Home

The optimization of underwater acoustic (UWA) communication protocols remains a great challenge owing to the harsh propagation conditions encountered at sea. The available bandwidth is limited and the time-varying multipath channel combined with the low speed of sound (1500 m/s) are responsible for strong signal distortions and high latency. In addition, in the absence of spectrum regulation, UWA communication systems may also suffer from unintentional jamming due to interference from other acoustic sources (that can be anthropogenic, biological or abiotic). All these varying phenomena contribute to drastic changes in the quality of UWA communication links from one environment to another.

In the field of UWA communications, I mostly contributed to the following topics:

Channel simulation

Among the various simulation strategies proposed in the literature, replay of time-varying impulse responses (TVIRs) measured in-situ has emerged as a relevant and accurate approach. The idea relies on the assumption that a TVIR measured at sea is an observation of an underlying ergodic random process. Based on this single observation, the objective is then to generate an infinite number of realizations of this process. This is what is called stochastic replay. The difficulty is to build a (not-too-complex) simulator that keeps the true space, time and frequency statistics of the real-world channel. Such a simulator lacks the universal applicability of usual parametric propagation models. However, this approach is of great interest to fully exploit sea experiments under controlled and reproducible laboratory conditions. From a single impulse response recorded at sea, it is thus possible to test a communication system in various noise conditions and for different types of modulations. It is also possible to compute channel capacity bounds or fading statistics useful for communication system design. I mostly contributed to the design of methods for non-WSSUS SISO or SIMO channels as well as the combination of parametric modeling with stochastic replay. I also contributed to the Watermark initiative led by FFI whose aim is to freely share at-sea measurements of time-varying impulse responses.

Channel capacity analysis

A few experiments over the past ten years have shown that transmission rates of tens of kbits/s can be achieved in some environments over ranges of several (tens of) kilometers. This is a big improvement compared to the few tens of bits per second achieved in the early practice. This improvement may either result from an efficient use of advanced signal processing techniques or from an advantageous environment, or from both. An assessment of the UWA channel capacity appears critical to determine whether state-of-the art signal processing techniques make the maximum rate offered by the channel achievable. In this context, I mostly contributed to the derivation of bounds on the Shannon capacity by considering realistic constraints such as rapidly varying multipath propagation, peak-power constraints and bandwidth limitations. Others works dedicated to OFDM signalling have been conducted in collaboration with Northeastern University.

Self-configurable modems

Software-defined UWA modems are now mature enough to motivate the development of adaptive optimization algorithms dedicated to on-the-fly configuration of physical (PHY) layers. Self-configuration of modems is a key enabler to improve efficiency and robustness of digital communications through underwater channels. The key parameters to consider in self-configuration problems are (i) the performance metrics to optimize, (ii) the set of actions available at the PHY layer to adapt to specific environments as well as (iii) the amount of environmental information available to modems. Among these three key parameters, the last one is surely the most specific to UWA communications. Information on the acoustic environment is often acquired by analyzing signals at reception and is then forwarded to the transmitter through a feedback link. Since sound propagates slowly and channels may exhibit short coherence times, adaptation with respect to instantaneous channel state information (CSI) is rarely possible. The CSI must either be predicted based on past information or averaged over time. Because the bandwidth is limited underwater, low signaling overhead is also desirable. The difficulty is then to find the right compromise between fine and coarse-grained feedback information in order to reach and track efficient operating points without too much overhead. I mostly contributed to the design of adaptive decentralized spectrum sharing method for non-cooperative systems in interference channels. This work relies on a game-theoretical framework based on either Nash equilibria or Satisfaction equilibria. We also collaborated with the University of York on the design of an adaptive channel selection method for underwater sensor networks.

In a broad sense, Passive Acoustic Monitoring (PAM) refers to the use of hydrophones (underwater microphones) to either detect, locate, classify sound sources or to use them as signals of opportunity to study the physical properties of the ocean. Passive acoustics is not intrusive because it is solely based on listening, no noise is put into the environment. Historically developed for underwater warfare, this type of monitoring meets new operational needs for physical oceanography, geophysics or biology. The success of PAM has led to an increasing deployment of underwater acoustic recorders across many oceans. As a result, the development of efficient and robust automatic signal processing methods is needed to analyze the growing amount of acoustic data generated by these recording systems.

My work on PAM has focused on two research topics. The first one concerns the localization of low-frequency sources when the ocean environment is poorly known. I contributed to the derivation of performance bounds [1,2,3] of localization techniques known as matched-field and matched-mode processing. A Bayesian localization method has also been proposed in collaboration with the University of Victoria to improve robustness to environmental mismatch. The second PAM topic I am interested in is the detection and classification of baleen whale calls. I contributed to the development of a subspace detector for Antarctic blue whale calls or to the use of sparse-representations for call detection or classification. Although this is mostly applied research, it is also related to fundamental aspects such as the asymptotic formulation of the well known Karlin-Rubin theorem.

RF spectrum monitoring refers to the characterization (detection, localization, classification) of any radio-frequency signal. It is  is an essential building block in signals intelligence systems. Historically developed by major defense manufacturers, these systems used to be very costly to design. With the development of software-radio, the cost of RF sensors has been considerably reduced and has made it possible to transpose monitoring techniques from defense to civilian applications (cognitive radio, surveillance of critical infrastructures, surveillance of sporting events, etc.). In parallel, the number of possible threats and the number of communications protocols have exploded.

Most of my work on that topic has been conducted in industry or through industrial contracts. I have been mostly interested in the design of efficient methods for detecting communication signals and blindly estimating their modulation parameters. A large part of my work is not accessible in the literature because of  security restrictions. However, some contributions have been published in the context of 

• Cyclostationarity-based detection of OFDM systems for cognitive radio [1], 

• Robust noise variance estimation [2,3,4]

• Detection of random signals with unknown distribution [5,6]