You are on page 1of 5

Filter Design Considerations for Acquisition of BOC-modulated Galileo Signals

Adina Burian, Elena Simona Lohan, Markku Renfors Institute of Communications Engineering, Tampere University of Technology P.O.Box 553, FIN-33101, Finland Email: adina.burian@tut., elena-simona.lohan@tut., markku.renfors@tut.
Abstract In this paper we study the effect of the transition band in designing both FIR- and IIR-types of digital lters, as possible bandwidth-limiting receiver lters, during the CDMA code acquisition of a BOC-modulated and oversampled Galileo signal. It is shown that using an asymmetric transition band (i.e., the band between the passband and stopband frequencies) with respect to one fourth of the sampling rate, the performance, in terms of Root Mean Square Errors (RMSE), can be improved for both FIR and IIR lters compared to the situation with symmetric transition bands. As an optimum bound we have employed the ideal rectangular pulse (i.e. no bandwidth limitation). All investigated ltering methods still suffer from some performance degradation, compared to rectangular shaping. The analysis is done here for both static and fading multipath channels.

I. I NTRODUCTION Since, in real communications systems the signal bandwidth is limited (due, for example, to some frequency regulations), the spectrum shaping of received signal is necessary. In context of GPS and Galileo satellite systems, the frequency spectrum represents the most important resource. For example, the Galileo Open Services (OS) and Public Regulated Services (PRS) signals are clustered around 1575 MHz of L1 frequency [1]. New challenges in the acquisition process are posed also by the Binary Offset Carrier (BOC) modulation, planned for future GPS and Galileo signals, which provide a higher degree of spectral separation with the current GPS signals [2]. An extended theoretical analysis of code tracking accuracy for a bandlimited GPS receiver (i.e., BPSK signals) is provided in [3], but no discussion about the lter design is included. The effect of receiver front-end bandwidth on the correlation loss for an GPS M-code signal is studied in [4] and different precorrelation bandwidths are dened for Galileo L1 signal, by assuming ideal rectangular bandpass ltering [1]. In Galileo OS (sine BOC(1,1) modulation case), 85.57 % of signal power is contained within 4 MHz (double-sided) bandwidth, while 92.51 % of the total signal power is included in 8 MHz bandwidth. For Galileo PRS (cosine BOC(15,2.5) modulation case), 77.45 % of signal power is included in 40 MHz bandwidth, and increasing the band above 40 MHz gives very slow improvement in the signal power (e.g., in 60 MHz bandwidth, we have 79.75 % of the total signal power). These values were computed according to [2]. In [5] the analysis of acquisition and tracking of BOC signals is done taking into account the lter effects. Here non-ideal lters, such as

Butterworth lters, are used, which have a non-constant (time) group delay. Another important parameter which affects the performance of the baseband receiver (and it is usually not mentioned in the research studies) is the oversampling factor (or the number of sub-samples per BOC sample or sub-chip interval). The oversampling factor has a major effect on the achieved timing accuracy, because the initial sampling point varies randomly, according to the channel delays. The effect of limiting the signal bandwidth is that, in time domain, the autocorrelation function (ACF) becomes smoother around the peaks, and it is no longer piecewise linear, as in the case of innite bandwidth. This smoothing of the ACF is prone to increase the distance and delay errors, in a similar way with lowering the sample rate, as it was reported in [6]. However, low sample rates are often used, because of a lower computational complexity. Also, the application of minimum-phase types lters (such as Butterworth) can have an impact on ACF. The symmetrical ACF may become skewed and delayed, depending on the cutoff frequency and on the lter order [6]. A brief analysis of basic digital ltering methods (nite impulse response FIR and innite impulse response IIR) for use in navigation GPS spread-spectrum receivers is provided in [7]. The two ltering structures are compared in time and frequency domain and a simple functional relation between them is provided. This paper considers the design of digital bandlimiting receiver lters for a BOC-modulated and oversampled Galileotype signal, taking into account also the transition band between the passband and stopband frequencies. The motivation of our work comes from the need of designing efcient ltering structures when the receiver bandwidth is to be limited, due, e.g., to frequency regulation. The studied structures are IIR digital lters (Butterworth, Chebyshev) and FIR (optimum equiripple design employing Parks-McClellan algorithm). This paper is organized as follows: the next section gives several theoretical considerations for the studied ltering methods, i.e. IIR and FIR lters. In Section III the signal model and the acquisition problem are introduced. Section IV presents the effects of symmetrical and asymmetrical transition bands on receiver performance, for IIR and FIR lters, in the presence of BOC modulation and oversampling. The examples are taken from OS and PRS specications of Galileo signals. Section V concludes the paper.

II. FIR

VERSUS

IIR FILTERING IN G ALILEO RECEIVERS


Normalized envelope of ACF

Autocorrelation functions, filtering with BW=4 MHz 1 Sim., unlimited BW Sim, limited BW Th., unlimited BW Th., limited BW

0.8

The bandwidth limitation produces a non-constant envelope, which is prone to degrade the performance. For digital IIR lters, an exact linear phase is not possible (in contrast to FIR lters) and the design is not so exible, because of the feedback necessary in the implementation. On the other hand, an IIR lter structure has fewer stored coefcients and a lower multiplication rate and requires less memory than a FIR lter with the same frequency domain specications [8]. An IIR lter can give a sharper cut-off than a FIR lter with the same order. Also, the phase and group-delay characteristics of IIR lters are generally not so good as those of FIR lters [8]. Very popular IIR lters are the Chebyshev type I lter (which has ripples in passband and is monotonic in stopband) and the Butterworth lter (which is monotonic everywhere). The Chebyshev lter amplitude response is superior to Butterworth at cut-off frequency and in stopband, but Butterworth amplitude response is better near the zero frequency, due to its maximally at propriety. But as the amplitude response is better, the phase response becomes more non-linear, so Butterworth lter can be superior. Also, lower-order Chebyshev and Butterworth phase responses are superior to those of higher order. FIR lters have the advantage of the linear-phase property compared to the IIR lters. Therefore, FIR lters have a constant group delay through the frequency spectrum and complete stability. For example, the Parks-McClellan optimal equirriple FIR lter algorithm provides the best approximation to the desired frequency response [8]. Moreover, the ease of implementation makes this lter very attractive. One disadvantage could be that the number of states required to meet a frequency specication may be much higher than that required by IIR lters. There are different lter structures employed in GPS/Galileo receivers. For example in [9] the equivalent RF/IF lter is simulated as a succesion of IIR (Butterworth) lters and an FIR lter. A steeper slope in stopband is achieved by FIR lter (simulating the IF ltering effect) and a lower slope away from passband is given by Butterworth lter (which acts as an RF lter). Usually, in the design of digital lters, the transition band between passband and stopband frequency is considered as a dont care region. However, by taking into account and by designing properly this region, the resulting approximation ripple or Gibbs phenomenon can be signicantly reduced [10]. The wider this transition band is, the better the approximations in passband and stopband are [8]. Also, for any FIR design algorithm, the lter order required to meet a given specication is inversely proportionally to the transition width allowed, if the peak ripple remains the same. In order to have a correct comparison with simulated results, theoretical power spectral densities (PSDs) and ACFs were computed based on formula given in [4]. The normalized power spectral density Gs (f ) of a baseband signal with sine-

0.6

0.4

0.2

0 1

0.5

0 Delay error [chips]

0.5

Fig. 1.

Effect of bandlimitation on the autocorrelation function


OS BOC(1,1) modulation

10 20 30 40 Unlimited bandwidth, theory Limited bandwidth, simulation

PSD

50 60 70 80 90 100 20 15 10 5 0 5 Frequency [MHz] 10 15 20

Fig. 2.

Power Spectrum Density for BOC-modulated PRN code

BOC(fsc , fc ) modulation is: sin Gs (f ) = fc

f 2fsc

sin
f 2fsc

f fc

f cos

(1)

where fc is the chip rate and fsc is the sub-carrier rate. Fig. 1 illustrates the effect of limiting the band to 4 MHz, for both theoretical (th) ACF curves (obtained from the inverse Fourier transform of eq. (1)) and simulated (sim) ones (with Butterworth lters). The spectrum of the bandlimited-signal is illustrated in Fig. 2, together with the unlimited-spectrum case. We observe that the bandwhidth-limitation rounds the peaks of the ACF function and also that the curves obtained through simulation match well with the theoretical ones. III. T HE SIGNAL
MODEL IN PRESENCE OF

BOC

MODULATION AND OVERSAMPLING

The simplied baseband block diagram of transmitted BOCmodulated CDMA signal is shown in Fig. 3. The PRN sequence is BOC-modulated, i.e., by a square subcarrier modulation, where the signal is multiplied by a rectangular subcarrier with sub-carrier frequency fsc , which splits the signal

Data symbols bn

Spreading

BOC modulation

Ns

D/A converter

Channel

PN code

Fig. 3.

Block diagram of the transmitter and channel

ltered with a digital LPF lter, whose design is the focus here. The lters considered here are: IIR Chebyshev lter(type I), IIR Butterworth lter, and a FIR lter employing the ParksMcClellan optimal equiripple algorithm. The situation with no ltering (rectangular pulse shaping) is kept as a benchmark. The next operation in the receiver is the code-Doppler acquisition (estimating the Doppler shift fD and channel delay l ), which is done via correlation with a reference signal sref (t, , fD , n1 ) including the PRN code and the BOC modulation:
SF NBOC 1

spectrum. The common baseline for OS structure (agreed by US and European negotiation in June 2004) employs the sine BOC(1,1) modulation, which uses a 1.023 MHz square-wave subcarrier modulated by spreading code chips at a chip rate fc =1.023 MHz [1]. For PRS services, both sine and cosine BOC(15,2.5) have been proposed. In this paper we have selected the sine BOC(15,2.5) modulation, but the results are expected to be very similar for cosine BOC(15,2.5) modulation as well. Next, the data sequence is oversampled with oversampling factor Ns , representing the number of sub-samples per BOC sub-chip interval. Therefore, one chip will consists of NBOC Ns sub-samples, where the BOC-modulation order NBOC is dened as: NBOC = 2fsc , fc (2)

sref (t, , fD , n1 ) = e

j2 fD t k=1

ck,n1
m=0

(1)m

p(t n1 SF NBOC Ns Ts kNBOC Ns Ts mNs Ts ),

(5)

After correlation, the signal is coherently averaged over Nc msec (with maximum coherence integration length dictated by the coherence time of the channel and by the stability of oscillators), and next non-coherently averaged over Nnc blocks. The decision statistic is then formed in a serial, parallel or hybrid fashion, splitting the code-Doppler search space into several code-Doppler windows. The measure of performance in our simulations is the root mean square error (RMSE) computed as: RM SE[chips] = E(| |), (6)

For clarity, the continous model is employed in the theoretical analysis, but all presented results are based on discretetime implementation. We denote by sn (t) the PRN code sequence corresponding to n-th code bit. The baseband model of received signal r(t) via a fading channel can be written as:
n=+ L

where E() is the expectation operator, is the true rst path delay, and is the estimated delay, in case of acquisition. The RMSE in meters was computed as RM SE[m] = RM SE[chips] *c, where c is the speed of light. IV. D ESIGNING
THE TRANSITION BAND FOR BANDLIMITING FILTERS IN PRESENCE OF OVERSAMPLING

r(t) =

Eb e+j2fD t
n=

bn
l=1

n,l (t)sn (t l ) +(t), (3)

AND

BOC MODULATION

where Eb is the bit energy of signal, fD is the Doppler shift introduced by channel, bn is the data bit corresponding to the n-th code bit (the same data bit is usually kept for 20 ms [1]), L is the number of channel paths, n,l (t) is the timevarying complex fading coefcient of the l-th path during the n-th code epoch, l is the corresponding path delay and () is the additive white noise added by the channel. The BOCmodulated and spread code sequence, corresponding to n-th code bit is:
SF NBOC 1

sn (t)

=
k=1

ck,n
m=0

(1)m p(t nSF NBOC Ns Ts (4)

kNBOC Ns Ts mNs Ts ),

where ck,n is the k-th chip value corresponding to the n-th data symbol, SF is the spreading factor or the code epoch length (e.g., for C/A signal of GPS, SF = 1023), p() is a train of rectangular pulses, and Ts is the sampling rate. The main operations performed at receiver side are shown in Fig. 4. After the analog-to-digital converter, the signal is

In bandlimiting operation, the lters use the following main design parameters: passband fp and stopband fs edge frequencies, passband ripple Rp (in dB), and stopband attenuation Rs (in dB). The normalized sampling rate corresponds to W = Ns NBOC fc bandwidth. The passband edge frequency can be written as fp = 11 W and the stopband edge frequency can 4 be written as fs = 1+2 W , where 1 and 2 are two design 4 parameters, which dene the width of the transition band. We distinguish two situations: 1) symmetrical transition band (sym): 1 = 2 (0, 1) 2) asymmetrical transition band (asym): 1 = 2 The symmetry of transition band is taken therefore with respect to W point. The desired amplitude in passband was 4 set to one. The minimum order lters for which the design specications are met are used here. In order to determine the best choice of parameters 1 and 2 for passband and stopband edge frequencies, for both FIR and IIR lters, we have tested, with a small step the combinations (1 , 2 ) in the intervals between 0.01 and 0.99.

A/D converter

rx sign
LPF

Coherent integration

.2

Non-coherent integration

Form decision statistic

z1

>

Yes

Declare acquisition

ref. code at Doppler frequency f D

Readjust est. delay and est. f D

No

Fig. 4.

Block diagram of receiver

Number of filter coefficients

The maximum passband ripple, and, respectively, stopband attenuation, were xed to Rp = 0.01 dB for passband and Rs = 20 dB for stopband. Figs. 5 and 6 illustrate the results for a static channel case with short coherent and non-coherent integration lengths Nc =5 and Nnc =1. In order to get reliable results, the simulations were run over 1000 random points.
Asymm. vs. Symm. filtering, IIR (Cheb1): CNR=30 dBHz, N =4, NC=5, BOC(1,1) s

Asymm. vs. Symm. filtering, IIR (Cheb1): CNR=30 dBHz, N =4, N =5, BOC(1,1)
s C

30 25 20 15 10 5
2

0 10 10 10
0 1 0

10

10

250
2

10

RMSE [meters]

200

150

Fig. 7.

Number of coefcients for IIR lter, for different 1 and 2

100 2 10 10
1 0 2

10 10 2 10 10
1
1

Fig. 5.

Optimum choice of parameters 1 and 2 , for IIR ltering

Asymmetric vs. Symmetric FIR filtering: CNR=30 dBHz, Ns=4, NC=5, BOC(1,1)

400 350

300 250 200 150 100 10


2

10

10
1

10

10 2

10

Fig. 6.

Optimum choice of parameters 1 and 2 , for FIR ltering

From Fig. 7 it can be seen that the minimum lter order is obtained for wide transition widths, i.e. for high 2 and/or high 1 . As it can be observed from Figs. 5 and 6, the best RMSE values are obtained for high values of stopband edge

frequency 2 and for small values of passband frequency 1 . Also, the worst results are obtained for small values of 2 and high values of 1 , as expected. Therefore, by using an asymmetric transition band, with high values for 2 and low values for 1 we get better results in terms of low RMSE and low lter order than using a symmetric transition band. The ACFs for symmetric and asymmetric transition bands are shown in Fig. 8. We note that, by using asymmetrical transition bandwidth (i.e. 1 =0.1, 2 =0.9) the sidelobes of ACF tend to be smaller and the width of the main lobe is narrower than for symmetric transition band. We also notice from Fig. 8 that increasing the oversampling factor (or, equivalently, increasing the receiver bandwidth) from Ns = 5 to Ns = 20, the ACF becomes closer to the ideal, piecewiselinear shape, in the absence of bandwidth limitation. In Figs. 9 and 10 we show the simulation results using asymmetric (1 = 0.1, 2 = 0.9) and symmetric FIR and IIR lters (1 = 2 = 0.3). The channels used in simulation are 3-path fading channels with decaying power delay proles. The rst path was Rician distributed and the last two paths were Rayleigh distributed. The maximum separation between succesive paths xmax was set to 4 chips, the mobile speed was 25 km/h, the coherent integration Nc was done over 80 ms and non-coherent integration was Nnc = 5 blocks. The oversampling factor was set to Ns =4, which corresponds to a passband edge frequency of of (1 1 )2.046 MHz for OS and of (1 1 )30.69 MHz for PRS. We observe that, in both cases (OS, BOC(1,1) and PRS,

RMSE [meters]

Asymmetric ( =0.1, =0.9) versus symmetric ( = =0.5) filtering


1 2 1 2

1 0.9

Ns=5, sym. transit. band Ns=5, asym. transit. band Ns=20, asym. transit. band

Normalized envelope of ACF

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1.5 1

BOC(15,2.5)), the ideal rectangular pulse-shaping (i.e., no bandwidth limitation) gives the best performance, as expected. When using an asymmetric-band lters, we improve the performance compared to symmetric-band lters, especially for low-CNR situations (this observation holds for both OS and PRS cases). Also, the best results among the considered lters are obtained using Butterworth lter (which may come very close to ideal, unlimited-bandwidth, situation, as seen in Fig. 10). V. C ONCLUSIONS The purpose of this paper is to analyze the inuence of the transition band in designing the receiver bandlimiting lters, for target application such as Galileo satellite system or modernized GPS. Different digital ltering structures were presented and compared with the ideal case where no bandlimitation is present. It has been shown that the performance in terms of RMSE can be improved if asymmetric transition bands are used (the symmetry point was with respect to one fourth of the sampling rate). The results were checked via simulations with BOC-modulated and oversampled signals modelled according to the Galileo proposals. We noticed that, for PRS signal, if suitable asymmetrical transition bands are chosen, we do not loose much in terms of RMSE performance compared to the case with no bandwidth limitation, while for both PRS and OS signals, using a Butterworth lter with asymmetric transition band gives the best results (in both static and fading channels) when the signal has to be bandwidth limited. ACKNOWLEDGMENT This work was carried out in the project Advanced Techniques for Mobile Positioning funded by the National Technology Agency of Finland (Tekes). This work has also been partially supported by the Academy of Finland. R EFERENCES
[1] G.W. Hein, M. Irsigler, J.A. Avila Rodrigues, T. Pany,Performance of Galileo L1 Signal Canditates, in Proc. of The European Navigation Conference GNSS, 2004. [2] J.W. Betz, The Offset Carrier Modulation for GPS modernization, in Proc. of ION Technical meeting, pp. 639648, 1999. [3] J.W. Betz, Extended Theory of Early-Late Code Tracking for a Bandlimited GPS Receiver, Journal of the Institute of Navigation, Vol. 47, No. 3, Fall 2000. [4] J.W. Betz, Design and Performance of Code Tracking for the GPS M Code Signal, MITRE Technical Paper, Sep. 2000 . [5] V. Heiries, D. Roviras, L. Ries, V. Calmettes, Analysis of Non Ambiguous BOC Signal Acquisition performance, ION GNSS 17th International Technical Meeting of the Satellite Division, 21-24 Sept. 2004, pp. 26112622. [6] A.J.R.M. Coenen, Combating short-delay multipath requires enhanced peak-detection, 2nd ESA Workshop on Satellite Navigation User Equipment Technologies NAVITEC , 2004. [7] T.A.J.R.M. Coenen, On Elementary Filters (IIR and FIR) for Acquisition and Tracking in GPS Receiver. http://wwwtvs.et.tudelft.nl/PN/people/ton. [8] T.W. Parks, C.S. Burrus, Digital Filter Design, WileyInterscience Publication, 1987. [9] F. Bastide, E. Chatre, C. Macabiau, B. Roturier, GPS L5 and Galileo E5a/E5b Signal-to-Noise Density Ratio Degradation due to DME/TACAN Signals: Simulations and Theoretical Derivation, ION NTM 2004, San Diego, 2004. [10] D. Morgan, A general characterization of lowpass FIR frequency response in the transition band, IEEE Transaction on Signal Processing, vol. 43, Jun. 1995.

0.5 0 0.5 Delay error [chips]

1.5

Fig. 8. ACF for symmetric and asymmetric transition bands, for two oversampling factors.

Rayleigh fading channel, 3 paths, xmax=4, 5 freq. bins 450 400 350 300 Rect Remez asym Remez sym Cheb asym Cheb sym Butt asym Butt sym

RMSE [meters]

250 200 150 100 50 0 25

26

27

28

29

30 31 CNR [dBHz]

32

33

34

35

Fig. 9. Performance of bandlimiting ltering, for a fading channel, for OS (BOC(1,1))

RMSE [meters]

PRS signal, modulated with BOC(15,2.5), fading channel with 3 paths 140 Rect Remez asym 120 Remez sym Cheb asym Cheb sym 100 Butt asym Butt sym 80 60 40 20 0 25

26

27

28

29 30 CNR [dBHz]

31

32

33

Fig. 10. Performance of bandlimiting lters, for a fading channel, for PRS (BOC(15,2.5))

You might also like