Enabling Early Sleeping and Early Data Transmission in Wake-up Radio-enabled IoT Networks

Wireless sensor networks (WSNs) are one of the key enabling technologies for the Internet of things (IoT). In such networks, wake-up radio (WuR) is gaining its popularity thanks to its on-demand transmission feature and overwhelming energy consumption superiority. Despite this advantage, overhearing still occurs when a wake-up receiver decodes the address of a wake-up call (WuC) which is not intended to it, causing a certain amount of extra energy waste in the network. Moreover, long latency may occur due to WuC address decoding since WuCs are transmitted at a very low data rate. In this paper, we propose two schemes, i.e., early sleeping (ES) and early data transmission (EDT), to further reduce energy consumption and latency in WuR-enabled IoT/WSNs. The ES scheme decodes and validates an address bit-by-bit, allowing those non-destined devices go to sleep at an earlier stage. The EDT scheme enables a sender to transmit small IoT data together with WuC packets so that the main radio does not have to be in full operation for data reception. We implement both schemes through a WuR testbed. Furthermore, we present a framework based on M/G/1 and assess the performance of the schemes through both theoretical analysis and simulations.


Introduction
The 5th generation (5G) wireless network aims to facilitate various unprecedented capacities such as enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), as well as ultra-reliable low latency communications (URLLC) [2]. Furthermore, a heterogeneous network integrat- 5 ing 5G with the Internet of things (IoT) paves the way for massive IoT applications [3] [4]. With the inter-network connectivity 5G IoT provides to small-size, low-cost, and often battery-powered devices, a variety of applications are envisaged, ranging from mission-critical services, smart home and smart city, to industrial automation and smart farming [5][6] [7]. In such massive IoT and 10 wireless sensor network (WSN) applications, performing energy-efficient communication for battery powered IoT/WSN devices is of vital importance [8][9].
Traditionally, duty-cycled (DC) medium access control (MAC) mechanisms, which allow WSN devices sleep and wake up periodically or aperiodically, have been adopted to reduce energy consumption. However, idle listening and over- 15 hearing occur in DC-MAC mechanisms during their active periods when a node senses the channel for receiving control messages and when an unintended node overhears the transmission of other nodes respectively. In recent years, wakeup radio (WuR) has emerged as a convincing solution to replace DC-MAC for providing energy-efficient communication in IoT/WSN networks [10] [11]. 20 In a WuR-enabled IoT/WSN node, an auxiliary wake-up receiver (WuRx) is attached to the micro-controller unit (MCU) of a main radio (MR). While the MR, which is responsible for data transmission, is active only when necessary, the WuRx is always on, waiting for detecting wake-up calls (WuCs) at any time. The power consumption of a WuRx is 1000 times lower than that of the 25 MR, i.e., the reception power of WuRx is in µW whereas it is in mW when a main radio is in full operation [10]- [16]. Upon the detection/reception of such a WuC sent by a transmitter, the destined WuRx triggers its MR to wake up from the sleep mode to perform data communication afterwards. In addition to this overwhelming energy saving, another advantage of WuR is that it works in a purely asynchronous manner. Such an on-demand communication operation remarkably reduces latency in comparison with DC-based MAC operations.
WuR was initially developed for energy-efficient data collection and reporting in WSNs. Over time, the application scenarios of WuRs have been expanded to diverse wireless networks including IoT, Wi-Fi, and mMTC. Despite the 35 great superiority on energy consumption that WuR provides, overhearing is not eliminated in WuR-enabled IoT networks. Indeed, overhearing occurs when a node decodes and validates an address of a WuC which is not intended to it. Although a WuRx operates in µW, the effect of overhearing in WuR-IoT cannot be simply ignored considering the effect of energy-hungry (decoding and 40 matching) components in WuRxs, long WuC duration, and network size. On the other hand, low latency communication is an important performance indicator in many IoT applications. Very recently, the third generation partnership project (3GPP) has started to work on standardizing early data transmissions (EDT) as one of the 5G new radio techniques to further support energy-efficient and 45 low latency communication for mMTC applications [17] [18]. So far, little work can be found in the literature with respect to reducing overhearing and latency in WuR-enabled IoT considering WuC decoding and EDT. This paper makes an effort towards this direction.
In this paper, we propose two schemes, referred to as early sleeping (ES) and 50 EDT respectively, tailored for eliminating overhearing and shortening latency in such networks. More specifically, ES decodes and matches the address of a WuC bit-by-bit so that the non-destined nodes can go to sleep at an earlier stage. EDT enables a transmitter to send IoT small data encoded with a WuC so that a data transmission may be completed without fully waking up the 55 MR. Both schemes are implemented through a WuR testbed. Furthermore, we develop a queuing model to evaluate the performance of the proposed schemes in WuR-IoT. Extensive simulations are performed to validate the accuracy of the analytical model. In brief, the main contributions of this work are as follows: Two energy-efficient schemes are proposed for WuR-enabled IoT, i.e., ES 60 and EDT. While both schemes reduce the energy consumption of such a network, the latter one also minimizes the latency of data transmission.
For proof of concept, the schemes are implemented in a small-scale WuR testbed and the functionalities of these schemes are demonstrated via the testbed. 65 To evaluate the performance of the ES and EDT schemes for a larger network, we present a generic framework based on an M/G/1 queuing model. The accuracy of the model is validated through discrete-event simulations.
Analytical expressions for three performance parameters including packet 70 delivery ratio (PDR), latency, and energy consumption are derived based on the proposed analytical framework.
The remainder of this paper is structured as follows. In Section 2, we summarize the related work and highlight the qualitative differences between our work and the existing ones. The network scenario and the WuR principle are 75 presented in Section 3. In Sections 4 and 5, we first propose the ES and EDT schemes and then develop a generic M/G/1 queuing framework for performance evaluation of these schemes. The performance metrics are defined and derived in Section 6, followed by testbed implementation and experimental validation presented in Section 7. The simulation results are explained in Section 8, before 80 the paper is concluded in Section 9.

Related Work
In this section, we summarize briefly the related work relevant to this study from four perspectives, i.e, WuR prototype implementation, WuR protocol design, theoretical frameworks for WuR-enabled IoT networks, and early data 85 transmission for 5G new radio. For more detailed surveys on the state-of-theart WuR techniques, please refer to [19] [20].

WuR Prototype Implementation
The nanowatt wake-up radio [11] is a WuR prototype which focuses on ultralow power consumption for WuRxs, with a WuRx power consumption level 90 around 1 µW when listening to the channel. The prototype can operate in different frequencies of the industrial, scientific and medical (ISM) band through on-off keying (OOK) modulation. Its minimum sensitivity is -35 dBm and the response time for an interrupt is 100 µs. Subcarrier modulation WuR [13] is another popular inband WuR circuitry design and its WuC can reach up to 100 95 meters. It also employs OOK modulation for sending addressable WuC with a sensitivity level of around -53 dBm. A near-zero power WuRx design with -69 dBm sensitivity was proposed in [21], and it was implemented based on an insulator complementary metal-oxide-semiconductor (CMOS) process using metal-oxide-semiconductor field-effect transistor (MOSFET) devices. The wake- correlator. Furthermore, a DC scheme and a packet structure were constructed to maintain the tradeoff between latency and power consumption while ensuring low false alarm rates.

110
Many WuR protocols have been proposed to show the benefits of WuR over traditional DC-MAC based WSNs or to improve the performance of existing WuR schemes. SCM-WuR [10] is a WuR scheme which relies on a single channel for WuC and data packet transmissions. No acknowledgment (ACK) is sent after the successful delivery of a WuC but it is needed when a data packet 115 is received. The performance of SCM-WuR was evaluated using simulations, showing its superior performance over DC-MAC protocols. Another energyaware cross-layer scheme is OPWUM [14] which opportunistically selects the best relay nodes for forwarding packets based on neighbors' energy. Moreover, ALBA-WuR [15] employs semantic addressing for relay selection, providing a 120 geographic cross-layer solution for WuR-enabled WSNs. Considering that multiple wake-up transmitters (WuTxs) may send WuCs at the same time, a backoffenabled WuR scheme, BoWuR, was proposed in our earlier work in order to reduce collision probability for transmitter-initiated (TI) data reporting [23].
For receiver-initiated (RI) data collection, a multiple packet transmission MAC 125 scheme has been proposed which reserves the channel for a node to send multiple packets consecutively [24]. Furthermore, the authors of [25] proposed to adopt a short local address to reduce the latency of WuC transmission. Their addressing scheme utilized partial functions of MCU to decode and match a full address. Improved performance over the correlator-based address decoding 130 approach is shown therein. As part of this study, we proposed in [1] a bit-by-bit WuC address decoding scheme which is known as ES in this paper.

Theoretical Frameworks for WuR-enabled IoT
Although a number of theoretical frameworks for modeling IoT or massive IoT applications exist, e.g., [4], there are few frameworks which deal with WuR-135 enabled IoT networks. Among them, an analytical model which is based on an absorbing Markov chain was presented in [27] to assess the performance of WuR-enabled IoT networks. In [23], the performance of the BoWuR protocol was analyzed using a discrete time Markov chain for a single hop SCM-WuR enabled IoT network under saturated traffic conditions. Based on duty-cycled 140 received-initiated WuR IoT networks, an analytical model was developed in [24] to evaluate the performance of aggregated packet transmissions. However, none of these frameworks considered an ES or/and EDT scheme for their performance evaluation. during the random access procedure (after the physical random access channel (PRACH) transmission and before the radio resource control (RRC) connection setup is completed). With EDT, a small amount of data exchange can be achieved during the random access procedure before the data link is formally established.

155
However, none of the aforementioned WuR prototypes and protocols considered eliminating overhearing for WuC transmissions. In Table 1, we make a qualitative comparison of our schemes with a few state-of-the-art WuR protocols. To the best of our knowledge, this work is first effort which applies EDT to WuR-enabled IoT networks.

Network Topology and WuR Principle
In this section, we illustrate the network scenario as well as the design and operating principle of WuR.

Network Scenario and Assumptions
Consider a WuR-enabled IoT/WSN consisting of N number of nodes as   arrivals follow a Poisson process with rate λ for each device. Assume further that the channel is error-free and no packet loss occurs due to buffer overflow.
However, packet loss occurs due to collisions.   follows, we focus on TI-WuR for performance evaluation of WuR-IoT. In TI-WuR, a node sends a WuC to the intended node once it has a packet to transmit.
After receiving the WuC, the intended node turns on its main radio for data 180 communication unless EDT without ACK, which will be presented in the next section, is employed. When both MRs are active, data communication is performed. A transmission cycle finishes with an ACK from the targeted receiver.
It is worth mentioning that no ACK is necessary to acknowledge the successful delivery of a WuC transmission to avoid frequent switching of radio operation 185 mode [10] [23]. The working principle of TI-WuR is shown in Fig. 2.

Enabling ES and EDT for WuR
In this section, we propose ES and EDT tailored for WuR-enabled IoT/WSNs and explain their working principles.

190
Typically a WuRx needs to decode the full address of a WuC before it decides whether to wake its MR up or not [25]. This procedure is referred to as full address decoding (FAD), which is a typical scheme for WuC address decoding.
In this paper, we propose to decode an address bit-by-bit in order to improve the performance of FAD.

195
More specifically, the ES scheme defines an address decoding and validating rule that is able to diminish WuC overhearing energy cost for unintended WuRxs. It uses partial MCU functions to decode and validate the address. Different from FAD, ES allows a node to decode and match the received address bit-by-bit instead of decoding and validating a complete address after all bits 200 are received. That means, the MCU decodes the first bit of the received address sequence and matches it with its corresponding address bit. If the first bit of the received address matches, then it will decode and validate the next bit. For an intended receiver, this process will continue until the whole address matches with its own address. For a non-intended receiver, whenever a bit of 205 the received address mismatches with its own, the node stops the decoding and matching process and goes to sleep immediately. In this way, ES reduces energy consumption in a network by letting overhearing nodes sleep at an earlier stage before the whole WuC decoding procedure is complete.
As an example, consider a network cluster consisting of 16 nodes and let 210 us use a 4-bit local address for illustration. All nodes in this cluster hear each other. Assume now that a sender (address: "0000") transmits a WuC to wake up a targeted node (address: "1110") for performing data communication. All unintended nodes overhear the WuC transmission. If ES is employed, 7 nodes with addresses "0XXX" will go to sleep right after decoding and validating the 215 first bit. The remaining 8 nodes will decode the second bit. Among these 8 nodes, 4 nodes with addresses "10XX" will go to sleep right after decoding the second bit. This process will continue as more overhearing nodes go to sleep, thus reducing energy consumption for the whole network. In the end, only the intended node will wake up after decoding and validating all 4 bits.

220
The hierarchical structure of the address decoding and matching stage of ES is illustrated in Fig. 3. Clearly, the number of sleeping nodes in a WuR with ES will be increased in comparison with WuR with FAD.

Early Data Transmission Together with WuC
In this subsection, we propose an early data transmission scheme which is 225 tailored to WuR-enabled IoT/WSNs. The scheme is still referred to as EDT.
The proposed EDT scheme defines a data transmission procedure in which a small-size data packet is jointly encoded with the WuRx address and transmitted through a WuC before the MR is fully waken up. That is, a WuR transmitter performs data transmission simultaneously while sending a WuC with the sup-230 port of partial operation 1 of the main radio of an intended node. Such a scheme reduces the latency it takes to transmit small data and improves the energy efficiency of WuRxs. Depending on whether an ACK upon the successful reception of a small data is needed or not, EDT can be operated in one of the following two modes.

EDT with ACK
In this mode, the to-be-transmitted small data is encoded via an errordetecting code, e.g., cyclic redundancy check (CRC), at the transmitter side.
The checksum for a given polynomial of CRC is appended to the small data.
After that, the appended data will be encoded with the intended WuRx address.

240
Then the output bits of the XOR operation will be transmitted as the WuC frame. After receiving a WuC, the MCU of the MR performs partial functions to decode and validate the received the WuC.
More specifically, a reverse operation will be performed at the receiver side.
That is, the receiver will perform the XOR operation between its address and 245 the received packet. The output bits of the XOR operation will be divided by  result of the division is not equal to zero, it will treat the WuC as overhearing 250 and the MCU will go to the deep sleep mode as soon as the decoding process is finished. The data transmission cycle of EDT with ACK is illustrated in Fig.   4(a) and its workflow chart is shown in Fig. 5 respectively.
As an example, consider a TI-WuR scenario where a sender intends to send a small data, e.g., humidity 53% (data "110101") to an intended receiver (WuRx 255 address "11111100"). For a given polynomial, g(x) = x 2 + 1 (101), the data packet with CRC will be "11010111". Then, the CRC appended data packet will be encoded with the wake-up receiver address "11111100" using the XOR operation. The output bits of XOR, i.e., "00101011", will be transmitted as an encoded WuC. At the receiver side, a reverse operation will be performed to ex-260 tract the actual data message. That is, the intended receiver will perform XOR operation between its address "11111100" and the received WuC "00101011".
The output of this operation, "11010111", will be checked with the same CRC polynomial using division. If the remainder is zero, the intended node removes the CRC bits to extract the actual data. The detailed procedures for this data 265 encoding and decoding process at the transmitter and the receiver are presented in Fig. 6  is not necessary in a network. Accordingly, we propose another variant of the EDT scheme, i.e., EDT without ACK, which disables ACK. In this case, no ACK is transmitted from the MR of the intended receiver back to the sender upon a successful data reception. As both data and ACK transmissions are 275 performed when the MR is operated in the active mode, EDT without ACK further decreases latency and reduces power consumption since such a small data exchange cycle is accomplished without fully waking up the MR, as shown in Fig. 4(b).
Furthermore, both versions of EDT can be used to transmit a data packet 280 based on a given WuC address length. In our prototype implementation to be presented in Section 7, we have designed the WuC address length as 16 bits, allocating 10 bits to represent data values. However, a shorter or longer address Step 3. Transmit output bits as WuC

Receiver
Step Step 3. Send an ACK message  Table 2.
In most of the existing unslotted carrier sense multiple access with collision avoidance (CSMA-CA) based models, e.g., [29], a state transition is based on the approximation of mini-slot/symbol duration. However, due to the on-demand nature of data transmission in WuR-enabled IoT/WSNs, a node can wake up as the period in which a device is continuously busy [30]. All packets arrived in a busy period will be served (successfully or unsuccessfully) in that busy period itself. The interval between two busy periods is an idle period. The regenerative cycle in our model is shown in Fig. 7.
A regenerative cycle consists of a busy and an idle period. For analysis 315 simplicity, we consider that all busy periods have a constant duration which is equal to its mean value, T B . For a given λ, this assumption is reasonable since no retransmission is allowed, the size of IoT small data packets (e.g., temperature, T tout ack ≈ T SIFS + T ack where T ack and T SIFS are the duration of ACK and short inter-frame space (SIFS) respectively, is needed for identifying an unsuccessful where Γ is the number of packets served in a busy period of the M/G/1 queuing system with traffic intensity α = λ T S. E[Γ] can be computed as Thus, P c can be approximated as (3)

By inserting the value of E[Γ], E[T B
] for a given λ, the value of P c can be obtained.

Performance Metrics
In this section, we define three metrics for performance evaluation, i.e., packet delivery ratio, latency, and energy consumption and derive the their expressions.

335
Denote by P s the PDR. It is defined as the ratio between the number of successful transmissions and the total number of transmission attempts during a regenerative cycle. Since there exists only one reason for packet loss in this study, i.e., packet loss due to collisions, P s can be computed from (3) as follows

Latency
The average latency of a successfully delivered packet, denoted by T d , is defined as the duration from the time instant that a packet arrives at the transmission queue of the data generation node until the moment the packet is successfully transmitted to the intended node. Since no retransmission is allowed, T d for an unsuccessful transmission is almost the same and it is obtained by where T ST and T FT are the duration of a successful transmission and a failed transmission respectively.
Furthermore, the obtained value of T d depends on the scheme employed in the network. For ES, T ST = T wuc +T AT +T data +2 T SIFS +T ack where T wuc , T AT , 340 and T data are the duration for WuC transmission, fully active MCU, and data transmission respectively. Since no separate transmission is required for sending data, the duration of a successful transmission for EDT with ACK or without ACK is T ST = T wuc + T AT + 2T SIFS + T ack or T ST = T wuc + T SIFS respectively.

345
Denote by E the energy consumption for the whole network consisting of one transmitting node, one destined node and (N − 2) unintended nodes. We where E T is the energy consumption of the sending node for a packet transmission (successful/unsuccessful). It is given by where E T c and E T x are the energy consumed by the sender for a successful and unsuccessful packet respectively. E R is the total reception energy consumed by the nodes in the network with N nodes for one packet transmission. It includes the energy consumed by the destined node for receiving a packet and the energy consumed by the other (N − 2) unintended, i.e., overhearing nodes. E R is given by where E R c represents the total reception energy consumed by a network for an unsuccessful transmission. Assuming that collisions happen due to the simultaneous transmissions of two nodes, then E R c can be estimated as E R c = (N − 2)E idle . E idle is the energy consumed by a node when it is actively monitoring the channel using WuRx. For a collided packet, a node (intended or 350 unintended receiver) consumes the same energy as needed for idle listening since it cannot detect and decode the packet. The total reception energy consumed by the network for a successful packet, E Rx , can be calculated as where E IN and E UN are the energy consumption of the destined node and the unintended nodes respectively. Lastly, the average reception energy consumed by a node for a packet transmission, denoted by E avg R , is obtained as , and E T ack are the energy consumption respectively for WuC reception, data packet reception, and ACK transmission. With an assumption that N is exponential to the power of 2, E UN for ES can be approximated as where P R wuc , T b , E R prm , and B represent the reception power of WuC, the time needed to decode and validate 1 bit of an address, the energy needed for pream-355 ble detection and partially switching on MCU, and the total number of bits in the received address, respectively.
In EDT, no separate transmission is required for sending a small data packet.
Thus, E T c , E T x , and E IN for EDT with ACK can be obtained as Furthermore, since no ACK is required for EDT without ACK, E T c , E T x , and E IN are obtained as follows Lastly, E UN is the same for both EDT with and without ACK and it is given

360
In this section, we first give a brief overview regarding the prototype implementation of our WuR testbed and then present the experiments which validate the functionalities of the ES and EDT schemes.

Implementation Overview
Based on the principles presented above, we have implemented a small-scale As illustrated in Fig. 9, the WuRx consists of four blocks: a matching network, an envelope detector, a comparator, and a preamble detector. The

Experimental Validation of the ES Functionality
Based on the developed WuR prototype, we have performed various experiments to measure the duration of a WuC address decoding procedure for both FAD and ES. In both cases, whenever a WuRx detects a WuC signal, it starts to process it. As soon as the preamble detector detects a preamble, it provides 400 an interrupt signal to the MCU to initialize a low frequency clock and generalpurpose input/output so that address decoding can start, with the MCU in partial operation.
We have measured the duration to decode and validate one bit of a received WuC address and find out that it is 10 ms. As shown in Fig. 10 Fig. 3 where a 4-bit address is used.

415
It would take 55 ms to decode this address if FAD is used. Fig. 11 reveals that the duration for a WuRx stops processing a WuC address after decoding and validating the first three bits of a 4-bit address. It is observed that the WuRx needs 43 ms (between A2 and A1) before it goes to sleep after decoding and validating the first three bits of a 4-bit WuC address "1110" since the fourth 420 bit mismatches with its own.

Experimental Validation of the EDT Functionality
Based on the same prototype, experiments to validate the functionality of EDT are also carried out. For EDT, a WuRx follows a similar procedure as needed for FAD for WuC address decoding and data processing. After demod-425 ulating the WuC packet, the XOR operation is performed between WuRx's address and the received packet. The output bits of XOR will be treated by the CRC polynomial for error check. In case of EDT with ACK, the WuRx will trigger the MR to wake it up from light sleep to active for sending an ACK upon each successful data reception. When EDT without ACK is employed, however, 430 the WuRx will not trigger the MR for ACK transmission. It is worth mentioning that the measured duration to decode and validate one bit of a received WuC packet in EDT is also 10 ms, indicating that EDT does not add any extra latency for data processing. Instead, shorter latency can be achieved since no data transmission is performed by the MR, in addition to energy saving. Based on the nRF UART 2.0 mobile APP development kit 5 , we have developed an Android APP programmed in Android Studio. Fig. 12 illustrates a screenshot for data sniffing of an EDT experiment when EDT without ACK was employed. In this example, a pair of WuR devices exchanged data and the developed APP is used as a sniffer to capture data transmission and reception.

440
As mentioned earlier, the receiving node uses its WuRx and MCU in the light sleep mode for data reception but the MR is not fully activated. The address of the WuRx has 16 bits and the CRC polynomial is 101. For a data packet which has also 16 bits after CRC, we use 3 bits to represent up to 8 different data types and 10 bits to represent data values.

445
For example, data types 000, 001, and 010 represent temperature, humidity, and light intensity respectively. To report a temperature of 24 degrees Celsius, i.e., binary '0000011000', data type 000 will be appended. Then the data packet will be '0000000011000' before CRC. Assume that the address of the WuRx is configured as '1010101110101011'. After the CRC and XOR operations, the to-

Numerical Results and Discussion
In this section, we first validate the theoretical framework by comparing the analytical results with the simulation results, and then evaluate the performance of the proposed schemes with respect to the defined metrics.  Table 3 unless otherwise stated.

Simulation Configurations and Theoretical Framework Verification
To perform simulations, we constructed a custom-built discrete-event simulator in MATLAB which is similar to the one we built in our early work [26].
The developed simulator mimics the behavior of the studied WuR-IoT with the ES or EDT capabilities implemented respectively. That is, a node wakes up as soon as a packet arrives in the queue and maintains a timestamp to track the time duration of every single state. The sending node works in an on-demand manner, like in ALOHA, for sending its WuCs. Upon receiving a WuC, the targeted node decodes the WuC and switches on its MR for data communication

Metric-based Performance Evaluation
In this subsection, we first evaluate the performance of the proposed schemes 495 with respect to three parameters, e.g., packet delivery ratio, average latency, and energy consumption, and then estimate node lifetime.  Furthermore, the duration of WuCs, which is decided by the adopted address length and WuC data rate, has an impact on collision probability. The longer the WuC duration, the higher the P c , and vice-versa. This is because a longer WuC occupies the channel for a longer period of time than a shorter WuC.   performance and the resulted latency does not vary with traffic load. This is 525 because the latency defined in this study is based on successfully transmitted packets. To receive a packet successfully, both FAD and ES have to decode all address bits in a WuC. Thus, a same data transmission procedure applies to both schemes.

Average Latency
Compare now the latency performance among FAD-16, EDT with and with-530 out ACK, in Fig. 16. It is clear that EDT without ACK achieves the best performance. This is due to the fact that EDT without ACK does not need to wake up the MR to transmit an ACK upon a successful data reception. Based on the parameter configuration mentioned above, approximately 2 and 1. and EDT with ACK respectively 6 . It is worth mentioning herein that, although these two figures report the results for the same metric, they are based on two different address lengths which correspond to different network sizes. While Fig. 15 represents the obtained latency when the WuC address length is 5 bits (WuC duration = 65 ms), the results presented in Fig. 16 are based on a 16 540 bit address (WuC duration = 175 ms). Clearly, the latency from the latter case will be longer than the one obtained from the former case.

Average Reception Energy Per Node
The average reception energy consumption per node for the FAD-5 and ES-5 schemes is shown in Fig. 17  that the EDT schemes perform better than the ES scheme despite that non-570 destined nodes can sleep earlier using ES. This is because no separate data packet transmission is required for data transmission in EDT. Note however that EDT applies only to small data, e.g., with 10 bits for data as mentioned earlier in our testbed experiments.  As mentioned in the previous subsection, EDT does not avoid overhearing.

590
To activate EDT, all nodes in the network need to decode and validate the full range of the WuC address. Thus, the overhearing energy consumptions for FAD, EDT with and without ACK are the same. When ES is employed, however, it will perform best in terms of minimizing overhearing since an unintended ES node needs only partial address decoding before going to sleep. 595

Lifetime Estimation
Let us now calculate the expected lifetime of the studied network. Assume an ideal homogeneous network in which all nodes behave identically under a collision-free condition. Then the network lifetime is the same as the lifetime of a randomly selected node, assuming that all nodes have the same amount 600 of initial energy, deplete their energy at the same rate, and have therefore the same node lifetime. To perform our lifetime estimation, we follow the packet arrival pattern for typical IoT applications considered in [2,3], i.e., one packet per every other hour. We further assume that each device is powered by a 3V button/coin cell battery with capacity 220 mAh. A decreasing rate of 2% for 605 battery self-discharging is included in our calculations. Table 4 depicts the network lifetime for the studies schemes, i.e., FAD, ES, and EDT with two flavors. From Table 4, we observe that ES performs slightly better than FAD. Among without EDT, EDT with ACK, and EDT without ACK, the EDT scheme without ACK is the best in terms of lifetime. The 610 lifetime of ES is longer than all other schemes since it uses short local WuC address (e.g., 5 bits for the network with N=32) and bit-by-bit address decoding scheme.
The reason for this insignificant lifetime extension of ES over FAD or EDT over without EDT is the fact that the energy consumed for one successful packet transmission is higher than the energy consumed for reception in a WuR-IoT device, i.e., WuC transmission energy is dominating the total energy consumption of a network. We argue, however, the proposed schemes are still meaningful since a real-life WSN/IoT application for environment surveillance may experience much lower traffic load as configured in this study.

Further Discussions
Finally, it is worth mentioning that in this study we regard concurrent transmissions of two or more devices as collisions regardless of the interference level.
When capture effect is taken into account, a transmission which is much stronger than other transmission(s) to the same receiver may survive. In such a case, the 625 achieved packet delivery ratio would be slightly higher but the obtained delay and energy consumption would be basically the same since our schemes do not allow retransmission and a collided packet consumes the same amount of energy.
To precisely analyze the performance under more realistic conditions, one may consider that packet losses due to capture effect and channel impairments are 630 statistically independent of losses due to protocol behavior [28].

Conclusions and Future Work
In this paper, we have proposed two schemes, i.e., early sleep and early data Another direction is to expand our testbed to a larger scale and perform more real-life experiments for metric based performance evaluation.