on the energy efficiency of wireless...

On the Energy Efficiency of Wireless Transceivers

Andrew Y. Wang1 and Charles G. Sodini21Analog Devices, Wilmington, MA 01887

2Massachusetts Institute of Technology, Cambridge, MA 02139Email: [email protected]

Abstract- Battery life-time is an important performance met-ric for many wireless networks and there is a growing need toimprove the life-time of these networks. However, system analysisis complicated by the fact that the transceiver electronics energycost is non-negligible and often has a major impact on the batterylife. This paper develops a simple metric to evaluate the energycost of both communication protocols and circuit electronics. Thismetric treats the electronics energy as an overhead cost on topof the EbINO requirement. This enables the joint optimizationof system and circuit parameters for energy limited wirelesstransceivers with arbitrary communication protocols.

I. INTRODUCTIONOver the past two decades the wireless communications

market has experienced explosive growth, and wireless tech-nology has become an integrated part of our lives providingservices from telephony to data transfer to sensing and mon-itoring applications. While awe-inspiring progress has beenmade to make wireless a ubiquitous technology, challengingissues remain in capacity, reliability, battery life, and cost. Aparticularly important performance criteria for mobile devicesis battery life. While some of these devices can be rechargedwithin a few days or a few weeks, other devices, such as thoseused in industrial wireless sensing applications, require batterylife of one year or better [1]. While long battery life is a plusfor devices such as cell phones and wireless LAN cards, it isa must for large-scale sensor networks due to the difficulty ofreplacing or recharging batteries for hundreds of sensor nodes.

For these reasons, there is a growing interest in the study ofnetwork life-time of wireless sensor networks. This problem isusually tackled from the network layer down. At the networklayer, better flow-control algorithms, such as task sharing [2]and multi-hopping [3], have been proposed to increase theaverage network life-time. At the MAC layer, various multi-access techniques have been evaluated in terms of energy cost[4], [5]. At the physical layer, modulation techniques havebeen investigated in terms of energy efficiency [1], [6].

Ultimately, it is the energy drawn from the battery thatdetermines the battery life. This energy is equal to the integralof the transceiver power consumption over time. Therefore,both the instantaneous transceiver power consumption andthe temporal behavior of the transceiver affect the batterylife. For instance, in sensor applications the RF transceiveris often operated in burst mode with duty cycle of 1% orless, which means the transceiver is shut off most of thetime. This transceiver would have a much longer battery lifeas compared to one that has lower power consumption butoperates continuously.

Much of the difficulty in analyzing battery life comes frommodelling the transceiver power consumption as a functionof time. An inaccurate model would lead to an incorrectconclusion. For example, a popular model assumes that thetransceiver power consumption is proportional to the RFtransmit power, which in turn scales with the data rate (theRF transmit power needs to be increased to maintain thesame transmission distance and BER at higher data rate). Thismodel leads to the conclusion that to increase the battery life,the data rate should be reduced. Fig. 1 shows the transmitterpower consumption as a function of data rate for a short-rangeRF transceiver employing binary FSK [7]. It is clear that thetransmitter power consumption is not a function of the datarate until the data rate exceeds 100kbps. As explained in [7],data rate does not scale with power consumption due to thefixed power consumption cost of the transmitter electronics.Since reducing the data rate increases the transmission time, orequivalently, the duty cycle, the battery life is actually reduced.

Data Rate (bps)

Fig.rate.

1. Transmitter power consumption does not scale linearly with data

For short-range RF transceivers operating in the Giga-Hertzcarrier frequency range, the circuit electronics dominate thepower consumption. Therefore it is important to model thecircuit power in battery life analysis. It is equally importantto model the communication protocols and parameters such asmodulation, data rate, and channel condition, as they determinethe RF transmit power, the transceiver operation time, andthe transceiver architecture, which in turn affect the circuitelectronics power consumption. This interplay between the

37831-4244-0355-3/06/$20.00 (c) 2006 IEEE

This full textpaper was peer reviewed at the direction ofIEEE Communications Society subject matter expertsfor publication in the IEEE ICC 2006proceedings.

communication protocols and the transceiver electronics iscomplicated and has rarely been investigated. However, byworking on the joint optimization, it is found that the batterylife can be improved by nearly an order of magnitude [7].This paper goes further and attempts to answer the morefundamental question: given choices of communication pro-tocols and parameters, how do we compare the energy cost ofthe transceivers designed for these protocols? In this paper asimple figure of merit is developed to enable this comparison.

This paper is organized as follows. Section II introduces theconcept of transceiver energy efficiency. Section III developsthe transceiver energy figure of merit and analyzes the energycost of a binary PSK transceiver and a binary non-coherentFSK transceiver. Section IV summarizes the results.

II. TRANSCEIVER ENERGY EFFICIENCY

Fig. 2 shows a simplified block diagram of a generic RFtransceiver. There are four major circuit building blocks withinthe transceiver. The transmit block (TX) is responsible formodulation and up-conversion (translating the basband signalto RF), the receive block (RX) is for down-conversion anddemodulation, the local oscillator block (LO) generates therequired carrier frequency, and the power amplifier block (PA)amplifies the signal to produce the required RF transmit powerPT.

Fig. 2. Generic transceiver architecture.

The power consumption in the receive path is

Prxr = PLO + PRX (1)

where PLO is the power consumption of the local oscillatorLO and PRX is the power consumption of the receive blockRX. Similarly, the power consumption in the transmit path is

Ptxr = PLO + PTX + PPA (2)Where PTX is the power consumption of the transmit blockTX and PPA is the power consumption of the power amplifierPA. The PA power consumption PPA depends on the PAefficiency, the link-budget, and the sensitivity requirement.Specifically,

PPA1

PTTiPA1 (4w)2dn (Eb \F.R.No.0

TiPA GTGRA2 No

(3)

(4)

where PT is the RF transmit power, '1PA is the PA efficiency, dis the transmit distance, n is the path loss exponent, GT andGR are the transmitter and receiver antenna gains, A is thewavelength, F is the receiver front-end noise factor, R is thedata rate, No is the available thermal noise power spectrumdensity, and EbINO is the required SNR per bit for a givenBER. Note that although Eb/NO is independent of data rate[8], PPA depends on the data rate linearly.From above, it is clear that PPA captures the effect of the

communication protocols and parameters such as Eb/No, datarate, and channel condition. The circuit blocks LO, TX, andRX are essential building blocks of a RF transceiver, but theirpower consumptions are often ignored in system analysis. Interms of energy cost, however, these blocks are non-negligibleas their power consumptions are significant.The power consumption of a RF transceiver can vary sig-

nificantly depending on the application. Table I shows typicalpower consumption numbers for GSM [9], 802.11b [10], andBluetooth transceivers [11]. PA efficiency can vary quite a bitdepending on PA type, output power, technology, and design.A fixed realizable PA efficiency of 40% is assumed here forthe ease of comparison. The transceiver energy efficiency T/pwill be explained shortly.

I[IGSM 802.1 lb BBluetoothPRX + PLO (mW) 240 60 30PTX + PLO (mW) 360 100 12PT (mW) 1000 100 1PPA (mW) 2500 250 2.5r1P 32% 24% 2%

TABLE IPOWER CONSUMPTION OF SHORT AND LONG RANGE TRANSCEIVERS.

The GSM transceiver has a transmission range greater than1 kilo-meter and has the most stringent system specifications.Its transceiver electronics power and RF transmit power arethe highest. The 802.1 lb transceiver has an intermediate rangeon the order of 10's of meters. The Bluetooth transceiver hasthe shortest transmission range and the most relaxed systemspecifications. Its power consumption is the lowest.The RF transmit power of the Bluetooth transceiver is three

orders of magnitude lower than that of the GSM transceiverdue to the reduction in transmission range. However, thereduction in transceiver electronics power is only an order ofmagnitude. What this means is that in long-range transceiversthe power is dominated by the RF transmit power, but in short-range transceivers the power is dominated by the electronicspower.

The energy overhead due to the transceiver electronics isbest characterized by the transceiver energy efficiency,whichis defined as

aT PTACR Prxr + AT Ptxr (5)

The terms aT and aR are the transmitter and receiver activity

3784


factors defined as

CTttx

ttx + trx

trx

(6)

(7)ttx + trx

where trx is the average receive time and ttx is the aver-

age transmit time in one receive-transmit cycle. This modeltakes into account systems that have asymmetric transmit andreceive requirements. Since the operation time is inherentlydefined in (5), qp is the energy efficiency and not the power

efficiency.When the transceiver operates equally between the transmit

and the receive mode, (5) simplifies to

PTTP = Prxr + Ptxr (8)

In this scenario, all the terms are power consumptions, so theenergy efficiency and the power efficiency are equivalent.The transceiver energy efficiency is a measure of the

communication cost over the electronics cost. When doesthe transceiver energy efficiency reach 100%? It is when allthe energy is used for RF transmission and the transceiverelectronics energy cost is negligible. Unfortunately, this isnever the case. In reality, the best transceiver energy efficiencyis achieved when the PA dominates the transceiver power

consumption (i.e. high-data-rate long-range systems). In thiscase, T/p can be approximated as

PTTIP PPA TIPA

In this scenario, the transceiver power consumption can bereduced through either better circuit design that improves thePA efficiency or more efficient system techniques that reducethe RF transmit power.

When the PA power consumption is small relative to thetransceiver electronics (i.e. low-data-rate short-range systems),T/p is approximately

PTTIP r(PRX + PLO) + (PTX + PLO)

to transceiver electronics power improves. Another viewpointis that we are increasing the data rate of the system with-out incurring a corresponding cost in the transceiver power

consumption. In contrast, a PA-power dominated system, thedata rate and the power consumption would increase at thesame rate and produce no change in the transceiver energy

efficiency.

0

0)

.>

U

0,

1k 10k 100kData Rate (bps)

iM 1oM

Fig. 3. Transceiver energy efficiency improves with data rate.

III. TRANSCEIVER ENERGY FIGURE OF MERIT

Although the transceiver energy efficiency is a good metricin evaluating the electronics overhead energy cost relative tothe RF transmit energy, it does not allow the comparison ofdifferent communication protocols. This is easily seen in thecase where the PA power consumption dominates. Here thetransceiver energy efficiency approaches the PA efficiency,which is a circuit parameter, thus hiding the effect of com-

munication protocols.To enable the comparison, we introduce the concept of

transceiver energy figure of merit (EFOM), which is definedas

(10)

In this scenario, the transceiver energy cost is dominated bythe electronics. PA design and efficient system techniques donot affect the transceiver energy cost. Instead, the focus shouldbe placed upon improving the ratio of PT versus PRX, PLO,and PTX.

For the systems shown in Table I, the maximum achievabletransceiver energy efficiency is limited by the PA at 40%. BothGSM and 802.11b transceivers have respectable efficiencies,but the Bluetooth transceiver has an efficiency of only 2%,which means that transceiver electronics power dominates.

Fig. 3 plots the transceiver energy efficiency as a function ofdata rate for the microsensor transceiver discussed in SectionI. As mentioned previously, the RF transmit power scaleslinearly with the data rate (to keep the same distance andBER), the LO, RX, and TX powers are fixed cost. Therefore,by increasing the data rate, the ratio of RF transmit power

"YP

Eb

( NT,IPtp

(1 1)

In simplest terms, EFOM can be interpreted as EbINOplus the overhead cost due to the transceiver electronics.By examining (11), (5), and (4), it is clear that EFOM isproportional to the transceiver energy cost CR -Prxr+T -Ptxr-Thus the goal is to design communication protocols andtransceiver hardware jointly to achieve the smallest EFOM.An attractive feature of EFOM is that it separates out the

problem of system design and hardware design. The systemdesigner can work out the Eb/NO requirement for a particularset of communication protocols. The hardware designer can

then design the transceiver and provide an estimate of thetransceiver energy efficiency. Putting the two together yieldsa measure of the transceiver energy figure of merit. In the restof this section, we will show an application where this conceptis applied.

3785


...

15,

~4 L~

(a) Direct modulation of FSK Signaling.

(a) I/Q transmitter.

_LNF

(b) I/Q receiver.

Fig. 4. Generic I/Q transmitter and receiver architecture.

The problem is choosing between a binary PSK transceiverand a binary non-coherent FSK transceiver for a short-rangeenergy-limited microsensor application. The required Eb/Nois 15dB for the PSK and 21dB for the FSK [8], [12]. Using(4), the required RF transmit power can be determined tobe 0.25mW for the PSK transmitter and 1mW for the FSKtransmitter, both at 1Mbps [7]. For a different data rate, thetransmit power needs to be scaled linearly to keep the sametransmission distance and BER.

Fig. 4 shows the architecture of a generic I/Q transceiverthat can be used to generate binary PSK constellation. Inthe transmitter, the I and Q components are generated withinthe modulator (Mod). They go through the digital-to-analogconverter (DAC), are up-converted to the carrier frequency bythe RF mixers, and are combined and sent out via the PA.Since a linear PA is required for PSK to adequately suppressout-of-band power, the PA efficiency is assumed to be 40%. Inthe receiver, the received signal is first amplified by the low-noise amplifier (LNA) and is down-converted by the mixersinto I and Q components. The intermediate frequency (IF)signal is then digitized through the analog-to-digital converters(ADC) and demodulated.

Although FSK can be implemented using the same I/Qarchitecture, the power consumption can be reduced by usingthe architecture shown in Fig. 5. In the transmitter, the outputis produced by directly controlling the frequency synthesizerinside the LO, which generates one of the two requiredfrequency tones [13]. Since FSK is a non-linear modulation

(b) FSK receiver with frequency discriminator.

Fig. 5. Low power non-coherent FSK transceiver.

technique, a non-linear PA with efficiency as high as 70%can be used. In the receiver, the received signal is first down-converted to an IF frequency, then the signal energy near thetwo frequencies at wo and w1 are compared.

Table II shows the power consumption of the circuit buildingblocks. The PSK transceiver is more power hungry due to theextra mixers required. It is assumed that the power consump-tions of TX, RX, and LO are constant (i.e., fixed cost), andthat the power consumption of the PA varies as in (4) [7].

| |2-PSK 2-FSKTX 7mW |mWRX 18mW 15mWLO 11mW 11mWTiPA 40% 70%PA PT /0.4 PT /0.7

TABLE IIPOWER CONSUMPTION OF TRANSCEIVER CIRCUIT BLOCKS

Fig. 6 plots the transceiver power consumption (sum oftransmitter and receiver power consumption) of FSK and PSKtransceivers as a function of data rate. At low data rate, theRF transmit power is negligible, so the transceiver powerconsumption is dominated by the electronics. In this case thePSK transceiver is more power hungry. At high data rate,the PA power dominates, which is a function of both theRF transmit power and the PA efficiency. PSK requires lowerRF transmit power (-6dB) but also has a lower PA efficiency(0.4/0.7=-2.4dB), so overall its PA power consumption is3.6dB lower than that of the FSK transceiver.

Fig. 7 shows the transceiver energy efficiency. Note thatthe FSK transceiver is more efficient at all data rate. This isbecause FSK requires more RF transmit power and less fixedelectronics cost as compared to PSK. So from the hardware

786


I'i0-F.-;:M

200336

34

32

m30

26

24

22

10Data Rate (Mbps)

100

Fig. 6. Power consumption of PSK and FSK transceivers.

point of view the FSK transceiver is more efficient. HoNthe transceiver energy efficiency does not reflect the facthe extra RF transmit power for FSK is due to itsmodulation efficiency (i.e. EbINo), so from a communi(point of view FSK is less efficient. To evaluate the comeffect of the communication protocol and the RF electrEFOM needs to be examined.

S 50-

0

<. 40

30

.> 20

(D10

1 10Data Rate (Mbps)

10

Fig. 7. Energy efficiency of PSK and FSK transceivers.

Fig. 8 shows the EFOM of the transceivers. We se

FSK and PSK cross over each other and form two distiregions. At low data rate, the transceiver energy efficis low, so EFOM is dominated by the transceiver electienergy cost. FSK has lower electronics cost and so

a lower EFOM. At high data rate, the transceiver e

efficiency is high, so EFOM is dominated by Eb/NO.has a more efficient Eb/No and so it has a lower ETherefore, for data rate lower than 10Mbps, binarycoherent FSK transceiver has less overall energy cost, ai

data rate greater than 10Mbps, binary PSK transceiver haoverall energy cost.

Interestingly, Fig. 8 shows that most of the gains in E

Fig. 8. FSK and PSK transceiver EFOM comparison.

wever, comes not from choosing between FSK and PSK but from:t that increasing the data rate. Although Eb/No does not vary withworse the data rate, the overhead cost due to the transceiver does.cation This overhead cost is quite significant - it is approximately[bined 16dB at 1Mbps and 7dB at 10Mbps. Thus by going fromonics, 1Mbps to 10Mbps, the overhead cost from the transceiver

electronics is reduced by 9dB.The above analysis works well for data rates up to a few 10's

of Mega-bits per second. For higher data rate, the assumptionthat LO, RX, and TX have constant power consumption no

longer hold. In fact, the power consumption of these circuitblocks will eventually increase with data rate. More carefulmodelling is needed for data rate beyond a few 10's of Mega-bits per second.

IV. CONCLUSION

This paper presents two new concepts. The transceiverenergy efficiency is a measure of the ratio between the commu-nication cost and the electronics cost. It is shown that this ratiocan be improved by increasing the data rate. The transceiver

o energy figure of merit, or EFOM, is a figure of merit that addsthe transceiver electronics energy cost as an overhead cost on

top of the Eb/No requirement. EFOM provides a means toallow the designer to evaluate the combined effectiveness of

~ethat system, architecture, and circuit techniques in the design of

nctiveenergy-limited RF transceivers.

viency V. ACKNOWLEDGEMENT

ronicsit has This work is funded, in part, by the MIT Center fornergy Integrated Circuits and Systems.PSKFOM. REFERENCES

nor

nd fis les

,FO

n- [1] A. Wang, S. Cho, C. Sodini, and A. Chandrakasan, "Energy efficientDr modulation and MAC for asymmetric RF microsensor systems," in Proc.SS Int. Symp. Low Power Design (ISLPED), 2001, pp. 106-1 1.

[2] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, "Anapplication-specific protocol architecture for wireless microsensor net-

\4 works," IEEE Trans. Wireless Commun., pp. 660-670, October 2002.

3787


¢ 180E

r 160.2

E 140n

0' 120

Io0

0, 80(-

0a 60

40

Eb/No dominated%&electronics dominated

2010

Data Rate (Mbps)100

(I

3-m

.n,

snJ.

1 '1

[3] J. Rabaey, J. Ammer, J. L. da Silva Jr., and D. Patel, "PicoRadio forwireless sensor networks: the next challenge in ultra-low power design,"in IEEE Int. Solid-State Circuits Conf: (ISSCC) Dig. Tech. Papers, 2002,pp. 200-201.

[4] J.-C. Chen, K. Sivalingam, P. Agrawal, and S. Kishore, "A comparisonof MAC protocols for wireless local networks based on battery powerconsumption," in Proc. Conf IEEE Comput. Commun. Societies, 1998,pp. 150-157.

[5] S. Cho and A. Chandrakasan, "Energy efficient protocols for low dutycycle wireless microsensor networks," in Proc. Int. Conf Acoustics,Speech, and Signal Processing, 2001, pp. 2041-2044.

[6] S. Cui, A. Goldsmith, and A. Bahai, "Energy-constrained modulationoptimization for coded systems," in Proc. IEEE Global Telecomm. Conf:,2003, pp. 372-376.

[7] A. Wang and C. Sodini, "A simple energy model for wireless microsen-sor transceivers," in Proc. IEEE Global Telecomm. Conf:, November2005, pp. 3205-3209.

[8] J. Proakis, Digital Communications. New York: McGraw-Hill, 1995.[9] A. Molnar and et al., "A single-chip quad-band (850/900/1800/1900

MHz) direct conversion GSM/GPRS RF transceiver with integratedVCOs and fractional-N synthesizer," in IEEE Int. Solid-State CircuitsConf (ISSCC) Dig. Tech. Papers, February 2002, pp. 232-233.

[10] R. Meyer, W. Mack, and J. Hageraats, "A 2.5 GHz BiCMOS transceiverfor wireless LAN," in IEEE Int. Solid-State Circuits Conf: (ISSCC) Dig.Tech. Papers, February 1997, pp. 310-311.

[11] N. Filiol, N. Birkett, J. Cherry, F. Balteanu, G. Gojocaru, A. Namdar,T. Pamir, K. Sheikh, G. Glandon, D. Payer, A. Swaminathan, R. Forbes,T. Riley, S. Alinoor, E. Macrobbie, M. Cloutier, S. Pipilos, and T. Vare-las, "A 22mW Bluetooth RF transeiver with direct RF modulation andon-chip IF filtering," in IEEE Int. Solid-State Circuits Conf (ISSCC)Dig. Tech. Papers, February 2001, pp. 202-203.

[12] J. Haartsen and S. Mattisson, "Bluetooth - a new low-power radiointerface providing short-range connectivity," in Proc. IEEE, 2000, pp.1651-1661.

[13] S.-H. Cho and A. Chandrakasan, "A 6.5 GHz CMOS FSK modulator forwireless sensor applications," in IEEE Symp. VLSI Circuits Dig. Tech.Papers, June 2002, pp. 182-185.

3788


on the energy efficiency of wireless...

Documents