A 96 Mbit/sec, 15.5 pJ/bit 16-QAM Modulator forUHF Backscatter Communication

Stewart J. Thomas and Matthew S. ReynoldsDepartment of Electrical and Computer Engineering, Duke University, Durham, NC

Abstract—We describe a low power vector backscatter modu-lator capable of transmitting 16-QAM at a rate of 96 Mbps whileconsuming only 1.49 mW (15.5 pJ/bit). While designed arounda center frequency of 915 MHz, the modulator is capable ofoperation over the worldwide 868 – 950 MHz UHF band. Wepresent experimental results from the modulator operating in 4-QAM/4-PSK, 4-PAM, and 16-QAM modes. Achieved data ratesare comparable to WiFi (IEEE 802.11) with a measured tag-sidepower consumption over 50 times lower than a WiFi chipset.Potential applications for low power, high bit rate modulatorsinclude biotelemetry, high-bandwidth data transfer from cameratags or audio tags, uplink from mass storage tags, and exchangeof large amounts of encryption or authentication data. Given a+36 dBm EIRP transmitter operating at 915 MHz, the semi-passive (battery-assisted) prototype tag is return link limitedand has a theoretical maximum operating range of 17.01 m at96 Mbps or 21.25 m at 40 Mbps.

I. INTRODUCTION

Ultra high frequency (UHF) radio frequency identification(RFID) tags achieve low-power wireless communication byrepartitioning the complexity and power of the radio link.Instead of generating and modulating a UHF carrier, UHFRFID tags reduce circuit complexity and power consumptionby backscattering a carrier generated by a remote reader.The scattered field is demodulated to reproduce the basebandwaveform. In a backscatter system, no transistors are biased forUHF operation. Instead, power is conserved by only switchingtag transistors at baseband data rates. For simple ID-only tagsemploying binary ASK/PSK communication, tag operatingpower can be less than 2.25 µW [1].

As shown in Fig. 1, passive UHF RFID tags based on globalUHF RFID standards [2] typically have sub 10 m operatingrange with data rates up to 640 kbps. While UHF RFIDoperating range can be increased using semi-passive (battery-assisted) technology, data rates remain low when compared toIEEE 802.11 (WiFi) or UWB [3]. The high data rate and longoperating range of these technologies come at an increase inpower consumption with reported values as high as ≈ 1 Wfor some WiFi implementations [4]. This is prohibitivelyhigh for low-power, remote or embedded sensor applicationsseeking high-bandwidth wireless communication. In contrast,the backscatter modulator described in this paper is capableof similar uplink data rates with measured power consumptionover 50 times less than a typical WiFi chipset.

In a typical passive tag employing binary backscatter mod-ulation, the circuit input impedance is switched between amatched, absorptive state designed to maximize RF-to-DC

Fig. 1. Comparison of wireless technology bit rates and communicationrange.

(a) (b)

(c)

Fig. 2. Block diagram of (a) traditional binary backscatter modulator, (b) PINdiode based vector modulator [5], and (c) high bit rate 16-QAM backscattermodulator presented in this work.

power conversion efficiency and a deliberate impedance mis-match causing an amplitude or phase change in the scatteredelectric field. Existing RFID standards [2] permit the circuitdesigner to choose between binary amplitude shift keying(BASK) or binary phase shift keying (BPSK). Historically the

Fig. 3. Photo of the high-bit rate 16-QAM tag. The PCB board measures6.8 x 4.9 cm.

Ideal Measured

Fig. 4. Comparison between ideal and measured impedance values for a16-QAM backscatter modulator plotted on a Z0 = 50Ω Smith chart.

modulation is measured in terms of the differential radar crosssection (∆RCS), a scalar value representing the distance be-tween input impedance states [6]. A typical binary backscattermodulator is shown in Fig. 2(a).

Prior work has shown that QAM backscatter can be pro-duced with careful control of the tag’s input impedance [5],[7], [8]. For example, the circuit shown in Fig. 2(b) producesIQ backscatter modulation using bias currents to change theimpedance of two PIN diodes connected to a quadraturehybrid [5]. Because this circuit requires a high bias powerconsumption of 80 mW and two DACs which introducesignificant power consumption and data rate limitations, it isnot compatible with low-power sensor or wireless applications.

We expand on previous work introducing a 4-QAM/4-PSK modulator [7], [8] and present a 16-QAM backscattermodulator shown schematically in Fig. 2(c). This modulatorincludes a 16-to-1 Mux to modulate the circuit impedancebetween 16 impedance states. It can be thought of as an“impedance DAC” that converts a 4-bit digital input to aspecified modulating impedance. The assembled tag, picturedin Fig. 3, is implemented as a semi-passive device using aCR2032 3.0 V coin cell battery for DC power and is capableof transmitting 96 Mbps with the modulator consuming only1.49 mW (15.5 pJ/bit). Using a +36 dBm EIRP transmitteroperating at 915 MHz, the tag has a theoretical maximumoperating range of 17.01 m.

TABLE IΓ∗ VALUES FOR 16-QAM MODULATOR WITH RC COMPONENTS,

Za = 50 Ω, α = 1

Ideal Γ∗ Measured Γ∗

Z1 -0.15 -0.19 + j0.03Z2 -0.45 -0.48 + j0.03Z3 -0.15 + j0.3 -0.16 + j0.3Z4 -0.45 + j0.3 -0.46 + j0.29Z5 0.45 0.44 + j0.04Z6 0.15 0.15 + j0.02Z7 0.45 + j0.3 0.44 + j0.33Z8 0.15 + j0.3 0.15 + j0.29Z9 0.45 + j0.6 0.42 + j0.63Z10 0.15 + j0.6 0.09 + j0.6Z11 0.45 + j0.89 0.44 + j0.9Z12 0.15 + j0.89 0.18 + j0.89Z13 -0.15 + j0.6 -0.19 + j0.62Z14 -0.45 + j0.6 -0.41 + j0.59Z15 -0.15 + j0.89 -0.14 + j0.9Z16 -0.45 + j0.89 -0.41 + j0.94Simulated EVM 868 MHz 2.96%

915 MHz 0%950 MHz 2.09%

Measured EVM 915 MHz 3.62%

II. QAM BACKSCATTER MODULATION

Hansen [9] previously analyzed a special case of radarscattering where the scatterer is an antenna connected toa load. Relating Hansen’s scattering results to backscattermodulation context, a set of equations for QAM backscattermodulation design is developed.

The reflected field from a tag can be decomposed intostructural scattering and load dependent scattering terms

Escat(ZL) = Escat(Z∗a) + Γ∗

EantI∗m

Iant(1)

where Γ∗ is the conjugate match reflection coefficient

Γ∗(ZL) =Z∗a − ZL

Za + ZL, (2)

I∗m is the antenna current when the load is conjugate matchedand Eant is the radiated field when the antenna is driven bycurrent Iant.

By substituting two different complex-valued reflection co-efficients from two complex-valued load impedances into (1),we find the complex-valued differential backscatter E-fieldthat is observed for binary backscatter

∆Escat = (Γ∗1 − Γ∗2)EantI

∗m

Iant(3)

which, when demodulated, may fall in any quadrant of theIQ plane. Careful load selection can result in the scattered E-field exhibiting amplitude modulation, phase modulation, orboth simultaneously and independently. This prediction is incontrast to scalar differential RCS analysis which can onlyrepresent the magnitude of change.

In the simplest implementation, M -ary QAM modulation isimplemented with M distinct load impedances switched acrossthe terminals of the tag’s antenna as shown in Fig. 2(c).

Fig. 5. Bistatic test setup for the high data rate QAM modulator

A. Received Signals in the Presence of Multipath and TagMotion

The E-field at the reader’s receiving antenna is composedof three components

Ercv = Eenv + E′scat(ZL) +N (4)

where distance-induced attenuation and phase shift of thescattered field from the tag is E′scat, Eenv represents scatteringof the transmitted field from all stationary objects in theenvironment, and N represents all environmental sources ofnoise including scattering of the transmitted field due toreflective objects in the environment. After expanding thisexpression into its constituent parts we note that the receivedE-field

Ercv = Eenv + E′scat(Z∗a) + Γ∗

E′antI′∗m

I ′ant+N (5)

can be treated as a sum of vector components composedof contributions from environmental multipath (Eenv), an-tenna structural scattering (E′scat(Z

∗a)), load-dependent antenna

mode scattering(

Γ∗E′antI

′∗m

I′ant

)and a noise vector (N ).

Since the data rate of existing UHF RFID tags is typicallymuch greater than the rate of tag physical motion or nearbyobject motion and load-dependent vector

(Γ∗

E′antI′∗m

I′ant

)changes

with the data rate, the contributions from environmental scat-tering and structural scattering (Eenv + E′scat(Z

∗a)) may be

assumed stationary and lumped into a constant. The receivedfield (5) then reduces to

Ercv = Estationary + Emodulated +N, (6)

the sum of a stationary field and a time-varying, modulatedfield. The stationary field component is mixed down to DCby the homodyne receiver and rejected by the receiver’s AC-coupling or DC offset compensation loop leaving only thedesired modulated field. Homodyne detection and DC rejectiondemodulates the load dependent scattering component to yieldthe desired data.

Movement of nearby scatterers results in a change in the en-vironmental multipath contribution, (Eenv), causing translationof the tag’s total scattered field within the IQ plane. Chang-ing the tag-to-reader separation distance causes scaling androtation of the tag’s total scattered field, (Eenv + E′scat(Z

∗a)),

due to a change in path loss and distance-dependent phase

shift. In all cases, a fixed phase relationship exists between thestructural scattering and antenna-mode scattering preservingthe transmitted constellation.

III. 16-QAM TAG MODULATOR DESIGN

Based on the observation from (3) that a loaded antennacan yield a scattered E-field component in any quadrant ofthe complex plane, a set of impedance states are found forproducing 16-QAM backscatter.

16-QAM impedance values are chosen by first writing eachsymbol of the desired IQ constellation in the form

Si = xi + jyi (7)

where xi represents the in-phase component and yi representsthe quadrature component of the i-th symbol. In order toproduce impedance values realizable with passive components,all reflection coefficients are confined within a circle aboutthe conjugate match with magnitude ≤ 1. These reflectioncoefficients are then scaled by a constant 0 < α ≤ 1.

Γ∗i = α · Simax |S|

(8)

Values of α closer to 1 produce constellations with greater av-erage backscatter signal strength by reflecting higher amountsof incident RF power toward the reader. For semi-passive(battery-assisted) tags, large α is desired for yielding long readrange by maximizing the backscatter signal strength. Passivetags, which harvest energy from incident E-fields for circuitDC power, are generally forward-link limited. Setting α 1allows a majority of incident field power to be absorbed bythe energy harvesting circuit at the expense of backscatterstrength. In passive tag design, choice of α is a tradeoffbetween backscatter strength and available DC power at thecircuit.

Rearranging the conjugate reflection coefficient Γ∗ from(2) yields a set of complex impedance values, normalized tothe conjugate of the antenna impedance Z∗a , for a given IQconstellation

ZLi

∣∣∣∣Z∗a

=Z∗a − ZaΓ∗i

1 + Γ∗i. (9)

This set of impedances values may fall in all 4 quadrants ofa Smith chart normalized to Z0 = Z∗a .

4-PAM 4-QAM/4-PSK 16-QAM

In−Phase

Qu

ad

ratu

re

−3 −1 1 3

−3

−1

1

3

In−Phase

Qu

ad

ratu

re

−1 0 1

−1

0

1

In−Phase

Qu

ad

ratu

re

−3 −1 1 3

−3

−1

1

3

(a) (b) (c)Fig. 6. Measured IQ backscatter constellations operating over a 915 MHz carrier at a separation distance of 2.75 m from a +23 dBm RF source. Data ratesfor the constellations shown are 1 Mbps (500,000 symbols/sec) for 4-PAM (a) and 4-QAM/4-PSK (b) and 2 Mbps (500,000 symbols/sec) for 16-QAM (c).

A typical antenna in a 50 Ω system has a resonantimpedance very near 50 + j0 Ω. For such antenna with ap-proximately real-valued Za, modulating impedances found bysolving (9) will require R, L, and C components. This createsa problem for monolithic implementation because standardCMOS processes yield good quality resistors and capacitorsbut physically large and low Q inductors. In order to makeCMOS integration feasible by eliminating L components, atransformation is found by enforcing the constraint that

Γ∗iC Only= β

(Γ∗i − jmin

[=(Γ∗)])

(10)

for all Γ∗i before re-applying the conversion to impedancevalues (9), where =

(Γ∗)

represents the imaginary portion ofthe reflection coefficients and

β =max

[=(ZL

)]max

[=(ZL

)]−min

[=(ZL

)] (11)

scales the reflection coefficients by the portion of impedancesabove the real line. This transformation compresses the IQ vec-tor space of the constellation reducing backscatter amplitudecompared to the RLC implementation, but yields the desiredQAM constellation using values compatible with a CMOSprocess.

The 16-QAM modulator in this work is designed for a50 Ω system (Za = 50 + j0 Ω) with β = 0.5 scaling toeliminate inductors. Conjugate reflection coefficients measuredat the design frequency of 915 MHz are compared to simulatedvalues in Table I and shown on a Smith chart in Fig. 4. Usingthe equation for error vector magnitude (EVM)

EVM =

√√√√√√ 1N

N∑k=1

|SkIdeal − SkMeasured |2

max [SIdeal](12)

that compares the experimental value of the kth symbol(SkMeasured ) to the ideal location of the kth symbol (SkIdeal )and normalizes by the maximum ideal symbol magnitude, a

Fig. 7. Measured DC power consumption of the 16-QAM backscattermodulator over a range of data rates.

symbol EVM of 3.62% was measured at 915 MHz. Simulationresults presented in Table I indicate small variation (< 3%added EVM) over the 868–950 MHz worldwide UHF band.

IV. EXPERIMENTAL RESULTS

Using the test setup of Fig. 5, a series of IQ backscatterconstellations are measured. In addition to 16-QAM backscat-ter, subsets of the 16 impedance states in Fig. 4 are selectedto yield 4-QAM/4-PSK and 4-Pulse Amplitude Modulation(PAM). Measurements in Fig. 6 are taken over-the-air using a+23 dBm RF source operating at 915 MHz and tag-to-readerseparation by 2.75 m. The symbol rate of the modulator washeld constant at 500,000 symbols/sec for all three constella-tions resulting in bit rates of 1 Mbps for 4-PAM and 4-QAM/4-PSK and 2 Mbps for 16-QAM.

The DC power consumption of the modulator was alsocharacterized. This measurement only includes power requiredfor the 1-to-16 Mux modulator implemented using AnalogDevices ADG904 CMOS SP4T RF switches and ignores datageneration logic feeding the modulator. As the data rate in-creases, DC consumption increases following 1

2CV2f CMOS

TABLE II16-QAM BACKSCATTER LINK BUDGET

λ Pt Gt, Gr ∆RCS, min. Pr rmax.96 Mbps 33 cm +28 dBm 8 dBi 0.03 m2 -63.2 dBm 17.01 m60 Mbps 33 cm +28 dBm 8 dBi 0.03 m2 -65.2 dBm 19.20 m40 Mbps 33 cm +28 dBm 8 dBi 0.03 m2 -67.0 dBm 21.25 m

Fig. 8. Measured baseband constellation of a 16-QAM modulator transmit-ting 96 Mbps at a distance of 1.24 m from a +23 dBm, 915 MHz reader.

switching speed as shown in Fig. 7. Continuous operation at96 Mbps requires 1.49 mW of DC power resulting in an energyper bit of 15.5 pJ.

Due to switching characteristics and state transition timeof the SP4T switches, a large settling time (≈ 25 ns) wasneeded to reduce intersymbol interference and constellationnoise. This resulted in a maximum usable symbol frequencyof 12 MHz corresponding to a bit rate of 96 Mbps. Maximumbandwidth for the QAM modulator of Fig. 2(c) is determinedby switching characteristics and state transition time.

A measured 96 Mbps 16-QAM constellation is shownin Fig. 8. The red dots showing demodulated data pointsare compared to the blue marks representing ideal symbollocations. The modulator was measured using a 915 MHz+23 dBm RF source placed 1.24 m away from the tag in atypical indoor environment. This link budget is summarized inTable II. During transmission of 196,736 bits, zero bit errorswere observed, for a corresponding bit error rate (BER) of< 5.08× 10−6 and measured EVM of 9.38%.

Equation (13) can be used to estimate range assuming freespace propagation, no polarization loss and a semi-passive tagthat is not limited by available power. Theoretical maximumoperating distances for varying data rates using a standard+36 dBm EIRP transmitter (Table II) are found by assumingspectral efficiency of 1 bps/Hz, a measured reader noise figureof 19 dB and a 12 dB Eb/No margin. For operation at96 Mbps, a theoretical maximum distance of 17.01 m is foundusing the minimum differential RCS of the constellation [6].

Fig. 9. Measured EVM of a 16-QAM tag placed 1.24 m away from a+23 dBm, 915 MHz reader with a varying data rate.

rmax. =

(PtGtGr∆RCSλ

2

Pr (4π)3

)1/4

(13)

Using the same test setup and conditions as the over-the-airtest for Fig. 8, EVM for varying data rates is measured. As thedata rate increases, the measured EVM in Fig. 9 also increases.This result is largely attributed to the switch transition timeof the SP4T CMOS switches dominating the bit period as thesymbol frequency increases.

V. DISCUSSION AND CONCLUSION

In this paper, a 96 Mbps 16-QAM UHF backscatter modu-lator was presented, consuming only 15.5 pJ/bit or 1.49 mWof power. The modulator was implemented as a semi-passivetransponder capable of producing 4-QAM/4-PSK, 4-PAM,and 16-QAM backscatter. Using a 915 MHz transmitter with+36 dBm EIRP, this backscatter data link has a theoreticalmaximum operating distance of 17.01 m at 96 Mbps or21.25 m at 40 Mbps. This device is suitable for ultra low-power wireless applications requiring high-bandwidth commu-nication such as remote camera sensors or wireless audio.

REFERENCES

[1] U. Karthaus and M. Fischer, “Fully integrated passive UHF RFIDtransponder IC with 16.7-µW minimum RF input power,” IEEE Journalof Solid-State Circuits, vol. 38, no. 10, pp. 1602–1608, October 2003.

[2] EPC Global US, “Class 1 Generation 2 UHF RFID protocol foroperation at 860MHz-960MHz, version 1.0.9,” Available online,http://www.epcglobalus.org/, 2005.

[3] V. Pillai, H. Heinrich, D. Dieska, P. Nikitin, R. Martinez, and K. Rao,“An ultra-low-power long range battery/passive RFID tag for UHF andmicrowave bands with a current consumption of 700 na at 1.5 v,” IEEETransactions on Circuits and Systems I: Regular Papers, vol. 54, no. 7,pp. 1500 – 1512, July 2007.

[4] Y. He, R. Yuan, X. Ma, and J. Li, “The IEEE 802.11 power savingmechanism: An experimental study,” in IEEE Wireless Communicationsand Networking Conference, 2008. WCNC 2008, 2008, pp. 1362 – 1367.

[5] M. Winkler, T. Faseth, H. Arthaber, and G. Magerl, “An UHF RFID tagemulator for preceise emulation of the physical layer,” in 2010 EuropeanWireless Technology Conference (EuWIT), Sept. 2010, pp. 273–276.

[6] P. Nikitin, K. Rao, and R. Martinez, “Differential RCS of RFID tag,”Electronics Letters, vol. 43, no. 8, pp. 431 – 432, April 2007.

[7] S. Thomas and M. Reynolds, “QAM backscatter for passive UHF RFIDtags,” in 2010 IEEE International Conference on RFID, April 2010, pp.210 – 214.

[8] S. J. Thomas, E. Wheeler, J. Teizer, and M. Reynolds, “Quadratureamplitude modulated backscatter in passive and semi-passive UHF RFIDsystems,” IEEE Transactions on Microwave Theory and Techniques, (toappear).

[9] R. Hansen, “Relationships between antennas as scatterers and as radia-tors,” Proceedings of the IEEE, vol. 77, no. 5, pp. 659–662, May 1989.