Copyright (c) 2009 IEEE. Personal use is permitted.

For any other purposes, Permission must be obtained from the IEEE by emailing [email protected].
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
1
Reducing Front-End Bandwidth may Improve Digital GNSS
Receiver Performance
James T. Curran, Daniele Borio, Colin C. Murphy, G erard Lachapelle
AbstractA novel evaluation of ltering and quantisation losses
for weak DS-CDMA receivers is presented. Using this method, joint
optimisation of lter center frequency and bandwidth is conducted for
one-, two- and three-bit quantisers. It is demonstrated that a joint loss-
analysis of these effects is necessary to optimise front-end design.
Index TermsCDMA, Correlation Receiver, Front-End Filtering, GPS,
Quantisation.
I. INTRODUCTION
M
ODERN global navigation satellite system (GNSS) receivers
are almost invariably digital receivers, being either dedicated
digital hardware receivers or, more recently, hybrid hardware/software
receivers [1] [2]. Typically, such receivers use a low resolution
quantiser (e.g. one or two bits) which results in a loss of signal
quality. The impact of this loss on the performance of a receivers
digital matched lter (DMF) has been of interest for many years (eg.
[3], [4]), and is further examined here. Previous work has focused on
the performance of baseband signals which are distorted by additive
white Gaussian noise (AWGN) [4]. In the context of GNSS, recent
simulation-based work has examined these effects in direct sequence
spread spectrum (DSSS) signal reception[5]. The impact of ltering
on the performance of a GNSS receiver has also been examined [6].
However, a joint analysis of ltering and quantisation appears to be
absent in the literature.
A joint analysis of ltering and quantisation loss has been devel-
oped in [7] which draws on early work on the effects of nonlinear
operations on correlated noise [8], [9], [10]. This loss model is
used here to examine the relationship between lter and quantiser
design parameters and the resulting DMF performance. Novel results,
pertaining to the optimum choice of lter bandwidth, are presented.
Section II introduces the signal model and denes the operation of
the down converter, lter and quantiser. The integrate-and-dump (ID)
matched lter and the characteristics of the AWGN are also dened.
An equivalent model of the receiver processing operations is dened
in Section III and the evaluation of key statistical properties of the
processed signal is detailed. Section IV denes the DMF performance
metric used in the analysis and presents the simplied front-end
lter design parameters. Theoretical results derived from the analysis
dened in Section III, for a range of receiver congurations, are
presented in Section V and are substantiated through Monte-Carlo
simulation in Section VI. A real signal validation of the theoretical
and simulation results is subsequently presented in Section VII.
Manuscript received July 6, 2009; revised September 29, 2009. This work
was supported by the IRCSET Embark Initiative, and by the PLAN Group.
Copyright (c) 2009 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to [email protected].
J. Curran and C. Murphy are with the Department of Electrical Engi-
neering, University College Cork, Ireland, e-mail: [email protected],
[email protected]
D. Borio and G. Lachapelle are with the Position Location and Navigation
Group (PLAN), University of Calgary, Canada, e-mail: [email protected],
[email protected]
II. SIGNAL AND SYSTEM MODEL
This section presents a model of the satellite signal and the
receivers front-end. The system model used in this analysis assumes
that a single satellite signal, distorted by AWGN, is incident on
the receivers antenna. Typically a receiver will perform initial
amplication and frequency selection at RF and, subsequently, mix
the signal to IF, in one or multiple stages, in the analogue domain
[1]. The signal incident on the antenna can be expressed as:
rRF (t) = sRF (t) +n(t)
sRF (t) =

2P sin(RF t +) c (t ) d (t ) , (1)
where P is the received signal power and RF and represent the
RF carrier frequency and phase respectively. Functions c (t ) and
d (t ) respectively model the signal spreading code sequence and
data sequence, delayed by time . For the sake of clarity, the data
term is omitted from the following expressions. The additive noise,
n(t), is white and has a two-sided power spectral density (PSD) of
N0/2 W/Hz.
The result of ideal frequency selection and down-conversion to
intermediate frequency (IF) is a continuous signal, rIF (t), centered
at . The center frequency, , is equal to the nominal IF center
frequency, IF , plus a residual Doppler frequency. This signal is
sampled at a rate of FS Hz (where the sample period is denoted TS),
producing the discrete time sequence, rIF [n]. Finally, this ltered
sequence is quantised by a B bit quantiser, Q
Ag
B
[x], producing the
signal:
r [n] = Q
Ag
B
[rIF [n]] , (2)
where the quantiser is dened as [11]:
Q
Ag
B
[x] =
_
2
B
1
_
+ 2
L

i=L
u(Agx i) , (3)
where Ag is a gain applied to the operand prior to quantisation, u(t)
is the unit step function and L = 2
(B1)
1.
Generally, a DMF is applied to this signal, which produces samples
at a rate 1/TI, where TI is an integer multiple of the spreading code
period. This lter, (or correlator), shown in Figure 1, operates on N
successive samples of r [n]. Estimates of , and , respectively
represented by ,

and , are adapted by the receiver so as to
maximise {y [k]}, the real part of the correlator output y [k].
)}

( exp{ 2 +
S
nT j ) (
S
nT c
] [n r ] [k y

N
1
Fig. 1. Block Diagram of Correlator
Authorized licensed use limited to: University of Calgary. Downloaded on January 20, 2010 at 15:27 from IEEE Xplore. Restrictions apply.
2009 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material
for advertising or promotional purposes or for creating new collective works for resale or redistribution to
servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the
IEEE.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing [email protected].
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
2
III. ANALYSIS
This section develops a mathematical model of the receiver signal
processing operations, from the antenna to the correlator output. The
performance of acquisition algorithms and tracking loops can be
directly related to the mean and variance of the correlator output
[12] and, therefore, it is these two moments that are of interest here.
A. Equivalent Front-End Model
To simplify the analysis of the receiver signal processing opera-
tions, it is convenient to describe this processing in one domain. The
ltering stage, therefore, is represented by an equivalent digital lter,
applied to the signal directly after sampling [13].
A block diagram of this equivalent front-end is given in Figure
2. The signal is amplied and frequency translated after the antenna
to produce a wideband signal centered at IF . Next, a zero phase,
brick-wall, anti-alias lter with a null to null bandwidth of FS/2
and centered at FS/4 +kFS, (k N), is applied to the signal [14].
This signal is sampled to produce the signal rw [n]. The signal is
then ltered by a digital lter, H
f
_
e
j
_
, to produce the signal
r
f
[n]. The subscript w, implies that the signal is white or un-
ltered signal, while the subscript f implies that the signal has been
ltered. The transfer function, H
f
_
e
j
_
, models signal ltering
incurred during the downconversion stages. The ltered digital signal
is then quantised to produce the nal IF signal, r [n], which can be
represented using discrete functions alone.
The signal rw [n] can be expressed as the sum of a signal
component, sw [n], and a white noise component, nw [n], which are
mutually independent. The signal r
f
[n] (see Figure 2) is expressed
as:
r
f
[n] = (rw h
f
) [n]
= (sw h
f
) [n] + (nw h
f
) [n]
= s
f
[n] +n
f
[n] (4)
where (a b) [m] is the convolution of a [n] and b [n m] and h
f
[n]
is the impulse response of the lter H
f
_
e
j
_
. Quantisation of the
ltered signal is expressed as:
r [n] = Q
Ag
B
[r
f
[n]]
= Q
Ag
B
[s
f
[n] +n
f
[n]] . (5)
This signal can be expressed in terms of a mean signal component,
rs [n], and a zero mean noise component, rn [n], such that:
rs [n] = E[r [n]]
rn [n] = r [n] rs [n] (6)
) (t r
] [n r
S
F
) (
j
f
e H

IF
Filter
] [n r
w ] [n r
f
] [x Q
g
A
B
Quantiser
Mixer and
Filter
Anti-alias
Amplifier,
Fig. 2. Block Diagram of Equivalent Front-End Model
B. Moment Evaluation
The correlator depicted in Figure 1 can be expressed [7] as the
convolution of r [n] and the impulse response:
hc [n] =
w[n]
N
c ((N n) TS )

2e
j( ((Nn)T
S
+kT
I
)+

)
, (7)
where:
w[n] =
_
1 for 0 n < N
0 otherwise
. (8)
Using this denition, the mean of the in-phase correlator output,
denoted i, can be found as:
i = {(hc rs) [0]} . (9)
The correlator output noise power, denoted
2
i
, can be evaluated in
a similar manner [7]:

2
i
=

n
Rn [n] {Rc [n]} (10)
where Rn [n] and Rc [n], respectively, denote the autocorrelation
functions of the correlator impulse response and the received noise.
C. Filtering and Quantisation Effects
In order to utilise (9) and (10), it is necessary to evaluate the signal
mean, rs [n] and the autocorrelation function, Rn [n]. Assuming that
the received signal-to-noise ratio (SNR) is low, it has been shown
[11] that the mean quantiser output is well approximated by:
rs [n] s
f
[n] KQ, (11)
KQ =
2
_
2
2
n
f
_
1 + 2
L

i=L
exp
_
i
2
2A
2
g

2
n
f
__
where
2
n
f
is the variance of the noise incident on the quantiser (given
by (13) evaluated at m = 0). For one-bit quantisation, (11) is, in fact,
exact, whilst for multi-bit quantisation this expression is based on the
approximation of integration by a summation [11]. Experimentally,
it has been found that this approximation is very accurate (1%)
for low SNR (< 18 dB) and gradually diverges for higher SNR
conditions (approximately 10% at 7 dB), for two- and three-bit
quantisers.
Being a memoryless operation, the effect of quantisation is simply
to attenuate the signal mean. This attenuation is a function of the
number of quantisation levels, the input noise variance and the gain,
Ag, and is represented here by the gain KQ.
Substitution of (11) into (9) yields:
i =

PKQ max

, ,

_
e
j()
(R
Code
( +nTS) h
f
[n])
_
=

PKQKF , (12)
where R
Code
() is the autocorrelation function of c (t), =

and = . The variable KF represents the attenuation of the

signal mean due to ltering. Equation (12) can be used to estimate
the mean of the correlator output.
From (4), Rn
f
, the autocorrelation of the noise incident on the
quantiser, is given by [7]:
Rn
f
[m] =
N0FS
2
R
f
[m] , (13)
where R
f
[m] is the autocorrelation of h
f
[n]. Unlike the signal mean,
rs [n], the noise component of r [n] is distorted by the non-linearity
of the quantiser [9]. Previous work on nonlinear operations on noise
[9] was employed in [7] to determine the autocorrelation of the noise
component of the signal after quantisation, R(), as a function of
the autocorrelation function of the noise incident on the quantiser,
denoted ():
R() =
2

i=L
L

k=L
_
()
0
exp
_

1
A
2
g

2
IF
i
2
+k
2
2rik
2[1r
2
]
_

1 r
2
dr. (14)
Authorized licensed use limited to: University of Calgary. Downloaded on January 20, 2010 at 15:27 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing [email protected].
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
3
0 0.1 0.2 0.3 0.4 0.5
0
0.2
0.4
0.6
0.8
1
Normalised Frequency
N
o
r
m
a
l
i
s
e
d

M
a
g
n
i
t
u
d
e

S
p
e
c
t
r
u
m
3 Bit
1 Bit
Continuous
2 Bit
Fig. 3. Effects of Quantisation on Magnitude Spectrum
A discussion and depiction of the relationship between correlation
coefcients of the noise incident on the quantiser and the quantised
noise for a range of quantisation levels is given in [7]. The effect
is a whitening of the signal, whereby power is spread across the
spectrum. The case of one bit quantisation, which can be compared
to extreme clipping, has been well documented in [10]. A qualitative
example of this effect is depicted in Figure 3 for a band-pass
continuous signal which is quantised with a variety of levels. It can be
seen that the magnitude spectrum is distorted as noise power spills
out of the band of the original, un-quantised signal.
For the purposes of this analysis, we dene a discrete time
correlation coefcient for use in (14):
[n] =
R
f
[n]
R
f
[0]
, (15)
and, under the assumption of low received SNR, where the autocor-
relation function of the signal r [n] is dominated by that of the noise
component, R[n], computed using (14). Thus, the variance of the
correlator output can be computed using (10).
In the case of GNSS signals, where the signal power occupies the
entire pass-band, the effective reduction in noise oor has a signicant
impact on receiver performance. It may be worthwhile, therefore,
considering this phenomenon during the design of a GNSS receiver
front-end. This issue is examined in the following sections.
IV. FRONT-END DESIGN
This section examines the impact of lter bandwidth, Bw, center
frequency, Fc, and, where applicable, the quantiser gain, Ag, on the
performance of a GNSS receiver. The theory developed in Section III
is used to evaluate the processing loss incurred by a range of ideal
lters and a number of quantisation levels.
To perform a comparative loss analysis, the performance of an ideal
receiver is chosen as a reference. This ideal receiver is assumed to
have an all pass front-end lter, H
f
_
e
j2f
_
= 1, and quantiser
given by Q[x] = x. The SNR of the real channel, referred to as
the coherent SNR, can be computed as:
SNRi =

2
i

2
i
. (16)
In the case of the ideal receiver, the SNR, SNR
Ideal
i
, is given
[15] by: 2PTI/N0 and, thus, the processing loss associated with a
non-ideal receiver is expressed as:
L
Total
=
SNRi
SNR
Ideal
i
= SNRi
N0
2PTI
. (17)
The lter H
f
_
e
j2f
_
is assumed to be a brick wall, zero-phase
lter. Although such a lter is non realisable, it provides a fair
comparison over a wide range of bandwidths and center frequencies,
as variations in gain roll-off and phase non-linearity need not be
considered. The frequency response of the lter is dened as follows:
H
f
_
e
j2f
_
=
_
1 for FP1 f FP2
0 otherwise
, (18)
where FP1 = Fc Bw/2 and FP2 = Fc + Bw/2. The discrete
impulse response of the lter, denoted h
f
[n], is given by the inverse
Fourier transform of (18).
Given a choice of Fc, Bw and quantiser conguration, h
f
[n], (11)
and (9) can be used to evaluate the signal mean. The correlator noise
power can be evaluated using h
f
[n], (13) and (14). Finally, the loss
coefcient dened in (17) can be numerically evaluated.
The Nyquist criterion restricts the range of Fc and Bw pairs to a
triangular region on the (Fc,Bw) plane. The aim of this analysis is to
examine the surface dened by the loss coefcient over this region in
order to gain an insight into the effects of ltering and quantisation
on the correlation operation.
V. THEORETICAL RESULTS
This section discusses the results of a theoretical optimisation of
the receiver front-end, for a variety of quantiser congurations. The
loss coefcient, (17), was evaluated for all allowable pairs of Fc
and Bw. In the case of two and three bit quantisation, the value of
Ag has been numerically optimised to maximise the processing loss
coefcient.
0.5 1 1.5 2 2.5 3 3.5 4
x 10
6
1
2
3
4
x 10
6
2.2
2
1.8
1.6
1.4
Fc (Hz)
Bw (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)
Fig. 4. Loss Surface for One Bit Quantisation GPS C/A (Fs = 10 MHz)
1
1.5
2
2.5
3
3.5
4
x 10
6
1
2
3
4
x 10
6
1.6
1.4
1.2
1
0.8
0.6
Bw (Hz)
Fc (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)
Fig. 5. Loss Surface for Two Bit Quantisation GPS C/A (Fs = 10 MHz)
Figure 4 depicts the loss surface for a one-bit quantiser and a
sample rate of 10 MHz. It can be seen that as the value of Bw
is increased from zero, the loss coefcient increases, up to a point,
before reducing smoothly as the bandwidth increases to Fs/2. The
choice of center frequency also has an impact on the loss coefcient,
which is found to be symmetrical about Fs/4.
Authorized licensed use limited to: University of Calgary. Downloaded on January 20, 2010 at 15:27 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing [email protected].
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
4
1
1.5
2
2.5
3
3.5
4
x 10
6
1
2
3
4
x 10
6
1.2
1
0.8
0.6
0.4
Bw (Hz) Fc (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)
Fig. 6. Loss Surface for Three Bit Quantisation GPS C/A (Fs = 10 MHz)
This response can be envisaged as the product of two factors, the
signal response and the noise response. The mean of the quantiser
output, i, increases with increased bandwidth, shown in (12), as
more of the signal is admitted into the correlator. As this bandwidth
is increased, the noise incident on the quantiser becomes whiter, and
the effective noise-oor reduction (depicted in Figure 3), caused by
the quantiser non-linearity, is reduced. As the bandwidth of the lter
approaches Fs/2, noise power which has been whitened and spread
out of the pass band, is aliased back into the band of the received
signal, and so contributes to the correlator output variance.
B Fc (MHz) Bw (MHz)
1 1.764 (3.236) 1.547
2 1.977 (3.023) 1.760
3 2.45 (2.55) 3.893
TABLE I
OPTIMUM FILTER PARAMETERS (Fs = 10 MHZ)
Theoretical results for two- and three-bit quantisers, depicted in
Figures 5 and 6 respectively, exhibit similar loss surfaces. Optimum
lter design parameters for one-, two- and three-bit quantisers are
shown in Table I. The optimum bandwidth increases with increased
quantiser resolution as the quantiser transfer characteristic, (3), be-
comes more linear. In the case of three-bit quantisation, the optimum
value approaches the maximum allowable bandwidth, where the gain
brought about by admitting more signal begins to outweigh the gain
of whitening the noise.
Results of further theoretical analysis using a higher sample rate
(Fs = 25 MHz) are depicted in Figures 7, 8 and 9. Here, the
maximum allowable bandwidth is far greater than the signicant
bandwidth (for example, the bandwidth within which 90% of the
signal power is contained) of the received signal. Optimum lter
design parameters for one-, two- and three-bit quantisers are shown
in Table II. In a similar fashion to the 10 MHz case, the one- and two-
bit loss surfaces increase to a maximum within 2 MHz and 4 MHz of
front-end bandwidth, respectively, before smoothly decreasing as the
bandwidth approaches Fs/2. The loss surface declines less steeply
with increased bandwidth in the two-bit case, as compared to the
one-bit case, due the quantisers more linear response.
Examining Figure 9, it can be seen that the loss surface increases
almost monotonically with increased bandwidth and is almost con-
stant with changing center frequency. This closely resembles classical
ltering loss theory developed for linear analogue receivers [12],
which is to be expected with a high number of quantisation levels
and a high sample rate.
Although this analysis focuses primarily on the GPS L1 C/A signal,
the model described in Section III is applicable to a variety of similar
2
4
6
8
10
12
x 10
6
2
4
6
8
10
12
x 10
6
2
1.8
1.6
1.4
1.2
Fc (Hz)
Bw (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)
Fig. 7. Loss Surface for One Bit Quantisation GPS C/A (Fs = 25 MHz)
2
4
6
8
10
x 10
6
2
4
6
8
10
x 10
6
1.3
1.2
1.1
1
0.9
0.8
0.7
0.6
0.5
0.4
Bw (Hz)
Fc (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)
Fig. 8. Loss Surface for Two Bit Quantisation GPS C/A (Fs = 25 MHz)
B Fc(MHz) Bw(MHz)
1 3.361 (9.139) 1.870
2 4.942 (7.558) 3.653
3 4.645 (7.855) 8.290
TABLE II
OPTIMUM FILTER PARAMETERS (Fs = 25 MHZ)
DSSS signals. As an illustration of this, the GIOVE A E1-B/C and
GIOVE B E1-B/C signals [16] were examined using sample rates of
25 MHz and 50 MHz with one-bit quantisation. It was found that
the loss characteristics are similar to those of the GPS C/A signal,
as depicted for the 25 MHz case in Figures 10 and 11. Relative to
the GPS case, the optimum design parameters, listed in Table III,
evaluate to approximately twice the bandwidth. This is due to the
binary offset carrier (BOC) spectrum having two main lobes, rather
than the single main lobe of the GPS, BPSK spectrum.
Signal Fs (MHz) Fc (MHz) Bw (MHz)
GIOVE-A E1 B/C 25.0 4.511 (7.989) 3.447
50.0 4.511 (7.989) 3.447
GIOVE-B E1 B/C 25.0 9.874 (15.130) 3.707
50.0 8.725 (16.270) 7.223
TABLE III
OPTIMUM FILTER PARAMETERS FOR GIOVE E1-B/C SIGNALS
VI. SIMULATION RESULTS
Validation, through simulation, of the theoretical results presented
in Section V is presented here. Taking the loss surface of Figure
7 as an example, a range of digital lters were designed and
implemented in a receiver simulation environment to investigate this
approximation.
Authorized licensed use limited to: University of Calgary. Downloaded on January 20, 2010 at 15:27 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing [email protected].
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
5
2
4
6
8
10
x 10
6
2
4
6
8
10
x 10
6
1.2
1
0.8
0.6
0.4
0.2
Bw (Hz)
Fc (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)
Fig. 9. Loss Surface for Three Bit Quantisation GPS C/A (Fs = 25 MHz)
2 3 4 5 6 7 8 9 10 11
x 10
6
2 4 6 8 10
x 10
6
2.2
2.1
2
1.9
1.8
1.7
1.6
1.5
Fc (Hz)
Bw (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)
Fig. 10. Loss Surface for One Bit Quantisation GIOVE-A E1 B/C (Fs = 25
MHz)
Simulated wideband IF samples of a GPS L1 C/A signal were
applied to the IF lter and, subsequently, quantised and correlated
with a synchronised carrier and code replica. The coherent SNR was
then estimated over simulation time of 100 s.
A total of 120 lters were designed and implemented, at points
along the most prominent features of Figure 7. Forty lters used the
optimal bandwidth (1.552 MHz) and a range of center frequencies
and forty were placed along each of two lines between the optimal
points (see Table II) and the maximum bandwidth of 12.5 MHz. The
lters used were eighth order Elliptic lters having a pass-band ripple
of 0.1 dB, a stop-band attenuation of -20 dB and a 3 dB bandwidth
of Bw.
Figure 12 shows the results of the simulation test points overlaid on
the theoretical loss surface of Figure 7. The theoretical and simulation
results have maximum and minimum percentage deviations of 4.94%
and -1.37%, respectively, with mean and standard deviations of
1.74% and 1.19%, respectively. This close agreement justies the
approximation of a real lter response with that of a Brick-Wall for
the purposes of loss estimation.
VII. REAL SIGNAL VALIDATION
A qualitative validation of this theory was conducted using a
recording of the GPS L1 C/A signal. This data was collected from
a rooftop environment under good signal conditions, with six high
elevation satellites in line-of-sight, and no signicant signal multipath
component. A National Instruments front-end
1
was used, congured
for a 25 MHz sample rate, a 6.42 MHz center frequency and a 14-
bit signal representation. Prior to digitisation, the signal was passed
1
The National Instruments [17] front-end consisted of a NI PXIe-1065
chassis containing a two-stage 3 GHz, NI PXI-5690 LNA, a 20 MHz band-
width, NI PXI-5600 downconverter and a 14-bit, NI PXI-5142 digitiser
and digital downconverter.
2
4
6
8
10
x 10
6
2
4
6
8
10
x 10
6
2.6
2.4
2.2
2
Fc (Hz)
Bw (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)
Fig. 11. Loss Surface for One Bit Quantisation GIOVE-B E1 B/C (Fs = 25
MHz)
0
2
4
6
8
10
12
x 10
6
0
2
4
6
8
10
12
x 10
6
2
1.5
1
Fc (Hz)
Bw (Hz)
P
r
o
c
e
s
s
i
n
g

L
o
s
s

(
d
B
)Simulation
Theory
Fig. 12. Loss Surface: Simulation Vs Theory (Fs = 25MHz)
through a wide anti-alias lter which limited the signal bandwidth
to approximately 10 MHz. While this lter has a linear phase
response and is sufciently wide-band, it exhibits poor roll-off at
low frequencies.
This digitised signal was re-ltered by a range of lters and,
subsequently, re-quantised at 1-bit resolution. Through this re-ltering
and re-quantisation, a range of front-end characteristics are impressed
upon the signal while preserving such real signal propagation effects
as thermal noise, signal Doppler, electromagnetic and co-channel
interference and ionospheric and tropospheric effects.
In all, 28 lters were applied to the signal, similar in design to those
described in Section VI, having a center frequency of 6.42 MHz and
3 dB bandwidths ranging from 1 to 8 MHz. The ltering operation
was carried out in software using 32 bit oating point precision.
For each of the ltered data sets, a single satellite signal was
acquired and then tracked using a 0.2 Hz, rst order delay-lock-loop
aided by a third order, 5 Hz phase-lock-loop [18]. Once the tracking
loops had locked and achieved bit-syncronisation, they were allowed
to settle for a further period of twenty seconds, at which point a
standard C/N0 estimator [12] was applied to the correlator outputs.
This C/N0 estimate was averaged over the remaining 180 s of the
data set.
Figure 13 shows a plot of the average C/N0 estimate versus Bw,
along with theoretical estimates of C/N0. The theoretical curve,
labeled Joint Theory, is the sum of a C/N0 of 49.43 dB/Hz, prior to
ltering and quantisation, and the processing loss shown in Figure 12.
Given a xed center frequency, Figure 13 is equivalent to a vertical
cross-section through the mesh depicted in Figure 12, along the plane
Fc = 6.42 MHz. The theoretical and real signal curves agree well,
having maximum values at 3.121 MHz and 3.333

MHz, respectively
and trending in a similar fashion with changes of Bw about these
points. A disparity between the curves is evident, however, for large
values of Bw. This is due to the assumption that the data set used
is white and that the front-end anti-aliasing lter is sufciently wide.
Authorized licensed use limited to: University of Calgary. Downloaded on January 20, 2010 at 15:27 from IEEE Xplore. Restrictions apply.
Copyright (c) 2009 IEEE. Personal use is permitted. For any other purposes, Permission must be obtained from the IEEE by emailing [email protected].
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
6
1 2 3 4 5 6 7 8
46.6
46.8
47
47.2
47.4
47.6
47.8
48
48.2
Front End Bandwidth (MHz)
E
s
t
i
m
a
t
e
d

C
/
N
0

(
d
B
)
Joint Theory
Measured
Independent Theory
Fig. 13. C/N
0
Vs Front-End Bandwidth using a real GPS signal
As mentioned above, this lter has a poor low-frequency roll off.
As a result, when a wide (5 - 8 MHz) lter is applied in the re-
ltering stage, the net bandwidth is less than expected. This reduced
bandwidth causes an increase in the signal C/N0 in this region.
Also depicted in Figure 13 is the curve labelled Independent
Theory. This curve represents an estimate of the C/N0 using the
classical approach to loss analysis which addresses the losses incurred
through ltering and quantisation independently. The discrepancy
between the two theoretical curves is discussed in detail in [7]. It is
worth noting here that not only does this curve exhibit a -1 dB bias, it
suggests that the C/N0 increases monotonically with increasing Bw
which, as shown by the measured data and joint theory presented
here, is not the case.
VIII. CONCLUSIONS
This paper presents a novel approach to front-end lter and B-Bit
symmetric quantiser design. Utilizing the ltering and quantisation
theory developed in [7], the effects of the choice of lter design
parameters, Fc and Bw, and the quantiser resolution, B, on the
processing loss of a receiver, have been examined. Novel results have
been presented regarding the relationship between front-end lter
bandwidth and processing loss for quantised signals. In the case of
one- and two-bit quantisation, these new results are, in fact, contrary
to the classical theory. It has been shown that an increase in Bw,
for a given sample rate and quantiser resolution, not only fails to
increase the effective SNR, but can actually reduce it.
REFERENCES
[1] G. B. Frank, Next generation digital GPS receiver, IEEE AES Maga-
zine, pp. 1015, 1990.
[2] D. Avagnina, F. Dovis, A. Gramazio, and P. Mulassano, Denition of
a recongurable and modular multi-standard navigation receiver, GPS
Solutions, vol. 7, no. 1, pp. 3340, May 2003.
[3] T. L. Lim, Non-Coherent Digital Matched Filters: Multibit Quantisa-
tion, IEEE Transactions on Communications, vol. 26, no. 4, pp. 409
419, April 1978.
[4] H. Chang, Presampling Filtering, Sampling and Quantisation Effects on
Digital Matched Filter Performance, Proceedings of the International
Telemetering Conference, San Diego, CA, pp. 889 915, 1982.
[5] J. W. Betz and N. R. Shnidman, Receiver Processing Losses with
Bandlimiting and One-Bit Quantization, Proceedings of the ION In-
ternational Technical Meeting, pp. 1244 1256, September 2007.
[6] J. K. Holmes, Noncoherent late minus early power code tracking per-
formance with front end ltering, Proceedings of the ION International
Technical Meeting, pp. 583 592, September 1997.
[7] J. T. Curran, D. Borio, and C. C. Murphy, Front-End Filtering and
Quantisation Effects on GNSS Signal Processing, Proceedings of the
IEEE Conference on Wireless Communications, Vehicular Technology,
Information Theory and Aerospace & Electronic Systems Technology,
pp. 227231, May 2009.
[8] R. F. Baum, The Correlation Function of Smoothly Limited Gaussian
Noise, IEEE Transactions on Information Theory, vol. IT-15, no. 4, pp.
448 456, July 1969.
[9] R. Price, A Useful Theorem for Nonlinear Devices Having Gaussian
Inputs, IRE Transactions on Information Theory, vol. IT-4, pp. 69 72,
June 1958.
[10] J. H. V. Vleck and D. Middleton, The Spectrum of Clipped Noise,
Proceedings of the IEEE, vol. 54, no. 1, pp. 2 19, January 1966.
[11] D. Borio, A statistical theory for GNSS signal acquisition, Ph.D.
dissertation, Politecnico Di Torino, May 2008.
[12] B. W. Parkinson and J. J. Spilker, Eds., Global Positioning System:
Theory and Applications. Progress in Astronautics and Aeronautics,
1996, vol. 163.
[13] K. Steiglitz, The equivalence of digital and analog signal processing,
Information and Control, vol. 8, no. 5, pp. 455 467, 1965.
[14] R. G. Vaughan, The theory of bandpass sampling, IEEE Transactions
on Signal Processing, vol. 39, no. 9, pp. 1973 1984, September 1991.
[15] G. L. Turin, An Introduction to Matched Filters, IRE Transactions on
Information Theory, pp. 311 329, 1960.
[16] Galileo Project Ofce, GIOVE-A+B Public SIS ICD,
http://www.giove.esa.int/images/userpage/ESA-DTEN-NG-
ICD02837 GIOVE-A+B PublicSISICD 1-1.pdf, [Accessed: 28 Sept
2009].
[17] National Instruments - Products and Services,
http://www.ni.com/products/, [Accessed: 12 June 2009].
[18] S. A. Stephens and J. B. Thomas, Controlled root formulation for digital
phase-locked loops, IEEE Transactions on Aerospace and Electronic
Systems, vol. 31, no. 1, pp. 7895, January 1995.
Authorized licensed use limited to: University of Calgary. Downloaded on January 20, 2010 at 15:27 from IEEE Xplore. Restrictions apply.