ALTIMETRY WITH REFLECTED GPS SIGNALS: RESULTS FROM A

LAKESIDE EXPERIMENT
Cinzia Zuffada, Robert Treuhaft, Stephen Lowe, George Hajj, Michael Lough, Lawrence Young,
Sien Wu, Mark Smith
Jet Propulsion Laboratory
California Institute of Technology
Pasadena, CA 91109 MS 238-600
Phone: (818) 354 0033iFax: (818) 393 4965E-mail: cinzia@cobra.iDl.nasa.eov
Jesse Lerma
Sony Precision Technology
La Habra. CA 90631
ABSTRACT
In the fall of 1999 an experiment was performed at Crater
Lake, Oregon, to demonstrate the feasibility of surface
altimetry with GPS. A GPS antenna was directed at the lake its axis pointing slightly downward - from a rocky precipice.
This arrangement allowed collection of both the direct GPS
signal as well as the signal reflected off the lake surface. The
relative delay and carrier phase rates between direct and
reflected signals are used to infer the height of the lake
surface. The site was chosen for its elevation, resulting in
clear separation of direct from reflected signal waveforms
much of the time. The paper discusses the experimental setup, the data processing steps and the findings of the
investigation to determine feasibility and accuracy of this
new type of altimetric measurement. Thermal-noise error
contributions of 1 cm in 40 seconds can be inferred based on
the analysis of carrier phase signals. Delay measurements
using the Coarse Acquisition (CA) signals give 1-cm thermalnoise error in about 13 hours. These measurements uncover
the systematic instrumentation, processing, and modeling
errors germane to future airborne and spaceborne
measurements over the ocean.
1. INTRODUCTION
The possibility of using the GPS signals scattered off the
ocean and sensed by an air- or spaceborne receiver in a
bistatic radar geometry, has been explored recently as a
means of doing ocean altimetry and scatterometry [ 1, 21. By
considering the constellation of 27 GPS transmitters and one
such receiver a multistatic system is obtained, capable of
intercepting reflections from several areas of the ocean
simultaneously. By extension of the traditional altimetry
approach, the bistatic path delay can be analyzed to derive the
important descriptors of the ocean surface such as height,
surface wind and significant wave height.

0-7803-6359-O/OO/$lO.OO
0 2000 IEEE

To assess the feasibility of this new type of measurement

(potentially very valuable), we need to understand its
accuracy as a function of the sea-state, the many relevant
system parameters and the observational geometry. To
accomplish this, it is advantageous to first perform a simple
altimetric experiment, with the minimum of complications,
where the receiver is not air-borne but rather at a fixed
location over a lake. The choice of a small lake reduces the
surface roughness effect when compared to the ocean,
providing experimental conditions closely resembling those
for the reflection of a plane wave from a (locally) planar
surface. Particularly true at low elevation angles. This set up
allows us to understand what are the limiting systematic
measurement errors associated with the technique.

2. EXPERIMENTAL SETUP
Crater Lake (located in Oregon) is a body of water
approximately round in shape, with a diameter of -10 km,
surrounded by steep rocky sides, and at a nominal altitude of
6178 ft. For our experiment, a large distance between the
receiver and the water is desirable in order to separate the
direct GPS signal from the reflected one, at least part of the
time. We are primarily interested in observing the GPS signal
scattered in the forward direction, where the return is the
strongest. Furthermore, it is desirable to minimize the effect
of the rocky sides on the signal multipath and to prevent the
collection of reflections coming from regions too close to the
rock (not representative of reflections from water).
The experimental equipment was placed on a rock, near
the Cloudcap lookout, directly overhanging the water at a
nominal altitude of 7780 ft. The relative height of the receiver
above the water was approximately 1602 ft, or about 489 m.
A Dorne-Margolin (choke ring) RHC-polarized GPS antenna
was oriented with its axis almost horizontal, looking across
the lake. The L1 GPS signal from this antenna, containing
both the direct and reflected portions, was fed into the frontend of a modified Turbo-Rogue GPS receiver (it only

2864

sampled the GPS signals). The sampled data was recorded

onto an AIT tape using a Sony SIR-1000 1-channel digital
data recorder at the sampling rate (20.456 MHz). A similar
zenith-pointing antenna, was connected to a standard TurboRogue GPS receiver. The GPS signals tracked by this second
combination provided measurements for direct positioning.

3. REFLECTION GEOMETRY
As the GPS satellites rise and set in the sky, the points of
specular reflection on the lake move at a non-uniform rate
along approximately straight lines, if we assume a fixed
receiver. The rate of motion depends on the elevation angle E.
At very shallow elevations the Rayleigh criterion is presumed
to be satisfied and a strong coherent signal is expected. Our
experimental set-up favors this situation since the direct and
reflected signals would arrive along paths almost aligned with
the main antenna axis (maximum gain). Also, the Fresnel
reflection coefficients for RHC-polarization (Fhc=Fpm+FF')is
greatest for low elevation angles. The drawback, however, is
that the separation in range between the direct and reflected
signal, R,p=2XH~~in(~),
becomes negligible, and makes the
data analysis considerably more difficult.

Doppler parameterize the model signal. During this time a

search is performed over a range of possible values for the
geometric parameters and is complete when the output of the
cross-correlation is maximized. The search simulates the
operations of a GPS receiver. Specifically, it simulates the
phase-locked loop, which first determines the value of the
delay, and then the Doppler, necessary to lock onto the peak
of the direct signal and keep its phase at 0'. Hence, every 20
msec, we have a time series for amplitude and phase,
encompassing both direct and reflected signals, which has
one value every (20.456~10~)"
seconds.
4.1 Recovering Height From Phase Data
From the complex time series one can locate the
maximum of the reflected signal by observing the occurrence
of the strongest change in the real and imaginary parts as a
function of time, i.e., at each 20 msec interval. By contrast,
the direct signal presents the weakest variation at the peak. In
fact, the phase-locked loop maintains the imaginary part
equal to zero at the peak. An example time series for phase
data is illustrated in Fig. 1, along with a model, for reflection
geometries associated with an elevation angle of 7'.

At high elevation angles scattering is expected to become

more diffusive, resulting in a decrease in signal strength and
change in polarization. The additional penalty of reduced
antenna gain at high elevation angles would lead to a further
reduction in received signal strength. Furthermore, as the
scattering becomes diffuse the footprint of the collection area
on the lake will increase from the Fresnel ellipse (order 10 m)
to the iso-range ellipse (order 1 km) with the possibility of
signal contamination from surfaces other than the lake. From
examination of the USGS area map, we inferred that the
rocky face between the chosen location and lake-shore had a
slope of about 45' with respect to the local horizontal. Since
this is a rough surface, the possibility that it is a source of
reflections that interfere with the received lake reflection
increase as the elevation angle approaches 45'. In fact, due to
multipath propagation, contamination from rock reflections is
expected to arise even before the onset of diffuse scattering.
The above considerations guided us in establishing priorities
for data analysis. Specifically, we selected reflections that
were as close as possible to the antenna pointing axis and
corresponding time intervals where elevation angles were
between about 5' and 25'.

.-$!

-50

52 -100
-150

-2oot'
1200

"

1400

'

'

'

1800

'

"

"

1800

"

"

2000

"
2200

time points (20msec)

Fig. 1: Phase data (broken line) versus model (continuous

line) for a 2 second sequence.
The phase is very well behaved over long time periods,
and this provides the strongest evidence that we observed
coherent reflections. To understand the phase behavior, we
constructed a model based on the assumption that the
reflected signal is simply a 'copy' of the direct one, modified
by a delay, reduction in amplitude and shifted, such that

Rec = A, e"'"A(t - t d )+ A , eio"A(t- t,)

(1)

4. DATA ANALYSIS PROCEDURES AND RESULTS

A description of the steps for processing sampled GPS
data is outlined in [2]. The first step consists of performing
cross-correlations between the sampled data and a model
(replica) of the expected GPS signal. The cross-correlations
are based on an integration time of 20 msec, while delay and

2865

where the complex received signal Rec is a function of lag t,

the receive time of the direct signal r,,, the receive time of the
reflected signal t , (r,- td=Rse,,/c),the amplitude of the direct
signal A, and the amplitude of the reflected signal A,. The
symbol A represents the GPS correlation function which

peaks at t = td for the direct signal. By referencing the phase

to the direct signal, we can write the phase as

Q, = arctan

Ar
Ad

Ad

'1

- t r ) sin{@ (tro- td)>

-td)
A(tr*- t r ) cos{w(tr~
-td)}
- td>

(2)

phase data, but may be more closely related to data used in

ocean applications. The parameters estimated were 1) an
overall phase, 2) amplitude Ad, 3) amplitude 4, 4) a lag
offset embedded in r,., 5 ) arrival time difference r,- rd. The
height is calculated using this last parameter.
Fig. 2 shows the amplitude data and the best model fit as
a function of lag for a single 0.02-second interval:

where we allow the phase to be evaluated at the time t,,, to

account for an error in the determination of r, i.e., the true
peak of the reflected signal. The ratio of amplitudes
multiplying the sin and cos terms is time dependent and will
decrease with increasing separation between peaks. The ratio
is not known a priori and its precise value is obtained in the
solution process.
The phase model is constructed by converting the
modeled range difference to phase, i.e., by multiplying by
2rr: /A, where h is the wavelength of the L1 GPS signal. This
forms the model for the argument in the sin and cos terms. In
this process a reference time is selected for the phase model,
resulting in a phase offset with respect to the data; the value
of this offset becomes a parameter to be determined in the
solution process. By differentiating the phase with respect to
time, we obtain a spectrum of Doppler frequencies, which
depend on the reflection geometry and can be used as a
detector of the surface height.
The first basic issue in this analysis is to determine the
sensitivity of the measurement to height variation. This
implies determining the thermal noise present on the data and
mapping it to an uncertainty in the Doppler and eventually to
a vector of uncertainty in the height. Consequently, we
analyzed the variation of the Doppler with height using our
model. Typical values of oo = 0.1 radians were obtained for
20 msec intervals, resulting in oA= 0.796 Hz,corresponding
to about 800 meters of height uncertainty. To improve on the
accuracy, a number of measurements (N) must be processed;
the corresponding improvement on oA scales like N-3'2 for
thermal noise. In our case the thermal-noise error is reduced
to 10 cm in 8 sec and to 1 cm in 37 sec; this error does
not account for expected systematic error sources.

4.2 Recovering Height From Delay Data

A second approach to estimating the height of the
receiver above Crater Lake involves fitting the complex
correlation (1) for r,- td, the difference in arrival time of the
reflected and direct signal. The fit is done over lags for a
single time point, and then repeated for each 0.02-second
interval. Ultimately, both phase and amplitude data will be
used in this fit, but in this first attempt, we used only the
amplitude data for 0.02-second intervals. Fitting the
amplitude data yields less precise heights. Moreover, the data
are probably sensitive to a different set of errors than the

2866

50

40
30
20
10
0
-

1
-400-200

0
0
0
200 400 600
Lag Offset (m)

Fig. 2: Amplitude data (+) and best model fit (-) for a
single 0.02-second interval at about 7.5 degrees elevation.
The reflected peak is located at about 120 m, that is, within
the central peak. The asymmetry of the large peak is because
the phase of the reflected peak is almost 180" out of phase
relative to the phase of the direct, so it subtracts from the
amplitude near 120 m. Analyzing many time points suggests
that the amplitude data yields a thermal-noise height error of
1 cm in about 13 hours. Correlated-noise contributions which
could degrade performance are under investigation.
5.0 CONCLUSIONS

The Crater Lake experiment will determine the limiting

accuracy of GPS altimetry. Current thermal-noise errors of 1cm seem attainable within -40 sec, by fitting the phase-rate
data type, and within -13 hrs by fitting the delay data type.
Both will be fit, perhaps simultaneously, in the future in order
to better understand both the thermal and systematic error
budget for the Crater-Lake measurement. This error budget
will form the instrumental and geometric limiting error for
ocean measurements. It may also bear on the ocean physical
model, which must account for increased surface roughness.
6.0 REFERENCES
[l] G. A. Hajj, C. Zuffada and J.B. Thomas, "Theoretical
Description of a Bistatic System for Ocean Altimetry Using
the GPS Signal," in Press.

[2] S. T. Lowe, J.L. Labrecque, C. Zuffada, L.J. Romans, L.

Young and G. A. Hajj, "First Spaceborne Observation of an
Earth-Reflected GPS Signal," in Press.