An Improved Alignment Method for the Strapdown Inertial Navigation System (SINS)

In this paper, an innovative inertial navigation system (INS) mechanization and the associated Kalman filter (KF) are developed to implement a fine alignment for the strapdown INS (SINS) on stationary base. The improved mechanization is established in the pseudo-geographic frame, which is rebuilt based on the initial position. The new mechanization eliminates the effects of linear movement errors on the heading by decoupling. Compared with the traditional local-level mechanization, it has more advantages. The proposed algorithm requires lower coarse alignment accuracy in both the open-loop and closed-loop KFs and hence can improve the system reliability and decrease the total alignment time. Moreover, for the closed-loop KF, it can decrease oscillation caused by the system errors and improve the closed-loop system stability. In addition, the proposed algorithm can also be applied to polar alignment. The performance of the proposed algorithm is verified by both simulations and experiments and the results exhibit the superior performance of the proposed approach.


Introduction
The inertial navigation system (INS) has been widely used for the positioning and navigation of vehicles because it has particular advantages, such as high autonomy, concealment, continuum, insusceptible climate, and the successive supply of position, velocity, and attitude (PVA) information. strapdown INS (SINS), which has the advantages of simple structure, small size, and convenient maintenance compared with plat INS (PINS), is especially popular [1,2]. However, an accurate alignment must be required before the start of navigation. Otherwise, it will directly degrade navigation accuracy [3][4][5]. The purpose of the SINS' initial alignment is to obtain the exact coordinate transformation matrix from the body frame to the navigation frame. Typically, this process consists of a coarse alignment and a fine alignment [6,7]. The objective of a coarse alignment is to provide a good condition for a fine alignment, where the misalignment angles are approximately estimated to a few degrees. In the fine alignment stage, the small misalignment angles are computed accurately and the precise initial transformation matrix is then formulated [8].
For a fine alignment, the Kalman filter (KF) is one of the most widely used methods due to its simplicity, optimality, tractability, and robustness [9,10]. Usually, the application of a KF in alignment can be categorized into two types: open-loop and closed-loop [11,12]. The estimation of SINS errors does not interfere with the operation of the SINS and is only subtracted from the output of the SINS in an open-loop system, whereas the estimation of the SINS errors is fed back into the strapdown computation processor in a closed-loop system. Therefore, one major drawback of an open-loop approach is that the system errors will be propagated (not courted to null). The large system errors may result in declining alignment accuracy or even lead to filter's divergence. In a closed-loop system, From [22] (p. 253), the static velocity and attitude error model in the fine alignment state can be formulated as follows: .
where Ω denotes the angular rate of Earth rotation; g denotes the local gravity; L and R denote the local latitude and the Earth radius respectively; and ∇ E , ∇ N , ε E , ε N , ε U denote the accelerometer zero-biases and the constant gyroscope drifts respectively. From Equations (1) and (2), we can know that the worse the accuracy of the coarse alignment (φ x , φ y , φ z ) is, the faster the level linear movement errors (δV E , δV N ) diverge. Then, the level linear movement errors will influence the misalignment angle's spread in turn from Equations (3)- (5). Therefore, poor coarse alignment will result in large system errors. For the open-loop and closed-loop KFs, the large system errors may result in alignment accuracy or even lead to a filter's divergence. As a result, the fine alignment system must exceedingly rely on coarse alignment. Furthermore, the fact that the heading error is not completely observable is widely accepted. Therefore, the influence of the level linear movement errors on heading error is more serious. On the other hand, from Equation (5), we can see that the propagation of the heading error is proportional to not only the linear movement error (δV E ) but also tangent function of latitude (tanL). Therefore, a small linear movement error may cause very large heading error in a high-latitude region. This is also one of the major difficulties for alignment in a high-latitude region, including polar alignment. Therefore, a decoupling mechanization between the linear movement errors and the heading error without much additional complexity is expected to be achieved.

The Definition of the Pseudo-Frame
In this paper, the decoupling mechanization is established in a pseudo-geographical frame. Suppose the Earth is spherical and the initial position (latitude and longitude) of the vehicle is (L 0 , λ 0 ) in e-frame. The p-frame (o p x p y p z p ) can be obtained by two successive transformations with two Euler angles from the e-frame (o e x e y e z e ), as shown in Figure 1. Additionally, the rotations are performed in the following order: According to the above description, the transformation matrix from e-frame to p-frame can be derived as Equation (6).
As shown in Figure 1, N p denotes the pseudo-North Pole; the origin of the pseudo-Earth-Centered Earth-fixed frame (o p x p y p z p ) is at the Earth's mass center; the o p x p axis points towards the North Pole, which is also known as pointing towards the pseudo-prime meridian; the o p y p axis is pointing from the center of the Earth to the projection point of the carrier's normal frame position S on the equator plane; the o p z p axis is perpendicular to the o p y p axis and conforms to the rules of the right-hand spiral in the equator plane, which is also known as pseudo-North Pole axis.
axis and conforms to the rules of the right-hand spiral in the equator plane, which is also known as pseudo-North Pole axis.
For the definition of the pseudo-coordinate system, it is similar to the normal coordinate system. p L and p  represent pseudo-latitude and pseudo-longitude, respectively. Moreover, the p p p t t t ox y z frame denotes p t -frame in pseudo-Earth frame as shown in Figure 1.  From the above pseudo-frames definition, the pseudo-equator overlaps the normal longitude circle in the initial position. Therefore, the carrier's initial position in pseudo-frame can be given as Equation (7).
Moreover, the transformation matrix from the t -frame to the p t -frame can be expressed as Equation (8).  (9) and (10) Substituting Equations (6), (7), (9) and (10) into Equation (8), the transformation matrix from the t -frame to the p t -frame at initial position can be given as Equation (11). For the definition of the pseudo-coordinate system, it is similar to the normal coordinate system. L p and λ p represent pseudo-latitude and pseudo-longitude, respectively. Moreover, the ox t p y t p z t p frame denotes t p -frame in pseudo-Earth frame as shown in Figure 1. The origin of ox t p y t p z t p frame is at the mass center of the vehicle; the oz t p axis is normal to the local-level frame (denoted as U p axis); the oy t p axis is pointing towards the pseudo-north direction (denoted as N p axis); and the ox t p axis completes a right-handed system (denoted as E p axis).
From the above pseudo-frames definition, the pseudo-equator overlaps the normal longitude circle in the initial position. Therefore, the carrier's initial position in pseudo-frame can be given as Equation (7).
r L p0 λ p0 s " r 0˝90˝´L 0 s Moreover, the transformation matrix from the t-frame to the t p -frame can be expressed as Equation (8).
where C e t " pC t e q T ; C t p p is a function of pseudo-latitude and -longitude; C t e is a function of latitude and longitude; and C t p p and C t e are shown in Equations (9) and (10), respectively.
Substituting Equations (6), (7), (9) and (10) into Equation (8), the transformation matrix from the t-frame to the t p -frame at initial position can be given as Equation (11). From Equation (11), it can be obtained that the t p -frame (E p N p U p ) overlaps the traditional South-East-Up (SEU) geographic frame at the initial position. As a result, the SINS alignment mechanization in t p -frame will be easier to implement without much additional complexity.

The Mechanization of the SINS in Pseudo-Frame
In the pseudo-coordinate system, the t p -frame is selected as the navigation frame (n-frame), and the SINS mechanization in the t p -frame is similar to the one in the t-frame. Therefore, the rotation angular rate of the p-frame relative to the i-frame (due to the Earth spin angle rate) decomposed in n-frame can be written as Equation (12).
where L p and λ p denote pseudo-latitude and pseudo-longitude in pseudo-Earth frame respectively. Moreover, the rotation angular rate of the local n-frame relative to the p-frame (due to vehicle movement over the Earth surface) decomposed in n-frame is expressed as Equation (13).
where V E p and V N p represent the level velocity vectors in t p -frame. Considering Equations (12) and (13), the total rotation angular rate of the n-frame relative to the i-frame decomposed in n-frame can be given as Equation (14).
According to the above analysis, the SINS mechanization in the pseudo-geographic frame is given as shown in Figure 2.
From Equation (11), it can be obtained that the p t -frame ( p p p E N U ) overlaps the traditional South-East-Up ( SEU ) geographic frame at the initial position. As a result, the SINS alignment mechanization in p t -frame will be easier to implement without much additional complexity.

The Mechanization of the SINS in Pseudo-Frame
In the pseudo-coordinate system, the p t -frame is selected as the navigation frame ( n -frame), and the SINS mechanization in the p t -frame is similar to the one in the t -frame. Therefore, the rotation angular rate of the p -frame relative to the i -frame (due to the Earth spin angle rate) decomposed in n -frame can be written as Equation (12).
sin sin cos cos cos where p L and p  denote pseudo-latitude and pseudo-longitude in pseudo-Earth frame respectively.
Moreover, the rotation angular rate of the local n -frame relative to the p -frame (due to vehicle movement over the Earth surface) decomposed in n -frame is expressed as Equation (13).
where p E V and p N V represent the level velocity vectors in p t -frame.
Considering Equations (12) and (13), the total rotation angular rate of the n -frame relative to the i -frame decomposed in n -frame can be given as Equation (14). sin sin cos cos cos tan According to the above analysis, the SINS mechanization in the pseudo-geographic frame is given as shown in Figure 2.  According to the above mechanization in pseudo-geographic frame, the navigation equations can be written as Equations (15)- (18).
. V n " C n b f b´p 2ω n ip`ω n pn qˆV n`gn (16) .
n ω n in , C b n " pC n b q T ; pω b nbˆq denotes a skew-symmetric matrix of ω b nb ; f b denotes the specific force measured by accelerometer; ω b ib denotes the body angular rate measured by gyroscope; and C n b denotes the attitude transformation matrix.

The Static Error Equations of the SINS in Pseudo-Frame
According to the SINS mechanization in Section 4.2, the pseudo-error equations can be easily derived. From Equations (15)- (18), the static PVA error equations of the SINS mechanized in pseudo-frame, which are derived in a small misalignment angles, can be described by The velocity error equations: The attitude error equations: .
The position error equations: where ∇ E p , ∇ N p denote the accelerometer zero-biases expressed in t p -frame; and ε E p , ε N p , ε U p denote the constant gyroscope drifts expressed in t p -frame.

The Filter Model of Kalman for Zero-Velocity Alignment
As the error equations of the SINS in the pseudo-frame have been derived in the above section, the static KF model can be given. In addition, since the alignment process of the SINS is short, the accelerometer and gyroscope errors here are regarded as random constant biases and white noise processes, namely, ε " 0. For the static alignment, the position errors and up velocity errors are usually ignored. Therefore, the level velocity errors, misalignment angles, the accelerometer zero-biases, and the constant gyro drifts are chosen as the state variables. The state vectors of the system error model are defined as: where Fptq denotes the state transition matrix; and Wptq denotes the white system process noise with the power spectral density Q.
In addition, L p0 is equal to zero in Equation (8), namely, the initial latitude in pseudo-Earth frame is zero. Therefore, referring to Equations (19)- (25), the static state transition matrix can be written as Equation (27).
In matrix F 21 , all elements in the third row are all zero. As a result, the proposed mechanization achieves the decoupling between the level linear movement errors and the heading error. It can also eliminate the influences of level linear movement errors on the heading error. Moreover, the initial latitude in pseudo-Earth frame is zero all the time. Consequently, the heading error would barely be influenced by the position (due to the mechanization). Therefore, with the innovative mechanization, the spread of the heading error influenced by the level linear movement errors (due to the latitude) can be removed without much additional complexity. For the fine alignment associated with the Kalman filter, this structural mode can eliminate the influences on the alignment performance and the loop stability from the heading error caused by the linear movement errors and the latitude, especially in the high-latitude region, including the polar region.
When the vehicle is in stationary base, the zero-velocity information in the navigation frame (namely, t p -frame) is usually selected as the external reference. Therefore, the measurement equation can be written as Equation (28).
where δV E p and δV N p are the measurement variables, which are equal to V E p and V N p from the SINS mechanized in pseudo-geographic frame. Therefore, the system measurement equation can be written as Equation (29).
where Vptq denotes the white measurement noise with the power spectral density R; and Hptq denotes the measurement matrix. According to Equation (28), the measurement matrix can be written as Equation (30).

Simulations and Experiments
In this section, both simulations and experiments are performed to verify the performance of the static alignment with the proposed mechanization, which is assisted by the open-loop and closed-loop KFs, respectively. Firstly, under different coarse alignment accuracy conditions, the simulations of the fine alignment with the proposed algorithm and the traditional one are done. Then, the experiments are also conducted In addition, the polar alignment of the SINS mechanized in the pseudo-geographic frame is performed to demonstrate the superior performance of the proposed approach.

The Simulations of Fine Alignment Assisted by the KF
In this section, two groups of 300 s static data are obtained by the SINS simulator to verify the alignment performance assisted by the open-loop and closed-loop KFs, respectively. The conditions of the simulation are as follows: The initial longitude and latitude are 45˝and 126˝, respectively; the true attitudes of INS are all set to zero; the constant and random drifts of gyroscope are 0.01˝/h and 0.001˝/h, respectively; the constant and random biases of the accelerometer are 100 µg and 10 µg, respectively, the update rate is 100 Hz, and the model of the static SINS simulator is described as follows: (31) where ω ib and f ib denote outputs of gyroscopes and the accelerometers; g n denotes the gravity; ε xcons , ε ycons , ε zcons and ∇ xcons , ∇ ycons , ∇ zcons denote the constant drifts and the constant biases, respectively; ε xrandn , ε yrandn , and ε zrandn are white noise, and its mean is equal to random drifts; ∇ xrandn , ∇ yrandn , and ∇ zrandn are also white noise, and its mean is equal to random biases.

The Simulations Based on the Open-Loop KF
Using the proposed and the traditional mechanization, the alignment simulation assisted by open-loop KF is conducted by intentionally adding initial attitude errors. The different initial conditions of fine alignment (namely, the different coarse alignment accuracy) are shown in Table 1  igure 3 shows the alignment error curves of the attitude by intentionally adding initial attitude errors (1˝for heading; 0.3˝for both pitch and roll), where the theoretical predictions of the heading (square root of P k ; for the matrix P k , if not explicitly stated, all matrices P k denote the error variance matrix in this paper) is denoted by the green solid lines. In Figure 4 in Figure 3. Due to the limitation of figures, under the other three cases, only the alignment results of the heading are shown in Figure 4.  Figure 3 shows the alignment error curves of the attitude by intentionally adding initial attitude errors (1° for heading; 0.3° for both pitch and roll), where the theoretical predictions of the heading (square root of k P ; for the matrix k P , if not explicitly stated, all matrices k P denote the error variance matrix in this paper) is denoted by the green solid lines. In Figure 4,    the heading are shown in Figure 4.  Figure 3 shows the alignment error curves of the attitude by intentionally adding initial attitude errors (1° for heading; 0.3° for both pitch and roll), where the theoretical predictions of the heading (square root of k P ; for the matrix k P , if not explicitly stated, all matrices k P denote the error variance matrix in this paper) is denoted by the green solid lines. In Figure 4,     Table 2. Under Case 1, the alignment results with the two mechanizations are shown in Figure 5. Due to divergence under the other two conditions, the alignment based on the traditional mechanization is conducted by setting larger initial values of the noise matrix to ensure the closed-loop system stability. As a result, the other two alignment results based on the proposed mechanization and the alignment result based on the traditional one in Case 2 are shown in Figure 6. n Figures 5 and 6 the dotted lines represent the estimations by the proposed algorithm and solid lines denote the estimations based on the traditional algorithm, and the partial magnifications of the alignment results are also shown. In Figure 6, the theoretical predictions of the heading (the square root of P k in Case 2) based on the proposed algorithm and the traditional one are denoted by a black solid line and a green solid line, respectively. The red line denotes the alignment results in Case 2, and the blue line denotes the alignment results in Case 3. Therefore, the performance of the proposed algorithm in the closed-loop system is superior to that of the normal one.  Under the other two initial conditions, with the coarse alignment becoming poor (Cases 2 and 3 in Table 2), the alignment based on the traditional mechanization would diverge. However, from Figure 6, the proposed one is still favorable, and the alignment results in 200 s are 1.7′ and −2.8′, respectively. Moreover, the estimation oscillations based on the proposed algorithm are still smaller. Therefore, the proposed mechanization requires less coarse alignment accuracy to Therefore, the performance of the proposed algorithm in the closed-loop system is superior to that of the normal one.  Under the other two initial conditions, with the coarse alignment becoming poor (Cases 2 and 3 in Table 2), the alignment based on the traditional mechanization would diverge. However, from Figure 6, the proposed one is still favorable, and the alignment results in 200 s are 1.7′ and −2.8′, respectively. Moreover, the estimation oscillations based on the proposed algorithm are still smaller. Therefore, the proposed mechanization requires less coarse alignment accuracy to maintain the desired performance of fine alignment, whereas the traditional method requires a  Table 2), the estimation result with the proposed algorithm is always within the theoretical bounds after 50 s. However, the one based on the traditional algorithm is outside the theoretical bounds after 150 s. In addition, the heading error based on the proposed mechanization is smaller, and the heading errors based on the proposed mechanization and traditional one are 1.6 1 and 3.6 1 , respectively, in 200 s. Moreover, the estimation oscillation based on the traditional mechanization is more obvious at the beginning of the alignment, and the maximum amplitudes based on the two methods are 86 1 and 52 1 , respectively. The violent oscillation, which may lead to the divergence of the closed-loop KF, will be very perilous. However, the proposed mechanization eliminates the effects of linear movement errors on the heading error by decoupling and hence can improve system performance. Therefore, the performance of the proposed algorithm in the closed-loop system is superior to that of the normal one.
Under the other two initial conditions, with the coarse alignment becoming poor (Cases 2 and 3 in Table 2), the alignment based on the traditional mechanization would diverge. However, from Figure 6, the proposed one is still favorable, and the alignment results in 200 s are 1.7 1 and´2.8 1 , respectively. Moreover, the estimation oscillations based on the proposed algorithm are still smaller. Therefore, the proposed mechanization requires less coarse alignment accuracy to maintain the desired performance of fine alignment, whereas the traditional method requires a higher accuracy of coarse alignment. The closed-loop estimated results would otherwise fail. Consequently, the proposed method can improve closed-loop system reliability and decrease total alignment time because of the low requirement of coarse alignment accuracy.
On the other hand, due to the fact that the closed-loop system with the traditional algorithm diverges in Case 2, a method of setting larger initial values of the noise matrix is used to maintain system stability. This alignment result is shown with a solid red line in Figure 6. The theoretical prediction 1 in Figure 6 is the square root of P k based on the traditional algorithm. And the square root of P k based on the proposed one in case 2 is denoted by the theoretical prediction 2 in figure 6. Compared with the alignment results with the two methods in Case 2 (the dotted red line and solid line), the estimation oscillation based on the traditional mechanization is still larger at the beginning. Moreover, compared with the theoretical predictions 1 and 2 in Figure 6, the alignment error of heading, which is obtained by using the traditional mechanization and setting larger initial values of the noise matrix, is larger at the end of alignment, and the speed of the convergence is obviously slower. Therefore, the closed-loop system stability can be ensured by setting larger initial values of the noise matrix, but at the expense of the alignment time and accuracy. As a result, the proposed algorithm based on the closed-loop KF has superior performance without much additional complexity.
In addition, the same behaviors assisted by the open-loop and closed-loop KFs have also been performed with the Monte Carlo simulation. Table 3 Table 4, and the mean error curves and error variance curves of heading error based on the Monte Carlo simulation are shown in Figures 7 and 8. From Table 3, under the above four initial conditions in Table 1, the open-loop mean errors based on the proposed algorithm are almost the same. However, the ones based on the traditional algorithm will be deteriorating with poor coarse alignment. For the closed-loop KF, the mean errors based on the proposed algorithm under the different conditions in Table 2 are also favorable from Table 4. From  Figures 7 and 8 for the closed-loop KF, the proposed algorithm can decrease oscillation, improve the closed-loop system stability, and reduce the requirement for coarse alignment accuracy without much additional complexity and at the expense of alignment performance. This is consistent with the results of the above analyses. Therefore, the proposed mechanization has a good capability in the SINS alignment assisted by the open-loop or closed-loop KFs. . "ˆ"denotes that alignment result diverges, and "´"denotes the experiment dose was not conducted.
traditional algorithm will be deteriorating with poor coarse alignment. For the closed-loop KF, the mean errors based on the proposed algorithm under the different conditions in Table 2 are also  favorable from Table 4. From Figures 7 and 8, for the closed-loop KF, the proposed algorithm can decrease oscillation, improve the closed-loop system stability, and reduce the requirement for coarse alignment accuracy without much additional complexity and at the expense of alignment performance. This is consistent with the results of the above analyses. Therefore, the proposed mechanization has a good capability in the SINS alignment assisted by the open-loop or closed-loop KFs.  "×"denotes that alignment result diverges, and "−"denotes the experiment dose was not conducted.

The Experiments of Fine Alignment Based on the KF
The experimental data were collected with the fiber optical gyroscope (FOG) SINS, fixed in a 3-axis high-precision turntable as shown in Figure 9. The inertial measurement unit (IMU) and the North Reference are also shown in the figure. The FOG SINS was made by our lab (namely, The Institute of Inertial Navigation and M&C Technology), and its sampling frequency is 100 Hz. The local longitude and latitude is 126.6778° and 45.7778°, respectively. Moreover, the attitude reference value of IMU is provided by the 3-axis high-precision turntable.

The Experiments of Fine Alignment Based on the KF
The experimental data were collected with the fiber optical gyroscope (FOG) SINS, fixed in a 3-axis high-precision turntable as shown in Figure 9. The inertial measurement unit (IMU) and the North Reference are also shown in the figure. The FOG SINS was made by our lab (namely, The Institute of Inertial Navigation and M&C Technology), and its sampling frequency is 100 Hz. The local longitude and latitude is 126.6778˝and 45.7778˝, respectively. Moreover, the attitude reference value of IMU is provided by the 3-axis high-precision turntable.

The Experiments of Fine Alignment Based on the KF
The experimental data were collected with the fiber optical gyroscope (FOG) SINS, fixed in a 3-axis high-precision turntable as shown in Figure 9. The inertial measurement unit (IMU) and the North Reference are also shown in the figure. The FOG SINS was made by our lab (namely, The Institute of Inertial Navigation and M&C Technology), and its sampling frequency is 100 Hz. The local longitude and latitude is 126.6778° and 45.7778°, respectively. Moreover, the attitude reference value of IMU is provided by the 3-axis high-precision turntable.  Table 1, and the closed-loop KF alignment initial conditions is described in Table 5. The alignment results (only heading error) based on the open-loop KF are shown in Figure 10. Figures 11 and 12 show the alignment results based on the closed-loop KF.   Table 1, and the closed-loop KF alignment initial conditions is described in Table 5. The alignment results (only heading error) based on the open-loop KF are shown in Figure 10. Figures 11 and 12 show the alignment results based on the closed-loop KF. Table 5. The different initial conditions for fine alignment.

Pitch Roll Heading
Case 1 0.3˝0.3˝1C ase 2 0.5˝0.5˝2C ase 3 1˝1˝3B y using the proposed mechanization and the traditional one, the open-loop alignment error curves in the four cases in Table 1 are shown Figure 10. In addition, due to the fact that the closed-loop alignment in the traditional method diverges in Cases 2 and 3 in Table 5, the closed-loop alignment errors, only including those heading alignment errors by using the traditional method in Case 1 and the proposed method in the three cases in Table 5, are presented in Figure 11. Moreover, Figure 12 shows the closed-loop alignment error, which is obtained with the traditional method in Case 2 by setting larger initial values of the noise matrix to ensure system stability. In Figures 10-12 the dotted lines represent the estimations of the proposed algorithm, and solid lines denote the estimations based on the traditional algorithm. The partial magnifications of the alignment results are also shown in the three figures, and the green solid lines denote the theoretical predictions of the heading, respectively. In Figures 11 and 12 the theoretical predictions of the heading are the square root of P k in Case 2 by using the proposed mechanization and the traditional one with the correction, respectively.
From Figure 10, under the above four initial conditions in Table 1, the open-loop estimated heading errors with the proposed mechanization in 300 s are´1.96 1 ,´1.32 1 , 0.25 1 , and 0.68 1 , respectively. There is almost no difference. However, the alignment error based on the traditional mechanization is smaller only when the coarse alignment is better (in Case 1 in Table 1). The other three alignment error curves display an obvious difference and diverge with the poor coarse alignment (in the other three cases in Table 1). Therefore, the traditional method requires more accurate coarse alignment. The open-loop estimated results would otherwise decline or even fail, whereas the proposed method can still maintain the desired performance. This is consistent with the above open-loop simulation results. k traditional one with the correction, respectively.
From Figure 10, under the above four initial conditions in Table 1, the open-loop estimated heading errors with the proposed mechanization in 300 s are −1.96′, −1.32′, 0.25′, and 0.68′, respectively. There is almost no difference. However, the alignment error based on the traditional mechanization is smaller only when the coarse alignment is better (in Case 1 in Table 1). The other three alignment error curves display an obvious difference and diverge with the poor coarse alignment (in the other three cases in Table 1). Therefore, the traditional method requires more accurate coarse alignment. The open-loop estimated results would otherwise decline or even fail, whereas the proposed method can still maintain the desired performance. This is consistent with the above open-loop simulation results. As seen from Figure 11, under the above three initial conditions in Table 5, the difference of the closed-loop alignment error curves based on the proposed method is very small, and the alignment results in 300 s are 3′, 3.4′, and 1.6′, respectively. However, the closed-loop alignment with the traditional method is convergent only when the coarse alignment is better (in Case 1 in Table 5), and the alignment result in 300 s is 4′. Compared with the alignment results with the two methods in Case 1 (the dotted red line and solid line in Figure 11), the estimation oscillation based on the traditional mechanization is very large at the beginning. However, the oscillations based on the proposed algorithm are all smaller under the conditions of the three different coarse alignments in Table 5. In addition, comparing Figure 11 with Figure 12, the closed-loop system stability based on the traditional algorithm can be ensured by setting larger initial values of the noise matrix but at the expense of alignment time and accuracy. The theoretical value and convergence time of the traditional alignment in Case 2 in Table 5 are 4.1′ and 200 s, respectively, whereas they are 3.4′ and 80 s when using the proposed alignment method. Consequently, the proposed algorithm can decrease estimation oscillation caused by system errors and improve closed-loop system stability. As seen from Figure 11, under the above three initial conditions in Table 5, the difference of the closed-loop alignment error curves based on the proposed method is very small, and the alignment results in 300 s are 3 1 , 3.4 1 , and 1.6 1 , respectively. However, the closed-loop alignment with the traditional method is convergent only when the coarse alignment is better (in Case 1 in Table 5), and the alignment result in 300 s is 4 1 . Compared with the alignment results with the two methods in Case 1 (the dotted red line and solid line in Figure 11), the estimation oscillation based on the traditional mechanization is very large at the beginning. However, the oscillations based on the proposed algorithm are all smaller under the conditions of the three different coarse alignments in Table 5. In addition, comparing Figure 11 with Figure 12, the closed-loop system stability based on the traditional algorithm can be ensured by setting larger initial values of the noise matrix but at the expense of alignment time and accuracy. The theoretical value and convergence time of the traditional alignment in Case 2 in Table 5 are 4.1 1 and 200 s, respectively, whereas they are 3.4 1 and 80 s when using the proposed alignment method. Consequently, the proposed algorithm can decrease estimation oscillation caused by system errors and improve closed-loop system stability. This is also consistent with the above closed-loop simulation results. Therefore, the proposed algorithm has superior performance in both open-loop and closed-loop KFs. This is also consistent with the above closed-loop simulation results. Therefore, the proposed algorithm has superior performance in both open-loop and closed-loop KFs.

The Simulations of Polar Alignment
The simulations of polar alignment based on the proposed mechanization and the closed-loop KF were conducted. Furthermore, two groups of static data in different positions were obtained by the SINS simulator. The conditions of the simulation are as follows: The initial longitude and latitude are (126°, 85°) and (126°, 89°); the true attitudes of INS are all set to zero; the constant and random drifts of gyroscope are 0.001°/h and 0.0001°/h, respectively; and the constant and random biases of the accelerometer are 50 μg and 5 μg, respectively. At the 85° latitude the three alignment results of heading with the different initial conditions in Table 5 are shown in Figure 13. Figure 14 shows the 89° latitude alignment results in Cases 1 and 2 in Table 5.
As seen from Figures 13 and 14, all the estimation curves converge with time. Figure 13 shows the polar alignment results based on the proposed mechanization under the three different initial conditions when the latitude is 85°, where the heading alignment errors are all less than 3.8′ in 1200 s. The heading alignment errors under the two cases are less than 18′ in 1200 s at 89° latitudes from Figure 14, and the alignment results do not display an obvious difference under different initial conditions in the two figures. On the other hand, the highly accurate coarse alignment is also difficult to achieve due to the smaller gyrocompass component. As a result, the proposed mechanization has a good ability in polar alignment because it requires less accurate coarse alignment. Therefore, the proposed mechanization can solve the problem of polar alignment coming from the coupling between the level movement errors and the heading error.

The Simulations of Polar Alignment
The simulations of polar alignment based on the proposed mechanization and the closed-loop KF were conducted. Furthermore, two groups of static data in different positions were obtained by the SINS simulator. The conditions of the simulation are as follows: The initial longitude and latitude are (126˝, 85˝) and (126˝, 89˝); the true attitudes of INS are all set to zero; the constant and random drifts of gyroscope are 0.001˝/h and 0.0001˝/h, respectively; and the constant and random biases of the accelerometer are 50 µg and 5 µg, respectively. At the 85˝latitude the three alignment results of heading with the different initial conditions in Table 5 are shown in Figure 13. Figure 14 shows the 89l atitude alignment results in Cases 1 and 2 in Table 5.
As seen from Figures 13 and 14 all the estimation curves converge with time. Figure 13 shows the polar alignment results based on the proposed mechanization under the three different initial conditions when the latitude is 85˝, where the heading alignment errors are all less than 3.8 1 in 1200 s. The heading alignment errors under the two cases are less than 18 1 in 1200 s at 89˝latitudes from Figure 14, and the alignment results do not display an obvious difference under different initial conditions in the two figures. On the other hand, the highly accurate coarse alignment is also difficult to achieve due to the smaller gyrocompass component. As a result, the proposed mechanization has a good ability in polar alignment because it requires less accurate coarse alignment. Therefore, the proposed mechanization can solve the problem of polar alignment coming from the coupling between the level movement errors and the heading error.    Furthermore, comparing Figure 13 with Figure 14, the convergence speed at 89° is slower than that at 85°, and the alignment accuracy is also lower than the one at 85°. This is consistent with the problem arising in polar self-alignment. The decline of gyrocompass component leads to a slow convergence speed and poor alignment accuracy in the polar region. Therefore, self-alignment will fail in poles or near the poles. Finding other solutions to achieve the alignment in the poles or near the poles will be pursued in future work.

Conclusions
The main objective of this paper is to solve the problems of KF alignment and polar alignment on stationary base. An innovative mechanization is proposed to achieve fine alignment for the SINS. The improved mechanization eliminates the effects of linear movement errors and latitude on the heading error by decoupling. Compared with the traditional mechanization, the proposed algorithm has superior performance without much additional complexity. The proposed algorithm would require lower coarse alignment accuracy in both the open-loop and closed-loop KFs and hence can improve the system reliability and decrease total alignment time. For the closed-loop KF, it can also decrease estimation oscillation caused by system errors and improve closed-loop system stability. Moreover, the proposed algorithm can also be applied to polar alignment. Finally, the simulations and experiments are conducted, and the results exhibit the superior performance of the proposed approach. As a result, the proposed algorithm is promising for polar alignment and KF alignment. On the other hand, though motivated by the SINS static alignment, it can be applied to quasi-static alignment and PINS alignment. For the PINS, only a switching control is required from the Furthermore, comparing Figure 13 with Figure 14, the convergence speed at 89˝is slower than that at 85˝, and the alignment accuracy is also lower than the one at 85˝. This is consistent with the problem arising in polar self-alignment. The decline of gyrocompass component leads to a slow convergence speed and poor alignment accuracy in the polar region. Therefore, self-alignment will fail in poles or near the poles. Finding other solutions to achieve the alignment in the poles or near the poles will be pursued in future work.

Conclusions
The main objective of this paper is to solve the problems of KF alignment and polar alignment on stationary base. An innovative mechanization is proposed to achieve fine alignment for the SINS. The improved mechanization eliminates the effects of linear movement errors and latitude on the heading error by decoupling. Compared with the traditional mechanization, the proposed algorithm has superior performance without much additional complexity. The proposed algorithm would require lower coarse alignment accuracy in both the open-loop and closed-loop KFs and hence can improve the system reliability and decrease total alignment time. For the closed-loop KF, it can also decrease estimation oscillation caused by system errors and improve closed-loop system stability. Moreover, the proposed algorithm can also be applied to polar alignment. Finally, the simulations and experiments are conducted, and the results exhibit the superior performance of the proposed approach. As a result, the proposed algorithm is promising for polar alignment and KF alignment. On the other hand, though motivated by the SINS static alignment, it can be applied to quasi-static alignment and PINS alignment. For the PINS, only a switching control is required from the alignment stage to the navigation stage. In addition, as the proposed mechanization is similar to the traditional local-level north mechanization, the new method has a high project value.