Active Sonar Data Set

Published: 9 October 2017| Version 1 | DOI: 10.17632/fyxjjwzphf.1
Contributors:
Mohammad Khishe,
Mohammad Reza Mosavi

Description

In this data set, 6 objects including 2 targets and 4 non-targets lay on the sea sand bottom. Upon this experiment, the transmitted signal is Wide-Band Linear Frequency Modulated Pulse (WLFM) which covers frequency range 5-110 KHz. Targets lay on the bottom rotate 180 degrees with 1 degree accuracy via electromotor. Off target to 10 meters backscattered echoes are accumulated. Fine dataset takes key role in sonar target classification. Regarding massive raw data obtained from previous stage, above massive calculation will be expected. To reduce calculation burden relating to classifying and extracting feature, it is essential to detect targets out of total received data. To implement this, the intensity of the received signal is used. It is inevitable to consider multi-path propagation, secondary reflections, and reverberation due to shoal of the region. The researcher attempts to eliminate artifact tract after detecting stage and before extracting feature by the use of a matched filter.

Files

Steps to reproduce

Afterwards, the researcher proceeds to regain main backscattered signal by means of inverse filter. The reason behind this stage is that separating artifacts in matched filter domain is much easier than time domain. The process of preprocessing takes place in four stages as follow: 1. Scaling: It converts raw signal into scaled signal due to eliminate filter and boost gain effect at accumulating stage. 2. Down Sampling: Main sampling rate is 2 MHz which highly surpasses main signal band width. To reduce sampling rate in such a way that useful data does not miss, reference [51] has been utilized. Based upon this reference, by means of environmental data such as water depth, functional frequency, area under exploitation, and etc. few fixed points get selected at sampling stage. Here 2048 points are chosen not to waste useful data extracting the feature. 3. Multi-path and Artificial Elimination Process: Within this method by means of cross-correlating the backscattered signal with incident, the position of maximum matched filter output, named x, is determined at every angle. Afterwards, a window which covers [x-left: x+right] exerts over signal. This area includes right=300 and left=211 that ultimately forms a window of 512 points. To maintain quantity of the main signal this signal gets segmented and zero padded and to eliminate the effect of transmitted signal inverse filtering is done by Eq. (5): (5) Wherein is Fourier transform of the transmitted signal and is added to equation to eliminate singularity problem. The output of recent signal is pure subtraction without artifact effect. 4. Normalization: Any target gets scaling in such a way that each takes same target strength finally. To do so, each backscattered signal gets dividend through Signal Reference Amplitude (SRA) which is the largest amplitude that is less than 90% of the maximum amplitude of whole aspects for the interesting target. Samples of backscattered signal from various targets and non-target are indicated in Fig.7 in which they are function of frequency and target bearing.