1
Introduction

Signal enhancement is a process to either restore a signal of interest or boost the relevant information embedded in the signal of interest and suppress less relevant information from the observation signals. Today, there is almost no field of technical endeavor that is not impacted in some way by this process.

Array signal processing manipulates the signals picked up by the sensors that form an array in order to estimate some specific parameters, enhance a signal of interest, or make a particular decision. The main purpose of this chapter is to:

• define the scope of the field that we call signal enhancement
• present a brief historic overview of this topic
• give some examples of fields where signal enhancement is needed and used
• discuss briefly the principal approaches to signal enhancement
• explain how array signal processing works.

1.1 Signal Enhancement

We human beings rely on our senses to sense the environment around us. Based on this information we build and expand intelligence in our brain to help make decisions and take actions. Similarly, we strive to build systems to help us "see" or "hear" distant events that cannot be reached by our senses. For example, nowadays, sonar systems can hear ships across hundreds of miles of ocean, radar devices can see airplanes from a thousand miles over the horizon, telecommunication systems can connect two or more users from different corners of the world, and high-definition cameras can see events happening on our planet from space. These systems use sensors to measure the physical environment of interest. Signal processing is then applied to extract as much relevant information as possible from the sensors' outputs. Generally, sensors' outputs consist of the signal of interest, which carries very important information, and also a composition of unwanted signals, which is generally termed "noise". This does not contain useful information but interferes with the desired signal. To extract the useful information in the presence of noise, signal enhancement is needed, the objective of which is to:

• enhance the signal-to-noise ratio (SNR)
• restore the signal of interest
• boost the relevant information while suppressing less relevant information
• improve the performance of signal detection and parameter estimation.

Signal enhancement is a specialized branch of signal processing that has been around for many decades and has profound impact on many fields. In the following subsections, we describe a few areas that routinely use signal enhancement techniques, particularly those developed in the following chapters of this text. Note that we can only cover a few applications, but this should leave the reader with no doubt as to the importance and breadth of application of signal enhancement techniques.

1.1.1 Speech Enhancement and Noise Reduction

In applications related to speech acquisition, processing, recognition, and communications, the speech signal of interest (generally called the "desired speech") can never be recorded in a pure form; it is always immersed in noise. The noise can come from very different sources. For example, microphones that we use to convert acoustic pressure into electronic signals have self-noise, even though the noise floor of popularly used capacitor microphones has been dropping significantly over the years. The associated digital signal processing boards, including preamplifiers, analog-to-digital (A/D) converters, and processors for processing the signals, may also generate noise. Most importantly, noise comes from ambient sources; the environment where we live is full of different kinds of sounds. While the sensors' self and circuit noise is generally white in spectrum, the noise from sound sources in the surrounding environment can vary significantly from one application scenario to another.

Commonly, noise from acoustic environments can be divided into the following four basic categories depending on how the noise is generated:

• Additive noise can come from various sources, such as cooling fans, air conditioners, slamming doors, and passing traffic.
• Echoes occur due to the coupling between loudspeakers and microphones.
• Reverberation is the result of multipath propagation and is introduced by reflections from enclosure surfaces.
• Interference comes from concurrent sound sources. In some communication applications, such as teleconferencing, it is possible that each communication site has multiple participants and loudspeakers, so there can be multiple competing sound sources.

Combating these four categories of noise has led to the development of diverse acoustic signal processing techniques. They include noise reduction (or speech enhancement), echo cancellation and suppression, speech dereverberation, and source separation, each of which is a rich subject of research [1-3]. This text presents many methods, algorithms, and techniques that are useful in dealing with additive noise, reverberation, and interference while its major focus, particularly the signal enhancement part from Chapter 2 to Chapter 6, is on reducing additive noise.

Additive noise and the desired speech signal are in general statistically independent. While the noise does not modify the speech characteristics directly, the characteristics of the observation signal are very different from those of the desired speech since it is a mixture of the desired speech signal and noise. Figure 1.1 plots a speech signal recorded in an anechoic (quiet and non-reflective) environment and the same speech signal but recorded in a conference room. The spectrograms of these two signals are shown in Figure 1.2. As can be seen, both the waveform and the spectrogram of the noisy signal are dramatically different from those of the clean speech. The effect of noise may dramatically affect the listener's perception and also machine processing of the observed speech. It is therefore generally required to "clean" the observation signal before it is stored, transmitted, or played (through a loudspeaker, for example). This problem is generally referred to as either noise reduction or speech enhancement.

Figure 1.1 (a) A speech signal recorded by a microphone in an anechoic environment and (b) the same speech signal recorded by the same microphone but in a conference room.

Figure 1.2 (a) The spectrogram of the speech signal in Figure 1.1a; (b) the spectrogram of the speech signal in Figure 1.1b.

1.1.2 Underwater Acoustic Signal Enhancement

Over the last few decades, ocean exploration activity for both military and civilian interests has been steadily increasing. As a result, there has been growing demand for underwater communication and signal detection and estimation technologies. Electromagnetic and light waves do not propagate over long distances under water (particularly sea water). In contrast, acoustic waves may propagate across tens or even hundreds of miles under the sea. Therefore, acoustic waves have played an important role in underwater communication and signal detection and estimation. For example, passive sonar systems can detect a submarine from tens of miles away by listening to the sound produced by the submarine, such as from the propellers, engine, and pumps; active sonars transmit sound pulses into the water and listen to the echoes, thereby detecting underwater features such as the location of fish, sunken objects, vessels, and submarines. Underwater wireless communication systems modulate useful information on acoustic carriers with frequencies between a few kilohertz and a few tens of kilohertz and transmit the modulated signal from one end to another through underwater acoustic channels.

However, processing underwater acoustic signals is by no means an easy task. First of all, underwater acoustic channels are generally known as one of the most difficult communication media in use today. Underwater acoustic propagation suffers from the time-varying multipath effect (due to sound reflection at the surface, bottom, and any objects in the vicinity, and also sound refraction in the water), frequency-dependent attenuation (due to absorption and signal spreading loss), and a severe Doppler effect (due to the low speed of sound and motion of the transmitter or receiver or the objects to be detected). Secondly, the ocean is filled with sounds, which interfere with the acoustic signal we are interested in. Underwater sounds are generated by both natural sources, such as marine animals, breaking waves, rain, cracking sea ice, and undersea earthquakes, as well as man-made sources, such as ships, submarines, and military sonars.

Marine animals use sound to obtain detailed information about their surroundings. They rely on sound to communicate, navigate, and feed. For example, dolphins can detect individual prey and navigate around objects underwater by emitting short pulses of sound and listening to the echo. Marine mammal calls can increase ambient noise levels by 20-25 dB in some locations at certain times of year. Blue and fin whales produce low-frequency moans at frequencies of 10-25 Hz, with estimated source levels of up to 190 dB at 1 m. Sounds generated by human activities are also an important part of the total ocean noise. Undersea sound is...