3.1. The Effect of SRB in Shallow Water
The SRT method with constant gradient is introduced in this section. If the sound speed changes little or uniformly, then we can consider the sound speed gradient to be a constant. As shown in
Figure 2,
represents the sound speed in water layer
, and
represents the depth of the water. The travel of sound in water follows Snell’s law [
10,
11]
where
represents incidence angle, and
is a constant. In each layer, we assume that the medium changes uniformly. In other words, the sound speed has a constant gradient, and the track of the ray is a continuous arc. The radius
of the arc can be expressed as [
27]:
where
represents the sound speed gradient. The horizontal distance of the sound ray
can be computed as follows
where
represents the length of layer
. In general, the interval between layers is wide if the variation of sound velocity between layers is small. The length of arc
in layer
can be computed by [
28]
Thus, the travel time is estimated by
In OBS underwater acoustic positioning, a fixed weighted mean sound velocity (WMSV) is usually used in the calculation, and the WMSV can be calculated by [
29]
where
represents the depth of water;
is the weight of each water column related to the incidence angle, sound speed structure and other factors. If
, the
represents the MSV.
The MSV assumes a single sound speed value for the entire localization area. In fact, the calculated MSV is different due to the incidence angles and sound ray, and can cause the SRB error. The SRB correction
can be modeled and absorbed by the MSV in many forms, such as a constant, 1st- or 2nd-degree polynomial with a time series [
15] as
The SRB correction
can also be modeled directly [
17] as
where
is the coefficient to be estimated, and
represents the travel time.
represents the minimum travel time among the measured epochs.
represents the incidence angle according to the calculation of the position of the transducer.
The above methods are usually used under conditions where there is a long observation time, and the SRB varies with time more significantly. Based on the above methods, we can analyze the SRB in shallow water, which helps us to find some useful models to describe it quantitatively.
The relationship among the horizontal distance (HD), geometrical distance (RD) and sound ray trace (SD) is shown in
Figure 3. According to the analysis of the above, the acoustic ranging is usually treated as a straight line, while the real sound track is a curve in inhomogeneous water. In this research, we ignore the complexity of the top surface sound speed, and consider the analysis of the steady acoustic environment in small areas within a short period, as the OBS measurement area is only a few hundred square meters, and a single operation period is not usually more than an hour. We temporarily ignore the complexity of those short sound speed changes, and combine all the observation data to estimate the SRB error in this region. We only consider the commonness between these observation data; for example, under the same incidence, the SIB error is approximately equivalent. We use these characteristics and combine all the observation data to estimate the bending error of the acoustic line in this region.
Based on the above equation, the SRB can be simulated using the measured SSP. Assuming that the coordinates of the transponder and the transducer are known, the beam incidence angle and travel time, as unknown parameters, can be solved by Newton iteration or other secant methods using the SSP.
As shown in
Figure 4, we can get some information from the above figures. The
,
,
and
increase significantly with the increase of incidence angle and water depth, and show little change when the incidence angle is smaller than 60 degrees in shallow water. The effect of SRB can reach 0.18 m at a depth of 100 m. The largest
can reach 450 m when the incidence angle is 80 degrees, and this determines the measured horizontal range.
From the above analysis (
Section 3.1), the SRB correction can be modeled approximately according to incidence angle at the same depth.
3.2. A Segmented Incidence Angle (SIA) Model
In this research, we assume that the SSP is inaccurate or the CTD has not been calibrated; thereby, the MSV and the SRB become the major factors affecting the accuracy of the acoustic positioning. This problem is also the most common for processing acoustic ranging data. In order to build an accurate fitting model, the parameters of SRB correction are divided into different groups by the incidence angles, and we can find a model through many simulation experiments. A new model is proposed to estimate the SRB correction according to the incidence angles
where
represents the incidence angle, and
represents the coefficient of the model.
represents the constant deviation due to inaccurate MSV.
represents the cosine of the incidence angle to the N. When the incidence angle is less than 60°, the SRB correction can be neglected according to the analysis in the previous section.
As shown in
Figure 5, the new SIA model can fit well with the SRB correction by selecting the parameters. The new model is used to fit the SRB correction, and N is selected as 3 or 4, corresponding to segments 60–70° or 70–80°, depending on the rising trend.
In shallow water OBS acoustic positioning, the area of multiple transponder positioning is normally small (one square kilometer). All transponders are located on the shallow and flat seabed (most shallow sea exploration terrain meets this condition); then, the water depth of the area can even be used as a fixed value. Assuming that the effect on SSP caused by the non-barotropic tidal flow or internal gravitational wave in a short time is small, the observation environment will be similar for all transponders within a short time. There are reasons to believe that the same incidence angles of acoustic ranging have the same SRB, which increases with the increase of the incidence angle. Naturally, the key issue is to select correct parameters, and the next section focuses on this problem. A high-precision underwater positioning algorithm for multiple targets with regard to acoustic line bending error is presented.
3.3. Calculation Method of Multiple Transponders with Sequential Least Square
For a single transponder, it is easy to cause ill-posed problems due to the introduction of many model parameters. From the above analysis, if the observations of different transponders have the same incidence angle, the same SRB correction can be estimated together. A large number of observations will place a huge burden on the computer. Based on the Sequential Least Square (SLS) method and the matrix orthogonal principle, a convenient solution is given in this research.
We first use the LS1 method or the LS2 method to calculate the initial transponder coordinates, and the observed values can be categorized according to the threshold values and incidence angles. Then, the unknown parameters
, and the new observation equation can be expressed as
where
represents the sum of random errors and modeling errors, and
.
represents the linearized observation.
and
represents the Jacobian matrix after linearization.
represents the coefficients of sound correction, and
.
represents the number of groups.
is the number of observations where the incidence angle is larger than 65 degrees, and
is the number of all transponder observations. According to the matrix orthogonal principle, the simplification of (17) is as follows
where
, and
represents an identity matrix.
represents the covariance of observation
, and
represents the SIA parameters.
Then the SLS method can be used to estimate the SIA parameters [
30].
where
represents the transponder ID, and
.
, and its variance-covariance matrix can be expressed by
As shown in
Figure 6, for the final transponder, SIA parameters and variance-covariance matrix can be calculated by solving Equations (21) and (22). When the final new model parameters of the SRB correction are solved, they are brought into Equation (17). Thus, the coordinates and incidence angles can be estimated at this time using the LS1 method again. The new results are brought into the sequential least squares algorithm and iterated until there are few differences from the previous results.
If the effect on SSP caused by the non-barotropic tidal flow or internal gravitational wave in a short time cannot be ignored, the new model parameters can be remodeled and estimated for sections of time. Only some new modeling parameters are added and estimated here. Certainly, we could also use the other models (see Equation (14) or Equation (15)) to reduce the impact of this change on positioning. Real-time and high-precision SSP is difficult to obtain in OBC positioning, and it is usually measured before or after positioning. The method proposed in this research is another way of improving its accuracy based on estimating the parameters of the acoustic bending model.