IEEE SIGNAL PROCESSING LETTERS, VOL. 11, NO.

5, MAY 2004 513

Direct Position Determination of Narrowband

Radio Frequency Transmitters
Anthony J. Weiss, Fellow, IEEE

Abstract—The most common methods for location of communi- electromagnetic emitter location papers and systems. Hence, the
cations or radar transmitters are based on measuring a specified required processing for traditional matched-field processing is
parameter such as signal angle of arrival (AOA) or time of arrival rather heavy. Moreover, the underwater source’s distance from
(TOA). The measured parameters are then used to estimate the
transmitter location. Since the AOA/TOA measurements are done the sensors is usually the same order of magnitude as the sensor
at each base station separately, without using the constraint that array size. Hence, the far-field assumption that is usually used
all measurements must correspond to the same transmitter, they in electromagnetic AOA does not hold for MFP.
are suboptimal. We propose a technique that uses exactly the same The direct position determination (DPD) method that we pro-
data as the common methods, except that the estimation of loca- pose takes advantage of the rather simple propagation assump-
tion is based on exact maximum likelihood, and the location de-
termination is direct. Although there are many stray parameters, tions that are usually used for radio frequency (RF) signals.
including the attenuation coefficients and the signal waveform, the This enables us to obtain a simple closed-form cost function.
method requires only a two-dimensional search. Monte Carlo sim- The cost function can be maximized using a two-dimensional
ulations indicate that the accuracy is equivalent to AOA, TOA, and (2-D) search for an emitter known to be located on a plane or a
their combination for high SNR, while for low SNR, the accuracy three-dimensional (3-D) search in general. The DPD belong to
of the proposed method is superior.
the least squares family if the noise statistics are unknown. If the
Index Terms—Angle of arrival (AOA), array processing, emitter noise is Gaussian, DPD is the exact maximum-likelihood esti-
localization, matched-field processing, maximum likelihood, time
of arrival (TOA).
mate of location. We demonstrate that DPD outperforms AOA,
TOA, and the combination of AOA and TOA. The DPD tech-
nique requires the transmission of the received signals (possibly
I. INTRODUCTION sampled) to a central processing location. However, AOA and
TOA require only the transmission of the measured parameters
T HE PROBLEM of emitter location attracts much interest
in the signal processing, vehicular technology, and
underwater acoustics literature. Defense-oriented location
to the central processing location. This is the cost of employing
DPD. This letter focuses on the single signal case. Extensions
systems have been reported since World War I. Perhaps the first to multiple signals will be published in the near future.
paper on the mathematics of emitter location, using angle of
arrival (AOA), is due to Stansfield [1]. Many other publications II. PROBLEM FORMULATION AND A POTENTIAL ALGORITHM
followed, including a fine review paper by Torrieri [2]. The Consider a transmitter and base stations intercepting the
papers by Krim and Viberg [3] and Wax [4] are comprehensive transmitted signal. Each base station is equipped with an an-
review papers on antenna array processing for location by AOA. tenna array consisting of elements. Denote the transmitter
Recently, Van-Trees [5] published a book that is fully devoted position by the vector of coordinates and the th base station
to array processing. Positioning by time-of-arrival (TOA) is position by the vector of coordinates . The signal observed by
well known in radar systems [6], and in underwater acoustics the th base station array is given by
[7]. In underwater acoustics, matched-field processing (MFP)
is viewed as a promising procedure for source localization [8]. (1)
MFP can be interpreted as the maximum a posteriori (MAP)
estimate of location given the observed signal at the output where is a time-dependent vector, is an un-
of an array of sensors [8], [9]. Another interpretation of MFP known complex scalar representing the channel effect (attenua-
is the well-known beamforming extended to wide-bandwidth tion), is the th array response to signal transmitted from
signals, nonplanar wave fields, and unknown environmental position , and is the signal waveform, trans-
parameters. mitted at time and delayed by . The vector rep-
In this letter, we discuss a method that has some similari- resents noise and interference, including multipath observed by
ties with MFP. While the concept is similar, the details are dif- the array.
ferent. The models of underwater acoustic propagation are usu- The sampled version of the signal in (1) is given by
ally more complex than the models used in most AOA/TOA
Manuscript received July 31, 2003; revised October 20, 2003. The associate
editor coordinating the review of this manuscript and approving it for publica-
tion was Prof. Dimitris A. Pados.
The author is with the Tel Aviv University, Tel Aviv 69978, Israel (e-mail:
ajw@eng.tau.ac.il).
Digital Object Identifier 10.1109/LSP.2004.826501 (2)

1070-9908/04$20.00 © 2004 IEEE

Authorized licensed use limited to: ULAKBIM UASL - Hacettepe Universitesi. Downloaded on November 28,2023 at 14:56:28 UTC from IEEE Xplore. Restrictions apply.
514 IEEE SIGNAL PROCESSING LETTERS, VOL. 11, NO. 5, MAY 2004

We observe that information about the transmitter position is Without loss of generality we assume that
embedded in the observed signal in two different ways. The first
is the array response. If the signal is in the far field (its distance
from the station is many times the array aperture), the array (9)
response becomes a function of the angle of arrival only. The
position is also reflected by the TOA of the signal at the array Substituting (8) and (9) in (7) we get
, which depends on the distance between the transmitter
and the station.
In order to obtain an attractive algorithm it is desirable to sep-
arate the delay and the transmit time from the signal waveform.
This happens naturally in the frequency domain representation
of the problem. Taking the discrete Fourier transform (DFT) of
(2) we get

(10)
(3)
Instead of finding the minimum of , we can find the max-
where the overbar indicate the DFT coefficient of the corre- imum of defined by
sponding time samples.
The least squares estimate of the position is given by mini-
mizing the cost function (11)

(4) Define the vectors

where stands for the Frobenius norm. Note that the cost
function can be represented by a sum over terms as follows:
(12)

Using these definitions, we can rewrite (11) as

(5)

Define the following vectors:

(13)
(6)
Under the common assumption that the signal waveform is not
where stands for the Kronecker product. Now (5) can be rep- known to the receivers, the cost function in (13) is maximized
resented by by selecting the vector as the eigenvector corresponding to the
largest eigenvalue of the matrix . Hence, (13) reduces to
(7) (14)

where the right side of (14) denotes the largest eigenvalue of ,

The estimate of that minimizes the cost function is given by
and the matrix is a function of the data, the array response
at each base station, the location of the base stations, and the
unknown emitter location . It is clear that the maximization of
(14) requires only a 2-D (or 3-D) search, although the estimator
knows neither the channel response nor the signal. It is inter-
esting to note that the dimensions of the matrix are ,
(8) which might be rather large for some cases. However, we can
replace with the matrix where

where stands for the Hermitian transpose operation. (15)

Authorized licensed use limited to: ULAKBIM UASL - Hacettepe Universitesi. Downloaded on November 28,2023 at 14:56:28 UTC from IEEE Xplore. Restrictions apply.
WEISS: DIRECT POSITION DETERMINATION OF NARROWBAND RF TRANSMITTERS 515

Fig. 1. RMS, mean, 67%, 95% of miss distance for four different methods, unknown signal.

Thus, (14) becomes on the desired resolution. The maximum element of is the
desired cost function.
(16)

This result holds for a single observation of the signal for a pe- III. NUMERICAL RESULTS
riod equivalent to samples. Extension to multiple observa- In order to examine the performance of the advocated
tions of the signal is straightforward. method and compare it with the traditional approaches, we
The case of known signal waveform (e.g., training signal or performed extensive Monte Carlo simulations. Some examples
a synchronization signals are known to the receivers) is of great are shown here. Consider four base stations placed at the
interest as was shown by Li and Compton in [10]. In this case, corners of a 4 km 4 km square. Each base station is equipped
we return to (13) and rewrite it as follows: with a circular array of five antenna elements. The radius of the
array is one wavelength. The transmitter location is selected
at random, uniformly, within the square formed by the base
diag stations. Each location determination is based on 32 samples
of the signal. The SNR is varied between 10 and 10 dB.
At each SNR value, we performed 100 experiments in order
to obtain the statistical properties of the performance. The
path-loss attenuation magnitude is selected at random using
(17) normal distribution (mean , std ), and the attenuation
phase is uniformly distributed in . We applied four
The unknowns are the transmit time , and the emitter position, different techniques in order to locate the transmitter.
. For any given , we can estimate by a one-dimensional
1) AOA estimation using maximum likelihood (also known
search or by FFT of the columns of . If we choose the later
as beamforming) and maximum-likelihood emitter loca-
method, we get the following cost function:
tion estimation using the AOA estimates as the data.
2) TOA estimation using maximum likelihood (under the as-
FFT sumption that the signal waveform is known at the base
stations and using all antenna elements) and maximum-
likelihood emitter location estimation using the TOA es-
timates as the data.
(18) 3) Maximum-likelihood emitter location estimation using
both AOA and TOA as the data.
Stated in words, perform FFT on each of the columns of , and 4) Direct position determination (DPD) according to (16).
sum the squared absolute value of the Fourier coefficient over The performance evaluation is based on the statistics of the
the results to obtain the vector . The length of corresponds miss distance, i.e., the distance between the true emitter position
to the FFT length, which may be a multiplicity of , depending and the estimated emitter position.
Authorized licensed use limited to: ULAKBIM UASL - Hacettepe Universitesi. Downloaded on November 28,2023 at 14:56:28 UTC from IEEE Xplore. Restrictions apply.
516 IEEE SIGNAL PROCESSING LETTERS, VOL. 11, NO. 5, MAY 2004

Fig. 2. RMS, mean, 67%, 95% of miss distance for four different methods, known signal.

We used four different criteria: ACKNOWLEDGMENT

1) Root mean square (RMS) of miss distance; The author would like to thank A. Amar and Y. Isbi for fruitful
2) mean of miss distance; discussions.
3) miss distance that upper bounds 67% of the errors;
4) miss distance that upper bounds 95% of the errors. REFERENCES
All the plots in Fig. 1 indicate that DPD is superior to AOA, [1] R. G. Stansfield, “Statistical theory of DF fixing,” J. Inst. Elect. Eng.,
TOA, and even combined AOA and TOA. The advantage of pt. 3A, vol. 94, no. 15, pp. 762–770, March 1947.
[2] D. J. Torrieri, “Statistical theory of passive location systems,” IEEE
DPD is at low SNR. At high SNR, all methods give excellent Trans. Aerosp. Electron. Syst., vol. AES-20, pp. 184–198, Mar. 1984.
results. Fig. 2 shows similar results for known signals. [3] H. Krim and M. Viberg, “Two decades of array signal processing re-
search,” IEEE Signal Processing Mag., vol. 13, no. 4, July 1996.
[4] M. Wax, “Model-based processing in sensor arrays,” in Advances
in Spectrum Analysis and Array Processing, S. Haykin, Ed. Upper
Saddle River: Prentice-Hall, 1995, vol. III.
IV. CONCLUSION [5] H. L. Van Trees, Detection, Estimation, and Modulation Theory, Part
IV: Optimum Array Processing. New York: Wiley, 2002.
We have proposed a direct position determination technique [6] M. I. Skolnik, Introduction to Radar Systems, 3rd ed. New York: Mc-
that has the same results as AOA, TOA, and their combina- Graw-Hill, 2000.
[7] G. C. Carter, Ed., Coherence and Time Delay Estimation. New York,
tion at high SNR but has better accuracy at low SNR. The NY: IEEE Press, 1993.
DPD is closely related to matched-field processing, but it is [8] J. A. Shorey and L. W. Nolte, “Wideband optimal a posteriori proba-
suitable only for RF signals and not for underwater emitter bility source localization in an uncertain shallow ocean environment,”
J. Acoust. Soc. Amer., vol. 103, no. 1, Jan 1998.
location. Further research is currently underway that explores [9] B. F. Harrison, “An L -norm estimator for environmentally robust,
the advantages and disadvantages of the proposed method shallow-water source localization,” J. Acoust. Soc. Amer., vol. 105, no.
for multiple signals and more complex propagation models. 1, Jan 1999.
[10] J. Li and R. T. Compton, “Maximum likelihood angle estimation for
Small-error analysis, threshold prediction, and comparison with signals with known waveform,” IEEE Trans. Signal Processing, vol. 41,
the Cramer–Rao bound will be published in the near future. pp. 2850–2862, Sept. 1993.

Authorized licensed use limited to: ULAKBIM UASL - Hacettepe Universitesi. Downloaded on November 28,2023 at 14:56:28 UTC from IEEE Xplore. Restrictions apply.