A Novel Signal Model and Estimation Method for Acoustic Doppler Current Meters

Sammanfattning: This thesis deals with the process of estimating underwater current from the backscattering signal transmitted from an ultrasonic transducer. The narrowband sinusoidal burst signal is Doppler-shifted due to the current, and this information is converted into current. The traditional mathematical model for this Doppler process is based on the assumption that the backscattering time signal is Gaussian, due to the Rayleigh backscattering amplitude with random phase. This is based on the assumption that the backscattering is due to many randomly distributed bubbles with about equal size. It is reasonable to question whether this assumption holds for real life signals. Therefore, part of the work discussed in this thesis has concentrated on looking at real life data, and has investigated whether the Gaussian assumption holds for the background noise and the Doppler signal received. It has been found that this is not generally the case. Thus, a different mathematical model has been sought which describes the Doppler signal in a more realistic way. Such a model has been developed, and is presented. When studying the work of other researchers, it has been noticed that some of the more recent modeling work relies on elaborate mathematics, but does not contribute much to the understanding of the physics, nor the effects on the estimation process. Again, and this is emphasized, the aim has been to develop a simple signal model and a simple estimation method that performs well. The latter is important since the complexity of the system needs to be low. This is because systems are often battery operated and a long battery life is needed as the systems are positioned out in the water. The new mathematical model for the backscattering signal is subsequently used to test different estimator approaches, and to investigate their behavior for different types of complexity in the Doppler signal. A new estimator, the Symmiktos Method™ estimator, has been developed, which is a non-linear estimation method. The estimator performance is compared with the covariance method, which is based on the Gaussian model assumption. The Symmiktos Method™ is based on a more complex Doppler signal model. The result is that the Symmiktos Method™ is more robust to changes in the signal complexity. Also, the method is quick and easy to implement, an important factor in real-life use. Otherwise the method becomes a theoretical method with very little practical use, and this has not been the intention with this work. The work should be viewed as a practical approach to solving a difficult signal processing application problem, where cost, size, simplicity and performance are very important for the end result. A large data set from four locations, Trubaduren, Almagrundet, Fladen and Ma-Wan, Hongkong has been collected. This data base was used when performing elaborate statistical analyses like ANOVA, higher order moments, histograms and normal probability plots. Classical signal processing has been performed as well as non-linear filtering using Multiple Peak Count Analysis, MPCA. The latter is presented in both a 2D and a 3D format. Different estimation methods are compared, including the Symmiktos Method™.