SUNFEST Projects Summer 1998
Jeffrey Berman (BE/EE) - University of Pennsylvania
A knowledge based feature extraction system for identifying the nasal and /r,w/ phonemes was developed using Matlab. The system is speaker independent and searches for identifying patterns of these phonemes in the speech signal and two physiologically based spectrograms. The /r,w/ semivowel phonemes were found to have characteristic features primarily visible in the time domain. However, the synchrony spectrograms used for this project can have poor frequency resolution, making detection of formant movement difficult. The nasal phonemes were found to have a distinctive pattern primarily in the frequency domain. A series of independent features including energy patterns, zero crossings, and duration were used to design a more robust method of finding nasals.
Alexis Diaz (Mechanical Engineer) University of Turabo
The concept of Binder extraction was thought of as an ideal way to remove some of the organic part of the LTCC tape by chemical etching. This would help to create channels and other marks on the LTCC tape, which in turn would be helpful in the design and construction of certain devices such as impactors, actuators and heat exchangers. Making binder extraction not only an innovative way to etch the LTCC green tape, but it also has a low cost manufacturing process involved.
Clara Dimas (EE) University of Pennsylvania
LTCC tapes are useful in Meso-scale EMS for their mechanical strength and high thermal conductivity. Since these structures require cavities and vias, patterning LTCC tapes is necessary. Current patterning methods process each ceramic individually. This paper discusses two solutions and advancements in LTCC patterning using photolithographic procedures, a batch process. One method is using e-beam resist on an LTCC and chemically etching a LTCC; another approach is to use an LTCC that is photosensitive. Research was conducted to work towards establishing procedure parameters, and determining resolution and minimum feature size. Finally two examples of LTCC applications, which would benefit from the above advancements, are a thermal accelerometer, and Silicon circuits mounted to LTCC cavities to dissipate heat. Thus the thermal accelerometers middle membrane was constructed, and bonding Silicon to LTCC directly and with a glass like polymer were explored.
David Friedman (Computer Science) - University of Pennsylvania
A vowel recognition algorithm was implemented for the University of Pennsylvania/ Corticon NP-4 neurocomputer. The NP-4 is a fully parallel, programmable analog neurocomputer. The algorithm was implemented in software and integrated into the control software for the neurocomputer. Preliminary testing was performed to find the responsiveness of trained neurons to vowels.
Christin Lundgren (EE) Bucknell University (with
Andrew Utada (Physics) Emory University)
Ultrasound transducers were studied in preparation for the design of a communication system for cooperative swimming robots. Different transducer materials and geometries were tested in order to determine which produced the best received pulse. The applied electrical pulse was also varied. It appears that piezoelectric PVDF film is the most feasible material for this application. It has good pulses, little ringing, and is flexible enough to accommodate multiple shapes and configurations.
Also, an amplifier circuit was built which will enable a robot to amplify a received signal and transmit it back to the sender. This will allow the sender to determine the time of flight, which in turn enables it to calculate the distance between the two robots.
Heather Anne Lynch (Mechanical Engineering) Villanova
As a material for meso-scale electro-mechanical devices, Low Temperature Co-Fired Ceramic (LTCC) shows great promise. This project undertook the fabrication of a pressure transducer. Work on chemical exfoliation of the ceramic material led to a good method for transforming the fired LTCC into the thin, elastic membrane needed in a piezo-resistive pressure transducer. Various other methods for machining the green tapes were explored in order to create a ceramic body for the device. Testing of the devices created showed their initial success and possibility for development to the point of being able to compete with existing commercial devices.
Andrew Utada (Physics) Emory University
In order to build a system for the communication and cooperation of swimming robots, planar circular piezo-ceramic ultrasonic disc transducers were used to create acoustic pulses underwater and their angular distributions were studied with and without two differently sized acoustic lenses. The wave spreads of unbacked polyvinylidene fluoride (PVDF) piezo-electric transducer films mounted on concave and convex mountings were also studied. The characteristics of the transducers generated pulse when measured on axis using another transducer as receiver, were observed to be closely related to the input signals pulse height and width. The planar ceramic transducers produced very narrow beams; however, when transmitting through a glass lens, the acoustic signals were more widely spread out. The concave and convex mounted transducer films also generated signals with a larger angular distribution than when mounted on a flat surface. Data acquisition was done using an A/D board and a software program called Gagescope. A transmit/ receive (t/r) amplifier circuit was also built to increase weak signals and to test its effectiveness for future applications.
Sancho Pinto (Computer Science and Mathematics) University of
Speech recognition is an area of research whose uses encompass multiple areas of life. Automatic speech recognition systems take either knowledge based or stochastic based approaches. In this project an isolated word recognizer using a stochastic approach, Hidden Markov Models, was implemented in Matlab. The vocabulary has been limited to 10 words.
Edain Velazquez (EE) - University of Pennsylvania
Much effort has been put to improve time-multiplex near infrared imaging systems to obtain tissue optical properties and detect tissue abnormalities such as tumours and breast cancer. These systems include Amplitude Cancellation Systems and PMS Systems. Thanks to these non-invasive optical methods the quality and quantity of information that can be obtained via X-ray and ultrasound mammography can now be surpassed. We have developed a more efficient sample-and-hold circuit that will allow a new amplitude cancellation system for breast cancer detection to determine the optical properties of breast tissue more accurately. In addition, our design of a new automatic gain control (AGC) for a novel PMS system (I & Q system) efficiently eliminates the phase-amplitude cross-talk of the system.
Tarem O. Ahmed (Physics) Middlebury College
This project involves the design and construction of an autonomous
robotic device that swims and follows a light beam. A BASIC Stamp processor in the robot
controls the voltage applied to a servo motor, which rotates a fin. Photoresistors act as
light sensors, and the center of oscillation of the fin determines the direction of