Optimization of the Geometry of
Communication for Autonomous
Missions of Underwater Vehicles
Candidato Alessio Micheloni 443219 Relatore Prof. Andrea Caiti
Controrelatore
Prof. Lorenzo Pollini
Anno Accademico 2014/2015
Outline
Introduction and Background
Acoustic Channel Simulations
Methods and Parameters Results and Discussion
Transmission Quality Analysis
Cross-Correlation Approach Source Signal Estimation
Source-Receiver Motion
Introduction
Autonomous Underwater Vehicles (AUVs)
AUVs navigation and localization
Acoustic-based sensors and communication
Underwater Acoustics
Influenced by speed of sound
Numerical models for sound propagation
Goals of the thesis
Characterization of a real acoustic channel Analysis of signals transmission
Underwater Acoustic Localization
Acoustic navigation techniques
AUV localization is achieved by measuring the
time of flight (TOF) of acoustic signals
The CommsNet13 Sea Trials
Framework: the THESAURUS project
Project specifications and results
Development of Typhoon AUV class
Localization: mixed USBL/LBL approach
Localization procedure
USBL-modem equipped on the vehicle
Acoustic modems deployed at the bottom Initialization and navigation phases
The CommsNet13 Sea Trials
Navigation Path
Irregularly spaced position fixes from acoustic
modems, how can we explain this?
The CommsNet13 Sea Trials
Spatial distribution of modem interrogations
There is a correlation between the AUV position and the unsuccesful modem interrogations.
The CommsNet13 Sea Trials
Spatial distribution of modem interrogations
Possible explanation: communication losses
The CommsNet13 Sea Trials
Underwater sound speed profiles measured
with a CTD sensor during the experiment.
Sound speed represents the most important
Acoustic Channel Simulations
Available Software
Ocean Acoustics Library (website)
Acoustic Toolbox (by Mike Porter) contains the
most common numerical models
AcTUP (A. Duncan, A. Maggi), is a MATLAB®
GUI-wrapper for Acoustic Toolbox
For our purposes we chose BELLHOP, a ray
Acoustic Channel Simulations
Simulation Parameters
Frequency: 30 kHz Bottom Depth: 30 m Source Depth: 29 m
Receiver Depth: 0.5 m (morning) and 5.5 m
(afternoon)
Channel Characterization
Acoustic waves paths
Acoustic Channel Simulations
Waves Paths Ray formalism
Waves modeled as complex amplitude-phase pairs
𝑝𝑝 𝑟𝑟, 𝑧𝑧 = 𝐴𝐴 𝑟𝑟, 𝑧𝑧 𝜃𝜃 𝑟𝑟, 𝑧𝑧
Snell’s Law: ratio cos 𝜃𝜃 / c is always constant
Arrivals (or eigenrays)
Impulse response of the acoustic channel
Waves Attenuation
Transmission Loss (measured in dB)
Results – Ray Paths
Results – Transmission Loss
Results – Arrivals
Morning, 200 m
Received signal calculation as the convolution with
the impulse response of the channel: 𝑟𝑟 𝑡𝑡 = ℎ 𝑡𝑡 ∗ 𝑠𝑠(𝑡𝑡)
Transmission Quality Analysis
We now consider a source-receiver pair in a
fixed configuration (no relative motion)
Transmitted signals encounter interference
Transmission Quality Analysis
We now consider a source-receiver pair in a
fixed configuration (no relative motion)
Transmitted signals encounter interference
and multipath effects due to reflections A simple example
Transmission Quality Analysis
To see what happens at the original signal, the
cross-correlation function can be calculated: 𝑅𝑅𝑠𝑠𝑠𝑠(𝑘𝑘) = � 𝑠𝑠 𝑡𝑡 𝑟𝑟(𝑡𝑡 + 𝑘𝑘)
Transmission Quality Analysis
Source Signal Chirp (sweeping sinusoid) Lowest frequency: 18 kHz Highest frequency: 34 kHz Duration: 40 ms
Cross-Correlation Approach
Conditions: morning, range 200 m
Cross-correlation emphasizes the interference of attenuated signal replicas.
Source Signal Estimation
Signal detection
Threshold value on cross-correlation
We are too sensible to signal fluctuations
Idea! Source signal estimation
Implemented on the receiver
Substantially improves signal detection
Two-steps estimation procedure
Impulse response (correlation analysis) Source signal (deconvolution)
Impulse Response Estimation
We suppose a linear system between input and output signals:
𝑦𝑦(𝑡𝑡) = � ℎ 𝑘𝑘 𝑢𝑢(𝑡𝑡 − 𝑘𝑘)
∞ 𝑘𝑘=0
The correlation equation then becomes:
𝑅𝑅�𝑦𝑦𝑦𝑦 𝑡𝑡 = � ℎ�(𝑘𝑘)𝑅𝑅�𝑦𝑦𝑦𝑦(𝑡𝑡 − 𝑘𝑘)
𝑀𝑀 𝑘𝑘=0
Writing out this equation for 𝑡𝑡 = 0,1, … , 𝑀𝑀 we get
a linear system so we can calculate the impulse
Deconvolution via Least Squares
Now we write the linear system in this form:𝒚𝒚 = 𝐻𝐻𝒖𝒖
To calculate the input signal estimate we perform the following regularized deconvolution:
𝒖𝒖� = (𝐻𝐻𝑇𝑇𝐻𝐻 + 𝜆𝜆𝜆𝜆) −1𝐻𝐻𝑇𝑇𝒚𝒚
The regularization parameter 𝜆𝜆 is needed because
convolution matrix 𝐻𝐻 is close to singularity (diagonal loading).
Deconvolution - Results
Results for morning, 200 m case when 𝜆𝜆 ≈ 10−5.
Signal detection can now be based on the cross-correlation between true and estimated signals.
Sensitivity to Regularization
Sensitivity to Regularization
Optimal value of 𝜆𝜆 selection:
𝑎𝑎𝑟𝑟𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑝𝑝𝑐𝑐𝑐𝑐𝑠𝑠𝑠𝑠 = ∑ 𝑝𝑝𝑃𝑃
𝑖𝑖
𝑃𝑃 = cross-correlation peak (highest value)
𝑝𝑝𝑖𝑖 = values above a specified threshold of 𝑃𝑃
The resulting correlation function becomes the
Sensitivity to Regularization
Optimal value of 𝜆𝜆 selection:
𝑎𝑎𝑟𝑟𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑝𝑝𝑐𝑐𝑐𝑐𝑠𝑠𝑠𝑠 = ∑ 𝑝𝑝𝑃𝑃
𝑖𝑖
Doppler Shifts
Actually, the receiver (USBL-vehicle) keeps
moving while communicating with the modem
Doppler shifts due to relative motion
Source-receiver relative speed 𝑣𝑣 Proportional to ratio 𝑣𝑣/𝑐𝑐
BELLHOP includes a specific algorithm
Vehicle speeds:1-1.5-2 knots
Doppler Shifts
Speed: 1 m/s
Doppler shifts do not change the envelope of
the signals. Signals estimation is not affected.
Conclusions and Future Work
What we have done
Acoustic channel characterization
Extensive analysis of transmission quality
We have shown that acoustic channel effects
can be eliminated using a simple reception
algorithm
Irregular position fixes – possible explanations
Acoustic modems were working at the limit of
their detection threshold
Fluctuations of sound speed and/or other physical