Improving Time to First Fix for GPS Receivers
Z.N. Low and C.L. Law
School of Electrical & Electronic Engineering
Nanyang Technological University
50 Nanyang Avenue, Singapore 639798
e-mail: [email protected]
Keywords: GPS, TTFF, Acquisition, CRPA, Correlator
Abstract
In the Global Position System (GPS), receivers spend a long
time to acquire the signals required for position. A study to
determine the factors affecting time to first fix (TTFF) and
methods to improve it is being conducted. This paper
addresses two key aspects, which are the acquisition process
and antenna design. Studies on advance correlation
techniques such as FFT, XMC and Sqaured-D are being done
over the conventional sliding serial/parallel correlator. At the
same time, we also investigate the advantages of CRPA
antenna over FRPA antenna as well as other new antenna
designs. Simulation and actual data logging of the satellite
view at a specific location is being done to study the
possibility and feasibility of implementing a look up table for
improving cold start. An actual field-test on three different
types of receivers coupled with three different antennas with
different LNA gain is being done at two different location to
simulate clear sky and actual operating conditions, to address
the validity of TTFF specifications provided by the
manufacturer.
1 Introduction
The Navstar Global Positioning System (GPS) is a spacebased radio positioning and time transfer system using a
minimum of 24 satellites. GPS provides accurate position,
velocity, and time (PVT) information to an unlimited number
of suitably equipped ground, sea, air and space users. Passive
PVT fixes are available worldwide in all-weathers in a
worldwide. Two levels of service are provided by the GPS,
namely the Precise Positioning Service (PPS) and the
Standard Positioning Service (SPS). The SPS is available to
the general public while PPS is available to selected users
such as the USA DoD. Currently, GPS receiver users are
plagued with poor TTFF due to various reasons. Although, it
is generally claimed that most GPS receivers are able to
perform cold start within 45 seconds, the actual field test
reflects time require an average of 1 min 30 seconds, under
clear sky condition. Many new and novel techniques
addressing to this issue have been presented in the recent few
years.
2 Definition of TTFF
TTFF to defined into three different variations, namely the
cold start, warm start and hot start.
Cold Start
A cold start occurs whenever there is a problem with these
almanac and ephemeris data elements. This is typical of a
receiver as delivered from a manufacturer, supply depot, or
repair depot. Date and time will not be maintained if the
receiver’s "keep alive" battery has been removed or drained.
If the receiver clock and memory remains active, the last
known position might be at a factory or depot thousands of
kilometres from the present position, and the almanac may be
several months old. Under such conditions, the receiver may
have to systematically "search the sky" until it can find a
satellite and retrieve time and a current almanac.
Warm Start
A warm or normal start is based on the assumption that the
receiver has an estimation of current time and position as well
as a recent copy of the satellite almanac data. Typically, time
should be known within 20 seconds of GPS time, position
determined within 100 kilometres, velocity within 25 metres
per second, and the satellite almanac should have been
collected within the past few weeks.
Hot Start
A hot start occurs when a receiver is provided with a standby
feature to maintain oscillator temperature, time, position, and
individual satellite ephemeris (as well as the almanac) or
powering down for less than 30 minutes. When the receiver is
commanded out of the standby mode, the time required to
achieve the next full position fix is usually Termed Time to
Subsequent Fix (TTSF) rather than TTFF. Typically, TTSF is
in the order of seconds to 10s of seconds for standby periods
of a few hours.
3. Acquisition Algorithms
GPS signal acquisition is a search process. The C/A code
dimension is associated with the replica code. The Doppler
dimension is associated with the replica carrier. The
uncertainty in both Doppler and C/A code phase suggests that
a 2-dimensional uncertainty region must be searched in order
to locate the received signal (Figure1). When the GPS
receiver has no almanac available (called by cold start), the
code uncertainty region is 1023 chips, and the Doppler
uncertainty region is about –11000Hz ~ +11000Hz typically.
demodulated in-phase and quadrature phase signals are mixed
with the combined replica of three C/A codes for despreading
the SS signal.
Figure1: Code/Doppler uncertain region
2.1 Serial/Parallel Acquisition
In a serial search, the searching procedure is executed one
cell at a time in sequence, while in a parallel search (Multicorrelator), groups of cells (M cells) are tested
simultaneously. Typically, the code bin size is 1/2 chip of
C/A code. The Doppler bin size is about 2/(3T), where T is
the PDI (pre-detection integration) time, which is usually take
to be 500Hz. The input signal is the combination of the
satellite signal (s(t)) and the noise (n(t)). This input signal is
mixed with the generated carrier and code, and passed
through an integrated-and-dump (IDF) with a pre-detection
integration time of T (Figure 2). For a fixed dwell scheme, a
large number, NB, of squared-law-detected samples of the
coherent integration are accumulated, summed, and compared
with a preset threshold. Thus, it tests 2046 half chips on the
code phase dimension for 41 carrier frequency search steps
(bin resolution=500Hz). This means searching approximately
80000 combinations, which is time consuming. We can
approximate the total number of computations to
41x2046x2046 additions and multiplications. In the case of a
parallel search (in the multi-correlator), several cells (M) are
tested concurrently for signal presence.
Figure 3: One correlation arm of XMC
It is noted that in [1] the pre-detection integration power in
XMC is stronger than that in conventional (single/multiple)
correlator. If D is a half chip (about 0.5 msec), then this value
is quadrupled. In XMC, several code phases can be tested in
just one correlator arm. Therefore, the number of total code
bins is decreased in XMC. [1] found that the number of total
code bins is about 71.5 % of that of conventional
(single/multiple) correlator through experiments results.
2.3 Squared-D
Squared-D is a new method for reducing the number of
Doppler bin and is based upon a signal squaring method for
L2 codeless tracking. A block diagram of Squared-D
searching method is shown in Figure 4. The main feature of
Squared-D searching method is shown in the shadowed block
of Figure 4. This method is to square the L1 signal, that is, to
multiply it by delayed itself, to remove the bi-phase C/A code
modulation and result in a demodulated. Therefore, this
correlator cannot detect the SV number because the C/A code
is removed. However, it can find twice of Doppler frequency.
Figure 4 Squared-D Searching Method
Figure 2: Serial/Parallel Acquisition
2.2 XMC (eXtended Multiple Correlator)
In XMC, the combined replica of several codes is used
instead of separate ones. A block diagram of XMC
correlation arm is shown in Figure 3. The main feature of this
correlation technique is shown in the shadowed block of
Figure 3. Differently from other correlation techniques, the
If GP2021 is used in the RF/IF chipset, the Squared-D
correlator has 12 channels, 8 visible satellites are presented,
and the Doppler uncertainty region is about –11kHz ~
+11kHz, it takes about 3.2 sec for finding Doppler candidates
[1]. By reducing the number of Doppler bin to 8, and the GPS
receiver using this Squared-D searching method can acquire
the GPS signal dramatically faster.
2.4 Fast Fourier Transform with DSP
The acquisition process time is shortened when the linear
search algorithm can be performed using FFT search
algorithm. Here 2046 samples of the received data x[n], are
correlated with the replica code, C/A[n], using different
Doppler shifts (Figure 5). Therefore, for a 500Hz resolution
for +/- 10Khz, we need to do 41 times of similar operations.
Figure 5 FFT Acquisition
There are three basic types of GPS antennas, a passive Fixed
Radiation Pattern Antenna (FRPA), a FRPA with an
integrated preamplifier, and a Controlled Radiation Pattern
Antenna (CRPA). The requirement to drive a long cable run,
with its associated signal loss between the antenna and the
GPS receiver has resulted in a FRPA with an integrated
amplifier. A CRPA is required to reduce the effects of RF
interference, which would otherwise jam the receiver's
operation. On top of it, novel designs such as the 3D choke
ring ground plane antenna as well as the Pinwheel antenna
with better performance are being presented.
3.1 CRPA System
CRPAs have been shown to be the only effective means of
protecting GPS receivers against multiple wideband jammers.
A CRPA has two components: an Antenna Control Unit
(ACU) and an antenna array (Figure 6, Ali System
Inc.).
The Fast Fourier transform algorithm is used to implement
the DFT and IDFT. The input signal FFT is being performed
on the input signal x[n] as well as the replica code, C/A[n].
After the FFT the two signals are mixed and an IFFT is
performed to the output. At every output of a operation we
get a similar output done by a 1 set of serial search algorithm
of a Doppler bin. The total number of computation is reduced
to 41x(2log24028+1)x4028 of additions and half the
multiplications. Although the total number of computation is
being reduced, the load on the processor is increased over a
large margin.
With the advent of powerful Digital Signal Processing (DSP)
chips, software receivers processing techniques and highly
augmented designs are becoming possible. DSP chips are
capable of performing long FFTs in very short times provide
the basis for employing sophisticated signal acquisition
techniques. Data collected in [2] is shown in Table 1. Using
TMS320C6414-600 DSP by TI as their baseline processor for
time estimate, it is clearly shown that it is able to simulate
2657 real time correlators, taking only 0.033 seconds for each
satellite search. Therefore, it is clearly seen that a full 32satellite search would take slightly less than 1 second.
Figure 6: CRPA Antenna Array
The ACU controls the array's radiation pattern by adjusting
the gains and phase from each antenna. It contains a series of
amplifiers and gain control systems for each channel, a set of
weights that make up a beam former and a microprocessor
and associated electronics that contains the control algorithm
and drive the weights in the beamformer. Each weight is a
phase shifter with gain control. A CRPA has one less degree
of freedom than the number of elements (N), allowing 1
independent jamming sources to be cancelled.
Test results shown in [3][4][5][6] indicate significant
improvement in CNR over conventional FRPA. An extract
from [3] illustrates the significant improvement is presented
in figure 7.
Table 1: FFT search performance using DSP chip
3 Antenna Design
two consecutive spiral arms is 30°; as well the electrical
phase length of the feeding network is set to 30°.
Figure 7: CRPA performance over FRPA
3.2 3D Choke Ring Ground Plane Antenna
A typical choke ring antenna is machined from a single billet
of aluminium. It consists of three to five concentric ring
structures (A,B,C,D of figure 8, source, Novatel Inc). The
choke rings are usually a quarter wavelength deep, in order to
create a high impedance surface that prevents propagation of
surface waves near the antenna and excitation of undesired
modes. The net effect is a very smooth controlled pattern with
low susceptibility to multipath. To improve the reception of
low elevation angle satellites, consecutive adjacent rings are
lowered (in z-plane) with respect to each other to create a
“pyramid”-like structure (E of Figure 8), in order to move the
apparent line-of-sight that joins the top of the rings from
horizon level to some angle below the horizon.
Figure 9: Pinwheel Antenna Design
The pinwheel antenna is made out of a flat printed circuit
board (PCB), with the upper layer being the ground plane
layer. The photonic crystal (air gaps in the substrate)
approach is implemented with vias located outside the
perimeter of the antenna. The PBG structure is implemented
with the use of eleven concentric slot rings located between
the vias and the outside edges of twelve spiral arms. This
antenna offers a performance similar to one achieved with a
choke ring antenna, but with much reduced size and weight.
4. Data logging of Satellite View
A data log of the satellites in view from the roof top of
PWTC at N 1°20’34.1022, E 103°40’45.7147, 83.52m, is
being conducted using the combination of simulated data
after office hours and actual data from the DGPS station at
PWTC, from 9th of December to 11th of December. Office
hours timing starts from 0030hrs to 1030hrs UTC. The
simulator being used is the GS1010, GPS L1 C/A Code
Dynamic Test Simulator at PWTC, shown in figure 10.
Figure 8: Various Choke Ring Antenna Design
Actual Field-testing results in [7] shows that the 3D choke
rings provide the best tracking and lowest susceptibility to
multipath. At the same time, a cumulative tracking capability
from 0 to 10degree elevation test is being done with 10%
improvement in low elevation tracking being observed.
3.3 High Performance GPS Pinwheel Antenna
The new, patented Pinwheel antenna discussed in [8][9], is an
array of multiple spiral slots that are electromagnetically
coupled to a feeding network. Figure 8 illustrates a board
layout of a 12-arm spiral pinwheel antenna. This antenna is
optimized for single frequency of operation; hence all spiral
arms are equal in length. The spatial difference between each
Figure 10: GS1010 GPS Simulator
The other set of simulated results is obtained from the website
of Interactive NAWCWPNS GPS/INS Section, GPS Satellite
Prediction. During office hours, the data is being
crosschecked every half an hour to ensure the validity of the
results. After office hours results are complied from the two
sets of simulated results, which display exact match for the
three days. The simulated data is then complied with actual
logged data during office hours to form the full 24 hours
satellite view data across 3 whole days. A chart illustrating
the view on 10th Dec 2003 is shown in figure 11.
In conclusion, we propose a soft look up table system. This
means that the receiver would automatically correct the look
up table information once it has received the full almanac
data from the satellites on a periodic basis. It would be done
in background when the receiver is providing positioning. By
computing various regions predicted satellite view, a user just
require to input the region he is in. For example when he
landed after a plane journey, a significant improvement in
time to first fix could be observed.
5. Actual Field Testing
th
Figure 11: Satellite View of 10 Dec 2003
From the 3 sets of charts complied, it can be observed that the
satellite view at different day is rather consistent and with
minor drift. This can be accounted for due to the fact that the
period of the satellite orbit is 11hours and 58minutes.
Therefore, it made two rounds around earth in 23hours and
56minutes. The drift is not extremely significant, as by using
the same chart, we are still able to predict 7 out of the average
of 10 satellites in view all of the time even after 1 month,
which is more than the minimum of 4 required for 3D
positioning
Another simulated data logging is being conducted to
investigate the consistency of the satellite orbit. Simulated
satellite view at PWTC rooftop is being logged on the 15th of
every month at 0000hrs UTC across a span of 4 years, from
2001 to 2004. The data is being crosschecked and consistency
is being proven. It is observed that the pattern of satellite
view is highly consistent and is in a cycle of 11 to 13 months
for different satellites. With such discrepancies, it is clearly
noted that even spanning across 4 years that we can still can
get a similarity of 6 out of 10 satellites in view, which is more
than the 4 satellites required. By using a resolution of half an
hour, a look up table for monthly-predicted satellite view can
be used in to improve the TTFF for cold start. The number of
bits required to store the predicted satellite for a single
location is 48x32x12, which is 18432bits, 2304 bytes. A 2MB
memory module would be able to store up to 900 different
locations. Since the look up table is only an approximate
guide for faster cold start, and a estimate of the GPS satellite
footprint would be approximately the whole of Southeast
Asia and Australia. Thus, one set of data is good enough for
the whole region of for example, Singapore, Johor and
Batam.
A simulated test is also being conducted on the Mitel GPS
Architect Development Kit. This is done by modifying the
sample codes provided by Mitel. The original codes, search
satellite ID 1 to 12 on a cold start basis. An array is created
storing the look up table for a day. On the event of a cold
start, the receiver would load the look-up table data and take
priority in acquiring the predicted satellites in view. A
tremendous increase in TTFF of 3D position of an average
time of 2minutes 57.4seconds to only 34.6 seconds is being
observed.
Actual field-testing is done at two different locations in NTU.
The first location is at the 10th storey rooftop of PWTC
shown in figure 12. This is selected to simulate full sky view
with as little multipath condition as possible. The antenna is
further mounted on top of the water tank above all
surrounding walls to ensure a clear view is obtained.
Figure 12: Setup of Test at PWTC rooftop
The other location is at the EEE carpark between block S1
and S2 show in figure 13, this is to simulate a typical start
point for most GPS users, as they would most probably start
up their receiver when they leave a building or drive off from
a carpark. The antenna is mounted on top of the pole shown
in the right side of figure 13, this is to ensure a decently good
view of the antenna. Due to lack of power source, two 2-hour
UPS are being used to provide power to the computer and
receiver for the whole test.
Figure 13: Setup of Test at EEE carpark
During the indoor simulation using the GS1010, it is noted
that cold start timing is getting faster and faster for the
Ashtech and Mitel receiver. Therefore, for every test for cold
start, fictitious data is being fed into the receiver to ‘blind’
the receivers effectively.
Hardware
Three different antennas and three different receivers with
different specifications are being selected to test out their
respective performance. However, no FFT correlator based
receiver is available.
Receivers
The three different receivers selected are Mitel GPS
Architect, Ashtech G12 GPS board and SiRFstarIIe
evaluation kit. They are selected due to their correlator
architecture and their claimed performance. The test is to
determine how accurate their TTFF claim is and the
advantage of using parallel over single correlator.
Antennas
The three different antennas selected are as followed with
their brief specifications obtained from the manufacturers’
website. Basically all 3 of them are patch antenna design but
with different LNA gain. This is to study the significance of
an antenna with higher LNA gain for TTFF.
Test Results
From an overview of the results presented in table 2, it is
clearly noted that the manufacturers are not being accurate on
their specification on TTFF. It is significantly longer than the
stated value even during open sky condition on the rooftop of
PWTC. The strobe correlator which mentioned in [10] is
suppose to reduce multipath conditions. However, no
significant advantage of TTFF is being observed during the
test between S1 and S2 block.
At the same time, there is a significant difference in TTFF
with respect to the antenna LNA gain. With a 2.5dB gain
with respect to the Matsushita antenna and the AeroAntenna
Technology antenna, we observed an average of reduction in
TTFF of approximately 2 minutes. Therefore, we can
conclude that by providing a higher gain LNA and improving
the SNR/CNR, a significant improvement in TTFF can be
observed, as it reduces the probability of false alarm and
misses.
PWTC Rooftop (All readings are in seconds)
Cold
Warm
6. Conclusion
Acquisition time can be reduced by a large margin, by
various techniques mentioned. The most significant technique
is by using FFT on DSP chips. At the same time, by using
CRPA over FRPA, CNR is improved by a considerable
amount. This increases the probability of acquisition as well
as mitigating the effects of intentional jamming. However, the
greatest impact for TTFF is the slow data rate of 50bps from
the satellite. No matter how much faster our acquisition
process may be, we still need to wait for the first 3 subframes
of the first frame to be acquired from the satellite. This takes
18 seconds minimum and 30 seconds maximum.
We also propose, the look up table technique to enhance cold
start timing. It is fast, as it requires no computations and
minimum memory space. However, more actual tests and
simulation should be done to fully determine its feasibility in
implementation.
Actual field-testing have also indicated that the manufacturer
specifications on TTFF is not accurate. Hence it is no
advisable to be used as a guideline when purchasing a
receiver.
References
[1] S.S. Hung, P. Chansik and J.L. Sang, A New Fast Acquisition
Algorithm for GPS Receivers
[2] L.Scott Rapid Signal Acquisition Techniques for Civilian &
Military User Equipments Using DSP Based FFT Processing,
ION Proceedings, (2001)
[3] A.Brown and R.Silva A GPS Receiver Designed For CarrierPhased Time Transfer, Navsys Inc
[4] A.Brown and D.Morley, Test Results of a 7-Element Small
Controlled Reception Pattern Antenna, ION Proceedings, (2001)
Hot
Ashtech
555
375
210
383
25
180
18.6
20.8
126
SiRF
254
305
116
27
10
22
15
4.4
0
Mitel
449
208
138
-
-
-
-
-
-
EEE Carpark Between S1 and S2
Ashtech
to design and build the most cost effective receivers tiwht
better performance.
X
810
123
X
70
528
X
44.2
30.6
SiRF
490
288
130
20
55
15
28.8
11
0
Mitel
X
200
142
-
-
-
-
-
-
: Function not supported by receiver
X
: Unable to acquire after 20 minutes
There are 3 columns of reading, 1st column represent antenna gain of 20dB,
2nd 26.5dB and 3rd 29dB
Table 2: Field-testing results of TTFF
More extensive field test on more varieties of receivers and
antenna designs can be conducted to further investigate their
performances. This would enable developers of GPS system
[5] A.Brown and N.Gerein,
Test Results of a Digital
Beamforming GPS Receiver in a Jamming Environment, ION
Proceedings, (2001)
[6] W.Kunysz Advance Pinwheel; Compact Controlled
Reception Pattern Antenna (AP-CRPA) designed for Interference
and Multipath Mitigation, ION Proceedings (2001)
[7] W.Kunysz A three Dimensional Choke Ring Ground Plane
Antenna, Novatel Inc.
[8] W.Kunysz, A Novel GPS Survey Antenna, Novatel Inc.
[9] W.Kunysz High Performance GPS Pinwheel Antenna, ION
Proceedings, (2000)
[10] L.Garin and J.M. Rousseau Enhanced Strobe Correlator
Multipath Rejection for Code & Carrier, ION Proceedings, (1997)