Adaptive Power Control for Single Channel Ad Hoc ... - IEEE Xplore

1 downloads 0 Views 101KB Size Report
Dept. of Computer Science. The Hong Kong University of Science and Technology. Clear Water Bay, Kowloon, Hong Kong. {junalex, zyfang, brahim}@cs.ust.hk.
Adaptive Power Control for Single Channel Ad Hoc Networks Jun Zhang, Zuyuan Fang and Brahim Bensaou Dept. of Computer Science The Hong Kong University of Science and Technology Clear Water Bay, Kowloon, Hong Kong {junalex, zyfang, brahim}@cs.ust.hk Abstract— Power control for mobile ad hoc networks has received increasing interest in recent years by the research community. Power control can be invoked to achieve different objectives, among which the most popular goals of either increasing the network throughput by mitigating the effects of interference on a node or ensuring minimal power consumption. In this paper, we propose a power control algorithm for ad hoc networks to improve network throughput. As such, we study and exploit the correlation that exists between the appropriate transmit power of successive RTS, CTS, DATA and ACK frames and propose an adaptive power control algorithm in the framework of the IEEE802.11 MAC protocol. As opposed to some alternative approaches our algorithm operates in a single channel and does not require additional hardware support, which makes it very appropriate for the existing IEEE 802.11 based networks.

I. I NTRODUCTION The Distributed Coordination Function (DCF) of the IEEE 802.11 protocol standardizes the MAC protocol of wireless LAN in both infrastructure mode and ad hoc mode. To reduce collision probability when collisions are most likely to occur, the DCF uses a CSMA/CA protocol [1]. In addition, to mitigate the effects of collisions of long data frames, the DCF invokes the well known a 4-way handshakes RTS/CTS/DATA/ACK on a per data frame-basis. Sender and receiver exchange Ready to send (RTS) and clear to send (CTS) frames before data is transmitted. RTS and CTS frames have the effect of inhibiting all the nodes on the floor of the sender and that of the receiver thus ensuring a successful completion of the 4-way handshake by the DATA frame and its acknowledgement (ACK) frame. Power control in MANETs targets the improvement of network throughput or reduction of energy consumption. The transmit power is adapted dynamically to the level of perceived interference to ensure successful transmission. Transmitting packets at a high power level results in large energy consumption and generates much interference to the rest of the network, thus impacting negatively the network throughput. Transmitting packets at a low power level increases the probability of collision of the current transmission. Therefore, many algorithms have been proposed in the literature [2], [3], [4], [5], [6], [7], [8] to integrate power control in the IEEE802.11 DCF protocol. This work is supported in part under Grant CERG HKUST6166/04E.

All of this previous work on power control only considers the assignment of the transmission power of each packet separately. Power control algorithms based on special equipment support, such as multiple channels or busy tone, determine the appropriate transmission power by the noise level at the receiver side at the time of packet transmission [2], [3], [4], [7]. However, the noise at the receiver side may change during the time the packet is transmitted. Thus the appropriate transmission power can not be calculated accurately. The adaptive transmission power assignment algorithm in [8] determines the transmission power of the current frame based on the status of the last frame it transmitted to the same destination. Nevertheless, the success of the transmission depends on the noise level at the receiver, which in turn depends on the transmission power of the last control packet sent by the receiver. Therefore, there exists a relationship between the appropriate transmission power of RTS and CTS, CTS and DATA, and DATA and ACK to fulfill one successful 4-way handshake. In this paper, we study this relationship and then propose an adaptive power control algorithm that relies on this relationship. The rest of the paper is organized as follows. Section II reviews basic notations in ad hoc networks, the IEEE 802.11 MAC protocol and the state of the art of power control algorithms. Section III presents our adaptive single channel power control algorithms. In Section IV we compare the performance of our power control algorithms with the IEEE 802.11 standard and other proposals, both in throughput and energy consumption. Finally, we conclude this paper in Section V. II. R ELATED WORKS A. Basic framework In mobile ad hoc networks, a frame sent from transmitter i can be decoded successfully at the receiver j if and only if the signal to interference ratio (SIR) at the receiver side is larger than a predetermined threshold ζ and the power of the desired packet is larger than a minimal power level κ. The SIR of a frame from transmitter i at the receiver j is given by: SIR(i, j) = Pj i /PN oise

(1)

where Pj is the power level of the desired frame at receiver j from transmitter i and PN oise is the sum of the total power

3156 0-7803-8938-7/05/$20.00 (C) 2005 IEEE

i

of other undesired frames and the power of the thermal noise at receiver j. The receiver cannot decode the packet when the SIR of a packet is lower than ζ at the receiver j or the power of the packet is lower than the κ. We call the above case a collision at receiver j. Each transmitter i determines its transmission zone and carrier sensing zone according to its transmission power. Assuming that there is no other transmitter, the power level of a frame from node i in node i’s transmission zone is higher than or equals to κ and the power level of a frame from node i in node i’s carrier sensing zone is higher than or equals to a carrier sensing threshold η. When the received power from node i is higher than η at node j, we say that node j can sense the transmission from node i. One node C is a hidden terminal to the transmission from node A to node B if node B is in C’s carrier sensing zone and C is outside A’s carrier sensing zone. Therefore, as C cannot detect such a transmission from A, the transmission from A and C may collide at B. This is the, now well known hidden terminal problem. In addition to carrier sensing, IEEE802.11 uses the so-called virtual carrier sensing. Each of the RTS, CTS and DATA frames headers carry a field called the duration field, set by the transmitter of the frame to the duration of the ensuing part of the 4-way handshake. Any node that overhears the RTS, CTS or DATA frames updates its network allocation vector (NAV) from the duration field of the frame and defers transmission for the duration of the NAV. No node is allowed to transmit an RTS when it senses the network is busy through physical carrier sensing or when the time in its NAV is set. B. Related work on power control for ad hoc networks Although the IEEE802.11 MAC protocol more or less solves the hidden terminal problem, it has its drawbacks. All packets are transmitted at the same power, generally the maximal possible transmission power, which is unnecessary. In the case where the network is dense, a low transmission power is sufficient to maintain network connectivity while a high transmission power brings about more interference to other transmissions, resulting in more retransmissions and more energy consumption. The throughput of the network is also poor in such a case because high transmission power reduces the spatial reuse of the network bandwidth. The power control algorithm called BASIC is presented in [3]. RTS and CTS are transmitted with the maximal possible power, and DATA and ACK are transmitted with the minimal necessary power to reach the destination. The BASIC power control algorithm claims a similar throughput to that of the IEEE802.11 protocol with lesser energy consumption. The hidden terminal problem exists in the BASIC power control scheme. Algorithms in [2], [4], [7] are variants of the BASIC power control scheme, with the support of multiple channels or busy tones. The signal in the control channels or the busy tone does not collide with the transmission in the common channel. Therefore, the information of how much noise each

receiver can tolerate is broadcast in the busy tone or the control channel, silencing hidden terminals and helping the sender to adjust its transmission power. Although these three algorithms claim to achieve good throughput and less energy consumption, the implementation of dual or multi-channels in the framework of IEEE802.11 faces both technical difficulties and market resistance as such algorithms would require a complete change of the standards. Agrawal et al. proposed a distributed power control algorithm in [8] where the transmission power for each sender/receiver pair is adjusted adaptively. Each transmitter increases/decreases its transmission power to its receiver if the last transmission to the same receiver fails/succeeds. This algorithm has higher throughput and lower energy consumption compared with the IEEE802.11 MAC protocol. However, it does not take into consideration the difference of the requirement in transmission power of different packets and the relationship between the transmission power of sequential packets between the transmitter and the receiver. Therefore, the transmission power changes frequently. In addition, this algorithm suffers from the hidden terminal problem because of the asymmetric transmission power, as pointed out in [3]. III. A DAPTIVE SINGLE CHANNEL POWER CONTROL ALGORITHM

Let (p, q) to be an element in the set of successive packet pairs: {(RT S, CT S), (CT S, DAT A), (DAT A, ACK)}. In each such pair of packets, the delivery of packet p has two possible goals: i) Handshaking; ii) preventing a collision of the ensuing transmission of packet q. To the extent of our knowledge, previous works on power control have only considered the assignment of transmission power to fulfill the first goal. However, if the assignment of transmission power fails to consider the second goal, it may fail to achieve the first one. Let us take the following example: A low transmission power is needed for a CTS to reach the transmitter; When the receiver transmits the CTS at a low power level, nodes around the receiver may not sense this transmission and may subsequently generate a large interference to the receiver; The transmission of the subsequent DATA packet may fail even with the maximal transmission power. This case is avoidable if the CTS is assigned a higher transmission power. Thus, we notice that there exists a relationship between the transmission power of packet p (in this example the CTS) and packet q (in this example the data). In the following, we analyze this relationship and propose our adaptive single channel power control algorithm based on the conclusions of this analysis. A. Relationship between the transmission power of adjacent packets We use the two ray ground propagation model in this paper. According to the two ray ground propagation model, the relationship between the transmit power Pt and the received power Pr at the receiver is as follows:

3157

Pr = Pt ∗ Gt ∗ Gr ∗ ht 2 ∗ hr 2 /d4 ,

(2)

where Gt and Gr are the antenna gains of the transmitter and the receiver respectively, ht and hr are the heights of the transmitter’s antenna and the receiver’s antenna respectively, and d is the distance between the transmitter and receiver. Without loss of generality, we discuss the relationship between the transmission power of CTS and DATA. We consider the following scenario, which is shown in Figure 1. Node A is the transmitter and node B is the receiver. All other nodes are densely distributed, which means that there exists a node at any location. All other nodes remain idle and transmit packets immediately after node A starts sending DATA to node B. For simplicity, we only consider single interference here, that is at most one node (other than node A) generates interference to node B. Before going into detail, we define several notations. Let us look at the example in Fig.1. We denote the transmission power of packet x as perceived at location y to be Px,y , where x is in {RT S, CT S, DAT A, ACK} and y is in {t, r}, where t is the transmitter of packet x and r is the receiver of the packet. The distance between node A and node B is denoted d. RCT S is the radius of the transmission zone of node B and RI is the radius of the interference zone of node B. The DATA packet for node B is lost if any other nodes in the interference zone transmits at maximal transmission power Pmax . ζ is the received signal to interference threshold, and κ is the received power threshold. The relationships between the transmission power of RTS/CTS/DATA are PRT S,r PDAT A,r PCT S,r

= PRT S,t ∗ Gt ∗ Gr ∗ h4 /d4 = PDAT A,t ∗ Gt ∗ Gr ∗ h4 /d4

(3) (4)

= PCT S,t ∗ Gt ∗ Gr ∗ h4 /d4 .

(5)

The radius of the transmission zone of CTS verifies κ

= PCT S,t ∗ Gt ∗ Gr ∗ h4 /RCT S 4 .

(6)

The interference radius verifies PDAT A,r = Pmax ∗ Gt ∗ Gr ∗ h4 /RI 4 . (7) ζ The received power of CTS and DATA must be greater than

According to [9], to guarantee the successful transmission of the DATA packet, RCT S should be larger than RI . RCT S

≥ κ

(8)

PDAT A,r

≥ κ.

(9)

(10)

From the above equations, after some manipulations, we obtain the following inequalities PDAT A,t ∗ PCT S,t PCT S,t

PDAT A,t



Pmax ∗ κ ∗ ζ PRT S,r /PRT S,t κ ∗ PRT S,t ≥ PRT S,r κ ∗ PRT S,t ≥ PRT S,r

(11)

(12) (13)

. Let Pw (CT S, DAT A) denote the right hand side of (11) and call it “the cross coefficient of the CTS-DATA pair”. Note that we only consider the simplest case here: this cross coefficient may be smaller than the above value in the case where the nodes are not densely distributed. In such a case, there may not exist a node at a distance between RI − δ and RI to node B, where δ is a very small value. Thus even though the transmission radius of CTS RCT S is slightly reduced, for example δ/2, the transmission zone of CTS still covers the interference zone. The cross coefficient may also be larger than the above value in the case where there exist multiple interferers. Because in this case Pmax is replaced by k ∗ Pmax in Equation (1), Pw (CT S, DAT A) is also replaced by Pw (CT S, DAT A) ∗ k, where k is an integer great than 1. Similarly we define the cross coefficient of the (RTS,CTS) pair to be Pw (RT S, CT S) and the cross coefficient of the (DATA,ACK) pair to be Pw (DAT A, ACK). The cross coefficient of the pair (p,q) defines the minimal possible value of the product of Pp,t and Pq,t to guarantee a successful transmission of packet q given that packet p has been successful. Assuming that the attenuation gain between the transmitter and the receiver is symmetric, that is, Pi,t /Pi,r is the same for all i in {RT S, CT S, DAT A, ACK}. Let g(t, r) denote Pi,r /Pi,t . Then the requirement of the assignment of transmission power to guarantee a successful transmission between the t and r is as follows: Pi,t ∗ Pj,t

κ. PCT S,r

≥ RI .

Pi,t

≥ Pw (i, j) ≥

κ g(t,r) ,

(14)

(15)

where i in {RT S, CT S, DAT A, ACK} and (i, j) in {(RT S, CT S), (CT S, DAT A), (DAT A, ACK)}. B. Discussion about the interference to the receiver

RI A

Fig. 1.

B RCTS

Interference zone of node B versus transmission zone of node B

The distributed power control algorithm in [8] is based on the following assumption: when the network is in a steady state, the necessary transmission power for each sender/receiver pair does not change much and is independent of the packet type. We show that this assumption is far from being true. The received power on each receiver varies vastly. Assume that nodes are uniformly distributed in the network with density ∆, that is, there are ∆ nodes in a unit space. All nodes transmit packets with the same transmission power Pt

3158

and their transmission ranges are R. Every antenna has the same height h and the same antenna gain of 1. Therefore the transmission power Pt of each transmitter equals to κ ∗ R4 /h4 . Given a node A, we denote r the distance to its nearest neighbor. According to the nodes’ density, r2 equals to 1/∆ on average. We analyze the received power on node A in three cases: 1) All nodes transmit at the same time. 2) All nodes outside A’s transmission zone transmit at the same time. 3) All nodes transmit at the same time, except the nodes that are inside any other node’s transmission zone. In the first case, the power received by node A equals to Pr (A)1

∞ Pt ∗ h4 (2 ∗ π ∗ x ∗ ∆ ∗ dx) ∗ x4

=

algorithms should not increase the transmission power right after a packet loss occurs. C. Adaptive power control algorithm In the case where the assignment of transmission power of packet i, i in {RT S, CT S, DAT A, ACK} fulfills the requirement in (14), the transmission is guaranteed with less power consumption compared to the IEEE802.11 DCF protocol. We propose our adaptive power control algorithm as follows. Each node s maintains a table for each destination d. The entry for each destination includes gain(d) and Pw (i, j, d) where (i, j) in {(RT S, CT S), (CT S, DAT A), (DAT A, ACK)}; gain(d) stands for the attenuation gain between s and d; Pw (i, j, d) stands for the cross coefficient of packet i and j for the node pair s/d. Initially, Pw (i, j, d) equals to

r

= κ ∗ R4 ∗ π ∗ ∆2

In the second case, the power received by node A equals to Pr (A)2

∞ Pt ∗ h 4 = (2 ∗ π ∗ x ∗ ∆ ∗ dx) ∗ x4 R

= κ ∗ R2 ∗ π ∗ ∆

(17)

In the last case, transmitters are separated from each other by a distance of at least R. Thus the density of the transmitters in the network is lower than 1/R2 . So the power received by node A is no larger than Pr (A)3

=

min(Pmax ∗ Pmax , Pmax ∗ κ ∗ ζ/gain(d)).

(16)

∞ Pt ∗ h4 (2 ∗ π ∗ x ∗ 1/R2 ∗ dx) ∗ x4

(19)

Whenever gain(d) is not known, it is simply set to 0. Two fields are inserted into the header of each packet i for the sender/receiver pair s/d. One is the transmission power of the packet. The other is the value of Pw (i, j, d) for sender/receiver pair s/d, where j is the subsequent packet of packet i. The algorithm is shown in Algorithm 1. For simplicity, we assume the local node is d and the incoming packets are from node s and the outgoing packets are also to node s. Therefore, we omit the index here. In addition, for ease of description, we call the packet pair (i, j) in {(RT S, CT S), (CT S, DAT A), (DAT A, ACK)} as the corresponding packet pair for packet i in {RT S, CT S, DAT A, ACK}, denoted as pair(i).

R

= κ∗π

(18)

Pr (A)1 is the highest possible interference at a receiver when it receives an RTS. Pr (A)2 is the highest possible interference at a receiver when it receives a CTS, a DATA or an ACK frame. The interference at a receiver when it receives CTS, DATA or ACK is lower than Pr (A)3 in general because of the presence of a MAC protocol that in general solves the hidden terminal problem. We observe that the interference to the receiver in general is much lower than that of the extreme case. According to the carrier sensing and collision avoidance schemes, the probability of all nodes transmitting packets at the same time is negligibly small. Therefore, transmitting a packet at the power level which guarantees a successful transmission in case 3 is ,in general, sufficient. The receiver sends a CTS only when it receives an RTS in the case where its NAV is not set and the channel is idle. Thus the requirement of the transmission power of RTS is only that the received power of RTS at the receiver side is larger than κ. Therefore, when the transmitter sends an RTS and does not get the corresponding CTS, it is very likely due to the fact that the receiver was busy at that time rather than that the transmission power of the RTS was too low. So adaptive power control

Algorithm 1 Adaptive Power control algorithm Notations. InP : packet being received OutP : packet being transmitted PP kt : transmission power of packet Pkt Pkt.T xP r : transmission power of Packet Pkt Pkt.RxP r : receiving power of Packet Pkt Pkt.Pw : packet’s correlation coefficient Pmin : κ/gain(s) k : cross coefficient increasing/decreasing rate BEGIN OnSendRTS PRT S =sqrt(Pw (RT S, CT S)) OnRecvPkt gain=InP.RxP r/InP.T xP r POutP =sqrt(max(InP.Pw ,Pw (pair(OutP )) On Pkt times out Pw (pair(P kt)) = min(Pmax ∗ Pmax , Pw (pair(P kt)) ∗ k) On 10 continuous pair(P kt) successful exchanges Pw (pair(P kt)) = max(Pmin ∗ Pmin , Pw (pair(P kt))/k) END

3159

IV. P ERFORMANCE E VALUATION

80 End to end throughput(kBps)

A. Simulation configuration We investigate the performance of our adaptive power control algorithm using ns2 simulations [10]. We modified the source code of ns2 to support simultaneous multiple interfering nodes. Both total end-to-end throughput and the energy consumption are considered as part of the performance metrics. We adopt two network topologies in the simulations. One is a 7 x 7 grid network with 7 end to end flows, each starting from the left most node and ending at the right most node of the same row (note that the routes are not necessarily the rows, they are in fact determined by the routing algorithm dynamically). The distance between each node pair is 100 m. The other scenario is a 50 nodes random network with 10 end to end flows and all nodes are inside a 1000 x 1000 m2 square. The flows are TCP flows or CBR flows. The packet size is 512 bytes. The packet sending rate is 100 packets per second for CBR. Therefore, there are four scenarios totally. 1) Grid network with TCP flows. 2) Grid network with CBR flows. 3) Random network with TCP flows. 4) Random network with CBR flows. We compare the performance of our adaptive power control algorithm with IEEE802.11 MAC protocol, BASIC [3] and distributed power control [8]. The total end to end throughput by the IEEE802.11 protocol and different power control algorithms are shown in Figure 2. The throughput of our adaptive power control algorithm is clearly higher than that of the others. The ratio of throughput to the average energy consumption per node different algorithms is shown in Figure 3. Our adaptive power control algorithm proves to be better than IEEE802.11 protocol in all cases and better than other power control algorithm in most cases. That is, it carries more bytes per joule of energy consumed. These results show that indeed by taking into account the correlation between the transmission power of successive frames, we can reduce power consumption and increase the throughput. V. C ONCLUSION AND F UTURE WORK In this paper, we studied the correlation between the necessary transmission power of RTS/CTS/DATA/ACK to deliver packets successfully. Based on this correlation, we developed and proposed an adaptive single channel power control algorithms that improves the network throughput and reduces the energy consumption in ad-hoc networks. It is shown by simulations that our algorithm has higher throughput and lower energy consumption compared to other algorithms in its class. R EFERENCES [1] IEEE Computer Society LAN MAN Standards Committee, Ed., IEEE Standard for Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, ser. IEEE Std 802.11-1997. The Institute of Electrical and Electronics Engineers, New York, 1997. [2] J. P. Monks, V. Bharghavan, and W. W. Hwu, “A Power Controlled Multiple Access Procotol for Wireless Packet Networks,” in INFOCOM, 2001, pp. 219–228. [3] E. Jung and N. H. Vaidya, “A Power Control MAC Protocol for Ad Hoc Networks,” in MOBICOM, 2002, pp. 36–47.

IEEE802.11 BASIC(c=4) Distributed Adaptive

60

40

20

0

1

2 3 Scenario

Throughput/Energy ratio(KBps/Joules)

Fig. 2.

Fig. 3.

4

End to end throughput

8 IEEE802.11 BASIC(c=4) Distributed Adaptive

6

4

2

0

1

2 3 Scenario

4

Ratio of Throughput by average energy consumption per node

[4] A. Muqattash and M. Krunz, “Power Controlled Dual Channel (PCDC) Medium Access Protocol for Wireless ad Hoc Networks,” in INFOCOM, April, Ed., 2003, pp. 470–480. [5] V. Kawadia and P. R. Kumar, “Power Control and Clustering in Ad Hoc Networks,” in INFOCOM, 2003, pp. 459–469. [6] T. ElBatt and A. Ephremides, “Joint scheduling and power control for wireless ad hoc networks,” in Wireless Communications, IEEE Transactions on 2004, ser. 1, vol. 3, Jan. 2004, pp. 74 – 85. [7] S. L. Wu, Y. Tseng, and J. Sheu, “Intelligent medium access for mobile ad hoc networks with busy tones and power control,” in JSAC 2000, ser. 9, vol. 18, Sept. 2000, pp. 1647 – 1657. [8] S. Agarwal, S. Krishnamurthy, R. Katz., and S. Dao, “Distributed power control in ad hoc wireless networks,” in PIMRC 2001, vol. 2, Oct. 2001, pp. 59–66. [9] K. Xu, M. Gerla, and S. Bae, “How Effective is the IEEE 802.11 RTS/CTS Handshake in Ad Hoc Networks?” in IEEE Globecom, 2002. [10] “The network simulator -ns-2,” http://www.isi.edu/nsnam/ns/,2.26.

3160