Poster Abstract: Scaling Capacity of a Dual Channel MAC Protocol for IEEE 802.11 Ad-hoc Networks Ping Chung Ng Department of Engineering Science, University of Oxford
[email protected]
David J. Edwards Department of Engineering Science, University of Oxford
[email protected]
Abstract — When an IEEE 802.11 ad-hoc network achieves capacity C by using a single channel, the targeted capacity by using two channels should be 2 C . However, most of the multichannel 802.11 protocols proposed in the literature only appear to be able to achieve less than 60% of the 2 C targeted capacity. In our paper [1], we proposed a link-directionality-based dual channel MAC protocol (DCP) to boost the network capacities up to 78% of our targeted capacities, 78%* 2 C =1 .56 C . However, DCP still failed to reach the 2 C capacity target due to the overheads incurred by the protocol. In this paper, we implement a power exchange algorithm on top of the DCP in an attempt to double the capacities of networks using the single-channel 802.11 protocol. This algorithm incurs relatively small overheads and significantly releases the protocol constraints imposed by DCP. Simulations show that the proposed scheme (DCPwPEA) can achieve more than 132% of our targeted capacities, 132%* 2 C = 2. 64 C . We believe this protocol can be further extended to allow multi-link simultaneous transmissions which can scale the network capacities of ad-hoc networks by using only two channels. Keywords — Wireless Networks, Ad hoc Networks, IEEE 802.11, Capacity, Scalable Performance.
I. INTRODUCTION When a wireless network uses more channel resources for transmissions, it should achieve a higher network capacity. If an IEEE 802.11 ad-hoc network achieves capacity C by using a single channel, the targeted capacity by using n channels should be n C . However, most of the multi-channel 802.11 protocols proposed in the literature simply compared their performance achievements with the original single-channel 802.11 protocol; without considering the channel resources they had used. In fact, most of them (e.g. [2][3]) only appear to be able to reach less than 60% of the targeted capacities ( n C ). This can be attributed to three reasons: i) an additional control channel is used to allocate transmission channels; ii) the overhead induced by the information added to the packet headers; and iii) the transmissions of RTS/DATA and the receptions of CTS/ACK by a node are assigned to the same channel which limit the potential of simultaneous transmissions (details will be explained in Section II). In the paper [1], we proposed a dual channel protocol which can achieve more than 78% of the 2 C targeted capacity, 0. 78 * 2 C 1 .56 C . Although pair-wise simultaneous transmissions are allowed by the protocol, the achievable capacities are limited by the overheads incurred. Thus, it fails to double the network capacities to 2 C . To compensate the penalties due to protocol overheads, in this paper, we further release the bundles of the DCP protocol by adopting a power exchange algorithm. II.
A LINK-DIRECTIONALITY -BASED DUAL CHANNEL MAC PROTOCOL (DCP)
A. The Concept In this section, we outline the concept and the protocol proposed in our paper [1]. To avoid simultaneous transmissions that may lead to collisions, the 802.11 protocol uses short request-to-send (RTS) and clear-to-send (CTS) messages to notify other nodes within the virtual carrier-sensing range (VCSRange) to update their Network Allocation Vector (NAV).
Soung Chang Liew Department of Information Engineering, The Chinese University of Hong Kong
[email protected]
The NAV includes the duration time of the ongoing transmission. Thus, no other nodes within the VCSRange can begin transmissions before the NAV expires. Figure 1a shows an example. Under the 802.11 protocol with RTS/CTS access mode, none of the links B, C or D can transmit at the same time with link A. This is because R A has to receive the DATA packet from T A while TA has to wait for the ACK from R A . Any other simultaneous transmissions within the VCSRange region of R A and T A in the same channel will lead to collision of the transmission between R A and TA . To release the above bundles, we can split the transmissions between two nodes of a link into two channels based on their directionalities. Let us consider the case where there are two channels, s and t. Nodes transmit RTS and DATA in one channel (e.g., channel s) as they are in the same direction (from TA to R A ) while CTS and ACK are transmitted in another channel (e.g. channel t). The channels are assigned dynamically based on the directionality, network topology, and who else are transmitting the neighborhood. RTS and DATA can be transmitted in either channel s or t, and thus CTS and ACK will be sent in the other channel (t or s). The main idea is to allow the simultaneous transmission of another link i within the VCSRange region of R A and TA provided that the transmissions of link i do not interfere with the receptions of the ACK on TA or the DATA on R A . There are two possible cases: Case 1: the transmissions of link i within the VCSRange use a different channel, and thus these do not affect the reception of R A ( T A ) in another channel. Case 2: the transmissions of link i using the same channel as the reception of R A ( T A ) but those transmissions are far enough from R A ( TA ). For Case 2, let d TARA be the distance between TA and R A , d TB RA be the distance between TB and R A , and assume the capture threshold (CPThreshold) is set to be 10dB. From [4], in a two-ray propagation model, assuming noise is negligible, if the signal-to-interference ratio (SIR) at R A is larger than the CPThreshold, R A can capture the signal from TA when T B is transmitting. That is, SNR ( d TBRA / d TA RA )4 CPThreshold
d TBRA 1.78 * d TA RA
(1)
In the worst case that TA and R A are separated by the maximum transmission range (250m), R A can capture the signal from TA if TB is located at more than 1.78*250m=445m away from R A . In our simulation, the VCSRange is set to be 550m. If TB can not receive the CTS from R A , TB must be far enough so that its signal can not interfere with the reception of signal from TA at R A . Our proposed MAC protocol in [1] utilized this property to assign transmission channels for links. Figure 1b shows the same scenario as Fig. 1a with the channel assignments based on the
Cases 1 and 2 for simultaneous transmissions of links A, B and C.
InRange varies with the distance between two nodes of a link. The closer the distance between T j and R j ( d Tj Rj ), the smaller
is the InRangej and simultaneous transmission links can then be packed closer to each other. This can drastically improve the network capacities.
a)
b)
Figure 1. A network topology using a) original 802.11 and b) our proposed scheme
B. Dual Channel Protocol (DCP) The proposed dual channel protocol (DCP) in [1] assigns the transmission channels of each link based on the availabilities of the receptions of RTS and CTS from other links. The protocol is modified from the original 802.11 MAC protocol and it attempts to seek opportunities for simultaneous transmissions. Assume all nodes use the same power for transmissions and each node has two half-duplex transceivers that are monitoring both channels at the same time. Consider two links, link i and link j. When a node (e.g., Ti ) of link i receives the RTS j but not the CTS j of another link j, it will assign RTS j to the same channel as that of RTS j . Thus, link iT R can transmit simultaneously with jT R beca use receiver R j is located far enough away from the transmitter Ti (as explained in Case 2 in Section II) and T j is receiving CTS or ACK in another channel (Case 1 in Section II). Thus, there is no collision between link i and link j. Similarly, when a node (e.g., Ti ) of link i receives the CTS j but not the RTS j of another link j, it will assign RTS j to the same channel as that of CTS j . If a node can receive both the CTS j and RTS j of another link j, it will fall back to the original 802.11 protocol and will resume transmissions only after NAV expires. In this case, links i and j have to take turn to transmit. Simulations in [1] showed that the proposed scheme (DCP) in [1] can significantly boost capacities up to 78% of the 2 C targeted capacities, 0 .78 * 2 C 1.56 C. III.
DUAL CHANNEL PROTOCOL WITH A POWER EXCHANGE ALGORITHM (DCPW PEA)
A. The Concept The previous section has outlined the DCP protocol which assigns transmission channels based on availabilities of receptions of RTS and CTS packets. In other words, pair-wise simultaneous transmission links must have one of their two nodes outside the VCSRanges of the nodes of the other link (see links B and C in Fig. 1b). This, however, induces unnecessary protocol constraints which limit the chances for simultaneous transmissions. Consider two links again, link i and link j. Deriving from the equation (1), we can define the interference range of link j as InRangej (CPThreshold )1 / d Tj Rj 1.78 * d Tj Rj
(2)
where the path loss exponent4 and CPThreshold 10dB . Simultaneous transmissions are actually permitted if a node of link i using the channel s (or t) for transmissions is apart from a node of link j using the same channel s (or t) for reception with a distance larger than InRange j . For example, in Fig. 2, links A and B can transmit concurrently as TB is located at more than InRange A away from R A . From equation (2),
Consider Fig. 2 again. Assuming T A uses channel 1 to transmit RTS and DATA to R A while R A uses another independent channel 2 to send CTS and ACK back to TA . Since TB is outside the interference range of link A ( InRange A ), it can also transmit RTS and DATA via channel 1. This does not affect the receptions of signals of R A from TA because the signal from TA at R A is more than 10dB stronger than the signal from TB . In addition, R B uses another independent channel (channel 2) for transmissions which can not interfere the signal receptions of R A . Therefore, links A and B can transmit at the same time. Comparing link B in Figures 1b and 2, simultaneous transmission links can now be packed closer to each other. This can significantly boost the network capacities. However, the protocol can no longer assign transmission channels based on availabilities of receptions of RTS and CTS packets as both nodes TB and R B can now receive the CTS packets from node R A . Thus, we implement a power exchange algorithm in the DCP by identifying possible simultaneous transmission opportunities within VCSRanges. B. DCP with a Power Exchange Algorithm (DCPwPEA) In this sub-section, we outline a power exchange algorithm for releasing the protocol constraints of the DCP protocol. To realize the concept mentioned in the previous sub-section, nodes seek simultaneous transmission opportunities based on the information included in the packets received from other links. As shown in Fig 2, we divide the regions for simultaneous transmissions into two parts: I.
a node of link j using channel s (or t) for transmission is more than VCSRange away from a node of link i using channel s (or t) for reception. (Section II)
II.
a node of link j using channel s (or t) for transmission is more than InRange but less than VCSRange away from a node of link i using channel s (or t) for reception. (Section III)
For region I, the DCPwPEA follows the DCP and keeps using the availabilities of receptions of RTS and CTS packets to decide channel assignments. For region II, we introduce a power exchange algorithm to assign channels based on the information included in the packets received from other links. The DCPwPEA first utilizes the information from the Power Exchange Algorithm to look for simultaneous transmission opportunities (for region II). In the case that one of the nodes of a link is located at region I (outside VCSRange) such that RTS/CTS packets and MAC headers of other packets can not be received, the DCP algorithm will then fall back to use the availabilities of receptions of RTS and CTS packets to select channels. Consider a pair of links, link i and link j, within region II. When R i ( Ti ) of link i receives a RTS (CTS) packet from Ti ( Ri ), it records the SIRTi Ri ( SIRRi Ti ) of the packet received and add an additional SIR header in its MAC header of the replied CTS (DATA) packet. When other nodes (e.g. T j ) receive the CTS (DATA) packet, they will extract the SIRTi Ri ( SIRRi Ti ) information from the MAC header. In addition, they will record the SIR Ri Tj ( SIRTi Tj ) of the CTS
(DATA) packets received. Owing to the reciprocal nature of the wireless radio link, we can state that SIRTj Ri SIRRi Tj ( SIRTj Ti SIRTi Tj ). The algorithm then compares SIRTj Ri ( SIRTjTi ) with SIRTi Ri ( SIR Ri Ti ). If SIRTj Ri SIRTiRi CPThreshold
( SIRTj Ti SIRRi Ti CPThreshold ), simultaneous transmissions are allowed and the protocol will then assign channels to nodes. Due to the space limitation, we defer the details of the protocol to the presentation in the conference.
data rate is set at 11Mbps while the basic rate (for transmitting RTS/CTS) sets at 2Mbps. All data sources are UDP traffic stream with fixed packet size of 1460bytes. Figure 5a shows three links in a string topology. As shown in Fig. 6a, using the original single-channel 802.11 protocol results in 4.85Mbps total network throughput, thus the targeted capacity for using dual channels is by definition 4.85Mbps*2=9.7Mbps. With our proposed scheme, a total network throughput of 12.85Mbps is achieved, which is 132% of the targeted capacity. Figures 5b and 6b shows another example of an irregular topology. Our proposed scheme obtains total network throughput at 21.71Mbps, which is 219% of the targeted capacity (4.95Mbps*2=9.9Mbps). In addition to the capacity enhancement, our proposed protocol achieves a fair bandwidth allocation in both cases. We will present more simulation results in the conference.
Figure 2. Closer packing of simultaneous transmissions are allowed after adopting a power exchange algorithm in DCP.
C. Overheads of DCPwPEA Comparing the DCPwPEA with the DCP, additional header overheads are incurred since DCPwPEA includes information about the SIR of the received packets in the headers of the replied packets (CTS or DATA). Figure 3 shows the structures of the DATA and CTS packets. However, the size of SIR (1byte) in the packet is much smaller than the size of the sum of the PHY header, MAC header and DATA/CTS. In contrast, the power exchange algorithm allows closer packing of simultaneous transmissions which significantly override the penalty incurred by header overheads.
a)
b)
Figure 5. a) Three links in a string topology and b) five links in an irregular network topology
a)
b)
Figure 3. SIR information added to headers of DATA and CTS p ackets Figure 6. Per-link throughput of the networks of Fig. 5 a and 5b with the original 802.11 protocol and our proposed MAC protocol (DCPwPEA)
Figure 4. Scaling the network capacity by our proposed channel assignment scheme (DCPwPEA)
V. CONCLUSION This paper has adopted a power exchange algorithm in the link-directionality-based dual channel protocol proposed in our paper [1] in an attempt to double the network capacities. The algorithm releases unnecessary protocol constraints imposed by DCP. The capacity improvements by allowing closer packing of simultaneous transmission links significantly override the overheads incurred by the algorithm. We have shown that our proposed scheme can achieve more than 132% of the targeted capacities, 132%*2C=2.64C. We have also demonstrated that the potential of scaling the network capacities by two channels. REFERENCES
Figure 4 shows another scenario with the channel assignments based on the directionalities of the links. Links 1 to 4 can transmit simultaneously and this can multiply the network capacity by using only two channels. In this paper, we first consider pair-wise simultaneous transmissions with the attempt to double the network throughput by using two channels. Channel assignments of simultaneous transmissions for multiple links require a more complicated MAC protocol. The details will be deferred to another paper. IV. S IMULATION RESULTS We have implemented our proposed MAC protocol (DCPwPEA) in the NS-2 [5] simulator. In our simulation, the
[1] [2] [3] [4] [5]
N. Choi et al., “Multi-channel MAC protocol for mobile ad hoc networks”, IEEE VTC 2003-Fall 2003. A.Baiocchi et al., “Why a Multichannel Protocol can boost IEEE 802.11 Performance”, ACM MSWiM’04. T. Rappaport, “Wireless Communications: Principles and Practice”, Prentice Hall, New Jersey, 2002 “The Network Simulator – ns2”, http://www.isi.edu/nsnam/ns. P.C. Ng, D. Edwards, S. C. Liew, “A Link-directionality-based Dual Channel MAC Protocol for IEEE 802.11 Ad-hoc Networks”, IEEE INFOCOM’06 Student Workshop, April 2006.
