Boston University researchers base noise-shaping technology on neuronal networks
(Boston, Mass.) - The human brain is the most elegant of receivers. It can discern the notes of a piccolo from amongst the multitude of tones in a symphony orchestra and identify the familiar outlines of a friend in the midst of a crowd of strangers.
Realizing this extraordinary ability of the neuronal networks of the brain to separate signal from noise led a group of researchers at Boston University's Center for BioDynamics (CBD) and Department of Biomedical Engineering to develop a biologically inspired model that would improve the fidelity of electronic devices. The work was supported by a research gift from Ray Stata, chairman of Analog Devices and conducted in collaboration with Carson Chow (University of Pittsburgh), Wulfram Gerstner (Swiss Federal Institute of Technology), and Robert W. Adams, (manager of audio development at Analog Devices, Inc. of Norwood, Mass.), who first suggested the concept. It is reported in the August 31 issue of the Proceedings of the National Academy of Sciences.
Spectral noise-shaping, a process by which engineers improve the signal-to-noise ratio in electronic systems, is used in audio applications such as digital telephones, theater sound systems, and CD players. Traditional noise-shaping relies upon digital averaging techniques that are effective only in a narrow range of frequencies.
The techniques developed by Douglas Mar, James Collins, and their colleagues have significant advantages in that they can be effective over a much wider bandwidth, and can tolerate a greater amount of variation in the system components. The system is based on large networks of interconnected circuits, similar to neuronal networks in the brain. It is known that neurons fire in an often noisy and irregular pattern in the brain. Also, neurons often fire slowly in comparison to many of the signals they need to encode - a pitched baseball, for example, is in the "hitting zone" for only a few milliseconds, whereas individual neurons often take ten milliseconds or more between firings. Despite this discrepancy in timing, major leaguers such as Mark McGwire and Sammy Sosa hit home runs regularly.
The researchers hypothesized that a more efficient flow of information is effected in neuronal networks by inhibitory coupling - a process through which activation of one neuron momentarily suppresses the level of activity in other neurons in the network. Inhibitory coupling reduces the clumping of information, avoiding information traffic jams that both make it difficult to "see" the information of interest and slow down the system. By connecting elements in this way the researchers were able to similarly smooth the flow of information and identify and shift the noise, or unwanted information, from the bandwidth of interest to higher frequencies where it can be filtered out.
The next step, according to Mar, is to take the theoretical results of this investigation and begin to apply it in actual devices to bring higher fidelity into a new generation of biologically inspired electronics.
The above post is reprinted from materials provided by Boston University. Note: Materials may be edited for content and length.
Cite This Page: