Abstract: Acoustic communication is an enabling technology for many autonomous undersea systems, such as those used for ocean monitoring, offshore oil and gas industry, aquaculture, or port security. There are three main challenges in achieving reliable high-rate underwater communication: the bandwidth of acoustic channels is extremely limited, the propagation delays are long, and the Doppler distortions are more pronounced than those found in wireless radio channels. In this dissertation we focus on assessing the fundamental limitations of acoustic communication, and designing efficient signal processing methods that can overcome these limitations.
We address the fundamental question of acoustic channel capacity (achievable rate) for single-input-multi-output (SIMO) acoustic channels using a per-path Rician fading model, and focusing on two scenarios: narrowband channels where the channel statistics can be approximated as frequency- independent, and wideband channels where the nominal path loss is frequency-dependent. In each scenario, we compare several candidate power allocation techniques, and show that assigning uniform power across all frequencies for the first scenario, and assigning uniform power across a selected frequency-band for the second scenario, are the best practical choices in most cases. We quantify our results using the channel information extracted form a recent experiment.
Next, we focus on achieving reliable high-rate communication over underwater acoustic channels. Specifically, we investigate orthogonal frequency division multiplexing (OFDM) as the state-of-the-art technique for communication over frequency-selective multipath channels, and propose a class of methods that compensate for the time-variation of the underwater acoustic channel. These methods are based on multiple-FFT demodulation, and are implemented as partial (P), shaped (S), fractional (F), and Taylor series expansion (T) FFT demodulation. They replace the conventional FFT demodulation with a few FFTs and a combiner. The input to each FFT is a specific transformation of the input signal (P,S,F,T). We design an adaptive algorithm of stochastic gradient type to learn the combiner weights for coherent and differentially coherent detection. The algorithm is cast into the framework of multiple receiving elements to take advantage of spatial diversity. Analysis of synthetic data, as well as experimental data, shows significant performance improvement over conventional detection techniques ($5$~dB$-7$~dB on average over multiple hours), as well as improved bandwidth efficiency.
Professor Milica Stojanovic, Advisor
Professor Tommaso Melodia
Sarah Kate Wilson