput of
2-byte payload. Each sender is assigned a unique preamble and a
WiChronos outperforms LoRa due to its reduced collisions (Fig.16),
postamble. On receiving valid bits on the serial interface, the MCU
and they converge at lower traffic. This is because, with reduced
correlates the first byte to detect a packet and decode the second
traffic load, data volume is low, inversely affecting goodput due to
byte to identify the sender uniquely and trigger the corresponding
lower utilization of channel capacity. We compare the goodput per-
timer. The senders are randomly placed inside a 15 m x 10 m office
formance of WiChronos and SigFox in Fig.18. As the scale increases,
space. The presented results are averaged over a 2-day period.
the increase in collision probability of SigFox is higher than that
We present the probability of collisions in a log scale Y-axis as a
of WiChronos. Similarly, with reduced traffic load, the decrease in
function of network scale under varying traffic conditions in Fig.16
collision is higher for WiChronos, leading to a better goodput.
The background grey shade darkens as the traffic load increases. We
MobiCom ’20, September 21–25, 2020, London, United Kingdom Yaman Sangar and Bhuvana Krishnaswamy

It must be noted that though WiChronos outperforms LoRa and Systems research on efficient low-power, high accuracy clocks have
SigFox in average network goodput, it trades off instantaneous the potential to maximize data-rate and minimize energy further.
goodput. In a 1000 node network where each node has a 2-byte pay- Anchor symbol loss: We use timeout to detect anchor loss when
load every 10 minutes, the instantaneous goodput of a WiChronos messages are shorter than the sensing period. However, when the
and a LoRa node are 15.0 bps and 517.27 bps respectively. Similarly, time between messages decreases, anchor symbol loss can go unde-
with 600 Hz BW, the instantaneous goodput of WiChronos and tected, leading to inaccurate reception. Further research to handle
SigFox are 12.08 bps and 46 bps respectively. WiChronos is there- multiple anchor losses is required to broaden our impact.
fore best suited for applications with short payloads, where it can Information security and authentication: In its current design,
deliver low-energy over long-range in a large-scale. a WiChronos receiver identifies the sender from the anchor symbols
LoRa can leverage the orthogonality of SF to enhance scalabil- and does not offer a mechanism for authentication. An adversary
ity [96]. Lower SF values offer high data rate, and low energy. High could be listening to the channel to decode data and it is there-
SF values help attain a greater range at the cost of data rate. While fore not secure. Encrypting anchor symbol bit pattern can be used
using different SF could help with scalability, it will also effect the to authenticate the sender whereas data security can be achieved
range and battery performance. Another option offered by LoRa by encrypting the data (D) to timing (Tinf o )mapping. Simple and
and SigFox is to use Frequency Division Multiple Access (FDMA). efficient security protocols for timing-based communication is a
FDMA can be used in addition to ALOHA to further reduce the branch of research that is open for further study.
collision rate of WiChronos as well. For a given receiver BW, the re- Anchor symbol design: We plan to explore anchor modulations
duced BW on each sender will increase the collision rate. However, to further decrease the energy consumption and increase the range.
this is only a function of the anchor symbols for WiChronos and We also plan to explore wideband modulation for energy-efficient,
the effect will be less pronounced compared to LoRa and SigFox. scalable, short-range communication scenarios such as on-body
The decrease in collision rate from the reduced number of devices sensors. Anchor symbol assigned to each sender is a unique se-
per channel outweighs the increase (Fig.6) due to lower BW. quence that serves as the sender address. Sequences with high auto-
correlation and low cross-correlation properties must used to com-
6.8 Environmental Conditions municate the addresses reliably. Designing such sequences within
We evaluated WiChronos outdoors under varying weather and the energy budget is challenging and requires further research. In
traffic conditions. Fig.15 shows the impact of temperature and wind a large scale set up, correlation for identification of senders can be
on the anchor symbol error rate at a distance of 250m. SER increases time consuming. We are exploring possible solutions of using chirp
by a factor of 15 at low temperatures and 19 when there are strong spread spectrum (CSS) where each sender is assigned a unique chirp
winds as well. While this is not a complete analysis, it indicates the Medium Access Control: A hybrid of FDMA- or CDMA-based
strong impact of environmental conditions on anchor symbol loss, ALOHA can potentially improve the reliability of WiChronos within
and in turn, bit errors at the receiver. We plan to explore the impact the energy budget. MAC algorithms leveraging spatial diversity
of environmental conditions and design antennas and protective and anchor symbol modulation is a future extension of this work.
casing to reduce the impact in our future work. Physical layer: WiChronos was implemented using OTS compo-
nents and we rely on their built-in signal processing for modula-
7 LIMITATIONS AND FUTURE WORK tion and demodulation, frequency spectrum, AGC timing among
WiChronos is the first step towards a low-power, long-range, scal- other features. Various other transceivers such as TI CC1101 and
able wireless network for low data-rate LPWANs. There are open TI CC1125 offer advantages such as lower AGC settling time, the
research challenges that need further investigation and new re- ability to detect a 4 bit preamble and a much lower active current.
search areas stemming from our work such as, Parameters for choosing appropriate hardware may be restricted
Timing error detection and correction: Existing error correc- by the application and requires further study. We are currently
tion codes are designed for bit errors. Timing errors in WiChronos working on exploring some of these modules from TI.
can occur due to distance, and channel conditions. Currently, we
transmit anchor symbol at low data rates of the order of few kbps,
8 CONCLUSIONS
and hence the impact of above conditions on anchor symbol timing We proposed WiChronos, a modulation technique that enables
is minimal. Error correction mechanisms will be needed for higher long-range communication in large-scale deployments of energy-
anchor symbol rates of 100s of kbps. We proposed a simple error constrained sensor nodes. By offloading the communication com-
correction mechanism for clock skew that uses the apriori knowl- plexity to timers present in every MCU, we reduce the energy
edge of the clock tolerance. Smarter, capacity achieving timing error consumption and the cost of each sender. We reduce the time-on-
correction can improve the data-rate performance of WiChronos, air by minimizing the number of symbols per message, in turn
making it suitable for a wider range of applications. improving the spectral efficiency. The reduced time-on-air enables
Data-rate energy tradeoff: WiChronos trades off data rate for an ALOHA-based MAC protocol to accommodate a large-scale net-
energy efficiency. We proposed two strategies to improve the data- work without significantly affecting the collision performance. We
rate performance of WiChronos. 1)A greedy algorithm that uses propose error detection and correction mechanisms for timing in-
prior knowledge of the source distribution. 2)Information-theoretic duced bit errors for bounded errors. Using off-the-shelf RF modules
and machine learning approaches to learn the source distribution and MCUs, we prototype and evaluated the energy, range, and scale
and encode data to clock cycles optimally; also an ongoing work. performance experimentally in indoor and outdoor scenarios.
WiChronos : Energy-Efficient Modulation for Long-Range, Large-Scale Wireless Networks MobiCom ’20, September 21–25, 2020, London, United Kingdom

REFERENCES (WMNC 2011), pages 1–9. IEEE, 2011.

[1] Mare Srbinovska, Cvetan Gavrovski, Vladimir Dimcev, Aleksandra Krkoleva, and [32] Gang Lu, Bhaskar Krishnamachari, and Cauligi S Raghavendra. An adaptive
Vesna Borozan. Environmental parameters monitoring in precision agriculture energy-efficient and low-latency mac for data gathering in wireless sensor net-
using wireless sensor networks. Journal of cleaner production, 88:297–307, 2015. works. In 18th International Parallel and Distributed Processing Symposium, 2004.
[2] JA López Riquelme, Fulgencio Soto, J Suardíaz, P Sánchez, Andres Iborra, and Proceedings., page 224. IEEE, 2004.
JA Vera. Wireless sensor networks for precision horticulture in southern spain. [33] Ilker Demirkol, Cem Ersoy, and Fatih Alagoz. Mac protocols for wireless sensor
Computers and electronics in agriculture, 68(1):25–35, 2009. networks: a survey. IEEE Communications Magazine, 44(4):115–121, 2006.
[3] Hanwook Chung, Jingjie Li, Younghyun Kim, and Christopher Y Choi. Continu- [34] Wei Ye, John Heidemann, and Deborah Estrin. Medium access control with coor-
ous and wireless skin contact and ear implant temperature measurements and dinated adaptive sleeping for wireless sensor networks. IEEE/ACM Transactions
relations to the core body temperature of heat stressed dairy cows. In 10th Inter- on Networking (ToN), 12(3):493–506, 2004.
national Livestock Environment Symposium (ILES X), page 1. American Society of [35] Andrea Goldsmith. Wireless communications. Cambridge university press, 2005.
Agricultural and Biological Engineers, 2018. [36] Injong Rhee, Ajit Warrier, Mahesh Aia, Jeongki Min, and Mihail L Sichitiu. Z-mac:
[4] Haider Jawad, Rosdiadee Nordin, Sadik Gharghan, Aqeel Jawad, and Mahamod a hybrid mac for wireless sensor networks. IEEE/ACM Transactions on Networking
Ismail. Energy-efficient wireless sensor networks for precision agriculture: A (TON), 16(3):511–524, 2008.
review. Sensors, 17(8):1781, 2017. [37] Naoto Kimura and Shahram Latifi. A survey on data compression in wireless
[5] National Soil Moisture Network. http://nationalsoilmoisture.com/. sensor networks. In International Conference on Information Technology: Coding
[6] National Ecological Observatory Network. https://www.neonscience.org/. and Computing (ITCC’05)-Volume II, volume 2, pages 8–13. IEEE, 2005.
[7] Decagon 10HS Soil Moisture Sensor. http://www.labcell.com/environmental/soil- [38] David Chu, Amol Deshpande, Joseph M Hellerstein, and Wei Hong. Approx-
moisture-sensors/10hs-soil-moisture-sensor. imate data collection in sensor networks using probabilistic models. In 22nd
[8] EC5 Soil Moisture Sensor. http://www.labcell.com/environmental/soil-moisture- International Conference on Data Engineering (ICDE’06), pages 48–48. IEEE, 2006.
sensors/ec5-soil-moisture-sensor. [39] Bhargav Kanagal and Amol Deshpande. Online filtering, smoothing and proba-
[9] TEROS 12: Avanced soil moisture sensing. https://www.metergroup.com/ bilistic modeling of streaming data. In 2008 IEEE 24th International Conference on
environment/products/teros-12/. Data Engineering, pages 1160–1169. IEEE, 2008.
[10] Shahin Farahani. ZigBee wireless networks and transceivers. Newnes, 2011. [40] Aria Nosratinia, Todd E Hunter, and Ahmadreza Hedayat. Cooperative commu-
[11] Carles Gomez, Joaquim Oller, and Josep Paradells. Overview and evaluation of nication in wireless networks. IEEE communications Magazine, 2004.
bluetooth low energy: An emerging low-power wireless technology. Sensors, [41] Zhong Zhou, Shengli Zhou, Shuguang Cui, and Jun-Hong Cui. Energy-efficient
12(9):11734–11753, 2012. cooperative communication in a clustered wireless sensor network. IEEE Trans-
[12] Colby Boyer and Sumit Roy. âĂŤinvited paperâĂŤbackscatter communication actions on Vehicular Technology, 2008.
and rfid: Coding, energy, and mimo analysis. IEEE Transactions on Communica- [42] Paolo Casari, Alessia Marcucci, Michele Nati, Chiara Petrioli, and Michele Zorzi.
tions, 62(3):770–785, 2014. A detailed simulation study of geographic random forwarding (geraf) in wireless
[13] Vincent Liu, Aaron Parks, Vamsi Talla, Shyamnath Gollakota, David Wetherall, sensor networks. In MILCOM 2005-2005 IEEE Military Communications Conference,
and Joshua R Smith. Ambient backscatter: wireless communication out of thin pages 59–68. IEEE, 2005.
air. In ACM SIGCOMM Computer Communication Review, volume 43, pages 39–50. [43] Jamal N Al-Karaki and Ahmed E Kamal. Routing techniques in wireless sensor
ACM, 2013. networks: a survey. IEEE wireless communications, 11(6):6–28, 2004.
[14] Yao Peng, Longfei Shangguan, Yue Hu, Yujie Qian, Xianshang Lin, Xiaojiang [44] Vamsi Paruchuri, Shivakumar Basavaraju, Arjan Durresi, Rajgopal Kannan, and
Chen, Dingyi Fang, and Kyle Jamieson. Plora: A passive long-range data network S Sitharama Iyengar. Random asynchronous wakeup protocol for sensor net-
from ambient lora transmissions. In Proceedings of the 2018 Conference of the works. In First International Conference on Broadband Networks. IEEE, 2004.
ACM Special Interest Group on Data Communication. ACM, 2018. [45] IEEE Standard 802.15.4. https://standards.ieee.org/standard/802_15_8-2017.html.
[15] Vamsi Talla, Mehrdad Hessar, Bryce Kellogg, Ali Najafi, Joshua R Smith, and [46] Benjie Chen, Kyle Jamieson, Hari Balakrishnan, and Robert Morris. Span: An
Shyamnath Gollakota. Lora backscatter: Enabling the vision of ubiquitous con- energy-efficient coordination algorithm for topology maintenance in ad hoc
nectivity. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous wireless networks. Wireless networks, 8(5):481–494, 2002.
Technologies, 1(3):105, 2017. [47] M. S. Trotter and G. D. Durgin. Survey of range improvement of commercial rfid
[16] Thomas M Cover and Joy A Thomas. Elements of information theory. John Wiley tags with power optimized waveforms. In 2010 IEEE International Conference on
& Sons, 2012. RFID (IEEE RFID 2010), pages 195–202. IEEE, 2010.
[17] Deepak Ganesan Pan Hu, Pengyu Zhang. Laissez-faire : Fully asymmetric [48] Mehran Abolhasan, Tadeusz Wysocki, and Eryk Dutkiewicz. A review of routing
backscatter communication. In Proceedings of the 2015 ACM Conference on Special protocols for mobile ad hoc networks. Ad hoc networks, 2(1):1–22, 2004.
Interest Group on Data Communication, pages 255–267. ACM, 2015. [49] Vikram Iyer, Rajalakshmi Nandakumar, Anran Wang, Sawyer Fuller, and Shyam-
[18] V. Pillai, H. Heinrich, D. Dieska, P. V. Nikitin, R. Martinez, and K. V. S. Rao. An nath Gollakota. Living iot: A flying wireless platform on live insects. arXiv
ultra-low-power long range battery/passive rfid tag for uhf and microwave bands preprint arXiv:1812.09419, 2018.
with a current consumption of 700 na at 1.5 v. IEEE Transactions on Circuits and [50] Evşen Yanmaz, Saeed Yahyanejad, Bernhard Rinner, Hermann Hellwagner, and
Systems I: Regular Papers, 54(7):1500–1512, 2007. Christian Bettstetter. Drone networks: Communications, coordination, and
[19] LoRa. https://www.semtech.com/lora. sensing. Ad Hoc Networks, 68:1–15, 2018.
[20] SigFox. https://www.sigfox.com/en/sigfox-iot-radio-technology. [51] Rapeepat Ratasuk, Nitin Mangalvedhe, Yanji Zhang, Michel Robert, and Jussi-
[21] NB-IoT. https://www.gsma.com/iot/narrow-band-internet-of-things-nb-iot/. Pekka Koskinen. Overview of narrowband iot in lte rel-13. In 2016 IEEE conference
[22] Juha Petäjäjärvi, Konstantin Mikhaylov, Matti Hämäläinen, and Jari Iinatti. Eval- on standards for communications and networking (CSCN), pages 1–7. IEEE, 2016.
uation of lora lpwan technology for remote health and wellbeing monitoring. In [52] Zigbee 3.0. https://www.nxp.com/products/wireless/zigbee/zigbee/zigbee-3.0:
2016 10th International Symposium on Medical Information and Communication ZIGBEE-3-0.
Technology (ISMICT), pages 1–5. IEEE, 2016. [53] Venkat Anantharam and Sergio Verdu. Bits through queues. IEEE Transactions
[23] Ruslan Kirichek and Vyacheslav Kulik. Long-range data transmission on flying on Information Theory, 42(1):4–18, 1996.
ubiquitous sensor networks by using lpwan protocols. In International Conference [54] Anand S Bedekar and Murat Azizoglu. The information-theoretic capacity of
on Distributed Computer and Communication Networks. Springer, 2016. discrete-time queues. IEEE Transactions on Information Theory, 1998.
[24] William Hart Hayt, John A Buck, et al. Engineering electromagnetics, volume 6. [55] Rajesh Sundaresan and Sergio Verdú. Robust decoding for timing channels. IEEE
McGraw-Hill New York, 1981. Transactions on information Theory, 46(2):405–419, 2000.
[25] MSP 430 user guide. http://www.ti.com/lit/ug/slau445i/slau445i.pdf. [56] Rajesh Sundaresan and Sergio Verdú. Capacity of queues via point-process
[26] Linx NT Series Transceiver User Guide. https://linxtechnologies.com/wp/wp- channels. IEEE/ACM Transactions on Networking (TON), 14(SI):2697–2709, 2006.
content/uploads/trm-fff-nt.pdf. [57] Balaji Prabhakar and Robert Gallager. Entropy and the timing capacity of discrete
[27] Gregory J Pottie and William J Kaiser. Wireless integrated network sensors. queues. IEEE Transactions on Information Theory, 49(2):357–370, 2003.
Communications of the ACM, 43(5):51–58, 2000. [58] Sarah H Sellke, C-C Wang, Saurabh Bagchi, and Ness Shroff. Tcp/ip timing
[28] Vijay Raghunathan, Curt Schurgers, Sung Park, and Mani B Srivastava. Energy- channels: Theory to implementation. In IEEE INFOCOM 2009. IEEE, 2009.
aware wireless microsensor networks. IEEE Signal processing magazine, 2002. [59] Pandurang Kamat, Wenyuan Xu, Wade Trappe, and Yanyong Zhang. Temporal
[29] Giuseppe Anastasi, Marco Conti, Mario Di Francesco, and Andrea Passarella. privacy in wireless sensor networks: Theory and practice. ACM Transactions on
Energy conservation in wireless sensor networks: A survey. Ad hoc networks, Sensor Networks (TOSN), 5(4):28, 2009.
7(3):537–568, 2009. [60] Wenyuan Xu, Wade Trappe, and Yanyong Zhang. Anti-jamming timing channels
[30] Tifenn Rault, Abdelmadjid Bouabdallah, and Yacine Challal. Energy efficiency in for wireless networks. In Proceedings of the first ACM conference on Wireless
wireless sensor networks: A top-down survey. Computer Networks, 2014. network security, pages 203–213. ACM, 2008.
[31] Ridha Soua and Pascale Minet. A survey on energy efficient techniques in wireless [61] Yujie Zhu and Raghupathy Sivakumar. Challenges: communication through
sensor networks. In 2011 4th Joint IFIP Wireless and Mobile Networking Conference silence in wireless sensor networks. In Proceedings of the 11th annual international
conference on Mobile computing and networking, pages 140–147. ACM, 2005.
MobiCom ’20, September 21–25, 2020, London, United Kingdom Yaman Sangar and Bhuvana Krishnaswamy

[62] Da-shan Shiu and Joseph M Kahn. Differential pulse-position modulation for [77] TI CC1101 datasheet. www.ti.com/lit/gpn/cc1101.
power-efficient optical communication. IEEE transactions on communications, [78] Theodore S Rappaport et al. Wireless communications: principles and practice,
47(8):1201–1210, 1999. volume 2. prentice hall PTR New Jersey, 1996.
[63] Zabih Ghassemlooy, AR Hayes, NL Seed, and ED Kaluarachchi. Digital pulse [79] Long-range RF communication:. http://www.ti.com/lit/wp/swry006/swry006.pdf.
interval modulation for optical communications. IEEE Communications Magazine, [80] TI CC1125 datasheet. http://www.ti.com/lit/ds/symlink/cc1125.pdf.
36(12):95–99, 1998. [81] Matthew Gast. 802.11 wireless networks: the definitive guide. " O’Reilly Media,
[64] Hisayoshi Sugiyama and Kiyoshi Nosu. Mppm: A method for improving the Inc.", 2005.
band-utilization efficiency in optical ppm. Journal of Lightwave Technology, [82] Kais Mekki, Eddy Bajic, Frederic Chaxel, and Fernand Meyer. A comparative
7(3):465–472, 1989. study of lpwan technologies for large-scale iot deployment. ICT Express, 2019.
[65] Hyuncheol Park and J. R. Barry. Trellis-coded multiple-pulse-position modulation [83] CC1120 High-Performance RF Transceiver for Narrowband Systems. http://www.
for wireless infrared communications. IEEE Transactions on Communications, ti.com/lit/ds/symlink/cc1120.pdf.
52(4):643–651, April 2004. [84] Chuhan Gao, Mehrdad Hessar, Krishna Chintalapudi, and Bodhi Priyantha. Blind
[66] Hyuncheol Park and J. R. Barry. Modulation analysis for wireless infrared distributed mu-mimo for iot networking over vhf narrowband spectrum. In
communications. In Proceedings IEEE International Conference on Communications International Conference on Mobile Computing and Networking. ACM, 2019.
ICC ’95, volume 2, pages 1182–1186 vol.2, June 1995. [85] Cory Beard and William Stallings. Wireless Communication Networks and Systems.
[67] Li Zhao and A. M. Haimovich. Multi-user capacity of m-ary ppm ultra-wideband Pearson Education, 2015.
communications. In 2002 IEEE Conference on Ultra Wideband Systems and Tech- [86] Pei Huang, Li Xiao, Soroor Soltani, Matt W Mutka, and Ning Xi. The evolution
nologies (IEEE Cat. No.02EX580), pages 175–179, May 2002. of mac protocols in wireless sensor networks: A survey. IEEE communications
[68] Li Zhao and A. M. Haimovich. Capacity of m-ary ppm ultra-wideband communi- surveys & tutorials, 15(1):101–120, 2013.
cations over awgn channels. In IEEE Vehicular Technology Conference, 2001. [87] James F. Kurose and Keith W. Ross. Computer Networking: A Top-Down Approach
[69] Zhi Zhang, Zhonghai Lu, Qiang Chen, Xiaolang Yan, and Li-Rong Zheng. Code (6th Edition). Pearson, 6th edition, 2012.
division multiple access/pulse position modulation ultra-wideband radio fre- [88] LoRa Airtime Calculator. https://www.loratools.nl/#/airtime,.
quency identification for internet of things: concept and analysis. International [89] 32 kHz crystal oscillator.
Journal of Communication Systems, 25(9):1103–1121, 2012. [90] Ultra-Small Low Power, Low-Jitter, 1 Hz to 2.5 MHz Oscillator. https://www.
[70] Francesca Cuomo and C Martello. Mac principles for an ultra wide band wireless sitime.com/products/1-hz-2-mhz-oscillatorstcxo/sit1579.
access. In IEEE Global Telecommunications Conference (Cat. No. 01CH37270), 2001. [91] Bernhard Korte, Jens Vygen, B Korte, and J Vygen. Combinatorial optimization,
[71] Xuemin Shen, Weihua Zhuang, Hai Jiang, and Jun Cai. Medium access control volume 2. Springer, 2012.
in ultra-wideband wireless networks. IEEE Transactions on Vehicular Technology, [92] Adafruit Feather 32u4 Radio . https://www.adafruit.com/product/3076.
54(5):1663–1677, 2005. [93] LoRa. https://www.semtech.com/products/wireless-rf/lora-transceivers/sx1272.
[72] Dong Sam Ha and Patrick R Schaumont. Replacing cryptography with ultra [94] Y. Chen, N. Chiotellis, L. Chuo, C. Pfeiffer, Y. Shi, R. G. Dreslinski, A. Grbic,
wideband modulation in secure rfid. In 2007 IEEE International Conference on T. Mudge, D. D. Wentzloff, D. Blaauw, and H. S. Kim. Energy-autonomous
RFID. wireless communication for millimeter-scale internet-of-things sensor nodes.
[73] Z. Lin and P. Wei. Pulse position modulation time hopping ultra wideband IEEE Journal on Selected Areas in Communications, 34(12):3962–3977, Dec 2016.
sharing signal for radar and communication system. In 2006 CIE International [95] FCC Regulations. https://transition.fcc.gov/Bureaus/Engineering_Technology/
Conference on Radar, pages 1–4, Oct 2006. Documents/bulletins/oet63/oet63rev.pdf.
[74] R. Scholtz. Multiple access with time-hopping impulse modulation. In Proceedings [96] Floris; Moerman Ingrid; Hoebeke Jeroen Haxhibeqiri, Jetmir; Van den Abeele.
of IEEE Military Communications Conference, 1993. Lora scalability: A simulation model based on interference measurements. Sensors,
[75] MSP 430 FR2355. http://www.ti.com/product/MSP430FR2353. 16(6).
[76] STM32L151x6. https://www.st.com/resource/en/datasheet/cd00277537.pdf.