当前位置：和泉文库 > 计算机 > A-Duplex：Medium Access Control for Efficient Coexistence Between Full-Duplex and Half-Duplex Communications

A-Duplex：Medium Access Control for Efficient Coexistence Between Full-Duplex and Half-Duplex Communications

文件格式：PDF，文件大小：1.76MB，售价：4.45元

文档详细内容（约15页）

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 14, NO. 10, OCTOBER 2015 5871 A-Duplex: Medium Access Control for Efficient Coexistence Between Full-Duplex and Half-Duplex Communications Aimin Tang and Xudong Wang, Senior Member, IEEE Abstract—As full-duplex wireless communication evolves into a practical technique, it will be built into communication nodes in many application scenarios. However, it is difficult to do so for legacy communication nodes. Thus, full-duplex communication nodes will coexist with half-duplex communication nodes in the same application environment. In this paper, a wireless local area network with a full-duplex access point (AP) and half-duplex clients is studied, and a media access control (MAC) protocol called asymmetrical duplex (A-Duplex) is developed to support ef- ficient coexistence between half-duplex clients and the full-duplex AP. A-Duplex explores packet-alignment-based capture effect to establish dual links between the AP and two different clients. In this way, the capability of a full-duplex AP can be utilized by half-duplex clients, which leads to much improved network throughput. Moreover, to ensure fairness of the MAC protocol, a virtual deficit round-robin algorithm is proposed for the AP to select appropriate half-duplex clients for dual-link setup. A-Duplex does not require any change in the physical layer of half-duplex clients; only an update of MAC driver is necessary. Thus, it is well suited for coexistence between half-duplex clients and a full-duplex AP. Both analysis and simulations are conducted to evaluate performance of A-Duplex. Results show that it improves the throughput by 48% and 188% and reduces the average packet delay by 26% and 22%, as compared to the IEEE 802.11 Distributed Coordination Function with and without RTS/CTS, respectively. Moreover, the throughput remains steady as the number of clients grows. A-Duplex also maintains a high level of fairness. Index Terms—Medium access control, full-duplex communication, wireless LAN, coexistence between full-duplex and halfduplex communications. I. INTRODUCTION F ULL duplex wireless communication is evolving into a more practical technique [1]–[5]. In [1], three antennas are used to achieve about 30 dB antenna cancellation. Combined with path loss attenuation, analog cancellation and digital cancellation, this design can achieve about 100 dB self-interference cancellation within 5 MHz bandwidth for IEEE 802.15.4. In [2], Balun cancellation is leveraged for an improved design of self-interference cancellation, in which only two antennas Manuscript received October 21, 2014; revised February 25, 2015 and June 6, 2015; accepted June 7, 2015. Date of publication June 12, 2015; date of current version October 8, 2015. The associate editor coordinating the review of this paper and approving it for publication was C.-F. Chiasserini. (Corresponding author: Xudong Wang.) The authors are with the UM–SJTU Joint Institute, Shanghai Jiao Tong University, Shanghai 200240, China (e-mail: wxudong@ieee.org). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TWC.2015.2443792 are used. With the help of path loss attenuation and digital cancellation, about 110 dB cancellation can be achieved within 20 MHz. In [3], an additional transmitting chain is added to generate an invert analog signal to cancel the self-interference. So far the best design is given in [4], where only one antenna is used to achieve almost perfect self-interference cancellation for full duplex WiFi radio. In this design, a passive circuit is first used to achieve 60 dB self-interference cancellation and then both linear and nonlinear digital cancellation are used to cancel the remaining self-interference. In [5], a balanced cancellation circuit is designed for RF front-end. This design can achieve 60 dB cancellation at the RF front-end. On the other hand, it can be integrated with the design in [4] to further improve the self-interference cancellation. Full duplex can nearly double the capacity of a point-to-point communication link and thus significantly improves the spectrum efficiency. However, to fully leverage the capability of full duplex communications in a network, an efficient media access control (MAC) protocol plays a critical role. So far a few MAC protocols have been proposed to support full duplex communications [6]–[9], but they assume that all nodes in the network are equipped with full duplex radios. However, in many application scenarios full duplex and half duplex radios have to coexist in the same network for two main reasons. First, despite fast progress in the development of full duplex radios, challenges still remain when applying full duplex radios to nodes such as smart phones or laptops. For example, in [1]–[3], [5], all radios need more than one antennas to achieve full duplex communications, and the distance between antennas are more than 20 cm. In [4], only one antenna is needed, but implementation of the circuit is rather complicated. These factors make it difficult for a smart phone or personal digital assistant (PDA) to adopt a full duplex radio. Second, many legacy devices only support half duplex communications. It is fine to replace a legacy AP by a full duplex AP, but replacing all legacy devices by full duplex ones is neither economical nor acceptable. Thus, coexistence between full duplex and half duplex communications becomes indispensable. For example, supporting a wireless LAN with a full duplex AP and half duplex clients is extremely meaningful for the next generation WiFi networks. In this paper, we study a full duplex wireless LAN where the AP supports full duplex communications but all clients are half duplex nodes. The full duplex AP is assumed to have perfect self-interference cancellation [4]. Since all clients have 1536-1276 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information

5s72 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS.VOL 14.NO.10.OCTOBER 2015 Preamble PACKET1 PACKET 2 AP to Client C will be longe Fig 1.A wireless LAN with a full duplex AP. AP ne packet om Chent A to the capture effect can help establish dual links but it cann no capability of full duplex communications.only asymr netric ensure throughput improvement.Asa result,to better utilize the capture effect n in Fig.I nds less LAN with a full duplex AP and hal AP the AP C).In [6] etric dual links (i.e.,AAP- c)do first:2)the received signal strength of the packet from the AP o a client is stronger than the interference from the packet sen cap from ano ther chent to th o fully leverage the capability of the full duplex AP.we develop that throug hput can be imp oved.To this end,the AP needs re effect at the client side to collect relative signal strength between nodes to build an be properly f full h clients can selecte n 1 is de ed to Thanks to the asymmetric dual links,our MAC pr ocol sur ports efficient co-existen ce between a full duplex AP and hal Ao be hidden to cachtke fom the AP to Client B.dug ex (A-Dup utilized to form full duplex communications at the AP.even though Clients A and B are not hidden from each other. improved by 48%and 188%as compared to DCF [14]wit RTS/CT 2线 proeL-align nt for hetter cantur A-Duplex is distinct with several features.First.it leverage Sant first.In either type.eket as shown in packet o fully uti lize the capa for the fir to ba cond packet (This e links example e first packet blist the based on this map asymm nks are e ba that of the sc environment.Third.a virtual deficit round robin algorithm is effect ma asymmetric dual links are established (eA ver update is peeded.To the best of our knowledge.this is the first MAC that holds these features the transmission rate from the AP to Client C is lower than nt sign of A-Dup links in A-Duplex are elaborated in Section IV.The fairness issue is studied in Section V.Simulation results are presented in

5872 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 14, NO. 10, OCTOBER 2015 Fig. 1. A wireless LAN with a full duplex AP. no capability of full duplex communications, only asymmetric dual links can be utilized to boost network throughput. For example, as shown in Fig. 1, while Client A sends a packet to the AP, the AP sends a packet to another client (e.g., Client C). In [6], asymmetric dual links (i.e., A → AP → C) do not allow collisions at both receivers, i.e., the signal from Client A does not interfere Client C. However, such a condition cannot be easily satisfied in a wireless LAN, so the capability of the full duplex AP cannot fully utilized. In order to fully leverage the capability of the full duplex AP, we develop a novel MAC protocol to explore capture effect at the client side to improve opportunity of full duplex communications in a wireless LAN. Capture effect [10]–[12] means, when a receiver receives two colliding packets, the stronger one can still be decoded correctly. For example, Client A and Client B may not be hidden to each other, but a packet from Client A to the AP may not corrupt the packet from the AP to Client B, due to capture effect at Client B. In this situation, capture effect is utilized to form full duplex communications at the AP, even though Clients A and B are not hidden from each other. The performance of capture effect can be significantly improved through alignment of two colliding packets. To leverage packet-alignment for better capture effect,1 there are two options: 1) the desired packet is sent first; 2) the interfering packet is sent first. In either type, if the preamble of the first packet does not collide with the second packet as shown in Fig. 2, it is relatively easy for the first packet to be decoded correctly. However, it is rather difficult to decode the second packet (This case is also called message in message (MIM) in [13]). For example, under the basic rate in IEEE 802.11a, the first packet can be decoded when its signal strength is 0 dB stronger than that of the second packet [12]. If we want to decode the second packet correctly, the signal strength of the second packet need to be 11 dB stronger than that of the first packet [12]. However, the asymmetric dual links supported via capture effect may not always improve the network performance. When asymmetric dual links are established (e.g., A → AP → C), the transmission rate from the AP to Client C is lower than that in the half duplex case since Client C is interfered by Client A. Thus, the transmission time of the packet from the 1If successive interference cancellation (SIC) is taken into account, then packet alignment is not necessary. However, the client side consists of legacy nodes, so any change to the physical layer is infeasible. Thus, SIC is not considered in this paper. Fig. 2. Capture effect without collision in preamble time. AP to Client C will be longer than that in the half duplex case. The concurrent transmission in dual links can save some time; however, if the transmission time from the AP to Client C is much longer than that of the packet from Client A to the AP, the dual links may reduce the throughput. In other words, the capture effect can help establish dual links but it cannot ensure throughput improvement. As a result, to better utilize the capture effect in a wireless LAN with a full duplex AP and half duplex clients, the MAC protocol must take into account three requirements: 1) the packet from the AP needs to be transmitted first; 2) the received signal strength of the packet from the AP to a client is stronger than the interference from the packet sent from another client to the AP; 3) the AP needs to choose a client with a proper transmission rate to establish dual links so that throughput can be improved. To this end, the AP needs to collect relative signal strength between nodes to build an information map from which clients can be properly selected to establish dual links effectively. In this paper, a dynamic information update scheme is designed to establish such a map. Thanks to the asymmetric dual links, our MAC protocol supports efficient co-existence between a full duplex AP and half duplex clients, so it is called asymmetrical-Duplex (A-Duplex) in this paper. A-Duplex not only improves the throughput but also reduces packet delay. Simulation results show that the throughput is improved by 48% and 188% as compared to IEEE 802.11 DCF [14] with RTS/CTS and without RTS/CTS, respectively. The packet delay is reduced by 26% and 22%, respectively. A-Duplex is distinct with several features. First, it leverages packet-alignment based capture effect to establish asymmetric dual links so as to fully utilize the capability of the full duplex AP. As a result, throughput is improved effectively as compared to the case of only using the hidden nodes to establish dual links. Second, a map of signal-to-interference ratio (SIR) is built at the AP; based on this map asymmetric dual links are established to take advantage of capture effect. The SIR map can be updated dynamically to capture the variable network environment. Third, a virtual deficit round robin algorithm is developed for the AP to select a downlink in dual link setup, which improves fairness of the MAC protocol. Finally, no physical layer change is needed at legacy nodes; only MAC driver update is needed. To the best of our knowledge, this is the first MAC that holds these features. The rest of the paper is organized as follows. Related work is introduced in Section II. The protocol design of A-Duplex is described in Section III. Detailed procedures of establishing dual links in A-Duplex are elaborated in Section IV. The fairness issue is studied in Section V. Simulation results are presented in Section VII. Compatibility and practicality issues are discussed in Section VIII. The paper is concluded in Section IX

TANG AND WANG:MAC FOR EFFICIENT COEXISTENCE BETWEEN FULL AND HALF-DUPLEX COMMUNICATIONS 5873 II.RELATED WORK A.Capture effect A当A mitted SFS ACK 1 destroy all the frames.If the receiving power of one of the arge eough than that other,the others NAVIRTS of receiving a frame from colliding frames is called (a) effect.The physical layer model for capture effect has been Delay ti ng de comes ear r has AP CTS DATA to node B ACK synchronized with the first frame.However.if a weaker frame B SIFS ACK 1 since the ver n be wit other be much larger than the weak one so that the receiver can re. 16].Since the receiver ta scenario:the transmis mak even e [121 the experimental results show that even two frame n time of the Howeve a pream ne th the SIRe and channel estimation which is consistent with [12].If packet alignment-based capture ofar the work in[]provides the best model for capture eftect is leveraged,then dB SIR is sufficient [12].For other des (ie. te),the pnbtknfhMaCp phor net ork It ador B.Full Duplex MAC noise(PN)sequence-based signatures toidentify the concurren links on-the-fly and explore the cha So far there exist a few mac protocols that supp full ance of exposed transmi duplex communications [6]-[8).In [6],both symmetric and ed.The ghput c are for by two no ng sig in Fig.1.A set of asymmetric dual links are esta IIL MAC PROTOCOL DESIGN ablished by three nodes in a two-hop setup.such as the links (BAP →C) In this section e the Ap full duplex C)in Fig. 11. full du ilit ork links need to be established.As explained before,to explore capture effect in an efficient way,the packet from the AP to downlin C)at the Client C:b)the A Howe since a lent is in th In 7.a reservation based MAC protocol was developed for the channel busy and cannot start a second link with the ap a network where all nodes have full du lex radios.All nodes are To resolve this issue,an RTS/CTS mechanism is added into the allocated wit tim ots by the Al In this M. pro to star dual links v re the AP-to-client lin dual lin -AP-C n00 an RTS if they are not hidden nodes,since the interference by Client Three cases are considered below.When a client gets the A is small at Client C.In this sche channel via the RTS frame,two cases need to be AP a pac ever packet-aigment based capture effect are not considered third case is that the Ap acquires the channel first.Note that

TANG AND WANG: MAC FOR EFFICIENT COEXISTENCE BETWEEN FULL- AND HALF-DUPLEX COMMUNICATIONS 5873 II. RELATED WORK A. Capture effect In random access, if more than one frames are transmitted concurrently, collision occurs. However, it does not always destroy all the frames. If the receiving power of one of the colliding frames is larger enough than that of the other, then the receiver can decode this frame correctly. Such a process of receiving a frame from colliding frames is called capture effect. The physical layer model for capture effect has been explored by various experiments in [10]–[12], [15]. Usually if the stronger frame comes earlier than other frames, it is easier to decode this frame since the receiver has already been synchronized with the first frame. However, if a weaker frame comes first, MIM is needed to decode the stronger frame. In this case, since the receiver has been synchronized with the weaker frame, the signal strength of the stronger frame must be much larger than the weak one so that the receiver can resynchronize with the stronger one to utilize capture effect [12], [13], [16]. Since the receiver takes a certain amount of time to achieve synchronization, even if the stronger frame arrives earlier, it is necessary to make sure the lead time is sufficient. In [12], the experimental results show that even two frames come at the same time, capture effect may still succeed in 802.11a networks. However, if the first frame is earlier by a preamble time, the success probability of capture effect can be improved, because a clean preamble is helpful to conduct synchronization and channel estimation. So far the work in [12] provides the best model for capture effect in 802.11a networks. It provides the measurement-based SIR capture thresholds for all the 802.11a bit rates. B. Full Duplex MAC So far there exist a few MAC protocols that support full duplex communications [6]–[8]. In [6], both symmetric and asymmetric dual links are considered. A set of symmetric dual links are formed by two nodes transmitting signal to each other simultaneously. An example is the link (A → AP → A) shown in Fig. 1. A set of asymmetric dual links are established by three nodes in a two-hop setup, such as the links (B → AP → C) shown in Fig. 1. In [6], the MAC protocol mainly explores symmetric dual links to improve network throughput, and in the asymmetric case (e.g., links (B → AP → C) in Fig. 1), dual links can be established only under the following conditions: a) the uplink transmission (B → AP) does not collide with the downlink transmission (AP → C) at the Client C; b) the AP transmits to Client C later. In [7], a reservation based MAC protocol was developed for a network where all nodes have full duplex radios. All nodes are allocated with some time slots by the AP. In this MAC protocol, asymmetric dual links are allowed even for non-hidden nodes. For example, dual links A-AP-C in Fig. 1 are allowed even if they are not hidden nodes, since the interference by Client A is small at Client C. In this scheme, AP first collects the interference information of clients and then schedules the transmission order according to the interference relationship. However, packet-alignment based capture effect are not considered Fig. 3. Dual link establishment. (a) The “AP-shorter” scenario: the transmission time of the packet from the AP to B is shorter than the packet transmission time of the other link plus a preamble time. (b) The “AP-longer” scenario: the transmission time of the packet from the AP to B is longer than the packet transmission time of the other link plus a preamble time. in this scheme. Thus, the performance of capture effect is limited. For example, in the basic operation mode (i.e., lowest rate with robust modulation/coding), the SIR needs to be 10 dB, which is consistent with [12]. If packet alignment-based capture effect is leveraged, then 1 dB SIR is sufficient [12]. For other modes (i.e., higher rate), the situation is similar. In [8], the MAC protocol addresses the exposed terminal problem in a full duplex network. It adopts the pseudo-random noise (PN) sequence-based signatures to identify the concurrent links on-the-fly and explore the chance of exposed transmissions. As a result, the throughput of a full duplex network can be improved. The protocol also assumes all nodes have full duplex capability. III. MAC PROTOCOL DESIGN In this section, a MAC protocol called A-Duplex is designed for a wireless LAN where the AP has full duplex capability, but all clients can only work in the half duplex mode. To utilize full duplex capability in this wireless LAN, asymmetrical dual links need to be established. As explained before, to explore capture effect in an efficient way, the packet from the AP to a client needs to start first. However, since a client is in the half duplex mode, once it detects such a transmission, it considers the channel busy and cannot start a second link with the AP. To resolve this issue, an RTS/CTS mechanism is added into the MAC protocol to start dual links where the AP-to-client link can be started first. As shown in Fig. 3, the procedure to set up a full duplex link always starts with an RTS frame by a client. Three cases are considered below. When a client gets the channel via the RTS frame, two cases need to be considered: 1) the AP can send a packet to another client to establish dual links; 2) the AP does not have a packet for another client. The third case is that the AP acquires the channel first. Note that

5s74 LEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,VOL 14.NO.10.OCTOBER 2015 RTS is only initiated by a client.so the AP does not g nerate A DATA to AP any RTS frame but respondstoaclient's RTS with a CTS frame an RTS cas Fig. CAP find AP others_ A and then transmits the packet to client B immediately.Note that in each frame the calle whic Fig.4.An example showing no dual link frame,it finds out that the duration in the CTS frame is longer The delay for Client A serves two purposes.First,for a legacy the pa time.The delay time can be computed by Eg.2).There are the packet to Client B is sent first two scenarios in this case.If the transmission time of the packet protected without interference rom th n the above procedur Client A in Fig.3 vaits a tim b busytone signal to ensure two transmissions finish at the same Frame Spacing (DIFS)time.Thus,the ACK may collide with me Otherwis as denoted by the "AP-longer h3(b).Client the packet from the other client.In our design.the CTS frame problem.I dura中 n in CI covers the enti and the AP finish their missions,Client B first returns an CTS fran update the NAV value ordins to the duration in ACK frame to the AP and then the AP returns an ACK frame the CTS frame.As a result.the ACK from the AP can also be to Client A upon re protected the aph L rly.Sin in the beeinnin the Client A does not know whether or not the AP will establish dual links.it computes the duration as if the A hepCteoraohercietomypiesaCTS es not set up From the cts frame Client A finds out that the duration in CT up a li he half me w A knows th duplex case.If the AP does set up dual links.it just uses th time and then starts a packet transmissio .When the packet is duration in the RTS frame to compute the duration of the CTS expla ed be transmission cove In the th and the AP As be set up i ot support full duplex communications.so symmetrical dual inks (1) selve the queue status as well as the length of the first packet in 3()the nC as follows.In the the AP knows exactly which client is about to send a pa et and the DurationcTs DurationgTS-TCTs-2TsIFS+Tp TACK. es how to establish dual links with another client.In [6] an业 time.In the case of Fig.3(b),the the packet is transmitted by OFDM symbols such as in 802.11a network,the AP cannot decode the packet header to get the Durationcrs =T2+TSIFs +2TACK ransm itting ent ad dress until it receives the wl e packet where t is the transmission time of the packet from the ap to AP Client B.When A receive the CTS frame,it decides the delay the packet is not interfered by the packet from Client A to the time according to can decod e the packet more easily throug Delay=DurationcTs-T-TsIFs-2TACK 21

5874 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 14, NO. 10, OCTOBER 2015 RTS is only initiated by a client, so the AP does not generate any RTS frame but responds to a client’s RTS with a CTS frame. The first case is shown in Fig. 3. Client A first transmits an RTS frame to the AP. When the AP finds that it has a suitable packet for Client B, it replies a CTS frame to Client A and then transmits the packet to client B immediately. Note that in each frame there is a field in the MAC header called duration, which is used for other nodes to update the network allocation vector (NAV) value. When Client A receives the CTS frame, it finds out that the duration in the CTS frame is longer than its duration. Thus, Client A knows that the AP intends to establish dual links. It then transmits the packet to the AP after a certain delay, which is at least longer than a preamble time. The delay time can be computed by Eq. (2). There are two scenarios in this case. If the transmission time of the packet from the AP to Client B is shorter than the packet transmission time of the other link plus a preamble time, as denoted by the “AP-shorter” scenario in Fig. 3(a), then the AP transmits a busytone signal to ensure two transmissions finish at the same time. Otherwise, as denoted by the “AP-longer” scenario in Fig. 3(b), Client A delays its transmission for enough time such that two transmissions finish simultaneously. When Client A and the AP finish their transmissions, Client B first returns an ACK frame to the AP and then the AP returns an ACK frame to Client A upon receiving the ACK frame from Client B. To ensure correct operation of A-Duplex, the duration in RTS and CTS needs to be computed properly. Since in the beginning Client A does not know whether or not the AP will establish dual links, it computes the duration as if the AP does not set up dual links. In case that the AP does not set up dual links, the duration field in the RTS frame works exactly as that in the half duplex case. If the AP does set up dual links, it just uses the duration in the RTS frame to compute the duration of the CTS frame as explained below. Since the AP’s transmission covers all clients, the duration field in the CTS frame can gracefully protect the transmissions between Client A and the AP. As a result, the duration in RTS frame is computed as DurationRTS = 3TSIFS + TCTS + T1 + TACK, (1) where TSIFS is the Short Inter-Frame Spacing (SIFS) time, TCTS is the time of CTS frame, TACK is the time of ACK frame, and T1 is the time of data packet to the AP. With the duration from RTS, the AP computes the duration in CTS as follows. In the case of Fig. 3(a), the duration in CTS is computed as DurationCTS = DurationRTS − TCTS − 2TSIFS + Tp + TACK, where Tp is the preamble time. In the case of Fig. 3(b), the duration in CTS is computed as DurationCTS = T2 + TSIFS + 2TACK, where T2 is the transmission time of the packet from the AP to Client B. When A receive the CTS frame, it decides the delay time according to Delay = DurationCTS − T1 − TSIFS − 2TACK. (2) Fig. 4. An example showing no dual links. The delay for Client A serves two purposes. First, for a legacy node, when it finishes a data packet transmission, it waits a fix time for ACK. If no ACK is received after the preset timeout, it starts the retransmission procedure. Second, the delay ensures the packet to Client B is sent first and its preamble can be protected without interference. In the above procedure, Client A in Fig. 3 waits a time period of TSIFS + TACK before it receives its ACK. This time period is longer than Distributed Coordination Function Inter Frame Spacing (DIFS) time. Thus, the ACK may collide with the packet from the other client. In our design, the CTS frame can solve this problem. The duration in CTS covers the entire period of the above procedure, so all other clients receiving the CTS frame update the NAV value according to the duration in the CTS frame. As a result, the ACK from the AP can also be protected. In the second case when the AP has no packet for a client, the operation procedure is shown in Fig. 4. When AP finds no suitable packet for all other clients, it only replies a CTS frame. From the CTS frame, Client A finds out that the duration in CTS frame is the same with its duration. Thus, Client A knows that the AP will not establish dual links, so it only delays an SIFS time and then starts a packet transmission. When the packet is received correctly, the AP returns an ACK to Client A. In the third case, the AP may get the channel first by sending a data packet without RTS/CTS exchange. No dual link can be set up in this case for two reasons. First, a client does not support full duplex communications, so symmetrical dual links cannot be formed. Second, clients cannot decide among themselves about which one can start asymmetrical links to explore capture effect. If the decision is done by the AP, then the queue status as well as the length of the first packet in queue of each client must be reported to the AP. To avoid such complexity, dual link setup is ignored in the third case. In A-Duplex, the RTS/CTS frame exchange procedures ful- fill the following functions. Firstly, from the RTS frame, the AP knows exactly which client is about to send a packet and then decides how to establish dual links with another client. In [6], the AP transmits packet to another client to establish dual links when it decodes the header of the received packet. However, if the packet is transmitted by OFDM symbols such as in 802.11a network, the AP cannot decode the packet header to get the transmitting client address until it receives the whole packet. Secondly, the RTS/CTS mechanism guarantees that the packet from the AP to Client B is transmitted first and the preamble of the packet is not interfered by the packet from Client A to the AP. Thus, Client B can decode the packet more easily through capture effect. Thirdly, the RTS frame contains signal strength information that can be used by the AP to select a proper client

TANG AND WANG:MAC FOR EFFICIENT COEXISTENCE BETWEEN FULL AND HALF-DUPLEX COMMUNICATIONS 5875 to set up a dual link Details of this functionre explained in the then they compute the SIR value with the two recorded values an A-Duplex may experience two types of collisions.In the first Client B)contends the channel successfully,it transr mits its type of collisions.more than on clients send an Rohe recorded SIR value via the RTS frame to the AP.Since the AP the last roun colision ha and do n the RTSfm ws th SIR In the second type of collisions,the AP starts a transmission given the interference from Client A.Thus,it updates such the SIR MAP.If the same client llican sim eive,it the AP stop cha value is igno and let the client's RTS frame pr channel successfully In this scenario all client just update ceed.so the client can still contend the channel successfullv thethe channe clents pdate the IV.ESTABLISHMENT OF ASYMMETRICAL DUAL LINKS the the The critical step of setting up dual links is to ensure capture STR value in the AP's SIR MAP By building up SIR MAP the effect to be properly utilized.Forexampe if du whe interfere other words,the AP needs to find out the signal-to-interference RTS frame from the interfering client.To achieve this goa ent ve use the transmit the RTS that othe an of SIR for all client what follow h is effec an information map is called the sir map wireless LAN the transmit power ofan RTS frame is usuall set the same as that of a regular data packet.Sec ond,an RT A.Establishment of SIR MAP rame uses the rate for tra n,so it The SIR MAP contains the SIR information of all clients a busy channel but can nnot receive an rTS frame it relies or Considering Clients A and B in a wireless LAN.SIR at Client wce from repre nted by sh le that ere was an RT nal str Cto A.A to B.Cto B.A to C.and B toC.because there are 6 interference signal strength en two clients all clients can find out whether 10 the a clie on happens via kI ge.M cally,if ing an RTS frame from a client,it knows collision has happened and thus skips updating its interference signal strength. ansmits sa packet toa client (B).th ue to channel f .an A)tre AP,the current client (i.e.Client B)can also get the inter- the value instead a moving average strategy is necessary one erence strength by o erhearing the pack et.Sinc we make n casy approach is to use the average of the most recent value change to the physic signal str the Al store to the Strength Indicator (RSSD when a client succeeds in receiving a needs to stores 5 SIR values for each client.In order to reduce packet in the MAC layer. the storage of AP.a simple moving average scheme is given as AC layer m follows d help buld SIR=SIROm x (1-0)+ (3 the channel.it first transmits an KTS frame to the AP All where 6 is a weight strength nter re ing the R 4,the the current the signal strength from AP according to the CTS frame,and scheme.the AP only needs to store one SIR value for each

TANG AND WANG: MAC FOR EFFICIENT COEXISTENCE BETWEEN FULL- AND HALF-DUPLEX COMMUNICATIONS 5875 to set up a dual link. Details of this function are explained in the next section. Fourthly, ACK frames in dual links are protected via the duration in RTS/CTS frames (i.e., the NAV mechanism). A-Duplex may experience two types of collisions. In the first type of collisions, more than one clients send an RTS to the AP. Without correct reception of an RTS, AP does not reply a CTS frame. As a result, the competing clients know that collision happens and do not start data packet transmission. In the second type of collisions, the AP starts a transmission simultaneously with an RTS frame from a client. Since the AP can simultaneously transmit and receive, it can find out collisions while it is transmitting. In this case, the AP stops transmitting immediately and let the client’s RTS frame proceed, so the client can still contend the channel successfully. IV. ESTABLISHMENT OF ASYMMETRICAL DUAL LINKS The critical step of setting up dual links is to ensure capture effect to be properly utilized. For example, if dual links (from A to AP and from AP to C) are to be set up, the AP needs to make sure that the signal strength at Client C is stronger enough than interference from Client A according to the transmission rate. In other words, the AP needs to find out the signal-to-interference ratio (SIR) for Client C and decides the transmission rate based on the SIR value. To this end, it is necessary for the AP to build an information map of SIR for all clients. In what follows, such an information map is called the SIR MAP. A. Establishment of SIR MAP The SIR MAP contains the SIR information of all clients. Considering Clients A and B in a wireless LAN, SIR at Client A given the interference from B can be represented by SIR of B to A. Thus, if a wireless LAN has 3 clients: clients A, B, and C, then SIR map table must have 6 items such as SIR of B to A, C to A, A to B, C to B, A to C, and B to C, because there are 6 different scenarios of interference between two clients. To measure the SIR value experienced by a client, two values are needed: the signal strength from the AP and that from the interfering client. However, an AP cannot get such information by itself, so it has to rely on clients to collect such information. When the AP transmits a packet to a client (e.g., Client B), the client can get the signal strength of this on-going transmission. When another client (e.g., Client A) transmits a packet to the AP, the current client (i.e., Client B) can also get the interference strength by overhearing the packet. Since we make no change to the physical layer of legacy nodes, the signal strength that can be collected in the MAC layer is the Received Signal Strength Indicator (RSSI) when a client succeeds in receiving a packet in the MAC layer. MAC layer messages are used to support SIR measurements at clients and help build the SIR MAP at the AP. As explained in the previous section, when a client (e.g., Client A) accesses the channel, it first transmits an RTS frame to the AP. All other clients who receive this RTS frame record the interference strength. After receiving the RTS frame, the AP replies a CTS frame and records this client address. All other clients update the signal strength from AP according to the CTS frame, and then they compute the SIR value with the two recorded values: the signal strength from the AP and interference strength. In the next round of RTS/CTS/Data/ACK, if one client (e.g., Client B) contends the channel successfully, it transmits its recorded SIR value via the RTS frame to the AP. Since the AP records which client (i.e., Client A) started the last round of RTS/CTS/Data/ACK, it knows that the SIR value carried in the RTS frame is the SIR of a new client (i.e., Client B) given the interference from Client A. Thus, it updates such an SIR value (of A to B) in its SIR MAP. If the same client contends the channel successfully, the SIR value is ignored. As explained in the protocol design, the AP may contend the channel successfully. In this scenario, all clients just update the signal strength from the AP and recompute the SIR value. If the AP continuously gets the channel, all clients update the SIR value until a client contends the channel successfully. Since then, the above procedure is followed to record or update the SIR value in the AP’s SIR MAP. By building up SIR MAP, the AP can have a dynamic view of SIR for each client. In the MAC layer, RSSI information is only available when a MAC packet is received. Given a client, in order to detect signal strength from an interfering client, it needs to get the RTS frame from the interfering client. To achieve this goal, we use the basic rate to transmit the RTS frame so that other clients can rely on this frame to detect strength of an interfering signal. This approach is effective for two reasons. First, in a wireless LAN the transmit power of an RTS frame is usually set the same as that of a regular data packet. Second, an RTS frame uses the basic rate for transmission, so it can mostly cover the large area of a wireless LAN. In case a client can sense a busy channel but cannot receive an RTS frame, it relies on a CTS frame to get RSSI. More specifically, when the client receives a CTS frame, it can conclude that there was an RTS frame before. Thus, the client can use the minimum required signal strength for decoding an RTS frame to approximate the interference signal strength. As for the collision case, all clients can find out whether a collision happens via RTS/CTS exchange. More specifically, if a client does not receive a CTS frame from the AP after receiving an RTS frame from a client, it knows collision has happened and thus skips updating its interference signal strength. Due to channel fading, an instantaneous SIR value cannot always accurately reflect the channel quality very well. Thus, when the AP updates the MAP table, it cannot simply substitute the value. Instead, a moving average strategy is necessary. One easy approach is to use the average of the most recent values. However, this demands the AP to store too many SIR values. For example, if we want to get the average of 5 SIR values, AP needs to stores 5 SIR values for each client. In order to reduce the storage of AP, a simple moving average scheme is given as follows: SIRNew = SIROld × (1 − θ ) + SIRupdate × θ , (3) where θ is a weight factor from 0 to 1, the value SIROld is from the current SIR MAP, SIRUpdate is newly learned from frame exchange, and SIRNew is the new SIR in the MAP. With this scheme, the AP only needs to store one SIR value for each

点击进入文档下载页（PDF格式）

共15页，试读已结束，阅读完整版请下载

您可能感兴趣的文档

点击购买下载（PDF）

下载及服务说明

购买前请先查看本文档预览页，确认内容后再进行支付；
如遇文件无法下载、无法访问或其它任何问题，可发送电子邮件反馈，核实后将进行文件补发或退款等其它相关操作；
邮箱：

文档浏览记录