Abstract- This paper proposes two GNSS signal detection front-end algorithms for GNSS receivers, thus saving valuable resources in chasing signals that are not available. These algorithms can also be used to develop a fast multi-signal GNSS software receiver. With the planed completion of Galileo and Glonass CDMA systems, a GNSS receiver can take advantage of these and GPS signals to aid localization in bad reception areas such as urban-canyons. Such receivers deploy all available resources to find signals even when signals are not available. This paper proposes two approaches that can rapidly detect, in a single view, any GNSS signal power present. The first approach analysis the three signals excitations to a nonlinear bandpass sampling (BPS) receiver that folds these three signals, with their harmonics, to their first Nyquist zone (FNZ). Then, it analyzes their behavior model based on a Volterra algorithm to obtain kernels of these three GNSS signals, if available. Because all three GNSS signals are transmitted with the same carrier frequency, the second approach filters out the right-side/lobe of the Glonass signal and the left-side/lobe of the Galileo signal. This will enable none overlapped folding, based on BPS, of these two signals with the 3rd GPS harmonic in FNZ. These approaches make any GNSS receiver well-informed of available signals, and so devote appropriate resources to acquire and track available signals only. Matlab simulation results prove these two approaches and show that much valuable overall processing time, especially on Smartphones, can be saved by adopting these approaches. Keywords - GPS, Galileo and Glonass signals, Bandpass Sampling (BPS), Volterra Series.
I. INTRODUCTION
The worldwide Global Navigation Satellite System (GNSS) market led by Location Based Service (LBS), including applications on Smartphones, is expected to grow by 13% year on year reaching €200 Billion in 2016 [1].
Multi-GNSS signal solutions on Smartphones will help provide better localization fix, accuracy and availability. We believe that such multi-signal GNSS (including GPS, Galileo and Glonass) solutions will roll out in most Smartphones as more LBS applications become available. However, developers should consider the technical aspects, including processing resources and battery efficiency as well as cost and size of the GNSS solution. Solutions that implement these various GNSS-receivers side-by-side (parallel processing scenario) will be costly (processing, power, area, etc.). Integration of some of the processing chain functions in hardware/software will achieve savings, but leads to complicated control algorithms and will still mean that only one signal type is processed at a time. The authors believe that the optimum solution lies in developing multi-signal processing algorithms to achieve considerable saving yet maintaining the benefit of looking at all available signals at any one point in time.
Furthermore, GNSS signals reaching a receiver on a Smartphone are weak even when outdoors, well under the noise level of around -130 dBm. In a typical GNSS receiver, acquiring the signal involves a lot of digital processing. When the GNSS-receiver is positioned in bad reception area or indoors, the received signal degrades by about (25-30) dB [2]. This causes the receiver to thrash all its available resources to find the signal, based on filtering or guessing algorithms, and dependent on that receiver architecture.
This paper proposes to focus on detecting these GNSS signals, by the receiver, at early stage (before conversion to digital). This will prepare the processing algorithms "acquisition and tracking" of the receiver for dealing with the available GNSS signals only. Thus, avoiding chasing any GNSS signals that do not exist. i.e. "quick-early detection" algorithms are proposed in this paper that sense multi-GNSS signals in a single view by measuring the power of all available received
signals prior to the acquisition stage. Our algorithms are based on the BPS technique [3].
By choosing appropriate sampling rate of 92.07 MHz, the first algorithm folds the whole bandwidth of the three GNSS L1- signals: 1. the GPS-C/A-BPSK, 2. Galileo-OS-BOC(1,1) and 3. the modernization Glonass-BOC(2,2) [4] to the FNZ, with isolation between signal frequencies and their harmonics. Volterra Series (VS) is then used to model the behavior of the BPSR. The resulting kernels of VS are used as an indicator to all signal availability based on pre-identified signal divergence.
Folding the Galileo and Glonass BOC signals with the GPS BPSK signal will result in overlapping of these frequencies when excited for a BPSR in the FNZ. This overlapping can be eliminated by filtering out the lower-sideband/lobe of the Galileo signal as well as the upper-sideband/lobe of the Glonass signal. Our second algorithm combines these filtered single-lobe signals with the 3rd harmonic of the GPS signal to avoid overlapping of these signals in FNZ of a BPSR. Therefore, we can detect the three signals directly.
For our simulations, we have used the mathematical representation of the GPS and Galileo signals that are presented in formula (1) and (2).
In equation (1), Si(t) represent the signal transmitted
from any ith GPS satellites (i= 1..32), P is the power of this signal, Ci(t) is the C\A code sequence assigned to this satellite, Di(t) is the navigation data sequence, is the modulo-2 adder and FL1 is the carrier frequency of L1 [5].
In equation (2), the SE1(t) represent the transmitted
Galileo signal, and scE1-B,a, scE1-B,b, scE1-C,a and scE1-C,b are sub carrier components for the data and pilot channels. Also, the general form representation of the sub-carrier (where SCx is x-channel sub-carrier, Rs,x represents the frequency of sub-carrier belonging to the x-channel) is:
The parameters α and β, in formula (2), are chosen such that, the combined power of the scE1-B,b and the scE1-C,b sub carrier components are: Furthermore, the data channel and the pilot channel representation [6] are:
and
respectively, where DE1-B is the navigation data stream, CE1-B and CE1-C are the ranging codes. The mathematical representation of the Glonass BOC(2,2) we used in this simulation has the same mathematical representation to the formula of the Galileo signal, but with different values for the bit rate being at 50 bits/s, chipping rate is 2.046 MHz, and the subcarrier frequency at 2.046 MHz.
The rest of this paper is organized as follows: Section II introduces the BPS concept and surveys other work in this area, while section III details our two proposed algorithms together with the simulation results that prove the feasibility of these two algorithms. Section IV concludes this work together with future work plans.
II. LITRATURE SURVEY OF BPS BASED SOLUTION
The key requirements, for any GNSS solution on a Smartphone, are to be integrated in a small size, take advantage of all the GNSS signals available while using minimum power, and to be low cost. Typical RF receivers front-end use the Super Heterodyne architecture, Zero-IF design, Low-IF design, or the BPS architecture. However, BPS receiver's architecture is a good fit, and so it is more likely to meet these requirements since it is designed to handle multi-signals in a single RF chain [7].
The BPS is a technique that do-away with analog mixers, as used in traditional receiver [8], by bringing the analog-to-digital converter (ADC) as close as possible to the antenna (direct conversion). This is achieved by folding the "information band" at the center frequency of the received signal (or at the "information band" at the center frequencies of the received signals in the case of a multi-signal BPSR) to the FNZ. Therefore, it is important to choose a suitable sampling frequency so to prevent overlapping of the signal with itself (or with other signals in a multi-signal BPSR scenario) in the FNZ. For a single signal, in the frequency domain, the folding process is a convolution between the FFT of this received signal and the summation of the shifted direct-delta function of its sampled pulses [9].
Actually, BPS is based on Nyquist second sampling theorem [10]. i.e. the minimum sampling frequency has
269
to be double the bandwidth of the received signal. This means that the sampling frequency is a fraction of the carrier frequency of the received signal. Equation (3) shows the mathematical relationship defining the folding of the carrier frequency to the FNZ (note that ffold is the folded frequency, fs is the sampling frequency, fc is the carrier frequency, fix(a) is the truncated portion of argument a, and rem(a,b) is the reminder after dividing a by b). Iterations of this relationship based on different values are used to calculate the sampling frequency suitable for our proposed solution focusing on GPS, Galileo and Glonass signals [3]. Multi-signal BPSR is a good candidate for use in software defined radio and cognitive radio [11].
A design of multi-GNSS GPS/CDMA and
Glonass/FDMA receiver [3] is achieved. These two signals are ideal for BPS concept based receiver, as their frequencies will not overlap in the FNZ. In the same vein, an L1+L2+L5 GPS transmission BPSR is combined with a Galileo E1+E5+E6 BPSR in the SDR solutions [12] and has been successfully implemented in the front-end.
A direct conversion reconfigurable front-end has been designed to handle GPS and Galileo (L1 and L5) signals [13]. This design is able to select these signals based on four operating modes. The switching between these 4 modes is based on changing the rate of the sampling frequency manually based on the required setup. Where, mode 1 and 2 handles GPS signals while modes 3 and 4 are either handles Galileo signals alone or GPS+Galileo signals. This solution has no automatic configuration to handle available signals as and when needed.
To overcome the limitation of the BPSR when the target signals share the same center frequency, VS is used to analyze the behavior model of the non-linear BPSR [14]. In this algorithm, obtaining the kernels of VS helps to characterize the output signals. i.e. the output of any nonlinear model/system that is described by VS will be linear with respect to the kernels. VS is defined as an approximation technique to model any nonlinear system, as shown in formula (4), where x(t) and y(t) are the input and output respectively, and h(t) are the Volterra kernels.
III. THE TWO PROPOSED ALGORITHMS Simulations of the two proposed algorithms are
implemented by using the MATLAB software. Seven scenarios were used to test each of two algorithms. These scenarios are based on satellite transmissions from GPS (C/A-BPSK), Galileo (OS-BOC (1,1)) and Glonass (BOC (2,2)) using code division multiple access (CDMA) with a center frequency of 1575.42 MHz, as shown in Table (1).
Note that, in all the following simulation results, the power spectrum density figures are estimated using the Welch algorithm available within Matlab.
A. First proposed Algorithm Figure (1) shows a block diagram of our first
receiver/front-end configuration, as implemented in Matlab.
Simulated GNSS signals, for each of the seven scenarios in Table (1), are fed to additive white Gaussian noise channels (AWGN). These signals are then processed by a BPSR implementation that includes a bandpass filter (BPF), low noise amplifier (LNA) and an analog-to-digital converter (ADC). The BPF is centered at L1-frequency of 1575.42MHz with a 10MHz band, to filter out undesired signals. A 10MHz band is chosen so to include all three GNSS bands (2MHz-GPS band, 4MHz-Galileo band and 8MHz-Glonass band). All passed signals are then amplified by the LNA (+10 dBm compression point, 35 dB gain and 3 dB noise figure). A 10-bit ADC, with 92.07 MHz sampling frequency, is then used to digitize these signals. This sampling frequency is chosen to prevent
270
overlapping between the fundamental frequencies of these signals and their harmonics in the FNZ.
Furthermore, the positions of the intermodulation signals produced from the non-linearity of our BPSR are located either at the fundamental frequency or at any of the harmonics frequencies in the FNZ. The frequencies for these signals are at 10.23 MHz for the fundamental (Fund), at 20.46 MHz for the 2nd harmonic (2nd H), and at 30.69 MHz for the 3rd harmonic (3rd H). This proves that only "in-band" signals are produced by our chosen folding property, as shown in Figure (2).
Fig. 2 Received and VS Estimation for GPS, Galileo and Glonass Signals
The output of the BPSR is then converted to an analog signal by using an "ideal digital-to-analog converter" [8], which result gets fed to the VS model together with the original analog signals. The VS is used to analyze the behavioral model of the BPSR. This is achieved by extracting the unknown kernels of the VS. The unknown parameters (kernels) are calculated by using the Multiple Linear Regression method [15].
This proposal finally detects the status of the available GNSS signals based on the extracted values of the kernels. i.e. the changing power distribution in the input and output signals of the BPSR "fundamentals and their harmonics" will result in different combinations of VS kernels unique to the input GNSS signals present. Therefore, for each of our 7 GNSS-signal scenarios, unique power distributions, with different kernel values, have been obtained.
Also, Figure (2) shows the frequency domain result of the first scenario of Table (1). The estimated behavioral model of VS is close enough to the BPSR behavior based on the extracted unknown kernels parameters. The normalized mean squared error (NMSE) is used as a parameter to evaluate the performance of the estimation between the original BPSR model and the VS model [16], as shown in Table (1). The NMSE of this scenario is around -38.19 dB.
Figures (3, 4 and 5) illustrate the frequency domain results of the second scenario, when the input signals are a combination of only two signals, scenarios (2, 3 and 4) as listed in Table (1).
Based on our algorithm design, if the Glonass signal
is one of the input signals to the BPSR, then the output signals will have peaks in the fundamentals band and the 3rd harmonics band only. The peaks in the 2nd harmonics band will fade out under the noise level.
Furthermore, the Galileo and Glonass signals use BOC modulation with different subcarrier frequency and the GPS signal uses BPSK modulation. Therefore, any combination of two signals that include GPS will have one peak power in the fundamental band. In contrast, if the combination does not include a GPS signal then two peaks power will be present in the fundamental band due to the BOC modulation technique.
The estimation of VS kernel behavioral of scenarios 1, 2, 3, and 4 is approximately identical to our simulation of the BPSR. The values of NMSE in Table (1) indicate that the assessments of the 4th scenario "GPS + Galileo" is better than the 2nd and 3rd scenarios due to present of the 2nd harmonic.
Fig. 3 Received and VS Estimation for GPS and Glonass Signals
Fig. 4 Received and VS Estimation for Galileo and Glonass Signals
Fig. 5 Received and VS Estimation for GPS and Galileo Signals
271
Figures (6, 7 and 8) show the frequency domain when the input signal to the BPSR is only one signal. The values of the extracted kernels will differ from one signal to another based on the distribution power and location of the peaks.
According to the modulation technique, the detection of GNSS signals in scenarios 5, 6 & 7 will be easier than in the previous scenarios This is because the number and position of the received signal power peaks, in the FNZ, determine the type of the available received signal.
Figure (6) shows that the power peak of received Glonass signal in the fundamental band is located away from the center frequency by ±2.046 MHz. The position of the power peak of the received Galileo signal in the fundamental band is shown in figure (7) to be away from the center frequency by ±1.023 MHz, whereas, the GPS signal has a single power peak at the center frequency, as shown in figure (8).
Fig. 6 Received and VS Estimation for Glonass Signal
Fig. 7 Received and VS Estimation for Galileo Signal
Fig. 8 Received and VS Estimation for GPS Signal
B. Second proposed Algorithm The second algorithm focuses on detecting the
power peaks of all GNSS signals present in the FNZ. This is achieved by removing the overlapping between all the folded GNSS signals in the FNZ so to ensure that the detection of the signals is easier and faster. Therefore, the single-sideband (SSB) of the Galileo and the Glonass BOC signals are used in this algorithm. The SSB is produced by using the subcarrier frequency in the BOC modulation. i.e. the subcarrier splits the power spectrum of the BOC signal into two symmetrical components around the center frequency that makes these SSB signals.
It is known that processing double-sideband BOC signals during multipath scenarios can cause ambiguity to acquire and track the signal. Splitting the DSB into SSBs will remove the effect of the subcarrier frequency and make each sideband as a BPSK modulation, just like the GPS signal [17]. In addition, the two SSB can be shifted to the center frequency, by ± subcarrier frequency, resulting in each sideband of this BOC is like the BPSK signal [18]. Based on this fact, our algorithm proposes a method to prevent the overlapping between our chosen GNSS signals. This algorithm filters out the left-sideband of the Galileo signal and right-sideband of the Glonass signals. The reverse is also possible. This filtering must ensure the correct choice of sampling frequency to guarantee non-overlapping between these two signals as well as the 3rd harmonic of the GPS signal.
This receiver front-end configuration of our algorithm is implemented in Matlab, as shown in figure (9). The simulated signals are passed through a nonlinear channel. The first three BPF are used to obtain right-sideband of the Galileo signal, left-sideband of the Glonass signal and the 3rd harmonic of the GPS signal. Then, the filtered signals are amplified by using an LNA (38 dB, 3 dB noise figure and IIP3 24 dBm). A 10-bit ADC converts the amplified signals to their digital form. This configuration uses a sampling frequency of 34.782 MHz to ensure non-overlapping between the three GNSS signals in the FNZ.
The seven scenarios shown in Table (1) are used to test this algorithm. In these tests, the BPSR will deal
272
with input signals as three distinct GNSS signals i.e. each signal has a separate folded frequency in the FNZ, as shown in figure (10). These signals have three distict power peaks present in the FNZ. The 1st power peak is centered at 4.092 MHz (GPS signal) with bandwidth of 2 MHz. The 2nd power peak is at 8.184 MHz (Glonass signal) with a bandwidth of 4 MHz. The 3rd power peak is at 11.253 MHz (Galileo signal) with a bandwidth of 2 MHz. There are no overlapping between these power peaks. Figure (10) also proves that three signals are simultaneously input to our BPSR.
Fig. 10 Power Spectrum of GPS, Galileo and Glonass Signals
The results of scenarios (2, 3 and 4) from Table (1) are illustrated in figures (11, 12 and 13). The figures prove that there are two separate power peaks in the frequency domain of any two signals processed by our BPSR.
Fig. 11 Power Spectrum of GPS and Glonass Signals
Fig. 12 Power Spectrum of Galileo and Glonass Signals
Fig. 13 Power Spectrum of GPS and Galileo Signals
The results of the remaining scenarios (5, 6 and 7) are presented in figures (14, 15 and 16). The power distribution of the received signals in these figures proves that a single signal power peak is present the FNZ from our BPSR. The position of this power peak determines the type of the received signal.
Fig. 14 Power Spectrum of Glonass Signal
Fig. 15 Power Spectrum of Galileo Signal
Fig. 16 Power Spectrum of GPS Signal
273
IV. CONCLUSION
This paper proposes two algorithms to rapidly detect any available GNSS signals at the RF front-end based on using BPS techniques.
The first algorithm is based on folding the carrier frequency of the three GNSS signals with their harmonics to the FNZ. Volterra Series is then used to analyze the behavioral model of this algorithm.
The second algorithm filters out the left-sideband of the Galileo signal and the right-sideband of the Glonass signal. This prevents the overlapping between these two folded signals with the 3rd harmonic of the GPS signal in the FNZ.
Simulation results show that our two proposed algorithms are good candidates for GNSS signals detection in the RF front-end. This eliminates the need to search and process signals that are not available at the time, thus saving resources and power.
REFERENCES
[1] G. Website, “GNSS MARKET REPORT,” vol. Issue 2, May
2012. [2] J. Chandrasekhar and C. R. Murthy, “GNSS Signal Detection
Under Noise Uncertainty,” in Communications (ICC), 2010 IEEE International Conference on, 2010.
[3] D. Akos, M. Stockmaster, J. Tsui and J. Caschera, “Direct bandpass sampling of multiple distinct RF signals,” Communications, IEEE Transactions on, vol. 47, no. 7, pp. 983-988, jul 1999.
[4] P. Ková andr, P. Kacmarik and F. Vejrazka, “Interoperable GPS, GLONASS and Galileo software receiver,” Aerospace and Electronic Systems Magazine, IEEE, vol. 26, no. 4, pp. 24-30, april 2011.
[5] K. a. C. D. Borre, “A Software-Defined GPS and Galilieo Receiver,” Birkhauser Boston, 2007.
[6] “European GNSS (Galileo) Open Service, Signal In Space Interface Contorl Document (OS SIS ICD),” Eurpean Union 2010, Sep 2010.
[7] D. Akos, A software radio approach to global navigation satellite system receiver design, Ohio University, 1997.
[8] R. Qi, F. Coakley and B. Evans, “Practical consideration for bandpass sampling,” Electronics Letters, vol. 32, no. 20, pp. 1861-1862, sep 1996.
[9] D. Manolakis and V. Ingle, Applied Digital Signal Processing: Theory and Practice, Cambridge Univ Pr, 2011.
[10] R. Vaughan, N. Scott and D. White, “The theory of bandpass sampling,” Signal Processing, IEEE Transactions on, vol. 39, no. 9, pp. 1973-1984, sep 1991.
[11] P. M. Cruz and N. B. Carvalho, “Modeling band-pass sampling receivers nonlinear behavior in different Nyquist zones,” in Microwave Symposium Digest (MTT), 2010 IEEE MTT-S International, 2010.
[12] D. Akos, A. Ene and J. Thor, “A prototyping platform for multi-frequency GNSS receivers,” in Proceedings of the 16th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GPS/GNSS 2003), 2001.
[13] A. Noroozi, C. Verhoeven, G. Monna and E. Gill, “A direct
conversion GPS/Galileo receiver front-end for space applications,” in Satellite Navigation Technologies and European Workshop on GNSS Signals and Signal Processing (NAVITEC), 2010 5th ESA Workshop on, 2010.
[14] A. Zhu, J. Dooley and T. Brazil, “Simplified Volterra series based behavioral modeling of RF power amplifiers using deviation-reduction,” in Microwave Symposium Digest, 2006. IEEE MTT-S International, 2006.
[15] S. Chatterjee and A. Hadi, Regression analysis by example, vol. 607, John Wiley and Sons, 2006.
[16] M. Muha, C. Clark, A. Moulthrop and C. Silva, “Validation of power amplifier nonlinear block models,” in Microwave Symposium Digest, 1999 IEEE MTT-S International, 1999.
[17] P. Fishman and J. Betz, “Predicting performance of direct acquisition for the M-code signal,” in Proceedings of the 2000 National Technical Meeting of The Institute of Navigation, 2000.
[18] N. Martin, V. Leblond, G. Guillotel and V. Heiries, “BOC (x, y) signal acquisition techniques and performances,” in Proceedings of the 16th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GPS/GNSS 2003), 2001.
