1. INTRODUCTION
The stellar inertial navigation system (stellar-INS) includes a star sensor and inertial navigation system (INS) which is fully dormant and widely used in near-Earth flight vehicles (Lewis et al., Reference Lewis, Hockbruckner and Reeve1991; AIAA, 2000; Malay et al., Reference Malay, Gaylor and Davis2005; Li et al., Reference Li, Needelman, Fowell, Tsao and Wu2006). The INS provides continuous attitude information with accumulated error over time, and the star sensor provides star vector or attitude reference to correct the accumulated attitude error (Zhang et al., Reference Zhang, Niu and Lu2017). There are three main factors that affect the accuracy of the stellar-INS integration attitude: star vector accuracy, INS accuracy and star observation scheme. For the stellar-INS, improvement of the star observation scheme has become an effective way to improve integration attitude accuracy.
Obviously, the aim of star observation scheme is to improve the accuracy of integration attitude by improving the observability of star sensor installation error, gyro error and initial platform angle error. Lu et al. (Reference Lu, Lei, Liang and Yang2017) proposed INS error calibration methods by using star vectors, which can improve the inertial navigation accuracy. Yang et al. (Reference Yang, Zhang and Lu2010) deduced the observability of stellar-INS. In fact, the star vector orientation in the field of view (FOV) also affects the observability of the above errors. Therefore, it is desirable to reveal the relationship between the star vector orientation and integrated navigation performance.
The star vector orientation in the FOV is decided by the star sensor field of view angle. In terms of the FOV, there are two kinds of star sensor: narrow field of view (NFOV) and large field of view (LFOV). Although stellar-INS with NFOV star sensor is widely used in near-Earth flight vehicles, there is still the problem of degradation of integration navigation performance due to the constrained star vector orientation in the FOV (Badshah et al., Reference Badshah, Yongyuan and Zhang2015; Ning et al., Reference Ning, Gui, Xu, Bai and Fang2016; Huaming et al., Reference Huaming, Di and Yonghui W2019). In its early stages, the platform-based stellar-INS conducted star vector orientation by turning the platform (Nash and Wells, Reference Nash and Wells1975). Later, the LFOV star sensor was proposed to solve the star vector orientation constraints problem with the development of Charge Coupled Device. Meawhile, the LFOV star sensor also reduced the requirements of attitude adjustment for star observation in carrier flight.
With the development of the LFOV star sensor (Aghaei and Moghaddam, Reference Aghaei and Moghaddam2016; Guo, Reference Guo2016; Pu et al., Reference Pu, Li and Guo2017), it has been widely applied in the satellite platform. In the existing studies on LFOV, the main research focuses on the methods for improving the performance of LFOV star sensors (Liu et al., Reference Liu, Li and Tan2010; Yang et al., Reference Yang, Zhang and Lu2010; Zhang et al., Reference Zhang, Niu, Zhang and Lu2018a). However, there are still two main disadvantages of the LFOV star sensor in application: one is the poor accuracy of the single star vector, the other is the selection of effective star vector based on the orientation in the FOV. For the first disadvantage (Qin et al., Reference Qin, Zhan, Zheng, Wu, Jia and Fu2016) proposed a gyro-assisted method to improve the star vector measurement ability. The second disadvantage has not been analysed prior to this study; in other words, the effect of star vector orientation in the FOV or the star sensor FOV angle on integrated navigation performance has not been studied. The aim of this study is to reveal the relationship between the star vector orientation in the FOV and integrated navigation performance.
In this paper, section 2 introduces the star vector observation principle of stellar-INS. In section 3, the sensitivity in three axes and the geometry factor of the star sensor is defined to feature the difference between the optical axis direction and the non-optical axis direction. The observation equation including the star sensor installation error and the platform angle error is established to reveal the relationship between the star vector orientation in the FOV and integrated navigation performance. Section 4 describes the simulations and experiments conducted to prove the revealed influence mechanism. The results of the analysis of this study are crucial for improvement of the design and accuracy of stellar-INS.
2. STAR VECTOR OBSERVATION PRINCIPLE OF STELLAR-INS
The star vector in the inertial frame is observed by the star sensor (Zhang et al., Reference Zhang, Xing and Sun2018b). The star vector can also be calculated with the help of the INS through navigation solutions. Through the Kalman filter, the two kinds of star vector can be fused to estimate the error states of stellar-INS, including the device errors and the platform errors. The estimated values of error states are used to correct the corresponding navigation parameters in the INS mechanical scheduling. Finally, the optimal navigation parameters of the carrier are output.
The corresponding observation equation for the star vector can be written as (Yang et al., Reference Yang, Zhang, Lu and Zhang2018)
$$\begin{aligned} {\textbf{Z}} & ={\textbf{L}}_s -{\textbf{L}}_c =-({\boldsymbol{\psi}}_s \times )(\hat{\textbf{C}}_{\rm e}^{c} {\textbf{C}}_i^e {\textbf{L}}_i ) \\ & =[(\hat{\textbf{C}}_{\rm e}^c {\textbf{C}}_i^e {\textbf{L}}_i)\times ]{{\boldsymbol{\psi} }}_s \end{aligned}$$
where 
${\textbf{L}}_{s} $ is the star vector information in the platform navigation frame (s-frame). 
${\textbf{L}}_{c} $ is the star vector information in the computer navigation frame (c-frame). 
$\boldsymbol{\psi}_{s}$ is the platform misalignments of INS in the platform navigation frame. 
$\hat{\textbf{C}}_{\rm e}^{c} $ is the matrix from the Earth frame to the computer navigation frame. 
${\textbf{C}}_{i}^{e} $ describes the relationship between the geocentric inertial frame and the Earth frame owing to the Earth's rotation. 
${\textbf{L}}_{i} $ is the star vector information in the geocentric inertial frame. 
$( \hat{\textbf{C}}_{\rm e}^{c} {\textbf{C}}_{i}^{e} {\textbf{L}}_{i} \times ) $ is the skew symmetric matrix of the vector 
$\hat{\textbf{C}}_{\rm e}^{c} {\textbf{C}}_{i}^{e} {\textbf{L}}_{i} $.
3. INFLUENCE MECHANISM OF STAR VECTOR ORIENTATION
The FOV angle, as a key technical index, ranges from a few arc-minutes to more than 20° and it depends on the size of the detector chip and the focal length of the optical system. The smaller the FOV angle, the higher the accuracy of the single star vector and the larger the lens aperture, resulting in a larger volume and weight of the star sensor. Conversely, the larger the FOV angle, the smaller the volume, and the greater number of stars, the higher accuracy of the attitude information can be output by multi-star vector fitting. In terms of the star sensor, whether it is the star vector information output by the NFOV or the attitude information output by the LFOV, the measurement accuracy along the direction of the optical axis is worse than that of the other two directions. Similarly, the influence mechanism is different in various star vector orientations.
In general, the astronomical angle of the star vector is resolved by the star point in the imaging plane. Owing to the focal length of the star sensor being much larger than the imaging plane size, the sensitivity is lower in the optical direction. Defining the star sensor imaging plane as the O xz plane, the optical axis is the y-axis. Taking the star vector projection in the O xz plane as an example, the imaging plane sensitivity of the star sensor in the non-optical axis direction can be analysed as below.
As shown in Figure 1, O xz is the star sensor imaging plane, f is the optical system focal length. The predicted imaging P-point is calculated with the help of INS navigation information, whose coordinates in the image plane are 
$( {0, \; z_{c} } ) $. The intersection angle between the star vector and the optical axis can be written as αz. The measurement imaging point P̂ is obtained by the star sensor, its actual coordinates in the imaging plane are 
$( {0, \; \hat {z}} ) $. The intersection angle between the star vector and the optical axis can be written as α̂. The sensitivity of the z-axis in the imaging plane is defined as follows.
$$ \eta _z =\frac{\Delta _z }{\Delta _\alpha }=\frac{\hat{z}-z_c }{\hat{\alpha }-\alpha _z } $$
where Δz is the variation of imaging point coordinates of the star sensor and 
$\Delta _{\alpha} $ is the star vector angle variation.
Figure 1. Imaging plane of the star sensor in non-optical axis direction.
The imaging point coordinate variation Δz can be expressed as
$$\Delta _z =\hat{z}_{c}-z_c =f\times \left( {\frac{\tan (\alpha_z )+\tan (\Delta_\alpha )}{1-\tan (\alpha_z )\times \tan (\Delta_\alpha )}-\tan (\alpha_z )} \right)$$ The angle variation of the output star vector 
$\Delta _{\alpha} $ can be given by
$$\left\{\begin{matrix} {\alpha_z \le 10^{\circ}\Rightarrow \tan (\alpha_z )\le 0.1763} \\ {\Delta_\alpha \le 1^{\circ}\Rightarrow \tan (\Delta_\alpha )\le 0.0175} \end{matrix}\right\}\Rightarrow \tan (\alpha _z )\times \tan (\Delta_\alpha )\le 0.0031$$Then, Δz can be approximated as follows.
$$\Delta_z \approx f\times (\tan (\Delta _\alpha ))$$When the angle variation is less than 1°, a small angle approximation is used. The observation information sensitivity of the z-axis can be simplified as
$$\eta _z =\frac{\Delta _z }{\Delta _\alpha }=\frac{f\times (\tan (\Delta _\alpha ))}{\Delta _\alpha }\approx f$$Similarly, the observation information sensitivity of the x-axis can be expressed as
$$\eta _x \approx f$$As shown in Equations (6) and (7), the larger the FOV, the smaller the focal length, and the lower the sensitivity of the orthogonal direction in the imaging plane. The sensitivity of the star sensor in the optical axis direction can be analysed as below.
As shown in Figure 2, the predicted P-point is calculated with the help of INS navigation information, whose coordinates in the image plane are 
$( {x_{c} , \; z_{c} } ) $. The measurement imaging point P̂ is obtained by the star sensor, its actual coordinates in the imaging plane are 
$( {\hat{x}, \; \hat{z}} ) $. 
$\Delta_{\alpha} $ can be defined as the error angle along the optical axis between the predicted P-point and the imaging point P̂. The distance between the imaging point and the actual measured imaging point is defined as
$$d=\left\| {\hat{P}(\hat {x},\hat {z})-P(x_c ,z_c )} \right\|$$
Figure 2. Imaging plane of the star sensor in optical axis direction.
The sensitivity on the optical axis can be expressed as
$$\eta _y =\frac{d}{\Delta _\alpha }$$The coordinate position variation of the imaging point can be obtained by
$$d\approx r\times \Delta _\alpha$$where r is the distance between the imaging point and the centre of the imaging plane, which is calculated by
$$ r=\sqrt {(f\times \tan (\alpha _x ))^2+(f\times \tan (\alpha _z ))^2} $$αx and αz is the ratio of the imaging point coordinates to the focal length, in other words, the intersection angle between the star vector and the x- and the z-axis respectively. The sensitivity on the optical axis can then be simplified as
$$\eta _y =\frac{d}{\Delta _\alpha }=r$$The distance r can be also expressed by the focal length f and the circular FOV Φ.
$$r=f\cdot \tan (\Phi /2)$$To sum up, the imaging plane sensitivity is only related to the focal length. The sensitivity on the optical axis is related not only to the focal length but also to the FOV, and the smaller the FOV, the worse the sensitivity. To describe the relationship between the sensitivity on the optical axis and the sensitivity in the imaging plane, the geometry factor of the star sensor is defined, which is represented by λ.
$$ \lambda =\frac{\eta _{x/z} }{\eta _y }=\frac{1}{\tan (\Phi /2)}$$where ηx/z is the sensitivity in the imaging plane.
For example, assuming that the circular field angle of the LFOV star sensor is 20° and the NFOV star sensor is 3°, the geometry factor of the LFOV star sensor and the NFOV star sensor are 5·63 and 39·19, respectively. Since the FOV star sensor ranges from a few corners to more than 20°, the geometric factor can be approximately linearly related to half of the circular field angle. Compared with the imaging plane, the sensitivity on the optical axis is poor, so the ability to measure small errors is poor. As a result, the installation error estimation accuracy along the optical axis is degraded.
The observability analysis of stellar-INS using the NFOV star sensor has been conducted as below. The star sensor installation error matrix can be defined as
$${\textbf{C}}_{st}^b \approx \left[ {{\begin{matrix} 1 & {\mu_{z }^b } & {-\mu_y^b } \\ {-\mu_{z }^b } & 1 & {\mu_x^b } \\ {\mu_y^b } & {-\mu_x^b } & 1 \\ \end{matrix} }} \right]=[{\textbf{I}}-({\boldsymbol{\mu }}^b\times )]$$ The star vector information 
${\textbf{L}}_{b} $ in the INS body frame can be expressed as below,
$$ {\textbf{L}}_b =\left[ {{\begin{matrix} {B_x } & {B_y } & {B_z } \\ \end{matrix} }} \right]^T=\left[ {{\begin{matrix} 1 & {\mu_{z }^b } & {-\mu_y^b } \\ {-\mu_{z }^b } & 1 & {\mu_x^b } \\ {\mu_y^b } & {-\mu_x^b } & 1 \\ \end{matrix} }} \right]\cdot {\textbf{L}}_{st}$$
where 
${\textbf{L}}_{st} $ is the star vector information in the star sensor body frame, which can be directly obtained by the star sensor. B x, B y and B z are the projection of the star vectors under the INS body frame, which are unit vectors and represent the direction information. Suppose the y-axis is the optical axis and O xz is the imaging plane.
$${\textbf{L}}_{st} =\left[ {{\begin{matrix} {S_x } & {S_y } & {S_z } \\ \end{matrix} }} \right]^T=\frac{\left[ {{\begin{matrix} x & f & z \\ \end{matrix} }} \right]^T}{\sqrt {x^2+f^2+z^2} }$$According to Equations (16) and (17), the ternary equation can be obtained by
$$\begin{cases} B_x =S_x +\mu_{z}^b \cdot S_y -\mu _y^b \cdot S_z\\ B_y =-\mu_{z}^b \cdot S_x +S_y +\mu _x^b \cdot S_z\\ B_z =\mu _{y}^b \cdot S_x -\mu _x^b \cdot S_y +S_z \end{cases}$$ Obviously, the star sensor characteristics is 
$S_{y} \ge S_{x} \approx S_{z} $. As shown in Equation (18), the larger the installation error coefficient, the better the observability of the corresponding installation error. Conversely, the smaller the installation error coefficient, the worse the observability of the corresponding installation error. It can be said that the coefficients S x and S z corresponding to the installation error 
$\mu_{y}^{b} $ on the optical axis are less than the installation error coefficients in both directions 
$\mu_{x}^{b} $ and 
$\mu_{z}^{b} $. Therefore, the observability of the installation error along optical axis is the worst.
Furthermore, considering the platform error angle, the relationship between the star vector measured directly by the star sensor and the star vector calculated in the navigation frame is satisfied by
$$\hat{\textbf{C}}_{\rm b}^n {\textbf{L}}_{st} =({\textbf{I}}-{\boldsymbol{\psi }}\times )({\textbf{I}}+{\textbf{T}}\cdot {\boldsymbol{\mu }}^b\times ){\textbf{L}}_c$$
where 
${\textbf{L}}_{\hbox{st}} $ is the star vector measured directly by the star sensor. It can be converted into the navigation frame and expressed as below,
$$\hat{\textbf{C}}_{\rm b}^n {\textbf{L}}_s =\left[ {{\begin{matrix} {B_{px} } & {B_{py} } & {B_{pz} } \\ \end{matrix} }} \right]^T$$ When the carrier conducts the first star vector observation, T is a unit matrix. Obviously, the platform error angles ψy and the installation error 
$\mu _{y}^{b}$ are coupled. Consequently, the observability of the platform error angle ψy is the same as the installation error 
$\mu _{y}^{b} $, and its accuracy is degraded. Certainly, the other platform error angles ψx and ψz are well estimated.
After the first observation, the attitude of the carrier is adjusted to the T matrix to conduct the second observation. The T matrix should be satisfied as below.
$${\textbf{T}}=\left[ {{\begin{matrix} 0 & 1 & 0 \\ {-1} & 0 & 0 \\ 0 & 0 & 1 \\ \end{matrix} }} \right]$$T is substituted into the star vector observation equation, and can be given by
$$\begin{align} &{\textbf{T}}\cdot ({\boldsymbol{\mu }}^b\times )={\textbf{T}}\cdot \left[ {{\begin{matrix} 0 & {\mu_{z }^b } & {-\mu_y^b } \\ {-\mu_{z }^b } & 0 & {\mu_x^b } \\ {\mu_y^b } & {-\mu_x^b } & 0 \end{matrix} }} \right]=\left[ {{\begin{matrix} {-\mu_{z }^b } & 0 & {\mu_x^b } \\ 0 & {-\mu_{z }^b } & {\mu_y^b } \\ {\mu_y^b } & {-\mu_x^b } & 0 \\ \end{matrix} }} \right]\end{align}$$
$$\begin{align} &{\textbf{L}}_{n(c)} ={\textbf{T}}\cdot {\textbf{L}}_b =\left[{{\begin{matrix} 0 & 1 & 0 \\ {-1} & 0 & 0 \\ 0 & 0 & 1 \\ \end{matrix} }} \right]\cdot \left[ {{\begin{matrix} {S_x } \\ {S_y } \\ {S_z } \\ \end{matrix} }} \right]=\left[ {{\begin{matrix} {S_y } \\ {-S_x } \\ {S_z } \\ \end{matrix} }} \right]\end{align}$$
$$\begin{align} &\left\{ {{\begin{array}{@{}l} {B_{px} =S_{cy} +\mu_{z }^b \cdot S_{cx} -\mu_y^b \cdot S_{cz} }\\ {B_{py} =-\mu_{z }^b \cdot S_{cy} +S_{cx} +\mu_x^b \cdot S_{cz} }\\ {B_{pz} =\mu_{y }^b \cdot S_{cy} -\mu_x^b \cdot S_{cx} +S_{cz} } \end{array} }} \right.\end{align}$$ As shown in Equation (24), the platform error angles ψx and the installation error 
$\mu _{y}^{b} $ are coupled. Consequently, the observability of the platform error angles ψx is the same as the installation error 
$\mu _{y}^{b} $, and its accuracy is degraded. Certainly, the other platform error angles ψy and ψz are well estimated. We can assume that the platform error angles can be considered as a constant value for a short time. After two observations, all the platform error angles ψx, ψy and ψz can be fully estimated. Instead, the installation error 
$\mu _{y}^{b} $ along the optical axis is still unobservable.
It can be concluded that the observability of the installation error on the optical axis is independent of star observation attitude. The observability of the platform error angle is related to star observation attitude, and through adjusting the attitude of the observation, all the platform error angles can be observed without being affected by the installation error on the optical axis.
4. SIMULATION AND EXPERIMENTS
Generally, the star vector orientation is determined by the FOV of the star sensor. To verify the impact of the FOV constraints on the stella-INS, mathematical simulation is performed. The simulation trajectory is a typical ballistic trajectory, which includes an overload acceleration section and a horizontal flight section in near-Earth space. After the initial alignment of the carrier is completed on the ground, the carrier reaches near-Earth space. The attitude adjustment sequence is shown in Figure 3; three axes of stellar-INS should be excited by angular motion to improve the observability. Attitude adjustment and star observation occur seven times, taking 11 s each time, including 6 s for attitude adjustment and 5 s for static star observation, 77 s in total.
Figure 3. The carrier attitude adjustment process.
The characteristics of the stellar-INS are described as follows. The gyro bias is 0·03°/h, the gyro scale error is 20 ppm, and the gyro random walk is 0·006°/h1/2. The accelerometer bias is 
$2\times 10^{-5}$ g, the accelerometer scale error is 50 ppm and the accelerometer white noise is 15 μ g. The gyro installation error is 20 arc-seconds, and the INS sample rate is 100 Hz. The star sensor white noise is five arc-seconds, the star catalogue error is one arc-second, the sample rate is 10 Hz, and the installation error of the star sensor is 20 arc-seconds.
The linear closed loop Kalman filter is used to estimate and compensate the full attitude errors in real time. The error states mainly include platform error angles, gyro bias errors, gyro scale factor, gyro installation errors, and star sensor installation errors. The filter parameter is set according to the sensor device accuracy. Time update is 100 Hz, and the measurement update is 10 Hz. The covariance convergence curves of the mainly error states are plotted in Figure 4.
Figure 4. Covariance convergence curves of the platform errors and the star sensor installation errors.
As shown in Figure 4(a)–4(c), the larger the star sensor FOV, the better the attitude precision. The platform errors basically converge to the order of several seconds in the condition of the narrow FOV. It is evident that the platform errors are weakly affected by the star sensor FOV angle. On the contrary, the estimated precision of the device errors along the optical axis mostly depends on the angle of FOV, as shown in Figure 4(d)–4(f). For example, the covariance convergence value of the star sensor installation error μy is close to 10 seconds. We can conclude that the device errors along the optical axis are seriously affected by the star sensor FOV angle. In summary, the characteristics of device errors and of platform errors are quite different. To further secure the conclusion, the covariance convergence values of the error states are listed in Table 1 in different FOV.
Table 1. Covariance convergence value of error states.
Table 1 shows that the attitude accuracy of the NFOV and the LFOV are at the same level, and the repeatability is approximately the same. However, the installation error estimation accuracy of stellar-INS along the optical axis is not substantially converged in the NFOV. It is verified that the geometric factor of the NFOV is poor. Furthermore, the relevant gyro errors along the optical axis are also not sufficiently converged.
To further compare the attitude accuracy of the NFOV and the LFOV, ground verification testing was carried out. Two stellar-INSs are used in the test, which are integrated by Institute of Optics and Electronics, Beihang University. One stellar-INS consists of the LFOV star sensor and laser Inertial measurement unit (IMU). The other consists of the NFOV star sensor and laser IMU. The ground test system mainly involves a high-precision turntable, a north-facing reference and two sets of orthogonally placed star simulators, and each set of star simulators includes two single star simulators. The characteristics of the stellar-INS are described as follows. The gyro bias is 0·03°/h, the gyro scale error is 20 ppm, and the gyro random walk is 0·006°/h1/2. The accelerometer bias is 
$2\times 10^{-5}$ g, the accelerometer scale error is 50 ppm and the accelerometer white noise is 15 μ g. The NFOV is about 3° and the LFOV is about 10°.
In the ground verification test, the star simulators are installed orthogonally in the horizontal plane to guarantee the calibration precision of the star simulators as shown in Figure 5. The stellar-INS is installed in the three axes turntable with over plate. The turntable then conducts the star observation sequence, as shown in Figure 6. Before beginning, the multi-position calibration should be conducted to ensure the output attitude of stellar-INS is consistent with the turntable. Consequently, the attitude of the high-precision three-axis turntable can be as referenced in a short time.
Figure 5. Schematic of ground verification test of stellar-INS.
Figure 6. Star observation sequence of stellar-INS in ground verification test.
In the experiment, the observability of all error items is guaranteed during the process of attitude adjustment and star observation. Six repeat tests are completed in the condition of static star vector measurement, and another six repeat tests are completed in dynamic star vector measurement. The dynamic angular velocity is about 0·2°/s. The results of repeated experiments in the NFOV and the LFOV are as follows.
In the ground test, the high-precision turntable can provide the attitude reference in two ways. One is the repeatability of several tests. Repeatability of the turntable's angular position is close to one arc-second. The other is the turntable attitude after optical calibration. The turntable attitude accuracy is less than three arc-seconds. According to Table 2, the integration navigation attitude angle repeatability of the NFOV is slightly better than the LFOV. The advantage of the LFOV has not been exerted because fewer star vectors are observed and noise smoothing is limited in the ground test system. In the NFOV, a single star vector has less noise and error based on the same imaging device, so the repeatability of the NFOV stellar-INS is better than that of the LFOV stellar-INS. Moreover, the navigation attitude average value in several tests can also be used to evaluate the navigation performance. We can conclude that the integration accuracy of the LFOV is better than the NFOV in terms of the attitude average value. The estimated value of installation error of the star sensor is shown as follows.
Table 2. Integrated attitude angles of NFOV and LFOV.
Table 3 shows that the installation error along the optical axis cannot be estimated due to weak observability of the NFOV star sensor. On the contrary, it can be fully estimated with the LFOV star sensor. So the estimation accuracy of the installation error on the optical axis in the LFOV star sensor is much better than in the NFOV star sensor. Obviously, the repeatability of the estimated value in the LFOV star sensor is degraded due to the limited star simulator in the ground test. Three conclusions can be drawn for engineering application.
(a) Observability of the gyro error and star sensor installation error along the optical axis is determined by the star sensor's FOV angle.
(b) The attitude accuracy of stellar-INS is not affected by the weak observability of the installation error and gyro error along the optical axis, when the attitude adjustment is unconstrained.
(c) In the condition of the same imaging device, the single star vector accuracy of NFOV is higher than that of LFOV. The high attitude accuracy of LFOV is dependent on the observation of multiple stars. The repeatability of stellar-INS integration navigation attitude of the LFOV star sensor is worse than that of the NFOV star sensor due to the limited star simulator in the ground test.
Table 3. Estimated installation error of NFOV and LFOV.
5. CONCLUSION
For certain stellar-INS, it is necessary to improve the accuracy of integration attitude to meet the increasing demand of higher accuracy applications. At present, there are two main methods to improve the accuracy of integration attitude. One is the promotion data fusion method to reduce the influence of device noise. The other is improving the observability of integration errors. This paper reveals the star vector orientation in the FOV mechanism of influence on observability of integration errors. It indicates that the larger the star vector orientation along the optical axis, the better the error estimation accuracy. For the NFOV stellar-INS, the optimal integration attitude accuracy can be achieved by the appropriate postures without the influence of the star vector orientation in the FOV. For the LFOV stellar-INS, not only can the optimal integration attitude accuracy be obtained, but also the optimal device errors estimation. These conclusions can offer universal guidance for the design and evaluation of stellar-INS with NFOV or LFOV star sensors.
