X-FDR: A Cross-Layer Routing Protocol for Multi-hop Full-Duplex Wireless Networks
IIEEE WIRELESS COMMUNICATIONS - ACCEPTED FOR PUBLICATION 1
X-FDR : A Cross-Layer Routing Protocol forMulti-hop Full-Duplex Wireless Networks
M. Omar Al-Kadri, Adnan Aijaz, and Arumugam Nallanathan
Abstract —The recent developments in self-interference (SI)cancellation techniques have led to the practical realization offull-duplex (FD) radios that can perform simultaneous transmis-sion and reception. FD technology is attractive for various legacycommunications standards. In this paper, after discussing theopportunities of FD technology at the network layer, we presenta cross-layer aided routing protocol, termed as
X-FDR , for multi-hop FD wireless networks.
X-FDR exploits a Physical (PHY) layermodel capturing imperfection of SI cancellation. At the mediumaccess control (MAC) layer,
X-FDR adopts an optimized MACprotocol which implements a power control mechanism withoutcreating the hidden terminal problem.
X-FDR exploits the uniquecharacteristics of FD technology at the network layer to constructenergy-efficient and low end-to-end latency routes in the network.Performance evaluation demonstrates the effectiveness of
X-FDR in achieving the gains of FD at higher layers of the protocolstack.
Index Terms —full-duplex, cross-layer, distributed networks,routing, energy-efficiency, MAC.
I. I
NTRODUCTION R ECENT advances in self-interference (SI) cancellationtechniques have made in-band full-duplex (FD) [1], [2]operation feasible for wireless communications. FD-capablenodes can perform simultaneous transmission and reception onsame resources in time and frequency domains. FD technologynot only offers the potential of (theoretically) doubling thecapacity and the spectrum utilization but also assists in solvingsome of the key problems in half-duplex (HD) networks, suchas the hidden node issues, loss of throughput due to highcongestion rates, and large end-to-end delays [1]. Existingefforts towards FD communications have mainly investigatedPhysical (PHY) layer aspects; however, solutions for mediumaccess control (MAC) and highers layers have also started toemerge [3]. In order to reap the maximum benefits of FDtechnology, optimizations are required at different layers ofthe protocol stack.On the other hand, energy saving in distributed wirelessnetworks is of significant importance due to the limited batterysupply of each node. Nodes in the network continuouslyparticipate in route construction, and act as relays for neigh-boring nodes. In addition to continuous variation in channelconditions, this leads to a large amount of control messagesbeing exchanged across the network, which potentially entails
M. Al-Kadri is with the School of Computing Science and Digital Media,Robert Gordon University, Aberdeen, UK.A. Aijaz is with the Telecommunications Research Laboratory, ToshibaResearch Europe Ltd., Bristol, UK.A. Nallanathan is with School of Electronic Engineering and ComputerScience, Queen Mary University of London, London, UK.Contact e-mail: [email protected] high energy consumption. Therefore, energy-efficiency in dis-tributed wireless networks is an important issue. Moreover,with the introduction of FD, the issue of energy efficiencybecomes critical owing to additional hardware and processingcapabilities of nodes.Research on routing protocols for FD wireless networks isstill in infancy. In [4], Fang et al. have proposed cross-layer op-timization for opportunistic multi-path routing in FD wirelessnetworks. The route selection problem has been solved undervarious resource competitions and node constraints. However,the proposed framework assumes perfect SI cancellation. Katoand Bandai [5] have proposed an on-demand detour routingprotocol for directional FD wireless networks. Although theuse of directional antennas mitigates the hidden terminalproblem, the protocol is not compatible with networks employ-ing omnidirectional antennas. Sugiyama et al. [6] designeda directional asynchronous FD-MAC protocol for mitigatingcollisions in multi-hop FD wireless networks, however theprotocol is not applicable to the omni-directional antennas,which are widely used in handheld devices. Ramirez andAazhang [7] addressed the problem of joint power allocationand routing in FD wireless networks through a modificationto Dijkstra’s algorithm. However, the paper assumes that anFD MAC is in place. Besides, the main focus of the paper issystem-level analysis. It is also important to mention that mostof the existing studies do not fully exploit the key opportunitiesprovided by FD technology, which have been discussed later.On the other hand, power-aware routing protocols [8] forconventional HD wireless networks have received significantattention over the last few years. It can be easily inferred thatdesign of routing protocols for FD wireless networks requiresfurther investigation from various aspects, which motivates thiswork.Our objective in this paper is to design a cross-layeraided routing protocol for imperfect FD wireless networks,where the notion of imperfection implies that SI is not fullycancelled at the PHY layer. The proposed protocol, whichis termed as
X-FDR , is particularly designed for minimizingenergy consumption and end-to-end latency in FD wirelessnetworks. The key features of
X-FDR can be summarized asfollows. First,
X-FDR accounts for residual self-interference(RSI) at the PHY layer. Second,
X-FDR adopts an optimized(not necessarily optimal) MAC protocol that implements apower control mechanism without creating the hidden terminalproblem. Third,
X-FDR adopts a novel energy cost metricand exploits the opportunities provided by the FD technologye.g., the ability to sense the medium while transmitting. Thisprovides immediate reaction to channel errors, and conse- a r X i v : . [ c s . N I] J un EEE WIRELESS COMMUNICATIONS - ACCEPTED FOR PUBLICATION 2 quently, nodes are able to send a burst of packets, constrainedby the minimum buffer size ( β min ) on the selected route.Moreover, nodes wait for the acknowledgement (ACK) of thelast received packet only instead of acknowledging from thereception of each individual packet. Fourth, nodes in X-FDR employ immediate forwarding, which is enabled by their FDcapabilities. A node does not have to wait for the reception ofthe full packet before it can forward it to the next hop. Thisfeature reduces the end-to-end latency of the network. Last,but not the least,
X-FDR employs a novel route maintenanceprocess that reduces the latency due to new route discovery.Performance evaluation demonstrates that
X-FDR provides aviable solution for multi-hop FD wireless networks.II. O
PPORTUNITIES OF FD AT N ETWORK L AYER
In this section, we describe the key opportunities providedby the FD technology that could potentially be exploited bynetwork layer protocols. • Immediate Forwarding – FD technology enables simul-taneous transmission and reception, which is particularlyattractive in multi-hop wireless networks. When a FDnode starts receiving a packet, it can simultaneously startforwarding it to the next hop. This provides a paradigmshift from conventional store-and-forward architecture inlegacy HD networks, to receive-and-forward architecture.For example, consider the scenario depicted in Fig. 1.With immediate forwarding, node A can start transmittingthe packet, which is being received from node S , tonext hop, as soon as it has processed the packet header.Immediate forwarding is particularly attractive to reduceend-to-end latency and improve throughput in multi-hopwireless networks. • Continuous Sensing – Another key advantage of FDtechnology is the ability to sense the medium while trans-mitting. In conventional HD networks, a node will not benotified of transmission errors, until after the transmissionis complete. With continuous sensing, FD nodes candetect an erroneous transmission as soon as it occurs,which leads to immediate termination of a transmission.This improves resource utilization and potentially enablesreduction of end-to-end latency. • Burst Transmission – The continuous sensing propertyfurther enables FD nodes to send burst of data packets,such that only the last packet is acknowledged. This isunlike conventional HD networks where packets are sentsequentially and each packet needs to be individuallyacknowledged. If properly exploited at the network layer,this feature has the potential to not only reduce end-to-endlatency, but also improve resource utilization (particularlyfor signaling resources) and throughput. • Faster Convergence – The above mentioned features,especially immediate forwarding, enable faster dissemi-nation of signaling information associated with routingprotocols. Hence, faster topological convergence can beachieved, especially for those routing protocols that relyon building a topology tree of the network. Besides, thesefeatures can also enhance the efficiency of flooding-basedrouting protocols. • Secure Routing – Having two simultaneous transmis-sions on the same frequency makes it difficult for a nearbynode to perform eavesdropping attacks as the receivedsignal would be a scrambled mix of both signals. Hence,such attacks on intermediate nodes become significantlymore complex to perform, thereby enhancing the securityof the routing protocol between source and destinationnodes.It is emphasized that some of the key opportunities likeimmediate forwarding, continuous sensing, and burst trans-mission have been exploited in the design of
X-FDR . Theseopportunities have been further explained while discussing theprotocol operation.III. N
ETWORK M ODEL
We consider a distributed network comprising N FD wire-less nodes indexed by the set N . Let, R denote the set of allpossible routes in the network. A route R ∈ R represents anordered set of nodes between a source node S and a destinationnode D . For example, Fig. 1 demonstrates a route comprisingfour nodes in the network.We assume that FD wireless nodes employ necessary SIcancellation techniques at the PHY layer. Since SI cancellationtechniques are not perfect in practice, a node experiences RSI.We use an experimentally characterized model [9] for RSI,based on which, the power of the RSI signal is given by P (1 − ρ ) t ∆ · χ ρ , where P t is the transmit power, ∆ is the interferencesuppression factor, χ depends on the SI cancellation technique,and ρ denotes the SI cancellation capability. Note that ρ = ∞ denotes perfect SI cancellation, resulting in zero RSI. More-over, ρ = 0 implies a constant reduction in transmission power.Realistically, < | ρ | < ; with ρ = 1 implying a constantpower, for RSI similar to noise.We assume that the received signal power at a node j , basedon a transmission from a node i at maximum transmit power P max is given by P r = P max ·| h i,j | · d − αi,j , such that h i,j is thechannel coefficient that accounts for small-scale fading, d i,j denotes the distance, and α denotes the path loss exponent.We assume that nodes in the network employ a power controlmechanism based on the received signal strength such that thecontrolled power level is determined by P ctrl = P max · ( P r ) − · ζ th · ˆ c , such that ζ th denotes the minimum required receivedsignal strength and ˆ c is a constant [10]. Please note that RSIis not part of P ctrl as FD communication is not yet initialized.The impact of RSI and cumulative interference is captured onlink-level. Further, our link-level model is based on signal-to-interference-plus-noise-ratio (SINR) which accounts for RSIand given by SIN R = P i | h i,j | d i,j − α RSI + I x + N , where P i denotes the transmit power of node i (either P max or P ctrl ) and N denotes the noise power. Moreover, I x isthe cumulative interference from neighboring nodes and isgiven by I x = (cid:80) x ∈N \{ i,j } P x | h x,i | d x,i − α , where P x is thetransmitting power of an interfering node x , h x,i is the channel EEE WIRELESS COMMUNICATIONS - ACCEPTED FOR PUBLICATION 3 DS SI SI SI AS h , BA h , DB h , A B
Fig. 1: Example of a route R = { S, A, B, D } . Straight lines representthe intended transmission, while dotted lines represent neighbouringinterference, and the red semi-circled arrows represent SI. FD-CTS SA µs µs µsDATA S → A SIFS SIFS SIFS DIFS
FD-CTSRTS D DATA A → D S A B D (a)
S A B D (b)
Z Z (c) P max P ctrl P max P ctrl SensingRangeTransmission Range
Fig. 2: (a) Ranges of nodes transmitting control signals using P max ;(b) ranges of nodes after application of power control; (c) illustrationof a uni-directional FD transmission at the MAC layer. coefficient between nodes x and i , and d x,i is the distancebetween nodes x and i .IV. MAC L AYER D ESIGN FOR
X-FDR
This section presents the MAC layer design for
X-FDR .In
X-FDR , we adopt the modified version of our recentlyproposed MAC protocol [11] for distributed wireless networks.The MAC protocol in [11] enables both bi-directional FDtransmissions and uni-directional FD transmissions. The for-mer enables simultaneous two-way transfer of two distinct data streams between a pair of nodes, whereas, the latter involvesthree nodes and same data stream is forwarded from one nodeto another via an intermediate relay node. In
X-FDR , we focusonly on uni-directional FD transmission. We also omit theMAC layer ACK procedure.We explain the protocol operation with the aid of Fig. 2.Let N = { S, A, B, D } be a set of nodes involved in theintended transmission, where S is the source node and D is thedestination node. After sensing the spectrum idle, node S startsthe transmission by sending a request-to-send (RTS) packet tonode A using P max . After receiving the RTS packet from S , node A waits for short inter-frame space (SIFS) durationbefore sending an FD clear-to-send (FD-CTS) packet [11]to both S and B . The FD-CTS packet includes the sourceand next hop addresses along with the transmission duration.Note that FD-CTS is also transmitted using P max to capturethe channel for forwarding. Using the received RTS from S ,node A calculates P ctrl as described in Section III. Node S calculates its P ctrl as well using the FD-CTS received fromnode A . Further, when node B receives the FD-CTS fromnode A , it replies with FD-CTS as well, and calculates its P ctrl based on the received power from A . After that, node A recalculates P ctrl based on the received FD-CTS from B and compares it with the previously calculated P ctrl , wherethe higher P ctrl is chosen to maintain connection with both S and B . Similarly, the rest of the relaying nodes attempt toacquire the channel until the intended destination is reached.During data transmission, nodes use P ctrl with periodicalincrease to P max , so that nodes in the carrier sensing zone,which cannot successfully decode the transmission and settheir Network Allocation Vector (NAV) to Extended Inter-Frame Space (EIFS) duration can sense the transmission.Note that the period between two successive power increaseintervals must be less than the EIFS duration . These periodicincrements preserve the channel, and ensure that nodes in thecarrier sensing zone will not attempt to initiate a transmission. A. Hidden Terminal Problem
Referring to Fig. 2b, consider that nodes S and A constitutea sender-receiver pair in HD mode. Node Z , which resides inthe carrier sensing range of S but not of node A , may act asa hidden node. In FD transmission, hidden nodes may affectthe reception of control signals at node S . Therefore, in theproposed protocol we adopt RTS-CTS handshake mechanism.Moreover, by sending FD-CTS using P max , the protocolensures that nodes in the carrier sensing ranges are aware of anongoing transmission. After power control is applied for datatransmission, node Z can again create a hidden node problem,which is why the periodic increments from P ctrl to P max arerequired. According to the IEEE 802.11n standard [12], µs is suitable for carriersensing, and µs is adequate to increase/decrease the power level from/to to/from . Therefore, a duration of 20 µs is deemed adequate fortransition of power level from P ctrl to P max and vice versa. Since EIFS isset to µs , nodes will transmit at P max every µs for a duration of µs , and the cumulative transmission duration is less than the EIFS duration. EEE WIRELESS COMMUNICATIONS - ACCEPTED FOR PUBLICATION 4 V. X-FDR : P
ROTOCOL O PERATION
This section explains the protocol operation of
X-FDR . Un-like conventional Adhoc On-demand Distance Vector (AODV)routing protocol [13], where the route cost relies mainly onhop count,
X-FDR uses energy consumption as the key metricfor route cost estimation.
A. Route Cost Estimation
Since
X-FDR is a cross-layer routing protocol, all relevantfactors must be accounted for in route cost estimation. Nodesin the network initiate connections using RTS/FD-CTS mes-sages with maximum power level P max , in order to restrainother nodes residing in the sensing range from initiating aninterfering transmission. Once the data transmission take placeusing controlled power level P ctrl , a periodic increase ofpower to P max takes place to stop potential interference andeliminate the problem of hidden nodes; therefore, the metricfor route cost shall account for different power levels.In a route N = { S, A, B, D } , the cost of energy for sendingdata from node S to node A can be estimated as ρ ( S,A ) = χ ( E data + E ctrl + E on ) , where χ = 1 / P f is the the number ofretransmissions attempts such that P f denotes the probabilityof transmission failure, and E data , E ctrl and E on denotethe energy consumed during data transmission, control signaltransmission and when the receiver is turned on, respectively.The energy consumption during data transmission phase canbe calculated as E data = P Sctrl ( β min /r − T inc ) + P max T inc ,where P Sctrl and P max denote the controlled power level andmaximum transmit power of node S , respectively, β min is theminimum buffer size (explained in Section V-B), and r is thedata rate. Moreover, T inc denotes the duration of the periodicincrease/decrease in power levels. The energy consumptionduring control signal transmission can be calculated as E ctrl = P max ( T RT S + T F D − CT S ) , where T RT S and T F D − CT S denotethe duration of RTS and FD-CTS messages, respectively.Assume that there exists a route R i = n o → n → ... → n k from the source node S to the destination D , where, withoutloss of generality, S = n and D = n k . Therefore, the totalcost, ¯ ρ i , along the route R i can be expressed as ¯ ρ i = k − (cid:88) j =0 ρ ( j,j +1) ( P j ) , where P j is the power level used by node n j to communicatewith node n j + l , and ρ ( j,j +1) ( P j ) is the relaying cost betweennodes n j and n j + l . Assuming that there are x routes fromsource to destination, the objective of the routing protocol isto select the route with minimum energy consumption i.e., R min = arg min( ¯ ρ i ) , ∀ i = 1 , , . . . , x. B. Route Discovery
The first stage of
X-FDR is route discovery. When a sourcenode S requires a route to destination node D , it broadcasts aRoute REQuest message (RREQ). Once neighbouring nodesreceive RREQ, they calculate the energy cost, ρ , add it toRREQ and broadcast it to the neighboring nodes. After that theneighboring nodes calculate the new ρ , add it to the previous S A B D RREQ
RREQ
RREQRREP
RREPRREP
R-ACKR-ACKR-ACK R o u t e D i s c ov er y D a t a D a t a D a t a D a t a A CK D a t a D a t a D a t a RERR D a t a A CK D a t a D a t a S A X D RRE Q RRE Q A CK R A CK RREP A CK RREP R U P D A CK R A CK R (a) Route Discovery Process (c) Route Maintenance Process S A B D D a t a T r a n s m i ss i o n Data Data Data Data ACK Data Data Data Data Data Data Data Data Data ACK ACK ACK ACK ACK (b) Data Transmission Process
Fig. 3: Example of (a): route discovery process; (b) data transmissionprocess; (c) route maintenance process in
X-FDR . cost received in RREQ, and broadcast it further until it reachesthe destination D . The destination node sets up a timer toallow several RREQ messages to arrive from different routes.After the timer expires, node D chooses the route R min withminimum energy consumption and replies with a Route REPly(RREP) message via R min . The routing table of each nodeis refreshed, whenever it receives RREQ/RREP messages.Each node maintains a received RREQ table and comparesthe new RREQ messages in order to eliminate the duplicateRREQ messages. Additionally, a Route-ACKnowledgement(R-ACK) packet is used by the nodes receiving RREP, inorder to confirm successful reception of the RREP packet andestablishment of the route. Fig. 3 demonstrates an example ofroute discovery performed by node S , where R min is foundto be { S, A, B, D } .Instead of sending packets sequentially and waiting foracknowledgements (ACKs) for each data packet, X-FDR sends
EEE WIRELESS COMMUNICATIONS - ACCEPTED FOR PUBLICATION 5 a burst of packets, such that the number of packets in eachburst is determined by the minimum buffer size, β min , of thenodes in the route R . For example, let β S denote the buffersize (in terms of the number of packets) of node S . Further,node S encapsulates β min = β S within RREQ and broadcastsit. When a neighboring node A receives RREQ, it compares β min with its own buffer size (i.e., β A ). If β A < β min , node A updates β min = β A in RREQ and broadcasts it forward.However, if β A > β min , node A keeps the buffer size as it isand forwards RREQ. When D receives RREQ, it compares itsbuffer size with the received β min , and sends the lowest of thetwo within RREP, which informs node S with the minimumbuffer size to be used for data transmission. Note that if anode in the route does not have a buffer enabled, β min willbe set to 1. C. Data Transmission
When node S receives the RREP message as a result ofroute discovery process, it becomes aware of the most energy-efficient route to the destination D . After the route discoveryprocess, node S starts transmitting a burst of data packetsto next hop (node A ), where the number of packet in eachburst is given by β min of the route. In conventional HDcommunications, when node S sends a burst of data, it willnot be notified of transmission errors, e.g., by receiving aRoute ERRor (RERR) message, until after the entire burstis transmitted. This incurs significant waste of time andresources. However, using continuous sensing offered by FDtechnology, node S can sense a problem in the transmission assoon as it occurs, which leads to immediate termination of thetransmission. Hence, node S continuously senses the packetsforwarded by node A and stops transmitting immediately if itreceives RERR message.Since node A is FD-capable, it can employ immediateforwarding , wherein it does not have to wait for the entirepacket to be received before forwarding. Once node A receivesall the packets, determined by β min , it replies with an ACKto acknowledge the reception of the last packet. If a packetis dropped while the route is not deemed faulty, node S getsnotified by the ACK packet sent by A , and it retransmits thelost packet. The same process is repeated at each hop until thedestination is reached. Note that each node in the route onlynotifies the previous hop with an ACK. This is because datais assumed to be buffered by the previous node in the routeas β min is known to all nodes. Therefore, if a node did notreceive all the packets, it would request these from previousnodes using ACK.For instance, assume that β min = 4 , and consider thescenario demonstrated in Fig. 3. Node S transmits data packets1 through 4 while continuously sensing the signal transmittedby A . Node A starts forwarding immediately; however it onlyreceives 3 packets. Therefore, it sends an ACK for data packet3, which notifies S that it needs to retransmit data packet 4.Note that if the buffer size of node S is larger than the amountof data packets that needs to be sent, it will include an end-of-queue (EQ) notification message with the last packet, in orderto avoid an unnecessary retransmission. D. Route Maintenance
The process of route maintenance is depicted in Fig. 3,where the transmission of packets 1 to 3 from source S to destination D is exemplified. First, node S transmits theburst of packets to node A . Node A receives the packetssuccessfully and responds with an ACK packet to the source S to confirm the successful reception. As node A receivesthe packets, it starts forwarding them to node B . However,node B fails to receive the data packet 3 successfully, despitemaximum number of retransmission attempts by node A due toa link error. Once the pre-set timer expires at node A withoutreceiving any ACK from node B , it infers that the link A − B isbroken and sends an RERR message to its previous hop, whichis node S in this case. The RERR message informs node S of a link failure and a new route discovery process. Node S updates its routing table and marks link A − B as broken, andthen acknowledges the RERR of node A . Since the route erroroccurred at node A , it initiates a new route discovery processby broadcasting a RREQ message. Intermediate nodes followthe same procedure as described earlier for route discovery.When the destination node D receives RREQ from node A ,prior to the full reception of packets in the same burst of β min ,it knows that the request is to complete the same data stream,and replies with RREP, piggybacked with an ACK packet toinform node A about the last packet node D had received.After receiving the RREP message, node A sends a RouteUPDate (RUPD) message, with the new β min , to the previoushop i.e., node S , to inform it of a new route. Finally, node A starts new data transmission to destination D from data packet3 onwards. If node S has new burst of data to send, it will usethe updated route towards node D , starting the transmissionafter sensing the last packet sent by node A . The process ofroute maintenance is summarized in Algorithm 1. (cid:4) Remark 1 – It is worth emphasizing that
X-FDR incurs lessoverhead and complexity as compared to its HD counterparts.First, it omits the MAC layer ACK procedure which reducesthe signaling overhead. Second, from the routing perspective,the overhead in most cases is reduced which simplifies thesystem design. For instance, in the route discovery process,ACK packets are only sent to acknowledge the RREP packetswhich reduces the overhead significantly as compared toacknowledging the RERR packets. Similarly, acknowledginga burst of packets instead of each packet reduces the overheadin the network.
Remark 2 – Some recent studies [14], [15] have investigatedthe problem of in-band wireless cut-through which is closelyrelated to the problem of multi-hop transmissions in FDwireless networks. To realize wireless cut-through transmis-sions, specialized hardware is required for cancellation ofall types of interference. It is worth emphasizing that theneed for MAC and routing protocols cannot be eliminated forrealizing wireless cut-through transmissions.
X-FDR adoptsa cross-layer approach for multi-hop transmissions in FDwireless networks and focuses only on SI cancellation whichcan be achieved through state-of-the-art FD radios.
X-FDR can directly benefit from additional hardware capabilities asrealized for wireless cut-through transmissions. Alternatively,
EEE WIRELESS COMMUNICATIONS - ACCEPTED FOR PUBLICATION 6 the cross-layer approach of
X-FDR can improve the efficiencyof wireless cut-through solutions.
Algorithm 1:
Route Maintenance Process in
X-FDR
Input:
Source: S , Destination: D , Nodes: N , R min Output:
New Route: Updated R min while S → D do Nodes forward incoming packets for each node i ∈ { R min \ D } do i transmits packets to i + 1 i sets timer t ACK if i receives ACK z while t ACK (cid:54) = then i forwards packets z + 1 else i marks link i → ( i + 1) as brokenSend RERR to node i − Nodes i − ( x + 1) , x ∈ { , , · · · , hops to S } traverse RERR back to Si broadcasts RREQ if D receives RREQ for the same stream then set a timer t max if t ≤ t max then continue receiving RREQ packets else stop receiving RREQ packets;compare received RREQ packets and select R min return RREP packet with R min and β min . if i receives new RREP then Update R min Send RUPD to node i − Nodes i − ( x + 1) , x ∈ { , , · · · , hops to S } update R min and send RUPD back to S Closest node to D with full β min received will resume transmission endendendend VI. P
ERFORMANCE E VALUATION
In this section, we conduct a performance evaluation of
X-FDR . We have implemented
X-FDR in OPNET. Neces-sary changes were made in the node and protocol modelsto implement simultaneous transmission and reception. Weassume that nodes are randomly distributed in an area of m . The buffer size is assumed to be fixed and set to kB. The maximum transmit power of a node is set to dBm ( mW). We assume a channel bandwidth of MHz. The path loss exponent is set to . We consider filetransfer protocol (FTP) application with packet size of kB.The RSI parameter, χ is set to dB. The simulation resultsare averaged over iterations. In each iteration, source anddestination nodes are randomly selected. We have modified thewireless model in OPNET to account for RSI and Rayleighfading. For performance comparison, we select two different baseline protocols: AODV and FD version of AODV, termedas FD-AODV, wherein nodes employ immediate forwardingand acknowledge each packet. Moreover, both AODV and FD-AODV do not employ power control.Fig. 4a shows the average power consumption of routesfrom source to destination nodes selected by different proto-cols. First, we note that the power consumption increases withthe number of nodes in the network. This is due to inclusionof more nodes in the routes selected by different protocols.Second, we note that X-FDR outperforms both baseline pro-tocols by performing up to 40% and 50% better than AODVand FD-AODV protocols, respectively. The performance gainof
X-FDR in terms of energy-efficiency is primarily due to theuse of power control at the MAC layer, which limits the effectof interference, and the adoption of energy-based routing costmetric. Third, we note that SI cancellation plays an importantrole in power consumption. A higher SI cancellation capability,corresponding to higher values of ∆ and ρ , reduces the powerconsumption due to less number of transmission failures dueto interference.Fig. 4b shows network throughput against the number ofnetwork nodes. We note that X-FDR outperforms AODV byperforming up to 50.2% and 21.2% better under high andlow SI cancellation scenarios, respectively. This is primarilydue to the FD features of
X-FDR . Further,
X-FDR achievesnearly 8.6% lower throughput than FD-AODV under high SIcancellation scenario. This can be attributed to the employmentof power control in
X-FDR as there is an inherent trade-off between power and throughput. Note that the presenceof SI, due to low SI cancellation capability, can degrade theperformance of FD-AODV to the extent that it achieves lowerthroughput than AODV. Such performance degradation is alsovisible in case of
X-FDR .Fig. 4c plots the average hop count between randomlylocated source and destination nodes, as a function of numberof nodes in the network. The average hop count increases withthe number of network nodes as more nodes are involved in theselected routes.We note that
X-FDR has higher average hopcount than the baseline protocols. This is because both AODVand FD-AODV use hop count as the routing metric. However,
X-FDR focuses on routes with minimal energy consumption,and therefore, it incurs higher hop count with lower totalenergy consumption.Fig. 4d plots the average end-to-end delay against thenumber of network nodes. We note that
X-FDR outperformsAODV by achieving up to lower delay, due to theuse of immediate forwarding, continuous sensing and bursttransmission mode. On the other hand, FD-AODV outperforms
X-FDR by achieving up to 12% lower delay. This is dueto the fact that
X-FDR incurs higher hop count. Althoughboth AODV and FD-AODV incur similar hop count, the latterachieves lower delay due to immediate forwarding feature.It is important to mention here that the results in Fig. 4dcorrespond to the scenario when the route does not suffer anyfailures along its path. In order to capture the impact of routemaintenance, we deliberately mark nodes to fail (an optionprovided by OPNET) across the route during transmissionprocess and evaluate end-to-end delay in Fig. 4e. Initially,
EEE WIRELESS COMMUNICATIONS - ACCEPTED FOR PUBLICATION 7
Number of Nodes A v e r a g e P o w e r C o n s u m p t i o n ( m W ) X-FDR, ρ = 0.8, ∆ = 38 dBX-FDR, ρ = 0.2, ∆ = 35 dBAODVFD-AODV, ρ = 0.8, ∆ = 38 dBFD-AODV, ρ = 0.2, ∆ = 35 dB (a) Number of Nodes N e t w o r k T h r o u g h p u t ( M B ) X-FDR, ρ = 0.8, ∆ = 38 dBX-FDR, ρ = 0.2, ∆ = 35 dBAODVFD-AODV, ρ = 0.8, ∆ = 38 dBFD-AODV, ρ = 0.2, ∆ = 35 dB (b) Number of Nodes A v e r a g e H o p C o u n t X-FDRAODVFD-AODV (c)
Number of Nodes A v e r a g e E n d - t o - E n d d e l a y ( s ) X-FDRAODVFD-AODV (d)
No Failure Failure at S+1 Failure at middle node Failure at D-1 A v e r a g e E n d - t o - E n d D e l a y ( s ) X-FDRFD-AODVAODV (e)
Number of Nodes A v e r a g e N u m b e r o f R e t r a n s m i ss i o n s Energy-FDRAODVFD-AODV (f)
Fig. 4: Performance evaluation of
X-FDR : (a) average power consumption; (b) network throughput; (c) average hop count; (d) averageend-to-end delay; (e) average end-to-end delay with node failures (number of network nodes = ); (f) average MAC layer retransmissions.The confidence intervals on different figures are also shown. we fail the first node after the source, then a node at themiddle of the route, and finally, a node right before thedestination for worst case scenario. As shown by the results, X-FDR outperforms both AODV and FD-AODV by performingup to 39% and 34% better than the former and the latter,respectively. The performance gain is due to the proposedroute maintenance procedure that initiates a route discoveryprocess at the last buffered node instead of starting new routediscovery process by the source.Fig. 4f shows the average number of MAC layer retrans-mission attempts against the number of network nodes. Theaverage number of retransmissions increase with the numberof network nodes due to higher probability of failures as aresult of higher inter-node interference. We note that
X-FDR incurs the lowest number of retransmissions than both AODVand FD-AODV. This is primarily due to an optimized MACprotocol that minimizes collisions due to hidden node problemwhile using power control.Finally, a qualitative comparison of
X-FDR against state-of-the-art protocols is given in TABLE I.VII. C
ONCLUDING R EMARKS
FD technology has the potential to play an important role inrealizing the capacity objectives of future wireless networks.Realizing the FD capability at higher layers of the protocolstack is particularly attractive to reap the full potential of FD technology. In this paper, we have designed a cross-layerrouting protocol, termed as
X-FDR , for multi-hop FD wirelessnetworks with imperfect SI cancellation.
X-FDR accounts forRSI at the PHY layer, adopts an optimized MAC protocol withpower control feature, and exploits the opportunities providedby FD technology at the network layer. Performance evaluationdemonstrates that
X-FDR outperforms baseline protocols interms of power consumption without a significant compromiseon network throughput. Besides, it achieves lower end-to-enddelay in the presence of route failures. Hence,
X-FDR providesa viable solution for multi-hop FD wireless networks.R
EFERENCES[1] J. I. Choi, M. Jain, K. Srinivasan, P. Levis, and S. Katti, “Achievingsingle channel, full duplex wireless communication,”
ACM MobiCom ,pp. 1–12, Sep. 2010.[2] D. Bharadia, E. McMilin, and S. Katti, “Full duplex radios,”
SIGCOMMComput. Commun. Rev. , vol. 43, no. 4, pp. 375–386, Aug. 2013.[3] D. Kim, H. Lee, and D. Hong, “A Survey of In-band Full-duplexTransmission: From the Perspective of PHY and MAC Layers,”
IEEECommun. Surveys Tuts. , no. 99, Feb. 2015.[4] X. Fang, D. Yang, and G. Xue, “Distributed algorithms for multipathrouting in full-duplex wireless networks,”
IEEE MASS , pp. 102–111,Oct. 2011.[5] K. Kato and M. Bandai, “Routing protocol for directional full-duplexwireless,”
IEEE PIMRC , pp. 3239–3243, Sep. 2013.[6] Y. Sugiyama, K. Tamaki, S. Saruwatari, and T. Watanabe, “A wirelessfull-duplex and multi-hop network with collision avoidance using direc-tional antennas,” in
International Conference on Mobile Computing andUbiquitous Networking (ICMU) , Jan 2014, pp. 38–43.
EEE WIRELESS COMMUNICATIONS - ACCEPTED FOR PUBLICATION 8
TABLE I: Qualitative Comparison of Different Routing Protocols for FD Wireless Networks
Feature / Protocol OMR [4]
D-FDW [5]
M-DA [7]
AODV FD-AODV
X-FDR
Residual SI No No Yes No Yes YesPower Control No No No No No YesDirectional Antennas No Yes No No No NoOptimized MAC Yes Yes No No No YesImmediate Forwarding No Yes No No Yes YesContinuous Sensing No No No No No YesBurst Transmission No No No No No YesEnergy Efficiency No No No No No Yes[7] D. Ramirez and B. Aazhang, “Optimal Routing and Power Allocationfor Wireless Networks with Imperfect Full-Duplex Nodes,”
IEEE Trans.Wireless Commun. , vol. 12, no. 9, pp. 4692–4704, Sep. 2013.[8] J. Li, D. Cordes, and J. Zhang, “Power-aware routing protocols in adhoc wireless networks,”
IEEE Wireless Commun. , vol. 12, no. 6, pp.69–81, Dec. 2005.[9] M. Duarte, C. Dick, and A. Sabharwal, “Experiment-driven characteri-zation of full-duplex wireless systems,”
IEEE Trans. Wireless Commun. ,vol. 11, no. 12, pp. 4296–4307, May. 2012.[10] E.-S. Jung and N. H. Vaidya, “A Power Control MAC Protocol for Adhoc Networks,”
ACM MobiCom , pp. 36–47, Sep. 2002.[11] M. Al-Kadri, A. Aijaz, and A. Nallanathan, “An Energy-EfficientFull-Duplex MAC Protocol for Distributed Wireless Networks,”
IEEEWireless Commun. Lett. , vol. 5, no. 1, pp. 44–47, Feb. 2016.[12] “IEEE Standard for Information technology–Telecommunications andinformation exchange between systems Local and metropolitan areanetworks–Specific requirements - Part 11: Wireless LAN Medium Ac-cess Control (MAC) and Physical Layer (PHY) Specifications,”
IEEEStd 802.11-2016
ACM MobiCom ,Sept. 2015, pp. 566–577.[15] L. Chen et al. , “Bipass: Enabling end-to-end full duplex,” in
ACMMobiCom , Oct. 2017. R EFERENCES[1] J. I. Choi, M. Jain, K. Srinivasan, P. Levis, and S. Katti, “Achievingsingle channel, full duplex wireless communication,”
ACM MobiCom ,pp. 1–12, Sep. 2010.[2] D. Bharadia, E. McMilin, and S. Katti, “Full duplex radios,”
SIGCOMMComput. Commun. Rev. , vol. 43, no. 4, pp. 375–386, Aug. 2013.[3] D. Kim, H. Lee, and D. Hong, “A Survey of In-band Full-duplexTransmission: From the Perspective of PHY and MAC Layers,”
IEEECommun. Surveys Tuts. , no. 99, Feb. 2015.[4] X. Fang, D. Yang, and G. Xue, “Distributed algorithms for multipathrouting in full-duplex wireless networks,”
IEEE MASS , pp. 102–111,Oct. 2011.[5] K. Kato and M. Bandai, “Routing protocol for directional full-duplexwireless,”
IEEE PIMRC , pp. 3239–3243, Sep. 2013.[6] Y. Sugiyama, K. Tamaki, S. Saruwatari, and T. Watanabe, “A wirelessfull-duplex and multi-hop network with collision avoidance using direc-tional antennas,” in
International Conference on Mobile Computing andUbiquitous Networking (ICMU) , Jan 2014, pp. 38–43.[7] D. Ramirez and B. Aazhang, “Optimal Routing and Power Allocationfor Wireless Networks with Imperfect Full-Duplex Nodes,”
IEEE Trans.Wireless Commun. , vol. 12, no. 9, pp. 4692–4704, Sep. 2013.[8] J. Li, D. Cordes, and J. Zhang, “Power-aware routing protocols in adhoc wireless networks,”
IEEE Wireless Commun. , vol. 12, no. 6, pp.69–81, Dec. 2005.[9] M. Duarte, C. Dick, and A. Sabharwal, “Experiment-driven characteri-zation of full-duplex wireless systems,”
IEEE Trans. Wireless Commun. ,vol. 11, no. 12, pp. 4296–4307, May. 2012.[10] E.-S. Jung and N. H. Vaidya, “A Power Control MAC Protocol for Adhoc Networks,”
ACM MobiCom , pp. 36–47, Sep. 2002.[11] M. Al-Kadri, A. Aijaz, and A. Nallanathan, “An Energy-EfficientFull-Duplex MAC Protocol for Distributed Wireless Networks,”
IEEEWireless Commun. Lett. , vol. 5, no. 1, pp. 44–47, Feb. 2016. [12] “IEEE Standard for Information technology–Telecommunications andinformation exchange between systems Local and metropolitan areanetworks–Specific requirements - Part 11: Wireless LAN Medium Ac-cess Control (MAC) and Physical Layer (PHY) Specifications,”
IEEEStd 802.11-2016
ACM MobiCom ,Sept. 2015, pp. 566–577.[15] L. Chen et al. , “Bipass: Enabling end-to-end full duplex,” in
ACMMobiCom , Oct. 2017.
M. Omar Al-Kadri (M’17) received the B.Eng. degree in Computer En-gineering from IUST, Syria, in 2010, the M.Sc. degree (with distinctoin)in Networking and Data communication from Kingston University, UK, in2013, and the Ph.D degree in Telecommunication engineering from King’sCollege London, UK, in 2017. He is now a lecturer in networking andcyber security at Robert Gordon University, UK. His current research interestsinclude security of wireless communications with application to healthcare,full-duplex communications, HetNets, and MAC/routing protocols.
Adnan Aijaz (M’14–SM’18) received his Ph.D degree in TelecommunicationsEngineering from King’s College London (KCL), UK, in 2014. After a post-doctoral year at KCL, he moved to Toshiba Research Europe Ltd. wherehe is currently a Senior Research Engineer. His recent research interestsinclude 802.11-based WLANs, 5G cellular networks, Industrial IoT, TactileInternet, and full-duplex communications. His publications have been featuredin internationally renowned conferences and journals.Prior to joining KCL, he worked in cellular industry for nearly 2.5 yearsin the areas of network performance management, optimization, and qualityassurance. He holds a B.E. degree in Electrical (telecom) Engineering fromNational University of Sciences and Technology (NUST), Pakistan.