Open Access
Add to BibSonomy
Abstract
An extended long-short ambiguity resolution (AR) approach to multi-antenna GNSS-over-fiber systems is proposed and experimentally demonstrated, providing enhanced attitude determination. GNSS signals are remotely acquired from M extended antennas and then transferred to a central node. Through virtual baseline processing, multiple baselines with intermediate length are constructed. Consequently, although the adjacent baseline ratio is limited, the overall aperture is expanded. The carrier phase measurements of equal length virtual baselines are averaged by following the parity detection process for reducing noise, leading to refined baseline estimation. Also, the extended long-short AR method is a single-epoch and search-free one, enabling both improved accuracy and real-time implementations. Both the AR success rate and accuracy of baseline estimation are remarkably enhanced, thus resulting in enhanced attitude determination. Full 3-D attitude determination is also possible with the proposed angle-type or cross-type planar array. In the experiments, the ambiguity is successfully resolved for more than 1400 epochs. The standard deviation of the baseline vector is reduced to almost one-third in both horizontal and vertical directions, without specific delay monitoring hardware.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
Attitude determination based on the Global Navigation Satellite System (GNSS) is an important application of satellite-based navigation, which has been widely used in land vehicles [1], Unmanned Aerial Vehicles (UAV) [2], and spacecraft [3,4]. The GNSS-based approach is capable of providing drift-free attitude information compared to the Inertial Navigation System (INS) counterpart. Moreover, it benefits from GNSS for high precision, high rate, low cost, and increasing availability from multiple GNSS constellations, e.g., Global Positioning System (GPS), GLObal NAvigation Satellite System (GLONASS), Galileo and BeiDou navigation satellite System (BDS) [5].
Conventionally, two antennas are rigidly connected to the underlying platform. The attitude information of the platform is obtained from the baseline vector of the two antennas using differential GNSS techniques [6]. It is well understood that the longer the baseline spans, the greater accuracy the attitude angles achieve. For achieving a better attitude estimation, a longer baseline is required, especially for large aircraft and ship platforms. However, due to the high loss, high weight, and complex thermal variation of the copper cables, the length of the baseline is limited. The GNSS-over-fiber system provides a promising solution to overcome this problem [7–11]. Photonic approaches were proposed for angle-of-arrival (AoA) estimation [12,13], whereas the GNSS-over-fiber system solves the inverse problem of estimating the receiver coordinate from the known angle of a signal. At the remote antenna end, the GNSS signal is used to modulate the optical carrier from a laser for long, remote transmission in a low-loss fiber link. In addition to the GNSS signal, auxiliary radio frequency (RF) signals are incorporated to monitor the path delay, e.g., local oscillator (LO) signal for relative delay monitoring and GNSS signal downconverting [8,9,14], linear frequency modulated (LFM) signal for delay monitoring [10], and a sinusoidal signal of $2$ GHz for phase-derived relative range measurement [11]. These architectures not only allow for a remote distribution between antennas and the processing center but also enable the use of the single difference algorithm for vertical precision improvement. However, for the improvement of vertical precision, special delay monitoring hardware is required.
Generally, the carrier phase is used in most high-precision relative positioning algorithms. The reason is that the carrier phase measurements have two orders of magnitude higher accuracy than pseudorange. However, the carrier phase has unknown integer ambiguities. Once the integer ambiguities are correctly resolved, the carrier phase measurements can be used for high-precision attitude determination. So, it is vital to resolve the carrier phase ambiguities, accurately, and reliably. There are two types of ambiguity resolution (AR) methods: search-based and search-free AR methods. The search-based AR methods are mainly based on the Least-squares AMbiguity Decorrelation Adjustment (LAMBDA) approach [15,16], while the search-free AR methods mostly rely on the long-short baseline method [17]. The former approach has the optimality [18], and success rate [19,20] rigorously studied as well. Further, arrays with fixed prior geometry are also introduced to obtain precise attitude and position solutions [21], which has no restrictions on the array geometry. However, the computation load of the search-based approaches is heavy. The later enjoy lightweight computational load and are suitable for high-rate implementations. The deployment difficulty from inter-sensor spacing less than half wavelength was solved by the construction of virtual baselines [22,23] and the combination of dual-frequency measurements [24]. However, a primary problem for this approach is that the short baseline is smaller than half wavelength. The ratio that the long baseline over the short one is also restricted or the ambiguities of the long baseline cannot be correctly resolved due to the amplification of noise within the measurements.
In this paper, an extended long-short Ambiguity Resolution (AR) approach to multi-antenna GNSS-over-fiber systems with enhanced attitude determination is proposed and experimentally demonstrated. Both the AR success rate and accuracy of baseline estimation are remarkably enhanced. In the proof-of-concept experiments, the ambiguity is successfully resolved in all 1458 valid epochs. The standard deviation of the baseline vector is reduced to almost one third in both horizontal and vertical directions, without specific delay monitoring hardware.
Notations: Scalars, column vectors, and matrices are expressed by regular, bold lowercase and bold uppercase letters, respectively. The superscript $^T$ corresponds to the transpose, and $cov(\cdot )$ denotes the covariance matrix of a random vector. Other conventional notations, e.g., $\|\cdot \|$ for the Euclidean norm of a vector, $|\cdot |$ for the absolute value of a scalar, $\mathbb {R}$ for the real numbers, and $\mathbb {Z}$ for the domain of integer numbers, are also adopted.
2. Principle
Figure 1 shows the block diagram of the proposed GNSS-over-fiber system for attitude determination. $M$ antennas are placed along a line with the position of the $m$-th antenna denoted by the scalar spacing to the first antenna as,
The position vector $\mathbf {p}_1$ is assumed to be the zero vector, hence the scalar spacing and the position vector of the $m$-th antenna is related as $p_m=\|\mathbf {p}_m\|$. The grid size is $d$ that is shorter than half wavelength. The minimum inter-antenna spacing is denoted as $Ad$, where $A$ is a design parameter. When $M=2$, the traditional two antenna attitude determination approach is obtained [25]. When $M=3$, the traditional three-antenna geometry is obtained. The carrier phase measurement model is,
Fig. 1. Block diagram of the proposed system for attitude determination. Ant., antenna; Bias-T, three-port bias-tee for power feed; LNA, low-noise amplifier; DML, directly modulated laser; Fiber, single-mode fiber; PD, photodetector; Receiver, GNSS receiver; DSP, digital signal processing block.
The baseline vector is related to the Double Difference (DD) measurements as,
The proposed AR method is based on the long-short baseline approach. Specifically, the AR for physical or virtual baselines with a length smaller than half wavelength is accomplished by integer rounding. The coarse estimation of the short baseline is then used to resolve the integer ambiguities within long baselines. For example, two virtual baselines with a length smaller than half wavelength are constructed in Fig. 1,
According to Eq. (3), the difference of unit direction vectors are also bounded in the norm. Those with norm smaller than one are selected out as valid DD carrier phase measurements, i.e., $\| \mathbf {e}_i - \mathbf {e}_1 \| \le 1$. Since $d\,<\,\lambda /2$, the true carrier phase variation for these selections are bounded by $0.5$ cycles, i.e., $\left |\left ( \mathbf {e}_i - \mathbf {e}_1\right )^T \mathbf {b}_{s} \right |\, < \,0.5$ for $\| \mathbf {b}_{s} \| = d \,< \,\lambda /2$. So they can be obtained by direct rounding. As a result, the short baseline vector is determined with more than three DD observations. For long baselines, with known designed length ratio $k$, coarse estimation of their baseline vector is obtained as $\mathbf {b}_{l} = k \mathbf {b}_{s}$. It is then substituted back into Eq. (3) and resolve the integer ambiguities by rounding. The subscripts $_s$ and $_l$ represent “short” and “long” respectively.Each element of the DD carrier phase is rounded to estimate the ambiguities in the traditional methods, which is abbreviated as Direct Rounding of Long-Short baseline (DRLS). However, the degrees of freedom in Eq. (3) is only $(M-1)+3$, which means the short baseline vector is determined by any subset of three DD carrier phase measurements. So, for better accuracy and robustness, Improved Rounding of Long-Short baseline (IRLS) is proposed for short baseline estimation. Specifically, the fractional residuals of valid DD carrier phase measurements are sorted by absolute value in an ascending way. The first $n$ elements are selected out to determine a rough estimate of the short baseline denoted as $\mathbf {b}_{n}$. The integer ambiguities of the $n+1$ measurements of the DD carrier phase with the smallest absolute fractional residuals are then resolved by Eq. (3). The short baseline and the integer ambiguities are resolved in this alternative way until all elements of valid DD carrier phase measurements are processed. The proposed alternative processing approach is denoted as IRLS(n).
It is called extended long-short AR processing when applying IRLS to the $M$ antenna system with $M>3$. An advantage of the proposed $M$ antenna attitude determination system is that multiple virtual short baselines can be constructed. The corresponding measurements are averaged to reduce noise. Before averaging, the parity of the unknown integer ambiguities within each measurement needs to be resolved, or a half-cycle in the fractional part results from the average of two carrier phase measurements with different parity. The artificial half-cycle breaks the integer characteristic of the ambiguities and makes the traditional AR algorithms fail. The measurement model is illustrated as,
where $\mathbf {e}$ is the corresponding direction vector of DD. It is proposed to detect the parity consistency of $N_{s,1}$ and $N_{s,2}$ by difference, The parity is the same if the rounding of the difference is even. Otherwise, the parity is different, and a half-cycle is compensated when averaging the measurements. The estimation of short baseline benefits from noise reduction, and the estimation of longer baselines depends on the estimation of the shorter one. Consequently, the multi-antenna GNSS-over-fiber system for enhanced attitude determination is achieved.The DD and baseline difference operations are linear transformations, which can be modeled by a transformation matrix $\mathbf {T}$. For example, the carrier phase difference operations of the two virtual baselines corresponding to Eq. (4) and Eq. (5) are modeled as,
Let $\mathbf {L}=\begin {bmatrix}\phi _1 & \cdots & \phi _4\end {bmatrix}^T$ denotes the carrier phase measurement vector of the four antennas. The covariance matrix of $\mathbf {L}$ is $cov(\mathbf {L}) = \sigma ^2 \mathbf {I}_4$, i.e., a four-dimensional identity matrix due to the i.i.d assumption. The equivalent measurement of the virtual short baselines $\phi _{s,1}$ and $\phi _{s,2}$ are then expressed as $\mathbf {V}= \begin {bmatrix}\phi _{s,1} & \phi _{s,2}\end {bmatrix}^T = \mathbf {T}\mathbf {L}$, and the covariance matrix is written as,Full attitude, possibly combined and integrated with positioning, instead of the one-dimensional attitude is generally required in practical applications [21]. For that purpose, at least two unparallel linear arrays are required. An angle-type configuration composed of the traditional antenna triplets was proposed in [25]. Furthermore, we propose that both the angle-type and cross-type planar arrays can be utilized in practice, as depicted in Figs. 2(a) and 2(b). In these configurations, antennas 1-2-3-4, 1-5-6-7, and 5-2-6-7 compose the proposed linear arrays, where the antenna at the cross is reused. By applying the previous techniques, two baseline vectors can be determined. The full 3-D attitude can be determined by transforming the known array geometry from the local to the global frame using the classic algebraic method as [21,25],
where $\mathbf {B}\in \mathbb {R}^{3\times (M-1)}$ is the baseline matrix composed of baseline vectors in the global frame (e.g. WGS84), $\mathbf {R}\in \mathbb {R}^{3\times 2}$ is the rotation matrix with orthonormal column vectors, and $\mathbf {F}\in \mathbb {R}^{2\times (M-1)}$ is the known array geometry matrix of the planar array expressed in the local frame. The full attitude information is parametrized by the rotation matrix $\mathbf {R}$, which can be solved in a Least-squares sense once $\mathbf {B}$ is solved by the proposed techniques [21].Fig. 2. Antenna configurations for 3-D attitude determination: (a) The angle-type array; (b) The cross-type array.
The geometric transformation methods are well studied, and the focus of this work is the extended long-short approach to the multi-antenna linear array. Therefore the following sections focus on the performance of estimating the individual baseline vectors.
3. Experiment setup and results
To verify the proposed system and processing algorithm, an experimental setup is established, as shown in Fig. 3. Four multi-frequency GNSS antennas (JNXNA-531S) are placed on a line strip. The minimum spacing between two antennas is $0.48$ m, and $d$ is set to be half wavelength of $\lambda _{B1}/2=0.096$ m. The antennas are connected to four specially made boxes, each of which contains a bias-tee (NMRF-KBT-1), an LNA (GNA-109T), a DML (KG-ELD-15-10G-SM-FA), and the corresponding power supply. This box accepts an input of RF signal and upconverts it into an optical carrier for long-distance and low-loss transmission. The GNSS signals are recovered using photodiode at the central processing place. Four GNSS OEM boards (two ComNav TH-BD05F and two NovAtel 628) demodulate the navigation messages and generate carrier phase measurements at the rate of $1$ Hz. The GNSS receivers are configured to gather the measurements using a MOXA hub. All measurement data of the GNSS systems are recorded, e.g., the pseudorange, carrier phase, Doppler, carrier noise ratio, and lock loss indicator information. The collected data are then post-processed with the proposed extended long-short processing algorithm using MATLAB.
Fig. 3. Experimental setup of the proposed real-time fiber-based GNSS attitude determination system. Ant., antenna; Bias-T, three-port bias-tee for power feed; LNA, low-noise amplifier; DML, directly modulated laser; Fiber, single-mode fiber; PD, photodetector; Receiver, GNSS OEM receiver; MOXA Hub, UPort 4-port USB-to-serial converter.
The satellite constellation is commonly visible by four receivers, as depicted in Figs. 4(a) and 4(b). The GPS satellites are numbered from $1$ to $32$, while the BDS satellites are numbered from $33$ to $67$. Four commonly visible geostationary satellites of BDS are $33$, $34$, $35$, and $37$. The $20^{\textrm {th}}$ and $21^{\textrm {st}}$ GPS satellites are only visible for a short while. For a successful estimation of the three-dimensional vector, a minimum number of four satellites need to be commonly viewed. In the experiment, the multi-frequency antennas support both GPS L1 & L2 and BDS B1 & B2. Therefore, both GPS L1 and BDS B1 are used to improve the satellite number and geometry. However, the proposed single-frequency processing is still applied, and no additional dual-frequency combination techniques as [24] are used.
Fig. 4. The satellite skyplot: (a) 2-D constellation variation; (b) 3-D constellation at the start of the test.
The traditional three-element antenna array is a special case of the proposed $M$ antenna array geometry when $M=3$. The performance of attitude determination based on three and four antennas are compared. The algorithms of traditional DRLS and proposed IRLS are compared as well. The results of short baseline estimation via DRLS and IRLS(4) are presented in Figs. 5(a)–5(d), where each component of the baseline vector is depicted. The results of long baseline estimation are presented in Figs. 6(a)–6(d). Another performance index is the AR success rate. More test results for both array schemes and more configurations of processing algorithm (from IRLS(3) to IRLS(5)) are summarized in Table 1.
Fig. 5. Short baseline estimation of the traditional three-element array (the first and third columns, (a) and (c)) and the proposed four-element array (the second and fourth columns,(b) and (d)) via traditional DRLS and proposed IRLS(4) algorithm.
Fig. 6. Long baseline estimation of the traditional three-element array (the first and third columns, (a) and (c)) and the proposed four-element array (the second and fourth columns, (b) and (d)) via traditional DRLS and proposed IRLS(4) algorithm.
Table 1. Baseline estimation with traditional three-element and proposed four-element array via DRLS, IRLS(3), IRLS(4), and IRLS(5). The Valid Epochs (VE) are those the algorithm can run. The Failure Counts (FC) of AR are obtained by comparing the AR results to the target value computed with the calibrated baseline vector. SD is the Standard Deviation. SR is the Success Rate. The S- and L- prefixes represent the quantity of short and long baselines, respectively.
From Figs. 5(a)–5(d), it can be seen that all components of the baseline vector are improved. The Standard Deviation (SD) of the East component decreases from $1.27$ cm to $0.44$ cm to almost one third comparing Figs. 5(a) and 5(b). Comparing Figs. 5(b) and 5(d), it can be seen the results of DRLS and IRLS(4) of the four-element array are the same. The reason is that all the integer ambiguities are correctly resolved. As a result, the same baseline vectors are estimated from the same unambiguous carrier phase measurements. However, in the case of the three-element array, the accuracy of the carrier phase measurements is not improved by averaging. The traditional DRLS failed several epochs in AR, whereas the proposed IRLS(4) algorithm correctly resolve all the ambiguities.
From Figs. 6(a)–6(d), the performance improvement from three-element array to four-element array is obvious. However, no clear improvement between DRLS and IRLS(4) is observed for the three-element array, although there is some (see Table 1). Because the estimation of long baseline mainly relies on the accuracy of the short baseline estimation. The scaling ratio $k$ also introduce uncertainties of the coarse estimation of the long baseline. So, more AR failures are present than the short baseline estimation (cf. Fig. 5(a) and Fig. 6(a)).
From Table 1, the proposed four-element array scheme outperforms the traditional three-element array scheme in both AR success rate and accuracy of baseline estimations. In both schemes, IRLS(4) and IRLS(5) have similar performance as DRLS and outperform IRLS(3). The quality of the carrier phase measurements is improved via the parity detection and averaging. As can be seen, the standard deviation of the short baseline estimation is reduced to almost one third in all directions. Although IRLS(3) requires the least satellite number, it shows a bad performance. From the results, IRLS(4) is preferable due to a better trade-off between accuracy and the requirement of satellite number.
4. Conclusion
We proposed and experimentally demonstrated a multi-antenna GNSS-over-fiber system, providing enhanced attitude determination. By extending the traditional three-element array to $M$ antenna array, enlarging the overall aperture. A novel search-free AR approach was proposed, benefiting from the averaging of multiple virtual baselines with the same length. Both the AR success rate and accuracy of baseline estimation were enhanced. Full 3-D attitude determination is also possible with the proposed angle-type or cross-type planar array. In the proof-of-concept test, the standard deviation of the baseline estimation was reduced to almost one third in the proposed scheme. The integer ambiguities of more than 1400 valid epochs were successfully resolved. Further accuracy improvement can be expected with single difference processing if the path delays are monitored by specialized hardware, which is the focus of our future work.
Funding
National Natural Science Foundation of China (61775185, 61901397); Sichuan Province Science and Technology Support Program (2018HH0002, 2019JDJQ0022); 111 Plan (B18045).
Acknowledgments
The authors would like to thank the anonymous reviewers for their valuable comments that help improve this paper.
References
1. W. T. Chen and H. L. Qin, “New method for single epoch, single frequency land vehicle attitude determination using low-end GPS receiver,” GPS Solut. 16(3), 329–338 (2012). [CrossRef]
2. G. Falco, M. C. C. Gutiérrez, E. L. Serna, F. Zacchello, and S. Bories, “Low-cost real-time tightly-coupled GNSS/INS navigation system based on carrier-phase double-differences for UAV applications,” in Proceedings of the 27th International Technical Meeting of the Satellite Division (ION GNSS 2014), (ION, 2014), pp. 841–857.
3. G. Allende-Alba and O. Montenbruck, “Robust and precise baseline determination of distributed spacecraft in LEO,” Adv. Space Res. 57(1), 46–63 (2016). [CrossRef]
4. P. J. Buist, P. J. G. Teunissen, S. Verhagen, and G. Giorgi, “A vectorial bootstrapping approach for integrated GNSS-based relative positioning and attitude determination of spacecraft,” Acta Astronaut. 68(7-8), 1113–1125 (2011). [CrossRef]
5. P. Xu, Y. Shu, X. Niu, J. Liu, W. Yao, and Q. Chen, “High-rate multi-GNSS attitude determination: experiments, comparisons with inertial measurement units and applications of GNSS rotational seismology to the 2011 Tohoku Mw9.0 earthquake,” Meas. Sci. Technol. 30(2), 024003 (2019). [CrossRef]
6. S. T. Goh and K.-S. Low, “Survey of global-positioning-system-based attitude determination algorithms,” J. Guid. Control. Dyn. 40(6), 1321–1335 (2017). [CrossRef]
7. L. M. Pessoa, J. M. B. Oliveira, D. Coelho, J. C. S. Castro, H. M. Salgado, and M. Fames, “Transmission of differential GPS signals over fiber for aircraft attitude determination,” in IEEE Avionics, Fiber-Optics and Photonics Digest CD, (IEEE, 2012), pp. 80–81.
8. D. Macias-Valadez, R. Santerre, S. Larochelle, and R. Landry, “Improving vertical GPS precision with a GPS-over-fiber architecture and real-time relative delay calibration,” GPS Solut. 16(4), 449–462 (2012). [CrossRef]
9. D. Macias-Valadez, “Improvement of vertical precision in GPS positioning with a GPS-over-fiber configuration and real-time relative hardware delay monitoring,” Ph.D. thesis, Université Laval, Canada (2011).
10. X. Jiang, X. Wang, and S. Pan, “A multi-antenna GNSS-over-fiber system with high vertical precision,” in 2018 International Topical Meeting on Microwave Photonics (MWP), (IEEE, 2018), pp. 1–4.
11. X. Wang, X. Jiang, S. Li, and S. Pan, “Multiantenna GPS-over-fiber system for attitude determination using phase-derived range measurement,” IEEE Photonics J. 11(4), 1–10 (2019). [CrossRef]
12. X. Zou, W. Li, W. Pan, B. Luo, L. Yan, and J. Yao, “Photonic approach to the measurement of time-difference-of-arrival and angle-of-arrival of a microwave signal,” Opt. Lett. 37(4), 755 (2012). [CrossRef]
13. P. Li, L. Yan, J. Ye, X. Feng, W. Pan, B. Luo, X. Zou, T. Zhou, and Z. Chen, “Photonic approach for simultaneous measurements of Doppler-frequency-shift and angle-of-arrival of microwave signals,” Opt. Express 27(6), 8709 (2019). [CrossRef]
14. D. Macias-Valadez, R. Santerre, S. LaRochelle, and R. J. Landry, “GPS-over-fiber architecture with relative cable delay monitoring for high precision GPS applications,” in Proceedings of the 22nd International Technical Meeting of the Satellite Division (ION GNSS 2009), (ION, 2009), pp. 622–634.
15. P. J. G. Teunissen, “The least-squares ambiguity decorrelation adjustment: a method for fast GPS integer ambiguity estimation,” J. Geod. 70(1-2), 65–82 (1995). [CrossRef]
16. Y. Li, “A novel ambiguity search algorithm for high accuracy differential GNSS relative positioning,” Aerosp. Sci. Technol. 78, 418–426 (2018). [CrossRef]
17. A. El-Mowafy and K. P. Schwarz, “Epoch-by-epoch ambiguity resolution for real-time attitude determination using a GPS multi-antenna system,” Navigation 42(2), 391–408 (1995). [CrossRef]
18. P. J. G. Teunissen, “An optimality property of the integer least-squares estimator,” J. Geod. 73(11), 587–593 (1999). [CrossRef]
19. P. J. G. Teunissen, “The success rate and precision of GPS ambiguities,” J. Geod. 74(3-4), 321–326 (2000). [CrossRef]
20. P. J. G. Teunissen, “Success probability of integer GPS ambiguity rounding and bootstrapping,” J. Geod. 72(10), 606–612 (1998). [CrossRef]
21. P. J. G. Teunissen, “A-PPP: array-aided precise point positioning with global navigation satellite systems,” IEEE Trans. Signal Process. 60(6), 2870–2881 (2012). [CrossRef]
22. L. Lu, Y. Li, and C. Rizos, “Virtual baseline method for Beidou attitude determination: an improved long-short baseline ambiguity resolution method,” Adv. Space Res. 51(6), 1029–1034 (2013). [CrossRef]
23. T. Ballal and C. J. Bleakley, “Phase-difference ambiguity resolution for a single-frequency signal,” IEEE Signal Process. Lett. 15, 853–856 (2008). [CrossRef]
24. S. Liu, L. Zhang, J. Li, and Y. Luo, “Dual frequency long-short baseline ambiguity resolution for GNSS attitude determination,” in 2016 IEEE/ION Position, Location and Navigation Symposium (PLANS), (IEEE, 2016), pp. 630–637.
25. T. Ballal and C. J. Bleakley, “GNSS instantaneous ambiguity resolution and attitude determination exploiting the receiver antenna configuration,” IEEE Trans. Aerosp. Electron. Syst. 50(3), 2061–2069 (2014). [CrossRef]
26. Y. Li, X. Zou, B. Luo, W. Pan, L. Yan, and P. Liu, “A 2q-order difference-set approach to eliminate phase ambiguity of a single-frequency signal,” IEEE Signal Process. Lett. 26(10), 1526–1530 (2019). [CrossRef]
