Bio-Inspired Science & Technology on the neuroscience of bat sonar
James Simmons, Dept. of Neuroscience
Echolocating bats can detect, localize, and classify targets-all technologically-desirable capabilities inherent in the use of wideband sonar signals, while also flying rapidly through complex surroundings and avoiding collisions with obstacles. The novelty is their ability to suppress interference from surrounding clutter without losing the ability to follow the flow of objects moving past on the left and right. This is a critical capability because guidance of movement is another desirable function of sonar, and clutter actually constitutes the bulk of most sonar scenes. The conventional solution involves narrow sonar beaming and range-gating to prevent formation or reception of echoes from off to the sides or from the wrong target distances. Scanning of the surroundings with a narrow beam is only feasible of there is plenty of time, however. Echolocating bats fly at several meters per second-too fast for scanning to be effective. They have evolved a very different solution whose discovery involves in-depth experimental and computational analysis of the entire process of forming and displaying perceptual images, using biosonar as a model system for addressing basic questions in cognitive and systems neuroscience. Knowing that bats emit FM sounds suggests pulse compression, knowing that the bat's inner ear segregates the FM sweeps into parallel frequency channels suggests a time-frequency version of matched filtering, and knowing that neuronal spiking occurs in the bat's auditory system suggests a point-process instantaneous-frequency representation, but these do not add up to a technological advance because the system uses arithmetic that is entirely unfamiliar. The bat's sonar receiver exploits a peculiarity of the auditory time-frequency representation: it transposes changes in signal amplitude at individual frequencies into changes in the timing of the neuronal spikes that register instantaneous frequency. (This effect is called amplitude-latency trading.) As a consequence, making estimates of echo parameters initially represented by amplitude values at different frequencies is converted into detecting timing disparities across frequencies. In the bat's processing scheme, making comparisons of analog time across frequencies and across time is the basis for the entire sonar imaging process. This kind of arithmetic is not comparable to familiar analog computation which operates on numerical amplitude values, and it is not digital computation. The only part of the bat's receiver that resembles the digital regime is the act of comparing analog times, which replaces cumbersome multiple-bit digital multiplication. Trying to design biomimetic sonar raises the prospect of a type of neuromorphic computer that excels at sensory imaging and motor control in real time.