0-7803-7016-3/01/$10.00 ©2001 IEEE
`Investigating the Energy Consumption of a Wireless
`Network Interface in an Ad Hoc Networking
`Environment
`Laura Marie Feeney, Martin Nilsson
`Abstract—Energy-aware design and evaluation of network protocols re-
`quires knowledge of the energy consumption behavior of actual wireless
`interfaces. But little practical information is available about the energy
`consumption behavior of well-known wireless network interfaces and de-
`vice specifications do not provide information in a form that is helpful to
`protocol developers. This paper describes a series of experiments which ob-
`tained detailed measurements of the energy consumption of an IEEE 802.11
`wireless network interface operating in an ad hoc networking environment.
`The data is presented as a collection of linear equations for calculating the
`energy consumed in sending, receiving and discarding broadcast and point-
`to-point data packets of various sizes. Some implications for protocol design
`and evaluation in ad hoc networks are discussed.
`Keywords—energy consumption, IEEE 802.11, ad hoc networks
`I. INTRODUCTION
`Because mobile devices are dependent on battery power, it is
`important to minimize their energy consumption. The energy
`consumption of the network interface can be significant, espe-
`cially for smaller devices. Most research in energy conserva-
`tion strategies has targeted wireless networks that are structured
`around base stations and centralized servers, which do not have
`the limitations associated with small, portable devices.
`By contrast, an ad hoc network is a group of mobile, wireless
`hosts which cooperatively form a network independently of any
`fixed infrastructure. The multi-hop routing problem in ad hoc
`networks has been widely studied in terms of bandwidth utiliza-
`tion, but energy consumption has received less attention. It is
`sometimes (incorrectly) assumed that bandwidth utilization and
`energy consumption are roughly synonymous. Recently, there
`has been some study of energy-aware ad hoc routing protocols,
`particularly for distributed sensor networks. In this context, en-
`ergy is often treated as an abstract “commodity” for purposes of
`minimizing cost or maximizing time to network partition.
`We believe that energy-aware design and evaluation of net-
`work protocols for the ad hoc networking environment requires
`practical knowledge of the energy consumption behavior of ac-
`tual wireless devices. In addition, it is important to present
`this information in a form that is useful to protocol developers:
`the total energy costs associated with a packet containing some
`number of bytes of data. Device specifications, which indicate
`the current draw while transmitting and receiving, are somewhat
`unhelpful in this respect.
`This paper describes a series of experiments which obtained
`detailed measurements of the energy consumption of a Lucent
`WaveLAN IEEE 802.11 wireless network interface operating in
`ad hoc mode. The data is presented as a collection of linear
`equations for calculating the energy consumed in sending, re-
`Both authors are at the Swedish Institute of Computer Science in Kista, Swe-
`den (http://www.sics.se). E-mail: lmfeeney@sics.se.
`ceiving and discarding broadcast and point-to-point data pack-
`ets of various sizes. While not intended to account for all IEEE
`802.11-based products or for all possible factors affecting en-
`ergy consumption at the wireless interface, the results provide
`a solid experimental basis for energy-aware design and evalu-
`ation of network-layer protocols operating in the IEEE 802.11
`environment.
`The results suggest new perspectives on design of ad hoc net-
`working protocols:
`First, it is clear that energy consumption and bandwidth uti-
`lization are not synonymous. It is necessary to consider not only
`the cost of transmitting a packet, but also of receiving, and even
`of discarding it. Protocol designers must therefore consider the
`proportions of broadcast and point-to-point traffic used by the
`protocol. Because channel acquisition overhead is large, small
`packets have disproportionately high energy costs. Promiscu-
`ous mode operation, which is irrelevant to bandwidth utilization,
`also incurs some energy cost.
`Second, the relationship between transmit speed and overall
`energy consumption is complex. Reduced data transmit and re-
`ceive times have only limited impact on per-packet energy con-
`sumption, due to the hight fixed overhead. There are other trade-
`offs involved: the reduced transmission range associated with
`higher data rates both increases the number of hops required in
`a multi-hop routing environment, but decreases the number of
`neighbors affected by each transmission.
`Third, ad hoc mode operation incurs an extremely high idle
`cost compared to operation in conjunction with a base station.
`Routing protocols which are structured to emulate some of the
`energy efficiency associated with BSS mode operation should
`be investigated.
`II. R
`ELATED W ORK
`Experimental Results
`There are few published measurements of the energy con-
`sumption of network interfaces. In particular, there are no de-
`tailed measurements of the per-packet energy consumption of
`an IEEE 802.11 network interface operating in ad hoc mode. In
`general, device specifications, which show current draw while
`transmitting and receiving aren’t sufficient to calculate per-
`packet energy consumption. For example, idle mode, switching
`from transmit to receive mode and the effects of internal energy
`management strategies may not be reflected.
`Experiments measuring the power consumption of IEEE
`802.11 PC cards are reported in [10]. Emphasizing operation
`in conjunction with a base station, the methodology is based on
`sampling the current draw of an otherwise idle laptop as it sends
`1548 IEEE INFOCOM 2001
`Authorized licensed use limited to: Austin Ginnings. Downloaded on October 08,2025 at 03:50:43 UTC from IEEE Xplore. Restrictions apply.
`SCT Exhibit 2026
`IPR2026-00098
`
`
`
`
`
`
`
`0-7803-7016-3/01/$10.00 ©2001 IEEE
`and receives traffic over an extended period of time. By compar-
`ing the total energy used while sending or receiving traffic with
`the energy consumed by an idle laptop, the total cost of process-
`ing traffic can be computed. This approach has the advantage of
`measuring total system cost, but provides little packet-oriented
`detail. Also, the results are potentially more dependent on par-
`ticular system-wide energy management techniques than on the
`behavior of the network interface.
`Packet-oriented measurements of the energy consumption of
`wireless interfaces are described in [7] and the experiments de-
`scribed below are very closely based on this methodology. How-
`ever, this work only includes measurements of the pre-IEEE
`802.11 WaveLAN interface (and several other devices). In addi-
`tion to updating these results to include more recent hardware,
`the current work examines ad hoc mode operation of the net-
`work interface and complexities introduced by the IEEE 802.11
`protocol. It is also unique in investigating the behavior of non-
`destination hosts which overhear wireless traffic.
`Energy-aware Protocol Design and Evaluation
`A detailed analytical study of the energy efficiency of a num-
`ber of MAC-layer protocols, including IEEE 802.11, is pre-
`sented in [2]. This probabilistic analysis examines the effec-
`tiveness of various media acquisition strategies in the presence
`of contention. Contention is dependent on many factors, such
`as communication patterns, node density and RF transmission
`characteristics, that are difficult to reproduce experimentally.
`We therefore attempted to minimize the possibility of media
`contention and retransmissions in our measurements.
`Most energy-conserving link-layer protocols are targeted to-
`ward the centralized, base station environment. Such protocols
`usually rely on a resource-rich base station to moderate com-
`munication among hosts, scheduling and buffering traffic to re-
`duce contention and allowing the more limited mobile devices
`to spend as much time as possible in a low-power consumption
`sleep state. Unfortunately, these strategies have limited applica-
`bility in the ad hoc environment, in which there are no fixed base
`stations and mobile hosts may have limited buffering capability
`and unpredictable connectivity.
`Evaluating the energy efficiency of network-layer protocols
`has proven to be a surprisingly subtle task. Energy consump-
`tion and bandwidth utilization are substantively different met-
`rics: the former must reflect costs for sending, receiving and dis-
`carding traffic and emphasizes the differences between broad-
`cast and point-to-point traffic. A simulation study [4] of the
`energy consumption of two well-known ad hoc routing proto-
`cols running over IEEE 802.11 demonstrated that an energy-
`oriented performance evaluation may come to quite different
`conclusions than a bandwidth-oriented one. This result, which
`was based in part on earlier versions of these experiments [6],
`has provided impetus for the more comprehensive investigation
`presented here.
`Application-level energy conservation strategies [7], [10] take
`advantage of usage patterns associated with activities such as
`email retrieval and web browsing. This allows the network in-
`terface to spend as much time as possible in an inactive, reduced
`power consumption state with minimal impact on the perfor-
`mance perceived by the user. Such techniques are not applicable
`to an ad hoc network. Because the hosts in an ad hoc network
`also form its routing infrastructure, it is difficult or impossible to
`predict when it is “safe” for a network interface to enter a low-
`power sleep state. Even entirely quiescent nodes may be (or
`become) essential to maintaining network connectivity or pro-
`viding other essential network services.
`There has also been some recent interest in energy-based
`techniques for ad hoc sensor networks, in which a collection of
`specialized sensors cooperatively forward sampled data to more
`powerful hosts for further processing or other action. Routing
`protocols which seek to maximize the connected lifetime of the
`network by energy-aware load balancing are presented in [1],
`[8]. Methods for selecting transmit power levels to maximize
`network lifetime and reduce spatial interference are presented
`in [1], [13]. However, energy is often treated as abstract com-
`modity and subtle issues such as those suggested above are not
`addressed.
`III. O
`VERVIEW
`The goal of this work was to investigate the energy consump-
`tion of a wireless network interface via direct measurement
`Is the network interface a significant factor in overall system
`energy consumption?
`A variety of measurements of the power consumption of an
`idle laptop computer are found in [3], [10], [7]: reported results
`range from 6 to 14 W. However, the growth of mobile computing
`is leading to the development of low-power “mobile” processors
`and systems. Detailed measurements of the energy consumption
`of Compaq’s experimental Itsy “pocket computer” are reported
`in [3]. This PDA has an idle power consumption of 100 - 200
`mW; most applications exhibit peak power consumption of 1
`- 1.5W. Transmeta’s proposed “all-day mobile computer” [16]
`is expected to consume 5 - 6W, of which the active CPU will
`account for 1-2W.
`Lucent IEEE 802.11 WaveLAN device specifications suggest
`transmit and receive power consumption of about 1.5W and 1W,
`respectively. A host generally spends only a small portion of its
`time sending and receiving traffic, so idle power consumption
`is important to overall energy costs. Operating in ad hoc mode,
`the idle power consumption is nearly as large as that of receiving
`data. Reducing the energy consumption of the network interface
`is therefore extremely important.
`What effect does ad hoc operation have on power consump-
`tion?
`IEEE 802.11 defines two primary modes of operation for a
`wireless network interface: base station (BSS) mode and ad
`hoc mode. Every mobile host operating in BSS mode must be
`in transmission range of one or more base stations, which are
`responsible for buffering and forwarding traffic between hosts.
`Hosts can send outgoing traffic to the base station anytime and
`periodically poll the base station to receive incoming traffic. The
`remainder of the time is spent in a sleep state, from which the
`interface must explicitly wake up in order to send or receive
`traffic. The base stations’ guaranteed availability and buffering
`and traffic management capabilities are required to support this
`energy conserving functionality.
`Ad hoc mode operation does not use any base station in-
`frastructure: nodes communicate directly with all other nodes
`1549 IEEE INFOCOM 2001
`Authorized licensed use limited to: Austin Ginnings. Downloaded on October 08,2025 at 03:50:43 UTC from IEEE Xplore. Restrictions apply.
`SCT Exhibit 2026
`IPR2026-00098
`
`
`
`
`
`
`
`0-7803-7016-3/01/$10.00 ©2001 IEEE
`that are in wireless transmission range. Because there are no
`base stations to moderate communication, hosts must always be
`ready to receive traffic from their neighbors. An network inter-
`face operating in ad hoc mode does not sleep; it has a constant
`idle power consumption which reflects the cost of listening to
`the wireless channel. This cost, which has been measured, but
`is not described in device specifications, is only slightly less than
`that of actually receiving traffic.
`How can we model per-packet energy consumption?
`In the simple case, the energy consumed by the network in-
`terface when a host sends, receives or discards a packet can be
`described using a linear equation
`/BX /D2/CT/D6 /CV /DD /BP /D1 /A2 /D7/CX/DE /CT /B7 /CQ/BM
`Trivially, there is a fixed component associated with device state
`changes and channel acquisition overhead and an incremental
`component which is proportional to the size of the packet. Ex-
`perimental results confirm the accuracy of the linear model and
`are used to determine values for the linear coefficients
`/D1 and /CQ
`for various operations.
`The model does not consider the case of link-layer fragmen-
`tation. It is expected that the linear model would continue to
`apply, with each instance of fragmentation introducing a small
`step discontinuity reflecting a fixed fragmentation overhead.
`The model also does not consider energy consumed in unsuc-
`cessful attempts to acquire the channel (media contention), or in
`messages lost due to collision, bit error or loss of wireless con-
`nectivity. Such effects are difficult to obtain for controlled ex-
`perimental measurements. While these phenomena are clearly
`important for the energy consumption behavior of a network,
`they are probably best examined probabilistically in the con-
`text of a specific model of host density, traffic load and wireless
`transmission environment.
`What are the relative costs of sending, receiving and discard-
`ing traffic? What are the relative costs of large and small pack-
`ets? Of broadcast and point-to-point traffic? Of promiscuous
`mode operation?
`In contrast with bandwidth metrics, which count the number
`of packets or bytes sent over the wireless media, energy con-
`sumption metrics must account for the reaction of every network
`interface within wireless transmission range of the participating
`hosts.
`The relative magnitudes of the various
`/D1 and /CQ coefficients
`also indicate the amount of per-packet energy consumption
`overhead.
`A packet may be sent as broadcast or point-to-point traffic.
`The former is received by all hosts within transmission range;
`the latter is discarded by non-destination hosts, unless they are
`operating in promiscuous mode, while the MAC traffic is pro-
`cessed by all hosts in range of either the sender or the destina-
`tion. In every case, energy will be consumed at the all relevant
`network interfaces. It is important to note that the costs of re-
`ceiving and discarding are multiplied by the number of hosts
`which receive or discard the traffic. Energy consumption is af-
`fected by node density.
`Protocols can be designed to use different combinations of
`large and small packets. Some use techniques which piggyback
`their data onto existing traffic. Protocols can also be designed to
`use different combinations of broadcast and point-to-point mes-
`sages. For the former, all nodes in wireless range of the sender
`must process the traffic. For the latter, non-destination hosts
`may discard traffic, unless they are operating in promiscuous
`mode. In addition, the collision avoidance and acknowledge-
`ment mechanisms used by the IEEE 802.11 protocol differ for
`broadcast and point-to-point messages. We note that bandwidth
`metrics do not address these issues.
`IV . M
`EASURING E NERGY C ONSUMPTION
`The test cards were popular and widely available 2.4GHz
`DSSS Lucent IEEE 802.11 WaveLAN PC “Bronze” (2Mbps)
`and “Silver” (11Mbps) cards. The test host was an IBM
`ThinkPad 560, running FreeBSD 4.0 and the freely avail-
`able WaveLAN IEEE 802.11 driver written by Bill Paul
`(wpaul@ctr.columbia.edu). The machine ran with
`power management turned off in order to avoid unexpected in-
`teraction with system facilities. UDP test traffic was generated
`at the rate of 10-20 packets per second on otherwise idle ma-
`chines running in single-user mode.
`Significant efforts were made to avoid interference which
`might lead to media contention or retransmissions. Thetcp-
`dumpfacility was used to ensure that no other traffic was present
`on the channel. Channels known to be in use by various nearby
`WLAN installations or which demonstrated unusually noisy idle
`power consumption were also avoided. It was probably impos-
`sible to avoid all sources of RF interference, including GSM
`and DECT phones, WLAN installations and various laboratory
`equipment. However, there was little evidence of significant re-
`transmissions.
`The circuit and methodology were closely based on the ex-
`periments reported in [7]. Energy consumption was determined
`by direct measurement of the input voltage and current draw at
`the network device. The latter was obtained by inserting a small
`resistance in series with the device.
`The test circuit was built using a Sycard PCCextend 140A
`CardBus Extender [15]. The extender is like a breakout box: it is
`inserted into the PC card slot on the host and the card to be tested
`is inserted into the card connector on the extender. The
`/CE
`/CR/CR
`line
`can be isolated: 0.5Ꜳ /A6 /BE/BM/BH/B1 test resistance was inserted at this
`point1. See Figure 1 for a diagram of the current measurement
`portion of the circuit. Data measurements were made using a
`Tektronix 100MHz digital oscilloscope and 15 MHz 1X probes.
`In this configuration, the voltages of interest can be measured
`with a scope accuracy of/A6/BH/BM/BE /A0 /BI/BM/BK/B1 .2
`The input current,/CX
`/CX/D2
`/B4/D8/B5 , was determined by measuring the
`voltage/DA
`/D6
`/B4/D8/B5 across the test resistance,/CA . The input voltage,
`/DA
`/CX/D2
`/B4/D8/B5 , is affected by fluctuations in/CE
`/CR/CR
`as well as by the vary-
`ing /DA
`/D6
`/B4/D8/B5 .F o r s m a l l/CA , the latter is small compared to/CE
`/CR/CR
`(/BO /A6/BM/BD V). The input voltage is therefore approximated by a
`constant,/CE
`/CX/D2
`.
`The instantaneous power consumption is the product of the
`/BD
`The extender actually allows all control and data lines to be examined using
`a scope or logic analyzer; it is usually used for testing and debugging PC cards.
`/BE
`The scope traces reproduced in Section V were made using 50MHz Fluke
`oscilloscope and 1Ꜳ resistance. The Fluke instrument is less sensitive, but has
`a more manageable downloading facility.
`1550 IEEE INFOCOM 2001
`Authorized licensed use limited to: Austin Ginnings. Downloaded on October 08,2025 at 03:50:43 UTC from IEEE Xplore. Restrictions apply.
`SCT Exhibit 2026
`IPR2026-00098
`
`
`
`
`
`
`
`0-7803-7016-3/01/$10.00 ©2001 IEEE
`PC CARDVcc VinR
`v(t)
`SCOPE
`Fig. 1. Test circuit
`input voltage and current:
`/C8 /B4/D8/B5/BP/CE
`/CX/D2
`/DA
`/D6
`/B4/D8/B5
`/CA
`/BM
`Total energy consumed over an interval/D8
`/BC
`/BM/BM/BM /D8
`/BD
`is the integral
`of power consumption over time:
`/BX
`/D8
`/BC
`/BM/BM/BM /D8
`/BD
`/BP
`/CE
`/CX/D2
`/CA
`/CI
`/D8
`/BD
`/D8
`/BC
`/DA
`/D6
`/B4/D8/B5/CS/D8/BM
`When operating in ad hoc mode, the idle power consumption
`of the network interface is constant. (This is confirmed experi-
`mentally in Section V.)
`/C8
`/CX/CS/D0/CT
`/BP
`/CE
`/CX/D2
`/CA
`/DA
`/D6
`/CX/CS/D0/CT
`/BM
`It is much easier to measure energy consumption over an ar-
`bitrary interval than it is to consistently identify the precise du-
`ration of a transmit or receive event. Instead, we measure the
`total energy consumed during an interval that roughly brackets
`the event of interest and subtract away the constant idle power
`consumption.
`If
`/D8
`/BC
`/BM/BM/BM/D8
`/BD
`is such an interval, then the additional energy con-
`sumed processing a packet is given by:
`/CE
`/CX/D2
`/CA
`/DA
`/D6
`/B4/D8
`/BD
`/A0 /D8
`/BC
`/B5 /A0 /C8
`/CX/CS/D0/CT
`/B4/D8
`/BD
`/A0 /D8
`/BC
`/B5/BN
`where
` /DA
`/D6
`is the mean value of/DA
`/D6
`in the interval/D8
`/BC
`/BM/BM/BM/D8
`/BD
`.T h i si s
`easily computed on-board modern digital oscilloscopes3.
`This technique can be used to measure the additional energy
`consumed by a network interface as it sends, receives or discards
`broadcast and point-to-point messages and it is in this form that
`our results are presented. Coefficients for the equations defined
`by the linear formulation above can be determined by perform-
`ing these measurements for packets of various sizes and apply-
`ing linear regression.
`V. V
`ISUAL AND A NALYTIC O VERVIEW OF THE IEEE
`802.11 PROTOCOL
`Each section below describes part of the IEEE 802.11 pro-
`tocol, defines coefficients for use in a linear formulation and
`/BF
`It is not appropriate to make A/C measurements because the data traffic is a
`very low frequency signal.
`/0/0/0/1/1/1
`/0/0/0/0/1/1/1/1
`/0/0
`/0/0
`/1/1
`/1/1
`source destination
`Fig. 2. Sending a packet
`0
`250
`500
`750
`1000
`1250
`1500
`power (mW)
`50 mV/div = 50 mA/div
`500 usec/div
`idle
`send
`data
`idle
`Fig. 3. Sending 2Mbps broadcast UDP/IP traffic (256 bytes)
`presents some representative oscilloscope traces of the measure-
`ments 4. The bottom and right axes show the scope measure-
`ments (time and/DA
`/D6
`/B4/D8/B5 , respectively). The left axis shows the
`corresponding instantaneous power consumption.
`Broadcast Traffic
`Before sending broadcast traffic, the sender listens briefly to
`the channel. If no signal is detected, the message is sent. Other-
`wise, the sender must back off and retry later. As noted above,
`we do not model this case.
`The network configuration for this case is shown in Figure
`2. Using a linear equation, where
`/D1 represents the incremental
`cost and/CQ represents fixed costs:
`/BX
`/CQ/D6 /D3/CP/CS/CR/CP/D7/D8/A0/D7/CT/D2/CS
`/BP /D1
`/D7/CT/D2/CS
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D7/CT/D2/CS /B4/CQ/CR/CP/D7/D8/B5
`/BX
`/CQ/D6 /D3/CP/CS/CR/CP/D7/D8/A0/D6/CT /CR /DA
`/BP /D1
`/D6 /CT/CR/DA
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D6 /CT/CR/DA /B4/CQ/CR/CP/D7/D8/B5
`This behavior is seen in the oscilloscope traces shown in Fig-
`ures 3 and 4. The most obvious feature is the extremely high
`idle mode power consumption in ad hoc mode: actually receiv-
`ing data requires only marginally more energy than waiting for
`it.
`In Figure 3, it is easy to examine the 2Mbps rating of the
`interface. The payload is 256 (228 (data) + 8 (UDP) + 20 (IP))
`bytes; MAC overhead includes the 24-byte PLCP header (which
`is transmitted at 1Mbps) and the 20-byte MAC frame header.
`This implies a 1.3 ms data transmit time, which is about what
`we see in the trace.
`/BG
`Note that these plots are not screenshots of the scope traces. They were
`obtained by downloading waveform coordinates from a Fluke ScopeMeter and
`re-plotting them using’gnuplot’.
`1551 IEEE INFOCOM 2001
`Authorized licensed use limited to: Austin Ginnings. Downloaded on October 08,2025 at 03:50:43 UTC from IEEE Xplore. Restrictions apply.
`SCT Exhibit 2026
`IPR2026-00098
`
`
`
`
`
`
`
`0-7803-7016-3/01/$10.00 ©2001 IEEE
`0
`250
`500
`750
`1000
`1250
`1500
`power (mW)
`50 mV/div = 50 mA/div
`500 usec/div
`idle
`recv
`data
`idle
`Fig. 4. Receiving 2Mbps broadcast UDP/IP traffic (256 bytes)
`A host that is not in transmission range of the sender cannot
`detect the sender’s signal when it senses the channel and will
`proceed to send its own transmission. Both signals will be re-
`ceived at any host that is in range of both senders. Depending
`on relative signal strengths at each receiver, one or both pack-
`ets will be lost due to collision. This is known as the hidden
`terminal problem.
`Point-to-point Traffic
`While the IEEE 802.11 protocol does not support media reser-
`vation for broadcast traffic, point-to-point traffic can use colli-
`sion avoidance techniques to reduce the impact of the hidden
`terminal problem.
`Before sending a point-to-point transmission, the source
`broadcasts a RTS (request-to-send) control message, specifying
`a destination and data size (duration). The destination responds
`with a CTS (clear-to-send) message. If the source does not re-
`ceive the CTS, it may retransmit the RTS message. On receiving
`the CTS, the source sends the DATA and awaits an ACK from
`the receiver.
`Any host that hears the RTS/CTS exchange must refrain from
`transmitting for the specified duration. This “virtual carrier
`sense” reduces, but does not eliminate, the possibility of col-
`lision at the destination node.
`The network configuration and linear formulation used for
`these measurements are analogous to the ones used for broad-
`cast traffic.
`/BX
`/D4/D3/CX/D2/D8/A0/D8/D3/A0/D4/D3/CX/D2/D8/A0/D7/CT/D2/CS
`/BP /D1
`/D7/CT/D2/CS
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D7/CT/D2/CS/B4/D4/BE/D4/B5
`/BX
`/D4/D3/CX/D2/D8/A0/D8/D3/A0/D4/D3/CX/D2/D8/A0/D6 /CT/CR/DA
`/BP /D1
`/D6 /CT/CR/DA
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D6 /CT/CR/DA /B4/D4/BE/D4/B5
`/BM
`The differences between the values of/CQ
`/D7/CT/D2/CS
`and /CQ
`/D6/CT /CR /DA
`for
`broadcast and point-to-point traffic reflect the difference be-
`tween the two kinds of channel access. The incremental cost
`of sending or receiving data once the channel is acquired is ex-
`pected to be the same both broadcast and point-to-point traffic.
`The oscilloscope traces are shown in Figures 5 and 6 and the
`various elements of the media reservation protocol can be easily
`identified.
`For small packets, this media reservation exchange represents
`considerable overhead. Therefore, for packets smaller than (a
`0
`250
`500
`750
`1000
`1250
`1500
`power (mW)
`50 mV/div = 50 mA/div
`500 usec/div
`send
`RTS
`recv
`CTS
`send
`data
`recv
`ACK
`idle idle
`Fig. 5. Sending 2Mbps point-to-point UDP/IP traffic (256 bytes)
`0
`250
`500
`750
`1000
`1250
`1500
`power (mW)
`50 mV/div = 50 mA/div
`500 usec/div
`recv
`RTS
`send
`CTS
`recv
`data
`send
`ACK
`idle idle
`Fig. 6. Receiving 2Mbps point-to-point UDP/IP traffic (256 bytes)
`configurable) “RTS threshold”, the sender simply senses the
`channel prior to sending the DATA message. The receiver also
`senses the channel before sending an ACK; ACK’s take prior-
`ity over other traffic due to their shorter sensing interval. As
`with broadcast traffic, this technique is vulnerable to “hidden
`terminal” collision, with the risk increasing as the message size
`increases. The linear formulation is:
`/BX
`/D4/D3/CX/D2/D8/A0/D8/D3/A0/D4/D3/CX/D2/D8/A0/D7/CT/D2/CS
`/BP /D1
`/D7/CT/D2/CS
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D7/CT/D2/CS/B4/BO/D8/CW/D6 /CT/D7/CW/B5
`/BX
`/D4/D3/CX/D2/D8/A0/D8/D3/A0/D4/D3/CX/D2/D8/A0/D6 /CT/CR/DA
`/BP /D1
`/D6/CT /CR /DA
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D6 /CT/CR/DA /B4/BO/D8/CW/D6 /CT/D7/CW/B5
`/BM
`In this case, the/CQ values are expected to have some interme-
`diate value between broadcast and point-to-point traffic. The
`overall effectiveness of this technique, taking into account the
`increased likelihood of collision and retransmission is outside
`the scope of this work.
`Discard Traffic
`As a consequence of operating in ad hoc mode, a network
`interface overhears all traffic sent by nearby hosts. It is there-
`fore important to consider not only the energy consumption of
`sending and receiving traffic, but also the energy consumed by
`an interface when it processes point-to-point traffic that it will
`discard after determining that it is not the intended destination.
`This case is interesting not only because it is the most amenable
`1552 IEEE INFOCOM 2001
`Authorized licensed use limited to: Austin Ginnings. Downloaded on October 08,2025 at 03:50:43 UTC from IEEE Xplore. Restrictions apply.
`SCT Exhibit 2026
`IPR2026-00098
`
`
`
`
`
`
`
`0-7803-7016-3/01/$10.00 ©2001 IEEE
`/0/0/0
`/0/0/0
`/1/1/1
`/1/1/1
`/0/0
`/0/0
`/1/1
`/1/1
`/0
`/0
`/0
`/1
`/1
`/1
`source
`discard destination
`Fig. 7. Discarding traffic (source and destination).
`/0/0
`/0/0
`/1/1
`/1/1/0/0/0
`/0/0/0
`/1/1/1
`/1/1/1
`/0/0
`/0/0
`/0/0
`/1/1
`/1/1
`/1/1
`destinationsourcediscard
`Fig. 8. Discarding traffic (source only).
`to energy conserving strategies, but also because it represents
`the “common case” – most of the traffic is for somebody else.
`The non-destination host’s location with respect to both
`source and destination determines what part of the IEEE 802.11
`protocol it overhears. Figure 7 shows the case of a discarding
`host in range of both the source and destination. The case of
`a non-destination node in range of the sender only is shown in
`Figure 8. A suitable machine configuration was found by walk-
`ing around with laptops running’ping’tests.
`In the equations,
`/CB and /BW represent the case of being within
`transmission range of the sender and destination, respectively).
`/BX /BP /D1
`/CS/CX/D7/CR
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D2/D3/D2/A0/CS/CT/D7/D8/B4/CB/BN/BW /B5
`/BX /BP /D1
`/CS/CX/D7/CR
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D2/D3/D2/A0/CS/CT/D7/D8/B4/CB/BN/BI/BW /B5
`/BM
`The values of/CQ
`/D2/D3/D2/A0/CS/CT/D7/D8
`reflect the cost of processing the
`MAC protocol messages in order to avoid contention with a
`transmission that is intended for another destination. For non-
`destination hosts in range of the sender, the sign of
`/D1
`/CS/CX/D7/CR
`indi-
`cates what kind of energy conservation strategy is used by the
`network interface. If/D1
`/CS/CX/D7/CR
`/BQ /BC , in particular, if/D1
`/CS/CX/D7/CR
`/BP /D1
`/D6 /CT/CR/DA
`,
`then the non-destination host effectively receives the traffic be-
`fore discarding it. If/D1
`/CS/CX/D7/CR
`/BP /BC , then the non-destination host
`maintains the network interface in the idle state during the data
`transmission. If
`/D1
`/CS/CX/D7/CR
`/BO /BC , then the interface must employ
`some energy-conserving strategy based on the presence of un-
`interesting data on the media. (Recall that all energy costs are
`defined relative to the idle power consumption of the interface.)
`The oscilloscope trace in Figure 9 shows the case of a
`non-destination host in range of both the source and destina-
`tion. While data is being transmitted, the Lucent interface uses
`slightly less power than it does in idle mode. While the source is
`transmitting data, the non-destination host can neither send nor
`receive
`5 point-to-point traffic because the source and destination
`have reserved the media via the RTS/CTS exchange. The non-
`destination host could, in principle, receive a broadcast message
`sent by a fourth host that is out of range of both the source and
`/BH
`It can’t send the CTS/ACK.
`0
`250
`500
`750
`1000
`1250
`1500
`power (mW)
`50 mV/div = 50 mA/div
`500 usec/div
`idle
`overhear
`low power (data)
`overhear
`idle
`Fig. 9. Non-destination (2Mbps) host in range of both sender and receiver
`discarding 2Mbps point-to-point UDP/IP traffic (256 bytes)
`destination. But because the non-destination host is in range of
`both senders, the packet has a high probability of being unintel-
`ligible due to collision. The non-destination host therefore loses
`little or no traffic by entering this reduced energy mode. The
`amount of energy that can be saved using this technique is de-
`pendent on the amount of time that the data transmission takes,
`as well as the amount of time and energy needed to return to the
`idle state.
`The network configuration for case of a non-destination host
`in range of destination only is analogous that of the non-
`destination host in range of the sender (Figure 8). This is ex-
`pressed as:
`/BX /BP /D1
`/D2/D3/D2/CT
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D2/D3/D2/A0/CS/CT/D7/D8/B4/BI/CB/BN/BW /B5
`The coefficient/D1
`/D2/D3/D2/CT
`refers to the network interface’s behav-
`ior while data is being transmitted, even though it is out of range
`of the sender. In principle,
`/D1
`/D2/D3/D2/CT
`need not be/BC , as the interface
`is aware of the data transmission via the CTS message.
`Promiscuous Mode Operation
`A non-destination host operating its network interface in
`promiscuous mode eavesdrops on all point-to-point data traffic
`that it overhears. The coefficients are therefore a combination of
`the receive and discard cases. The data is treated as in the case
`of point-to-point receive, but the control traffic is treated as in
`the case of point-to-point discard.
`In range of both sender and destination:
`/BX /BP /D1
`/D6 /CT/CR/DA
`/A2 /D7/CX/DE /CT /B7 /CQ
`/D2/D3/D2/A0/CS/CT/D7/D8/B4/CB/BN/BW /B5
`In range of sender only:
`/BX /BP /D1
`/D6 /CT/CR/DA
`/A2 /D7