See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224214985

Radio Frequency Time-of-Flight Distance Measurement for Low-Cost

Wireless Sensor Localization

Article in IEEE Sensors Journal · April 2011

DOI: 10.1109/JSEN.2010.2072496 · Source: IEEE Xplore

CITATIONS READS
155 9,752

3 authors, including:

Steven Lanzisera Kristofer S. J. Pister

Lawrence Berkeley National Laboratory University of California, Berkeley
15 PUBLICATIONS 1,179 CITATIONS 341 PUBLICATIONS 25,475 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Steven Lanzisera on 02 June 2014.

The user has requested enhancement of the downloaded file.

IEEE SENSORS JOURNAL, VOL. 11, NO. 3, MARCH 2011 837

Radio Frequency Time-of-Flight Distance

Measurement for Low-Cost
Wireless Sensor Localization
Steven Lanzisera, David Zats, and Kristofer S. J. Pister

Abstract—Location-aware wireless sensor networks will enable demonstrated. Although UWB transmitters are simple to imple-
a new class of applications, and accurate range estimation is ment and extremely low power, UWB receivers have proven to
critical for this task. Low-cost location determination capability be highly complex and consume a large amount of power when
is studied almost entirely using radio frequency received signal
strength (RSS) measurements, resulting in poor accuracy. More providing communication performance comparable to narrow-
accurate systems use wide bandwidths and/or complex time-syn- band radios. Some narrowband methods have been proposed
chronized infrastructure. Low-cost, accurate ranging has proven that require time-synchronized and/or high-performance, spe-
difficult because small timing errors result in large range errors. cialized base station devices, and this added complexity and cost
This paper addresses estimation of the distance between wireless limit the application of these systems [2], [3]. The second phase
nodes using a two-way ranging technique that approaches the
turns these ranges into locations and has been widely studied
Cramér–Rao Bound on ranging accuracy in white noise and
achieves 1–3 m accuracy in real-world ranging and localization [4]. There is a need for low-cost, simple ranging technology that
experiments. This work provides an alternative to inaccurate RSS provides the meter-level accuracy required for many localiza-
and complex, wide-bandwidth methods. Measured results using a tion problems.
prototype wireless system confirm performance in the real world. This paper presents a burst mode, two-way ranging (TWR)
Index Terms—Real-time location systems, sensor networks, method that closely approaches the theoretical lower bound for
two-way ranging (TWR). ranging accuracy in a noise-limited environment and achieves
meter level accuracy in multipath environments. All nodes in
the network are identical, simple to implement, and do not re-
I. INTRODUCTION quire time-synchronized infrastructure. The code modulus syn-
chronization (CMS) method has an online measurement com-
IRELESS networks have become a part of daily life,
W and the addition of location awareness can change the
application landscape. Mobile phones have low resolution capa-
ponent and an offline range extraction component, and this sep-
aration simplifies implementation and improves performance.
Measurements are taken at several carrier frequencies and com-
bilities today, and this is changing the way people plan, navigate, bined together to mitigate the impact of multipath channel char-
and consume information. Today’s indoor wireless networks are acteristics. These techniques are not specific to an individual
almost universally unaware of device location, but the combina- standard, modulation scheme, bandwidth, or RF platform. They
tion of data communication, location awareness, and low power can easily be added to the digital baseband processor of most ex-
will enable a new host of applications. Battery-operated wire- isting transceivers, thereby adding time-of-flight (TOF) ranging
less devices for tagging, locating, and sensing data in factories, capability.
hospitals, and other environments will be widespread, reducing A prototype of the system was implemented using a commer-
costs and improving quality. cially available 2.4 GHz radio, analog-to-digital interface elec-
Determining device location has two parts. The first phase tronics, a field-programmable gate array (FPGA), and a micro-
involves measuring a relationship between nodes (e.g., distance controller. A 2 MHz bandwidth, frequency shift keying ranging
and angle), and the second phase uses these relationships to esti- scheme was implemented that is compatible with the common
mate location [1]. Radio frequency (RF) received signal strength IEEE 802.15.4 standard. Measurements over a noisy channel
(RSS) measurements are commonly used to estimate range, but show that the Cramér–Rao Bound (CRB) is nearly achieved at
the accuracy of this technique is poor even in the best of con- moderate signal-to-noise ratios (SNRs). Measurements taken in
ditions [2]. The primary alternative is the use of ultra-wideband several environments show 1 m accuracy outdoors and 1–3 m
(UWB) RF ranging, and good ranging performance has been accuracy indoors.

Manuscript received January 30, 2010; revised June 09, 2010; accepted Au-
gust 19, 2010. Date of publication January 24, 2011; date of current version
II. LOCALIZATION AND TIME-OF-FLIGHT (TOF) RANGING
February 02, 2011. The associate editor coordinating the review of this paper
and approving it for publication was Prof. Henry Leung.
Determining the location of a device is called localization,
S. Lanzisera is with the Lawrence Berkeley National Laboratory, Berkeley, and the localization problem typically consists of estimating the
CA 94720 USA (e-mail: [email protected]). distance between nodes and then using these ranges to estimate
D. Zats and K. S. J. Pister are with the University of California, location. The accuracy of a localization system is limited by the
Berkeley, Berkeley, CA 94720 USA (e-mail: [email protected];
[email protected]). accuracy of the range estimates and the geometry of the network
Digital Object Identifier 10.1109/JSEN.2010.2072496 devices to be localized. This section contains an introduction to
1530-437X/$26.00 © 2010 IEEE
838 IEEE SENSORS JOURNAL, VOL. 11, NO. 3, MARCH 2011

low-power system, it is difficult to achieve this accuracy. Typ-

ical radios can only resolve the time of events at the rate of their
reference clock, resulting in resolution on the order of 50 ns (15
m), and special techniques and hardware must be employed to
enable the required accuracy. RF TOFranging has seen wide-
spread use in the global positioning system (GPS), but it has
seen limited use in terrestrial systems due to problems with time
accuracy, multipath channel effects, and system cost and com-
plexity.

C. Application Requirements
Tagging and locating assets in buildings is the primary appli-
cation for this technology, and room-level accuracy ( ) is
typically sufficient. Deployments must be low cost and should
not require dedicated wiring. This prevents the widespread use
of time difference of arrival techniques, which are typically
Fig. 1. CDF of location error normalized by the RMS ranging error. time synchronized and require dedicated wiring. To further
reduce cost, the transceiver should be compatible with the IEEE
localization, range estimation techniques, and typical applica- 802.15.4 standard widely used in wireless sensor networks
tion requirements. (WSNs) [5].

A. The Localization Process and Accuracy Limitations

III. SOURCES OF TOF RANGING ERROR
Multilateration is the basic process where range measure-
ments between a device with unknown location and three or TOF range estimation accuracy is primarily limited by
more reference devices are used to estimate a location in two di- clock synchronization, noise, sampling artifacts, and multipath
mensions. The best performing localization algorithms are sim- channel effects. This section addresses each error source and
ilar to a minimum squared error (MSE) optimization with three currently available techniques for error mitigation.
or more range estimates. Fig. 1 shows the relationship between
location error and range measurement error using MSE local- A. Clock Synchronization
ization and randomly placed nodes. Increasing the number of TOF ranging systems need to estimate the time of transmis-
reference nodes improves accuracy in the presence of range er- sion and arrival using a common time reference. In a simple
rors, therefore more reference nodes should be used than strictly case, two wireless devices, and , measure their separation
necessary. The geometry of the reference nodes plays a signif- with measuring the time of arrival of a signal sent by . If the
icant role in device location. For example, if the reference de- clocks are not perfectly synchronized, and ’s notion of
vices are collinear, a unique location cannot be determined. If is offset from ’s, then this adds as an error to the measurement.
there are more reference nodes than required, geometry rarely The required time synchronization ( ) is too strin-
limits localization accuracy. This area has seen significant re- gent for most systems.
search over the years, and the primary area for continued re- Two-way time transfer (TWTT) is a TWR method that mit-
search in the second phase involves determining location when igates the effect of clock synchronization error [6]. It allows
some measurements are highly erroneous [4]. the time offset between and to be ignored. Both and
are responsible for measuring a time delay accurately using a
B. Range Estimation Techniques local clock. If the time sends the signal is , the time
Measuring the range between two wireless devices has receives the signal is , the time replies to is , the
proven to be a challenging problem which has limited the use time receives the signal back is such that
of location-aware wireless systems. RF RSS methods have been , then measures and measures
widely used due to simplicity. The RSS in free space decreases . The TOF , , can be estimated by combining
with the square of the distance between the transmitter and these two measurements
receiver, providing a one-to-one mapping from RSS to distance.
In real environments, constructive and deconstructive interfer- (1)
ence cause the RSS to be unpredictable as a function of range,
and this problem has been widely reported and recognized. In TWR, the measurement takes place over a relatively long
Range measurements using RSS have proven to be unsuitable period of time, so if the reference frequencies at the two nodes
for most localization problems. are not identical, an unknown bias will be added to the signal.
Measuring the RF signal TOF between nodes avoids many The system must account for the clock frequency offset (i.e.,
of the problems of RSS methods, but it is challenging on its clock drift) error. Mitigation methods have been developed for
own. RF signals travel at the speed of light, and one meter accu- wireless systems, and one of these algorithms is required for
racy requires approximately 3 ns time resolution. On a low-cost, TWR [5].
LANZISERA et al.: RADIO FREQUENCY TIME-OF-FLIGHT DISTANCE MEASUREMENT 839

RF time difference of arrival (RF TDOA) techniques also

combat time synchronization issues by having wired infrastruc-
ture time synchronized to better than 1 ns. The mobile devices
transmit a signal, and the RF TDOA at the base stations is used
to estimate range. The interested reader is directed to [2] and
[7].

B. Noise
A range measurement degraded only by white noise is limited
in accuracy by the signal energy-to-noise ratio, , at the
receiver and the occupied bandwidth, . Ranging is a problem
that has been studied in the context of radar applications, and
the CRB provides a lower bound for the variance of the range
estimate in white noise. For a one-way ranging system using
IEEE 802.15.4 modulation, the CRB is

(2)
Fig. 2. CRB as a function of bandwidth.

The variance of the range estimate is , is the speed of light,

and is the occupied signal bandwidth in Hertz [8]. The SNR
is related to in that

(3)

where is the signal duration during which the bandwidth is

occupied. In many common signals, the bandwidth and dura-
tion are tied together such that . Therefore, the
ratio is approximately equal to the SNR. Signals with
would exhibit better noise performance at lower SNR values,
and ranging signals with this characteristic are called pulse-
compressed waveforms. For a fixed-signal energy and noise den-
sity, increasing the bandwidth provides improvements in noise
performance. This is one argument for wide bandwidth ranging
systems, but the bandwidth required to achieve reasonable noise
Fig. 3. Comparison of CRB to sampling induced error as a function of sampling
performance is not large. The CRB can be closely approached frequency.
in many cases where , and this is the intended
target area for most communication systems. Both bandwidth
and play significant roles in determining noise-limited divided up into range bins that are wide. Sampling adds
performance [8]. uniform range uncertainty in each bin of
In TWR systems such as radar or TWTT, the noise limit is
reduced by the round trip nature of the measurement. In radar (4)
systems a single measurement is made that is twice the desired
range, thus reducing by a factor of 4. In TWTT, two measure- In the case of the IEEE 802.15.4 example, with sampling at ,
ments are made and averaged to get the range estimate resulting , the variance due to sampling can be calculated
in a reduction of 2. These effects simply add constants to the to be . Continuous tracking, filtering, or averaging can
denominator of (2). be used to improve the resolution, but this is not bandwidth or
Fig. 2 shows the CRB as a function of bandwidth for power efficient. Using just averaging, over 1000 measurements
of 10 dB and 26 dB. Signals with products up to 1000 are are required to achieve a variance of , and an improved
easily achievable enabling large . It is interesting to note TWTT method would require over 30 measurements [3]. To re-
that noise alone does not prevent 1 m accuracy for bandwidths duce this error, the signal can be oversampled. Fig. 3 shows the
down to a few megahertz. CRB for a 2 MHz bandwidth signal with of 26 dB, the
standard deviation of the range error due to sampling, and the
C. Sampling Artifacts combined effect of both error sources. One must sample very
It is commonly believed that the resolution of a TOF measure- fast to have the error dominated by the CRB rather than sam-
ment depends directly on the sampling rate [3]. Known as range pling when is at the high values possible in communi-
binning [9], this occurs when a matched filter is used to estimate cations. As is reduced, the sampling rate required re-
the time of arrival with a sampling rate of up to . Sam- mains higher than twice the signal bandwidth, down to
pling adds error to the estimate because the estimate space is of about 3 dB. In IEEE 802.15.4 systems is typically
840 IEEE SENSORS JOURNAL, VOL. 11, NO. 3, MARCH 2011

between 15 dB and 30 dB [10], enabling reasonable noise per-

formance.
If the signal is sampled above Nyquist ( ), the signal’s
entire information content is captured, and better time resolution
than is possible. Interpolation between samples can yield sig-
nificant improvements in resolution [5], but a major challenge
is that many systems would need to perform this interpolation
in real time, increasing system complexity and power consump-
tion beyond reasonable limits [11].

D. Multipath Channel Effects

RF signals bounce off objects in the environment, causing
the signal to arrive at the receiver through many paths. This is
Fig. 4. Measured RSS as a function of frequency.
common indoors, and it is possible that the indirect paths have
higher power than the direct path [12]. The communication en-
vironment is called the channel, and multipath channels are spe-
cific to the environment (e.g., office and outdoors) and the spe- because the paths will interfere differently [13]. In some local-
cific transceivers’ geometry in that environment. The channel ization systems, the devices and environment may move slowly
impulse response can be modeled as a series of complex delta such that ranges taken over a short period of time can be con-
functions in time sidered to be taken in a static channel.
The frequency dependence of the channel can be observed
by measuring the RSS profile across carrier frequency in an in-
door environment for fixed transmitter and receiver geometries,
as shown in Fig. 4. At some carrier frequencies, the signal expe-
where , , and are the amplitude, time, and phase delay riences deconstructive interference (referred to as fading), while
of the th path, with representing the direct path. , , at others it has much higher signal strength due to construc-
and are random parameters, and a variety of distributions are tive interference. Without knowing the channel characteristics,
commonly applied to them [12]. The transmitted signal, , knowledge of the RSS at one frequency tells you little about the
is given as follows in phasor notation: RSS at a nearby or distant frequency. Communication signals
with bandwidth larger than the reciprocal of the time between
the first and last significant paths (the delay spread) are largely
immune from fading because it is sufficient for a most of the
In , the time-dependent phase term represents frequency signal bandwidth to be observable at the receiver.
or phase modulation, and the signals of interest have constant In a ranging system, however, the delay spread is not the crit-
amplitude that can arbitrarily be set to unity. The received signal ical parameter. If a receiver can estimate the first path arrival,
is the convolution of the transmitted signal and the channel with this will be the shortest length, and thus the desired estimate.
additive white noise If the system is unable to resolve the individual paths, the esti-
mate is blurred by the multipath effects, resulting in estimation
error. Therefore, the typical interpath delay, , is the critical
value. Indoors, inter-path delays of 5–10 ns are common and
The noise term will be ignored in this analysis, as it has negli- must be resolved if accuracy is to be better than a few meters
gible impact on multipath performance. If consists of only [14]. The bandwidth required for this is greater than , or
two paths, we can write the entire received signal .
There are several techniques available for reducing the
impact of multipath, and they fall into three basic categories:
1) increasing bandwidth; 2) estimating the channel impulse re-
(5) sponse; and 3) multipath bias reduction. The first two methods
have received some attention from researchers with varying
Although , , and are random variables, they are fre- degrees of success. The third category has not received signifi-
quency-independent over a given RF communication band. cant attention in the literature, and a basic algorithm to reduce
Over small periods of time, these parameters can be considered multipath bias is presented in Section IV.
constant, and (carrier frequency) can be used to manipulate Typical WSN radios do not use large bandwidths capable of
the relative phase of these paths. resolving the individual paths, but the recent IEEE 802.15.4a
Changing carrier frequency even by a few megahertz can dra- standard has a UWB physical layer option. UWB transceivers
matically affect the apparent multipath environment in narrow- use more than 500 MHz of bandwidth, which is sufficient to re-
band systems. Moving one transceiver by just a fraction of a solve the individual paths in the channel. Unfortunately, contin-
wavelength ( at 2.4 GHz) will cause the receiver uing research shows that although UWB transmitters are simple
to see what looks like an entirely new multipath environment and low power to implement, UWB receivers with comparable
LANZISERA et al.: RADIO FREQUENCY TIME-OF-FLIGHT DISTANCE MEASUREMENT 841

source of the code. Node D transmits two copies of the circu-

larly shifted code it received back to C, and this transmission is
shown in the shaded box over line 3. Node C receives the signal
and records it synchronized to its local reference shown on line
1. Because of the round-trip nature of the system, the circular
shift that occurred going from C to D is exactly undone going
from D to C. After C has received the code, the transceivers are
shut off, and all of the real-time processing is completed.
Node C then computes the cross correlation between the code
it recorded and the code that it sent; the measured code offset
Fig. 5. Baseband signals used in code modulus synchronization. is the time of flight. Because this system relies on sampling
the signal above Nyquist, the received code can be interpo-
performance (communication range and linearity) consume dra- lated to improve resolution up to the noise limit of the system.
matically more power than their narrowband counterparts [14]. The correlation and code offset estimation are not done in real
The second method for attempting to mitigate the impact of time, enabling the computation to be done at any time using any
multipath interference is through indirect channel estimation, method the user desires. This system can approach the CRB in
either through a super-resolution method or frequency domain a single measurement, substantially improving over other TWR
channel characterization. A super resolution algorithm is one methods.
that provides range resolution that is better than when there Multiple copies of the code can be sent in order to increase
is sufficient to resolve meaningful channel information. . The receiving system can accumulate (average) mul-
The interested reader is referred to [2], [13], [15] for more infor- tiple copies of the code to increase , but each is exactly one
mation. In IEEE 802.15.4, the achievable accuracy appears to be copy of the code that is circularly shifted in the same way as the
insufficient; the estimated time resolution would be or other received copies. Averaging of multiple copies is important
. for achieving good noise performance, and it does not change
the system’s ability to resolve the TOF accurately.
In TWTT, the time-of-arrival must be determined at both
IV. RANGING ERROR MITIGATION TECHNIQUES nodes involved in the range estimation, but in CMS only one
This section presents two new methods that, when combined node performs this calculation. Therefore, while CMS enables
together, combat the error sources discussed in Section III. Code better sampling performance, the full processing gain of the
modulus synchronization is a new method of round-trip TOF system is not realized at the second node. This causes an
apparent noise penalty. At the same time, CMS consists of a
ranging that mitigates the effects of sampling and poor time
synchronization. A frequency diverse range estimation method single range estimate just like in radar, resulting in the same
is also presented that successfully improves range estimation factor of 2 noise benefit compared to TWTT. Ignoring the
accuracy, while not requiring time-synchronized infrastructure, impact of the transmitter and receiver transfer functions for
simplicity, the effective for TWTT is
complex base stations, or special wide bandwidth transceivers.

A. Code Modulus Synchronization (CMS)

Code modulus synchronization (CMS) emulates a full duplex
where is the number of code copies averaged and is the code
ranging system using half-duplex radios such as those used in
length. The time-of-arrival is not estimated at node C in CMS,
WSNs.The delay between reception and retransmission must be
and the signal sent from D to C contains noise from the first leg
managed carefully. CMS uses a periodic signal (such as a square
of the trip. For CMS, then, is
wave or a pseudorandom code) modulating an RF carrier as the
ranging signal so that large is possible. Fig. 5 shows the op-
eration of the CMS using a square wave baseband signal. The
first node, C, generates a local baseband ranging signal, shown
on the top line (C REF/TX). This code is used to modulate the The last factor in (5) represents the noise penalty of CMS versus
carrier and, in the shaded region, is transmitted to the second TWTT under the constraint . This term is unity at
node D. D has a local clock with the same period as at C, but the infinite SNR because there is no penalty (processing gain pro-
phase of the clocks are offset. As a result, D knows the length of vides no benefit without noise). At very low SNR ( ),
the incoming code, but it does not know the phase offset in the the penalty term is approximately 1/2 if no averaging is used
clocks. D samples and demodulates this signal, and exactly one ( ). The worst-case performance degradation is at low
circularly shifted copy of the code is stored in memory (shown SNR, and this factor is cancelled by the factor of 2 difference
on line 2, D RX, in the shaded region). At this point, D has a between the TWTT averaging effect and the CMS single mea-
local copy of the code that is circularly shifted due to the clock surement effect. For moderate to large values of , the penalty
phase offsets between C and D, and this reference code is shown term approaches unity (no penalty). CMS with averaging pro-
on line 3 (D REF/TX). After C has sent the code and D has re- vides better noise performance than TWTT, and it is easy to
ceived the code, the transceivers switch states, and D is now the avoid the sampling penalties common in TWTT.
842 IEEE SENSORS JOURNAL, VOL. 11, NO. 3, MARCH 2011

Fig. 6. Impact of multipath signal on range error for relative mulitpath (a) phase, (b) delay, and (c) amplitude.

After a single measurement, the variance, , for TWTT or only depends on (see (5)). In the case of IEEE 802.15.4, min-
the enhanced version of TWTT presented in [3], is given by (4). imum shift keying (MSK), a version of frequency shift keying,
Comparing (4) to the CMS bound, given by is used. In MSK the modulation signal is square and changes
between and . In Section III-C, we assumed that was
constant for the two paths. Immediately after a to change,
however, the direct path is at and the second path is at .
Now, the composite received signal is the sum of two sinu-
soids at different frequencies and has rapid magnitude and phase
we find that CMS has a better single measurement variance changes. These changes are nonsinusoidal and affect the demod-
ulator output.
The multipath induced bias is a function of the relative mul-
tipath amplitude, phase, and delay, and the bias can be positive
Substituting for the factor where represents how or negative depending on the relative phase of the paths. This
much faster the sampling is than the signal bandwidth, we find is an important fact because it is intuitive to believe that only
that if positive biases are possible. Simulations of the two path multi-
path environment are presented in Fig. 6. Fig. 6(a) shows how
varying impacts the range estimate when the relative delay and
amplitude are fixed to 20 ns and 1/2, respectively. The trend as-
sociated with varying relative multipath delay when is set for
then CMS provides better performance than TWTT. This result maximum error and the relative multipath amplitude is set to
is directly in line with Fig. 3, where signals must be highly over- 1/2 are shown in Fig. 6(b). As the relative delay increases, the
sampled to achieve performance approaching the CRB unless magnitude of the bias increases. Eventually the delay is large
CMS is used. enough that it can be differentiated from the direct path, and the
error decreases. Fig. 6(c) shows the case when is set for max-
B. Frequency Diverse Range Estimation imum and minimum error, relative delay is fixed at 20 ns, and
We present a multipath mitigation technique that requires for varying relative multipath amplitudes. The key conclusions
minimal processing and relies on the properties of the multipath are that both positive and negative biases occur, and the magni-
environment and the signal demodulator. Measurements taken tude of the biases increases with delay.
at several carrier frequencies are combined to reduce the bias in From the trends in Fig. 6, it is instructive to consider how
the TOF estimate. The impact of multipath on the demodulator to best estimate the true time of flight when presented with a
output is critical to understanding this technique. series of measurements taken over the same channel with dif-
The signal demodulator is a simple digital frequency detector. ferent phase relationships. To generate measurements with dif-
The most common receiver for IEEE 802.15.4 is a low inter- ferent phase relationships, the measurements are taken at dif-
mediate frequency (low-IF) receiver with FM demodulation at ferent carrier frequencies. From the figures, it appears that an
the low-IF. The incoming modulated sinusoid has its period average value will reduce the overall bias. A more detailed study
measured with a counter from rising edge to rising edge and of the bias over a wider set of conditions is required to develop
falling edge to falling edge. The counter output is applied to a a heuristic for reducing overall bias.
lookup table to determine the demodulation value. This struc- Simulated multipath channels generated by the IEEE
ture is simple, produces a multibit output, and has reasonable 802.15.4a working group for indoor office and residential en-
noise performance. vironments were used to simulate 200 multipath environments
The simplest multipath situation has a direct path and a single [16]. For each channel, range was estimated on each of the 16
other path that arrive with some relative time delay, , and car- IEEE 802.15.4 carrier frequencies in the 2.4 GHz band, using
rier phase, . Both and affect the demodulator output, but an algorithm consistent with the methods presented in this
LANZISERA et al.: RADIO FREQUENCY TIME-OF-FLIGHT DISTANCE MEASUREMENT 843

to estimate the time offset for each carrier frequency. The

estimated time offset varies from frequency to frequency, as
described in Section IV-B. All 16 measurements are reported,
and the median of these 16 values is used as the TOF estimate.
In this implementation, 32 copies of the 2 chip signal are av-
eraged after demodulation for a product of 64, while main-
taining code modulus synchronization. The peak estimation al-
gorithm uses a linear regression across several correlation points
on either side of the peak and calculates where these two lines
intersect. This method is faster and has equivalent noise perfor-
mance to a low-pass interpolation of the correlation data fol-
lowed by a traditional dual correlator peak search.

VI. PROTOTYPE EXPERIMENTAL RESULTS

The implemented system is capable of performing range
measurements with noise performance (repeatability) meeting
Fig. 7. Photograph of the 7.5 2 6 cm Waldo software defined radio platform. or exceeding those demonstrated by systems with greater
instantaneous bandwidth and/or sampling rate [17], [18], while
having better than 3 m accuracy in ranging and localization
work. The median of the 16 estimates had the best error per- experiments.
formance, with over 80% of the simulated channels producing
errors less than 3 m. The mean value produced worse estimates A. Noise Performance
( ). The median value performs better To verify performance in a noise channel, two Waldo de-
because it is less influenced by single measurements that are far vices were connected via a RF cable and a variable attenuator.
from the median. These “outlier” measurements occur when Fig. 9 shows the standard deviation of ranging measurements
the channel is in deep-fade. as a function of baseband SNR, the CRB for this system, and
The median of the estimates provides a reduced bias estimate the range binning limit (previous work) [3]. One thousand mea-
of the TOF, is simple to compute, and requires no phase-co- surements were taken at a single frequency and baseband SNR
herent measurements, greatly simplifying implementation. This to generate each point in the figure. At high values of SNR, the
method is selected for use in the implemented ranging demon-
system does not achieve the CRB because of the limited dy-
strations.
namic range of the digital baseband processor. CMS performs
V. PROTOTYPE RANGING SYSTEM within a factor of 2 of the CRB, demonstrating that at an SNR of
6 dB ( ) equivalent performance using TWTT
The presented ranging system is a combination of new algo-
would have required a sampling rate of 60 MHz for a 2 MHz
rithms that require custom, yet simple, hardware to implement.
bandwidth signal.
To demonstrate these ideas, a software-defined radio platform,
dubbed Waldo, was developed (see Fig. 7). This platform con- B. Ranging Demonstrations
sists of a 2.4 GHz radio, digital to analog interfaces, an FPGA,
a microcontroller, and the corresponding Verilog and embedded Ranging experiments were performed both indoors and out-
C code required for correct system operation. Waldo is designed doors to verify the performance of the proposed algorithms and
for battery-operated field use, and no external PC or other hard- the Waldo platform. A localization experiment was also per-
ware is required. formed to show that several identical, non-time-synchronized,
The presented ranging system was implemented on Waldo, battery-powered Waldo nodes could be used to form a network
and the test signal occupies a 2 MHz RF bandwidth using binary and localize a node in the network.
frequency shift keying at a deviation of at 1 Mchip/s Two Waldo nodes were used to perform ranging estimates in
(similar to IEEE 802.15.4). The received signal is sampled at a a parking lot with some cars but mostly open space. The two
low IF of 5 MHz and demodulated in the digital domain. The nodes were not connected together in any physical way, and
demodulated data is limited to 2 MHz bandwidth and is sampled the only method of communication was through the wireless
at 16 MHz yielding range bins of 19 m. link. A range estimate using the proposed TOF method and RSS
The entire ranging procedure from the perspective of the were taken at distances ranging from 1 to 45 m. The TOF and
node originating the ranging operation (Node C in Fig. 5) is RSS range estimates are shown in Fig. 10. The TOF estimate is
shown in Fig. 8. The ranging operation starts with the exchange simply multiplied by the speed of light to yield the slope of unity
of a packet and acknowledgement between the two nodes. shown in the plot. The equation that empirically minimized the
This packet contains configuration information required for mean-squared ranging error in this set of measurements was
ranging. A range operation following the procedure outlined in used for the RSS estimates. Even with this advantage, RSS per-
Section IV-A using each of the sixteen carrier frequencies avail- formance is very poor compared to TOF. Approximately 80% of
able to IEEE 802.15.4 transceivers. After these measurements the TOF measurements are accurate to within 1 m, but not even
have been completed, the resulting data is analyzed in software 20% of the RSS based estimates are accurate to within 1 m.
844 IEEE SENSORS JOURNAL, VOL. 11, NO. 3, MARCH 2011

Fig. 8. Activity at the initiating node for a ranging operation including setup, CMS ranging operations (dark gray), and data processing (white).

Fig. 9. Measured noise performance as a function of SNR.

Fig. 11. Measured indoor ranging performance.

Fig. 12. Results of a four reference node localization experiment.

Fig. 10. Measured outdoor ranging performance.

Indoor range estimates using the same setup were performed The localization experiment was performed in a relatively
to verify that reasonable ranging accuracy can be achieved in en- open area between two buildings. The approximate dimensions
vironments typical to local area and sensor networks. The TOF of the space are 50 m by 40 m with some trees and bushes in the
and RSS measurements shown in Fig. 11 were taken in a clut- area and buildings along two sides. Internode distances of up to
tered hallway. The achieved accuracy for TOF was better than 70 m were available, and communication and ranging could be
1 m 50% of the time and better than 3 m 80% of the time. RSS performed at these distances. Four static nodes were setup on
achieved 8 m accuracy less than 50% of the time. There were no tripods, and a node was carried through the field. The results of
calibration steps or changes to the system firmware, software or the localization experiment are shown in Fig. 12, where the dia-
calculation methods between this environment and the outdoor monds are reference nodes, the boxes are ground truth, and the
environment. circles are estimated location. Localization accuracy is better
LANZISERA et al.: RADIO FREQUENCY TIME-OF-FLIGHT DISTANCE MEASUREMENT 845

than 2 m for 80% of the estimates using a simple MSE estimate [13] T. Watteyne, S. Lanzisera, A. Mehta, and K. Pister, “Mitigating multi-
for location. path fading through channel hopping in wireless sensor networks,” in
Proc. IEEE Int. Conf. Commun., May 2010, pp. 1–5.
[14] B. Lachartre et al., “A 1.1 nJ/b 802.15.4a-compliant fully integrated
m
UWB transceiver in 0.13  CMOS,” in Proc. Int. Solid State Circuits
VII. CONCLUSION Conf., San Francisco, CA, 2009, pp. 312–313.
[15] N. Dharamdial, R. Adve, and R. Farha, “Multipath delay estimations
Code modulus synchronization, a burst mode, TWR method, using matrix pencil,” IEEE Wireless Commun. Networking, vol. 1, pp.
632–635, Mar. 2003.
approaches the CRB without excessive over sampling, an [16] A. F. Molisch et al., IEEE 802.15.4a Channel Model – Final Report,
improvement over previously published methods. Frequency Tech. Rep. Doc. IEEE 802.15-04-0662-02-004a, 2005.
diverse ranging is an easily implemented strategy that improves [17] T. C. Karalar and J. Rabaey, “An RF ToF based ranging implementa-
tion for sensor networks,” in Proc. IEEE Int. Conf. Commun., 2006, pp.
ranging performance in multipath environments. Combined, 3347–3352.
these techniques achieve 1 m ranging accuracy outdoors and [18] S. Schwarzer, M. Vossiek, M. Pichler, and A. Stelzer, “Precise dis-
1–3 m accuracy indoors. A localization experiment further tance measurement with IEEE 802.15.4 (ZigBee) devices,” in Proc.
EEE Radio and Wireless Symp., 2008, pp. 779–782.
verifies performance. In communication systems where the
is typically large, the effect of sampling has dominated Steven Lanzisera received the B.S. degree in elec-
noise-induced error in TWR systems, but CMS avoids this trical engineering from the University of Michigan,
Ann Arbor, in 2002 and the Ph.D. degree in electrical
pitfall. Complex hardware and networks limit the application of engineering and computer sciences from the Univer-
location-aware networks, but the system presented here avoids sity of California, Berkeley, in 2009.
this complexity without the need for specialized base stations, He was an Engineer with the Space Physics
Research Laboratory, University of Michigan, from
time synchronization, UWB, or other expensive and complex 1999 to 2002, where he worked on spacecraft
equipment. integration and testing. He is currently a Researcher
in the Environmental Energy Technologies Division
at Lawrence Berkeley National Laboratory, where
he studies energy use in buildings with a focus on distributed sensing, controls
REFERENCES and appliance energy efficiency. He has published research on embedded
systems, wireless communication, networking, integrated circuits, building
[1] N. Patwari et al., “Locating the nodes: Cooperative localization in wire- energy efficiency, and public policy.
less sensor networks,” IEEE Signal Proc. Mag., vol. 22, no. 4, pp.
54–69, Jul. 2005.
[2] M. Pichler, S. Schwarzer, A. Stelzer, and M. Vossiek, “Multi-channel
distance measurement with IEEE 802.15.4 (Zigbee) devices,” IEEE J.
Sel. Topics Signal Proc., vol. 3, no. 5, pp. 845–859, Oct. 2009. David Zats received the B.S. degree in computer science and engineering
[3] K. Ahmed and G. Heidari-Bateni, “Improving two-way ranging preci- from the University of California, Los Angeles, in 2007 and the M.S. degree
sion with phase-offset measurements,” in Proc. IEEE Global Commun. in electrical engineering and computer sciences from the University of Cali-
Conf., 2006, pp. 1–6. fornia, Berkeley, in 2009. He is currently working towards the Ph.D. degree
[4] K. Pahlavan et al., “Indoor geolocation science and technology,” IEEE at the University of California, Berkeley, where his research focus is energy
Commun. Mag., vol. 40, no. 2, pp. 112–118, Feb. 2002. consumption in datacenter networks.
[5] S. Lanzisera and K. S. J. Pister, “RF ranging methods and performance As Research Assistant at the Center for Embedded Networked Sensing
limits for sensor localization,” in Localization Algorithms and Strate- (CENS), he worked on networks of wireless image sensor nodes.
gies for Wireless Sensor Networks, G. Mao and B. Fidan, Eds. New
York: Information Science Reference, 2009, p. 526.
[6] D. Kirchner, “Two-way time transfer via communication satellites,”
Proc. IEEE, vol. 79, no. 7, pp. 983–990, Jul. 1991. Kristofer S. J. Pister received the B.A. degree in ap-
[7] C. Hoene and J. Willmann, “Four-way TOA and software-based trilat- plied physics from the University of California, San
eration of IEEE 802.11 devices,” IEEE Personal, Indoor and Mobile Diego, in 1982, and the M.S. and Ph.D. degrees in
Radio Commun., pp. 1–6, 2008. electrical engineering from the University of Cali-
[8] H. L. Van Trees, Detection, Estimation, and Modulation Theory. New fornia, Berkeley, in 1989 and 1992.
York: Wiley, 2001. From 1992 to 1997, he was an Assistant Professor
[9] M. Richards, Fundamentals of Radar Signal Processing. New York: of Electrical Engineering with the University of
McGraw-Hill, 2005. California, Los Angeles. In 1997, he joined the
[10] S. Lanzisera, A. Mehta, and K. Pister, “Reducing average power in Department of Electrical Engineering and Computer
wireless sensor networks through data rate adaptation,” in Proc. IEEE Sciences, University of California, Berkeley, where
Int. Conf. Commun., Jun. 2009, pp. 1–6. he is currently a Professor and a Co-Director of the
[11] S. Srirangarajan and A. Tewfik, “Localization in wireless sensor Berkeley Sensor and Actuator Center. He coined the term Smart Dust and
networks under non line-of-sight propagation,” in Proc. IEEE Global pioneered the development of ubiquitous networks of communicating sensors.
Commun. Conf., 2005, pp. 3477–3481. During 2003 and 2004, he was on industrial leave as CEO and then CTO of Dust
[12] Q. H. Spencer, B. D. Jeffs, M. A. Jensen, and A. L. Swindlehurst, Networks, a company that he co-founded to commercialize low-power wireless
“Modeling the statistical time and angle of arrival characteristics of an mesh networking for sensors. In addition to wireless sensor networking, his
indoor multipath channel,” IEEE J. Sel. Areas Commun., vol. 18, no. 3, research interests include MEMS-based micro-robotics and low-power circuit
pp. 347–360, Mar. 2000. design.

View publication stats