Impact of using CSS PHY and RTS/CTS Combined with Frame Concatena-tion in the IEEE 802.15.4 Non-beacon Enabled Mode Performance

: This paper studies the performance improvement of the IEEE 802.15.4 non-beacon-enabled mode originated by the inclusion of the Request-To-Send/Clear-To-Send (RTS/CTS) handshake mechanism resulting in frame concatenation. Under IEEE 802.15.4 employing RTS/CTS, the backoff procedure is not repeated for each data frame sent but only for each RTS/CTS set. The maximum throughput and minimum delay performance are mathematically derived for both the Chirp Spread Spectrum and Direct Sequence Spread Spectrum Physical layers for the 2.4 GHz band. Results show that the utilization of RTS/CTS significantly enhances the performance of IEEE 802.15.4 applied to healthcare in terms of bandwidth efficiency.


Introduction
IEEE 802.15.4 is the de-facto communication standard [1], [2] that provides low-power and low-data-rate communication for Wireless Personal Area Networks (WPANs) and defines both Physical (PHY) and Medium Access Control (MAC) layers.Various Working Groups within IEEE 802.15.4 have been putting great efforts on developing new spectrum resource usage mechanisms or include the best of the best already existing ones for WPANs at Industrial, Scientific and Medical (ISM) and unlicensed bands [2].The idea has been to respond to the demands of the evolution of Wireless Sensor Network (WSN) applications, offering low power consumption but also higher data rates when it is needed.
The Request-To-Send/Clear-To-Send (RTS/CTS) scheme has not been considered in any of the existing IEEE 802.15 standards but facilitates to shorten the duration of frame collisions, as shown in [3].The proposed scheme involves the exchange of short RTS and CTS control frames prior to the exchange of the actual data frames.The RTS/CTS mechanism enables to reserve the channel and gives away from repeating the backoff phase for every consecutive transmitted frame.The fields of WSN applications in which the use of RTS/CTS assumes particular importance include industrial manufactory, healthcare and augmented reality.In these fields there is a need of sharing bursts of information with low collision probability.Although the proposal of employing RTS/CTS is not new and has already been standardized and implemented in legacy Wi-Fi (since it shortens frame collision duration, as shown in [3]), this reservation scheme has not been considered in any of the existing IEEE 802.15.4 standards, e.g., in the context of Higher Rate IEEE 802.15.4t or the developments of Wireless Next Generation Standing Committee (SCwng).In later editions of IEEE 802.11, the RTS/CTS mechanism has been enhanced in order to perform channel reservation in a more efficient manner.
The research developed in the context of this work shows that inclusion of the RTS/CTS mechanism significantly improves network performance, clearly demonstrating that its omission is not beneficial for the IEEE 802.15.4 standards.The non-beacon-enabled mode of the IEEE 802.15.4 MAC sub-layer enhancement by employing the RTS/CTS handshake scheme combined with frame aggregation concatenation is evaluated.Unlike the Direct Sequence Spread Spectrum (DSSS) PHY layer for the 2.4 GHz frequency band, which only supports data rates up to 250 kb/s, the Chirp Spread Spectrum (CSS) PHY enables speeds up to 1 Mb/s.Thus, apart from supporting off-body WPAN communications, it can provide on-body networking similarly to IEEE 802.15.6, as in [4], not supporting however in-body communications [5] for healthcare.Inspired by [6], we considered RTS/CTS combined with frame concatenation in our initial work published in [7], while authors in [4] have considered our approach to introduce RTS/CTS in the beacon-enabled MAC protocol of IEEE 802.15.6 supporting unobtrusive medical services to individuals with chronic health conditions.
One assumes that wireless nodes use equal backoff procedure from the IEEE 802.15.4 basic access mode.Nonetheless, this procedure is only repeated for each RTS/CTS set and not for each individual data frame.As such, channel utilization is optimized by decreasing the deferral time before transmitting each data frame.Differently from [7] and [8], in this work, the performance enhancement of applying RTS/CTS is studied for both the CSS and DSSS PHY layers that operate in the 2.4 GHz ISM band.
The remaining of this paper is organized as follows.Section 2 explores the formulations for time delay after addressing aspects of the MAC sub-layer and control messages flowchart for the non-beacon-enabled mode of IEEE 802.15.4.Section 3 presents the MAC sub-layer system model for the minimum delay and maximum throughput.Section 4 presents performance results for delay, throughput and bandwidth efficiency.It also extracts lessons from the comparison between the application of CSS and DSSS PHY layers for healthcare services support.Finally, conclusions are drawn in Section 5.

MAC Sub-layer
In the IEEE 802.15.4 basic access mode [8], nodes use a non-beacon-enabled CSMA-CA algorithm for accessing the channel and transmit their frames.The unslotted Carrier Sense Multiple Access with Collision Avoidance (CSMA-CA) in the non-beacon-enabled mode facilitates a better flexibility for large-scale IEEE 802.15.4-compliant peer-topeer networks [9].Before each transmission the MAC sublayer exchanges messages with the PHY layer for frame transmission (TX)/reception (RX).Figure 1 presents the algorithm's flowchart showing the interaction between the different frame types (e.g., DATA and ACK) and the control messages involved in frame transmission/reception.
The PHY and MAC layers exchange control messages every time an event occurs as follows: PHY -> MAC  RX_START: Start of message indicator;  RX_FAIL: Failed to receive message after RX_START.The message can fail because Cyclic Redundancy Check (CRC) or collision;  TX_END: Message being transmitted has completed;  TX_FAIL: End of transmission (like TX_END) but the message transmission has failed.For most radio transceivers this should never happen (but there are valid cases for packet-based radios, e.g., CC2420).After starting carrier sense, one of the following messages must be sent to the MAC sub-layer:  CHANNEL_IDLE: If the specified "frame length" has been processed, and the carrier sense returns channel not "busy";  CHANNEL_BUSY: If the carrier sense returns channel "busy".

MAC -> PHY
 SET_TRANSMIT: Switch the PHY layer to the transmit mode; If slotted CSMA-CA is used each operation (channel access, backoff counter and Clear Channel Assessment, CCA) can only occur at the boundary of a Backoff Period (BP).Additionally, the BP boundaries must be aligned with the slot boundaries of the superframe time [10].
In non-slotted CSMA-CA the backoff periods of one node are completely independent of the backoff periods of any other node in a PAN/Body Area Network (BAN).The backoff phase (generally called contention window in 802.15.4) algorithm is implemented by considering basic units of time called backoff periods.The backoff period duration is equal to TBO = 20 × Tsymbol (i.e., 0.32 ms), where Tsymbol = 16 μs is the symbol time [4].Before performing CCA, a device shall wait for a random number of backoff periods, determined by the backoff exponent (BE).Then, the transmitter randomly selects a backoff time period uniformly distributed in the range [0, 2 BE − 1].Therefore, it is worthwhile to mention that even if there is only one transmitter and one receiver, the transmitter will always choose a random backoff time period within [0, 2 BE − 1].Initially, each device sets the BE equal to macMinBE, before starting a new transmission and increments it, after every failure to access the channel.In this work we assume that the BE is not incremented since we are assuming ideal conditions.
Table 2 from [11] summarizes the key parameters for IEEE 802.15.4 both employing and not employing RTS/CTS with frame concatenation in the 2.4 GHz band, by considering the DSSS PHY layer with the O-QPSK modulation (250 kb/s).IEEE 802.15.4 [8] nodes support a maximum overthe-air data rate of 250 kb/s.However, in practice, the effective data rate is lower due to the protocol timing specifications [8].This is also explained by the various mechanisms that are employed to ensure robust data transmission, including channel access algorithms, data verification and frame acknowledgement.In this work, unicast data transmissions with ACKs are addressed and the channel access time is a dominant factor in the overall performance of the network.The non-beacon-enabled mode is considered.The regular procedure of the IEEE 802.15.4 non-beacon-enabled mode is presented in Figure 2. When a device attempts to transfer data, it simply transmits its data frame, using unslotted CSMA-CA, to the coordinator.The coordinator acknowledges the successful reception of the data by transmitting an ACK control frame.The beacon-enabled mode is not considered because collisions can occur between beacons or between beacons and data or control frames, making a multi-hop beacon-based network difficult to be built and maintained [12].Another important attribute is scalability, an intrinsic characteristic of multi-hop WSNs.Changes in terms of network size, node density and topology may occur.Nodes may die over time mainly due to energy depletion.Other nodes may be added later, and some may move to different locations.Consequently, for such kind of networks, the non-beaconenabled mode better adapts to the scalability requirement than the beacon-enabled mode.In the former case, all nodes are independent from the PAN coordinator and the communication is completely decentralized.
Moreover, for beacon-enabled networks [8], there is an additional timing requirement for sending two consecutive frames, so that the ACK frame transmission should be started between the TX/RX or RX/TX switching time, TTA, and TTA+TBO.Hence, there is time remaining in the Contention Access Period (CAP), for the message, as well as appropriate Interframe Space (IFS) and ACK. Figure 3 presents the timing requirements for transmitting a frame and receiving an ACK for the beacon and non-beacon-enabled modes.
In IEEE 802.15.4 [8], [10], the CSMA-CA algorithm is significantly different from the one used in IEEE 802.11e [13].The main differences are related to the backoff algorithm.While in IEEE 802.11e [13] the value of the Contention Window (CW) depends on the number of failed retransmissions for the frame, in the basic access mode for IEEE 802.15.4,this value (denoted as backoff phase) depends on the Backoff Exponent (BE), and Number of Backoffs (NBs).Moreover, in IEEE 802.11e, the backoff time counter (BOc) is decreased as long as the channel is sensed idle and is frozen when a transmission occurs.In the IEEE 802.15.4 basic access mode, nodes do not continuously monitor the channel during the backoff phase and the sensing phase (i.e., CCA) only occurs at the end of the backoff phase.
According to the IEEE 802.15.4 standard [10], a sensor node that sends a data or a MAC command frame with its ACK Request subfield set to one shall wait for at most an ACK wait duration period, TAW, for the corresponding ACK frame to be received.The TAW already includes the time for the ACK frame itself.The transmission of an ACK frame in a non-beacon-enabled PAN or in the Contention Free Period (CFP) shall start aTurnaroundTime symbols (i.e., 192 μs) after the reception of the last symbol of the DATA or MAC command frame ([10], Section 7.5.6.4.2).
The ACK wait duration period, TAW, is given by: Assuming the DSSS PHY layer for the 2.4 GHz band, the maximum ACK wait duration period, TAW, is given by:   By assuming a DATA and an ACK frame with 18 and 11 bytes, respectively (including the PHY and MAC overhead), the transmission time is 576 μs and 352 μs, respectively.Besides, Figure 4 also includes the ACK wait duration period, TAW.For every DATA frame transmitted, there is a random deferral time period, DT, before transmitting, given by: The initial backoff period, Initialbackoff Period, is given as follows: whereas the time delay, due to CCA, is given by:  =  +  (5) The rxSetupTime is the time to setup the radio from a previous state to the transmission or reception states, and it mainly depends on the radio transceiver used.During the TCCA, the radio transceiver must determine the channel state within 8 symbol duration (i.e., 128 μs, which corresponds to one symbol duration of 16 μs).In a normal transmission, for every DATA frame sent an ACK must be received, as shown in Figure 5. Details on the analytical model for the maximum throughput and minimum delay are given in [11].

Brief overview of the MAC sub-layer system model
The main reasons for which IEEE 802.15.4 basic access mode does not consider the adoption of the RTS/CTS handshake mechanism are the following: a) The introduction of RTS/CTS frames adds protocol overhead and, in low traffic load cases, short frame sizes could have the same order of magnitude of a RTS/CTS frame; b) The absence of a RTS/CTS handshake mechanism allows to reduce the system complexity.Although these assumptions are true for some particular cases, we argue that in the presence of link layer errors the additional protocol overhead due to the use of RTS/CTS frames is mitigated by the resulting concatenation mechanism.
In our proposal, we assume that both the RTS and CTS frames have the structure of an ACK frame, which is assumed to have a limited size of 11 bytes, as shown in Table 2 from [9].The maximum data payload for IEEE 802.15.4 depends on the application (maximum payload could range between 102 and 118 bytes).Consequently, the length of the data frames could be approximately ten times larger than the control frames length.In reality, IEEE 802.15.4 employing RTS/CTS with frame concatenation is composed by the following time periods: backoff phase, CCA mechanism, time needed for switching from receiving to transmitting, RTS transmission time, time needed for switching from transmitting to receiving and CTS reception time.
Both the IEEE 802.15.4 basic access and the proposed RTS/CTS schemes consider acknowledgment (ACK) frames to confirm successful frame reception.Aiming at overhead reduction, the use of RTS/CTS frames enables channel reservation and avoids the replication of the backoff phase for every consecutive transmitted frame and implies zero backoff exponential.Moreover, by considering RTS/CTS, nodes avoid frame collisions, which often take place due to the hidden terminal problem.Hence, IEEE 802.15.4 performance is considerably enhanced, since the number of retransmitted frames is significantly decreased.Differently from [11], block acknowledgement is not considered, and such frame concatenation mechanism is not applied.
In reality, authors in [7] have demonstrated that one fundamental reason for IEEE 802.15.4 MAC inefficiency is overhead, originated, e.g., by inter-frame spaces from the protocol, backoff period, transmission of PHY/MAC headers and ACKs, interference and retransmissions (due to unsuccessful reception of data frames).The un-slotted CSMA/CA algorithm and the backoff phase are characterized by using the formulation presented in [14].N.B.: we designate the backoff phase by Contention Window (CW).
To determine the maximum average throughput, Smax, for the basic access mode, the minimum average delay, Dmin is first derived.Figure 6 presents the frame structure for the IEEE 802.15.4 basic access mode in the absence of RTS/CTS while considering retransmissions [14].As mentioned above, under IEEE 802.15.4 employing RTS/CTS and frame concatenation, nodes use the same backoff procedure as in 802.15.4 but only for each RTS/CTS set.Hence, the channel utilization is enhanced by decreasing the deferral time before transmitting each data frame, as shown in Figure 7.
The minimum delay due to CCA, Dmin_CCA_RTS, (enabled to estimate if the channel state is busy or idle after the backoff phase), and prior to each RTS/CTS set, is given by: As in [15], the number of backoff periods is given by NB ∈ [0, NBmax].The time delay due to CCA is given by ( 5).Equation ( 6) considers that nodes only determine the channel state once per RTS/CTS exchange, i.e., if the total number of transmitted data frames is, for example, n=100 and the number of aggregated frames is Nagg=10, nodes only determine the channel state n/Nagg=10 times, i.e., once per exchange of RTS/CTS, plus the time needed for transmitting the frames (until the maximum limit for the number of retries, NBmax=4, is reached).
If the estimation of the channel is idle during CCA and, if after sending a data frame an ACK is not received within a duration equal to TAW, the retransmission process does not consider a backoff phase between two consecutive data frames.This simplification decreases the total overhead, as shown in Figure 7. Since any other station will receive all RTS/CTS/DATA/ACK frames, in the first transmission attempt, it will set the Network Allocation Vector (NAV).The NAV is responsible for defining the time duration for channel access deferring in order to avoid collisions.
With erroneous channels under IEEE 802.15.4 employing RTS/CTS, if the channel estimation is idle during CCA, there is data transmission and an ACK is not received within a duration of TAW, the delay due to frame retransmissions RTXs), is given by: where j is the number of RTXs, which varies up to MaxRet [1].
From the analysis of Equation ( 1), we can conclude the following: a) After CCA, if a node determines that the channel is found to be idle and an ACK is correctly received for each sent frame, the minimum delay, DminDataRetRTS, is determined by: where TTA is the TX/RX or RX/TX switching time.TDATA, TACK, and TIFS are the durations of the data frame, ACK frame and inter-frame spacing (IFS), respectively.Since transmission errors do not exist, the number of retransmissions is j=0.As such, in Equation ( 8), there is no need to consider the ACK wait duration period, TAW, which represents the longest time needed to receive an ACK control frame.b) After CCA, if a node estimates that channel is idle and an ACK has not been received within the duration TAW, for one or more transmitted frames (since we consider frame concatenation), the minimum delay due to frame RTXs, DminDataRetRTS, is determined by:  =  +  +  The term The minimum average delay, Dmin_RTS, accounting for channel state and frame RTXs is obtained by combining Equations ( 6)-( 9): The maximum average throughput, by considering frame RTXs, Smax_RTS, in bits per second, is then given by: which means that the maximum average throughput is easily achieved by knowing the minimum average delay.

Analytical and Simulation Results
We have compared IEEE 802.15.4 employing and not employing RTS/CTS by using the MiXiM framework of the OMNeT++ simulator [15].A two-hop network, with two sources nodes, one relay and two sink nodes has been considered.Two interferers are responsible for sending broadcast frames that collide with the frames sent by the sources and central node.The DSSS and CSS PHY layers performance analysis considers several runs with five different random seeds and a 95 % confidence interval.A perfect match between analytical and simulation results was obtained.
Table 1 compares channel access times and overhead for the DSSS and CSS PHY layers.Respective maximum data rates are 250 kb/s and 1 Mb/s [1].Likewise in [7], we analysed both Smax and Dmin.
Although we consider the 2.4 GHz band, the proposed formulation can be also applied to other frequency bands.A fixed payload size LDATA=3 bytes is considered.In [17], authors proved that, for short frame sizes, IEEE 802.15.4 achieves poor performance.Here, it is shown that the proposed mechanism can significantly improve channel efficiency even with retransmissions (RTX of 10 % of frames is assumed).The performance of CSS (1 Mb/s) is clearly better than the one of the DSSS PHY layer (250 kb/s).Moreover, performance results for Dmin as a function of the number of TX frames, for the DSSS PHY, show that when RTS/CTS with frame concatenation is considered, for 5 and 10 aggregated frames, Dmin decreases (Smax increases) 8 % and 18 %, respectively.For more than 28 aggregated frames, Dmin decreases (Smax increases) ~30 %, as shown in Figure 11.On the other hand, for the CSS PHY layer, by using RTS/CTS with frame concatenation, for 5 and 10 aggregated frames, Dmin decreases (Smax increases) 33 % and 59 %, respectively.For more than 28 aggregated frames, Dmin decreases (Smax increases) ~71 %.

Lessons Learned
The use of the RTS/CTS mechanism improves channel efficiency by decreasing the deferral time before transmitting a data frame.Although the RTXs are addressed here in a somehow rigid approach, the proposal shows that, even for the case with RTXs, if the number of aggregated frames is lower than five, IEEE 802.15.4 employing RTS/CTS combined with concatenation achieves higher values for the throughput in comparison to IEEE 802.15.4 without RTS/CTS.The advantage comes from not including the backoff phase into the RTX process like in the IEEE 802.15.4 basic access mode (i.e.BE = 0).results show that the CSS PHY (1 Mb/s) efficiency gain is clearly more evident.

Conclusion
This paper proposes a retransmission model for the non-beacon-enabled mode of IEEE 802.15.4 that employs RTS/CTS and frame concatenation, in which the backoff procedure is not repeated for each data frame sent, but only once for each RTS/CTS set.Performance results clearly show the substantial benefits of using RTS/CTS, in terms of bandwidth efficiency.In particular for the CSS PHY layer, the proposed MAC sub-layer protocol shows a clear reduction of the minimum delay, enhancement in maximum average throughput, and improved bandwidth efficiency, from ~2.5 % to ~4.2 % when the CSS PHY is considered.This enhancement will be beneficial for WBAN off-body and on-body communications in the healthcare ecosystem.The study of the energy efficiency of the proposed MAC sub-layer enhancement under specific application of machine-type communications is left for further study.

Figure 4
Figure 4 presents the ACK timing required for the IEEE 802.15.4 standard, by considering the DSSS PHY layer for

Fig. 6 .Fig. 7 .
Fig. 6.IEEE 802.15.4 basic access mode with retransmissions when channel is assessed as a) busy and b) idle.
of the Nagg−m transmitted frames whose ACK response was successful, where m denotes the number of transmitted (TX) frames that need retransmission.Due to lack of ACK frame reception, each individual frame can be retransmitted more than once.The term ji represents the number of RTXs until MaxRet has been reached.The last term corresponds to successful reception of the ACK.

Figures 8 ,
Figures 8, 9 and 10 present Dmin, Smax and the bandwidth efficiency, , as a function of the number TX frames [16] for the DSSS and CSS PHYs:

Fig. 8 .
Fig. 8. Minimum average delay versus the number of TX frames for the basic access and RTS/CTS modes.

Fig. 9 .
Fig. 9. Maximum average throughput versus the number of TX frames for basic access and RTS/CTS modes.

Fig. 10 .
Fig. 10.Bandwidth efficiency versus the number of TX frames for IEEE 802.15.4 basic access and RTS/CTS modes.

Fig. 11 .
Fig. 11.Increase of Smax/decrease of Dmin as a function TX frames for IEEE 802.15.4 in the RTS/CTS mode.As another proposal for future work, we are planning to consider the enhancements of the RTS/CTS mechanism to perform channel reservation in a more efficient manner.

Table 1
Comparison of the values and time parameters between the DSSS and CSS PHY layers for IEEE 802.15.4 (2.4 GHz band).