Snapping Shrimp Noise Reduction using Convex Optimization for Underwater Acoustic Communication in Warm Shallow Water Dayan Adionel Guimaraes˜ Lucas Silvestre Chaves Rausley Adriano Amaral de Souza National Institute of National Institute of National Institute of Telecommunications - Inatel Telecommunications - Inatel Telecommunications - Inatel Santa Rita do Sapuca´ı, MG, Brazil Santa Rita do Sapuca´ı, MG, Brazil Santa Rita do Sapuca´ı, MG, Brazil Email: [email protected] Email: [email protected] Email: [email protected] Abstract—We propose an ambient noise reduction technique channels, representing a challenging limitation for UWAC for underwater acoustic communication systems in warm shallow system design. The main limiting factors are the propagation water, where the snapping shrimp impulsive noise is a common delay, multipath, attenuation and ambient noise, all with strong impairment. This noise is emitted at the cavitation bubble collapse caused by the claw shut, and can produce very high acoustic randomness. pressures which are responsible for the impulsiveness of the noise. The channel is subjected to a three-dimensional variability The denoised signal is the solution of an unconstrained scalarized of the refractive index, leading to ray bending (refraction). bi-criterion convex optimization problem. One of the criterions is This is one of the main causes of multipath and shadowing. based on the robust least-squares signal reconstruction approach, Multipath propagation is also caused by reflections on sea for which the sum of the Huber penalty function of the residuals is minimized. The other criterion uses a quadratic smoothing surface and sea floor. Small coherence times combined with regularization term, acting on the first-difference of the denoised large propagation delays also impose severe limitation to the signal samples. It is shown that the proposed technique can system design. For example, in a 1 km range the signal can significantly reduce the ambient noise and, thus, reduce the bit take up to 0.66 seconds to hit the receiver. On the other hand, error rate of an underwater digital communication system. channel characteristics can change drastically within 0.5 sec- onds or less. Delay spreads can reach up to 60 ms, producing I. INTRODUCTION severe frequency-selective fading. Doppler shift and Doppler The high attenuation of electromagnetic waves, especially spread also appear, and are caused due to surface and internal by sea water, is one of the most limiting factors for using radio wave motion and due to relative motion between receiver waves for underwater communications. On the other hand, and transmitter. Additionally, the sound speed varies in water, these attenuations are significantly smaller for acoustic waves, mainly with depth, and depends on the water characteristics which motivates the use of acoustic signals in underwater and weather conditions. communication systems [1, pp. 15-47], [2], [3]. The main mechanisms of signal loss in UWAC channels Underwater acoustic communication (UWAC) systems have are the spreading of the signal energy with distance, similar been used in many applications, from sound navigation and to what occurs in radio communications, absorption and scat- echo detection (sonar), to diver-to-diver and ship-to-diver tering. Absorption is the conversion of acoustic energy into communications, for communications with submersibles and heat and it is a frequency-dependent and chemical-dependent other underwater vehicles for sea exploration, in telemetry phenomenon. Scattering occurs at sea surface and sea floor, and in underwater control systems. Applications related to but also on water bubbles and on small water volumes with underwater sensor networks are one of the most recent ones different temperatures. Moreover, since the signal-to-noise [4]–[7]. The main characteristic of UWAC systems is the use ratio is strongly dependent on range and frequency, higher of acoustic carrier waves typically above 5 kHz. The upper ranges mean lower available bandwidths for data transmission. limit depends on the range. As an example, for a 10 km range Besides the thermal noise generated at the receiver, typical this upper limit is about 10 kHz. For a 100 m range, signals ambient noise impairments in UWAC systems are generated up to 100 kHz can be used. Some modern high speed UWAC by breaking waves, rain, marine life and ships. Among theses systems for short-range communication use acoustic carriers ambient noise, in warm shallow water it is common the with even higher frequencies. Transmission powers typically existence of an impulsive noise generated by snapping shrimps are on the order of 30 W or less. Transmission rates are up to [8]–[11]. In this paper we propose a technique for combating a few kilobits per second in a 10 km range. the compound noise resultant from the addition of the thermal The impairments found in underwater acoustic channels are and the ambient snapping shrimp noise. Throughout the paper even more severe than those encountered in wireless radio we interchangeably refer to this technique as noise reduction 2 or simply denoising. The denoised signal is sometimes referred a to as the reconstructed signal. 1 The remaining of the paper is organized as follows: in 0 Section II we describe the proposed noise reduction scheme. -1 -2 Section III presents the digital communication system model 0 250 500 750 1000 1250 1500 1750 2000 20 adopted for assessing the performance of the denoising pro- b 10 cess, in terms of the bit error rate (BER) versus the ratio 0 between the average energy per bit and the noise power -10 spectral density (Eb=N0). Section IV provides the numerical -20 0 250 500 750 1000 1250 1500 1750 2000 results and Section V concludes the paper. 2 c 1 II. THE PROPOSED DENOISING SCHEME 0 An n-dimensional transmitted signal vector x, under fading, -1 is added to a vector n representing the compound (thermal -2 plus ambient) additive noise, resulting in the received vector 0 250 500 750 1000 1250 1500 1750 2000 r = x+n. From this vector, the noise reduction process tries to Fig. 1. Examples of: a) the transmitted signal vector x, b) the corrupted reconstruct the transmitted signal, resulting in the estimate xˆ, received signal vector r, and c) the reconstructed signal vector xˆ which is then used to detect the transmitted data. The denoised signal xˆ is the solution of the unconstrained scalarized bi- criterion convex optimization problem so far for combating the ambient noise in underwater acoustic Xn communication systems. minimize ϕhub (^xi − ri) + δϕquad(^x); (1) For illustration purpose only, Figure 1 shows 2000-point i=1 vectors from this signal denoising process. Notice that the where δ > 0 parameterizes the trade-off between fitting received vector is under a heavy noise condition, with strong (governed by the first term) and smoothness (governed by impulsiveness. Nonetheless, the denoised signal is noticeably the second term). The Huber penalty function with shape free from impulsive noise and from most of the thermal noise. parameter M is [12, p. 299] Yet, the reconstructed signal is free from the typical time { dispersion the would be produced if conventional filtering were u2 juj ≤ M used as a means for noise reduction. An efficient filtering (in ϕhub(u) = ; (2) M (2 juj − M) juj > M terms of noise reduction) would inevitably be accompanied by and the quadratic smoothing function is [12, p. 312] intersymbol interference, which would prevent the reduction of the symbol error rate. nX−1 2 ϕquad(^x) = (^xi+1 − x^i) : (3) III. SYSTEM MODEL FOR PERFORMANCE ANALYSIS i=1 The vector x represents a transmitted signal whose rate of The reasoning behind the proposed denosing technique goes amplitude variations is much slower than those present in the as follows: it is known that when the Huber penalty func- compound noise vector, so that the denoising process works as tion is applied to a linear regression problem, the result desired. This condition can be met by modulated and filtered is robust against data outliners, and that is the reason for UWAC signals due to the fact that, besides the smoothness giving the name robust least-squares to such approach. From produced by filtering, the carrier in these systems is an acoustic the perspective of the compound noise present in UWAC wave, typically with frequency on the order of a few tens communications, the impulsive noise samples can be seen as of kilohertz [1, pp. 15-47]. Moreover, the frequency content outliers within the received signal samples. The solution of of the snapping shrimp impulsive noise easily reaches 250 the least squares in the linear regression problem can be seen kHz [9], which means that, indeed, its rate of variation is as a linear approximation of a smoothed reconstructed signal much larger than the rate of variation of typical modulated segment. Then, by combining a robust least-squares fitting and filtered transmitted signals used in UWAC systems. with a smoothing process, the reconstructed signal will not Besides the compound noise, UWAC signals are also sub- depart too much from the smoothed one by the influence of jected to fading, even in short-range communications where the impulsive noise. In other words, the quadratic function in the observed fading is similar, but slightly less severe than (1) will produce a smooth reconstructed signal that, due to the predicted by the Rayleigh distribution [3, p. 73]. Thus, it influence of the first term, does not care too much about the is mandatory to consider the effect of the fading in the presence of the impulsive noise, depending on the value of δ: performance of an UWAC system. a smaller δ acts in favor of a better fit between the corrupted Aiming at adhering to the above requirements, we have and the reconstructed signal, in the Huber measure sense; a adopted the system model shown in Figure 2.