As an undergraduate student, you can work side-by-side in the lab with some of the best faculty in engineering, medicine, and the sciences to solve problems, take entire courses focused on design, or create your own prototypes in our maker spaces and machine shop. You also can work with students from across WashU, including the medical school, through IDEA Labs, a student-run bioengineering design incubator that solves health care problems.
If you are looking for an opportunity for Undergraduate Research, please look at this list to see what projects are available. Do not hesitate to contact the professor listed below to continue working on these projects or if you have ideas of your own related to these projects. Typically, students will take ESE297 (Introduction to Undergraduate Research) first to get a broad, practical, hands-on introduction to signal processing implementations. Those students will take 2 additional semester of ESE497 where they work directly with a mentor on their research projects.
Cocktail Party Hearing Aid by Microphone Array (Professor: Arye Nehorai)
Hearing in the presence of background noise is challenging enough for people with normal hearing. The problem is much worse for the hearing impaired. It is also a situation where traditional hearing aids don't perform well. In this project, the student will use a 64 element microphone array. We will place the array in the center of a table while several people sitting at the table carry on a normal conversation to produce the Cocktail Party Effect http://en.wikipedia.org/wiki/Cocktail_party_effect . Using data collected from the microphone array, the student will develop algorithms to remove the background noise and amplify the current speaker. The current speaker can be identified based on signal strength or from the pose of the listener identified with a camera. This filtered signal will then be transmitted to the smart phone of a hearing impaired person wearing one or two headphones.
Robotic Obstacle Avoidance using Kinect and UST (Professor: Arye Nehorai)
This project was undertaken by Yifan Wang and Stephen Gower over the summer of 2015, in an attempt to implement avoidance using a paired sensor arrangement. The idea is simple: the Kinect has excellent mapping of it’s surroundings, but a large blind spot in its near-field, while the Ultrasonic Transducer(UST) does not have this same problem, yet lacks in its ability to gather detailed information about the surroundings especially at far-field. Using this knowledge, it would make sense then that the two sensors could be intelligently paired to mask each other’s failings while allowing for the strengths to shine for each sensor. The scope of this project is limited to getting a working model for a simple avoidance field. Neither I nor Yifan had extensive experience in robotic obstacle avoidance prior to undertaking this project and as such there was a lot of learning to be done. With this in mind, we are publishing the results of our project and the code with detailed explanation as to how we reached our result. In the case of our robot, we were satisfied with the relatively consistent performance and we managed to at least regularly avoid arrangements of at least three obstacles regularly.
Automated Sleep Stage Classification (Professor: Arye Nehorai)
The objective of this project is to develop an efficient automated algorithm for sleep-wake staging. This project is motivated by recent research showing that Alzheimer's Disease (AD) is associated with sleep disruption and sleep disorders. AD is a progressive neurodegenerative disease that affect memory, cognition, and ability to carry out daily activities. Even in the very earliest stages of AD, there are abnormalities of sleeping such as excessive daytime napping and reduced sleep quality at night. Hence, a method of accurately detecting and scoring sleep in the community setting, consistent with American Academy of Sleep Medicine standards, is needed to identify individuals at high risk of having or developing AD. Ideal background for student candidates includes significant Matlab programming experience and prior coursework in signal processing.
Parkinson's Tremor Detection using Foot Mounted Inertial Movement Units (Professor: Arye Nehorai)
A group of researchers from KTH (Royal Institute of Technology) in Sweden have developed a tracking system based on foot mounted inertial movement units (IMU)*. This project is to use this technology to detect tremors in Parkinson's patients. You will the typical tremor patterns in Parkinson’s patients. Then, attach the IMUs to your body and alternate between imitating the tremor and normal movement while streaming the IMU sensor data (Sampling Rate = 1 KSample/sec) and video to the computer over USB. Next, synchronize the video and the sensor data and develop signal processing algorithms to automatically detect the tremor. Then develop a real-time system that implements the algorithm. Another aspect of the project is that currently, the IMUs can stream data over USB at 1 KSample/sec but are limited to 100 Samples/sec over Bluetooth. However, this project can only be used in a clinical setting if it is completely wireless. The source code for this project is all open-source and readily available but will need to be modified to increase the sample rate over bluetooth to the 1 KSamples/sec.
*2014 International Conference on Indoor Positioning and Indoor Navigation, 27th-30th October 2014, “Foot-mounted inertial navigation made easy”, John-Olof Nilsson, Amit K Gupta, and Peter Handel
Predicting local renewable energy generation using machine learning (Professor: Arye Nehorai)
Solar power has become a popular source of renewable energy for both commercial and presidential use. Unlike weather, solar power generation varies significantly across different regions, due to shade, surroundings, etc. Therefore, it is important to help solar power users accurately predict local generation amount. In this project, students will build real weather stations in multiple locations to measure local weather conditions (e.g., temperature, humidity, wind speed) and solar power generation, and predict local solar power generation based on weather forecasts. Machine learning techniques will be utilized to implement the prediction model, and measured data will be used to validate the designed prediction method.
Optimal and Distributed Demand Response Strategy Under Duopoly Competition (Professor: Arye Nehorai)
In a duopoly market, electricity firms concern not only about minimizing the cost generated from the fluctuations of market demand, but also about optimizing their own payoffs under competition. In this project, we will investigate the decisions of choices of loads and production quantities made by firms under different assumptions of the market. We will also examine how the pricing schemes change under various situations. Possible models include Cournot’s model, Stackelberg’s model, and Bayesian game model with incomplete information. Students will learn both concepts and applications of game theory, and convex optimization methods. Knowledge of microeconomics is not required but can be helpful.
Face Recognition Algorithms (Professor: Arye Nehorai)
Face recognition can be a very basic tool for smart phone applications and also home security system. In this project, we will investigate the face recognition problem with massive data of human face images. Robust face recognition methods will be built based on classification techniques. We will also focus on the sparsity and low-rank nature of the face images to improve the recognition accuracy. Several algorithms will be implemented in Matlab and results will presented in a technical paper and a presentation at the Undergraduate Research Symposium.
Car Security System: Detecting an intruder via a Smartphone (Professor: Arye Nehorai and Ravin Kodikara)
The objective of the project is to design a security monitoring system for vehicles which can be controlled and monitored via a Smartphone. Two or more small video cameras with night vision capability will be places inside a vehicle. The user will use an application on the phone to activate the cameras from a distance and to monitor the surrounding before approaching the vehicle. A suggestion is to use Bluetooth signals (range enhanced) for the communication between detectors and the phone.
Identification and Modeling of Rhythms in Neural Recordings (Professor: ShiNung Ching)
Brain activity often exhibits interesting dynamical patterns. We are interested in characterizing and modeling one such example of these dynamics: rhythms and oscillations. Our goal is to better understand their connection to neurological disease and basic brain function. We have several opportunities for using signal processing and dynamical systems theory to study rhythms and oscillations in recordings of human brain activity.
Development of Portable Visualization Suite for Low Cost EEG Systems (Professor: ShiNung Ching)
Electrical activity in the brain is most commonly measured noninvasively using Electroencephalography (EEG). Low cost EEG systems have recently entered the marketplace making accessible consumer-level recording of neural activity. We are interesting in developing a visualization suite for portable devices that will provide real-time information from a low-cost EEG system.
Machine Learning Basics with Applications to Email Spam Detection (Professor: Arye Nehorai)
Machine learning, a branch of artificial intelligence, concerns the construction and study of systems that can learn from data. For example, a machine learning system could be trained on email messages to learn to distinguish between spam and non-spam messages. After learning, it can then be used to classify new email messages into spam and non-spam folders. In this project, students will be exposed to several important aspects of machine learning techniques and popular methods. This project is organized in the form of (i) discussion seminars and (ii) applications. The discussion seminars will cover topics such as dimensionality reduction: feature extraction and feature selection; supervised classification: nearest mean classifier/k-nearest neighbor, logistic regression, binary decision tree, support vector machine, naïve Bayesian; performance evaluation; and unsupervised classification. Several sets of homework related to the classification methods will be assigned to the students. The students will then implement an email spam detector using the machine learning techniques as the "application."
Sensor Signal Processing on Mobile Robots (Professor: Arye Nehorai)
We are interested in applying statistical signal processing algorithms for adaptive source detection using an array of sensors mounted on a mobile robot with communication capability for transmitting and receiving data to and from a host computer. The ultimate goal is find the location of the source.
Adaptive source detection is the simultaneous process of sensing the environment (actively or passively) and modifying the sensing device in order to improve the performance of the target position estimator given a certain criteria. We would like to study scenarios based on stationary targets as well as moving targets with and without interference.
In particular, we would like to implement sensor arrays on a robotic device, such as the Lego Mindstorm. This robotic device is simple to work with and it can be controlled remotely via USB or Bluetooth. Moreover, it has several sensors that could be used for robot navigation (i.e. path planning) through the particular scenario. We would use commercially available USB data acquisition devices connected via a battery powered wireless USB hub. However, in the first stage, it is sufficient to use a wired USB hub or a direct connection to any computer USB port. Finally, a unit for signal conditioning should be considered based on the sensing modality.
Among the passive sensing modalities, we would like to use acoustic, magnetic, and thermal sensors. Also, we would like to apply source detection algorithms using forward models and measurement models based on simple scenarios for each sensing modality, respectively. This approach will allow for comparing the experimental results with theoretical results in each case.
For example, one performance criterion to be considered could be the Cramer-Rao lower bound (CRLB) on the mean-squared error of an unbiased estimator of target position. The CRLB, given a particular scenario and sensing modality, depends typically on the number of sensors, number of samples, array geometry and signal-to-noise ratio. We could modify the sensing device such that the CRLB of an unbiased target position estimator is lowered. This can be done by changing the array geometry configuration, changing its position with respect to the target, and changing the transmitting waveform in the case of active sensing devices, such as radar. Another performance criterion to be considered could be the probability of target detection which, based on the scenario, might depends on same parameters as the CRLB. For example, for energy constrained scenarios we could keep one sensor active until the presence of a source is detected, then, the additional sensors can be activated up to guarantee the desired performance.
Note: Sensing systems which measure energy that is naturally available, for example, generated by the target of interest, are called passive sensors. Sensing systems which provide their own energy source for target illumination, are called active sensors. Active sensors emit energy which is directed toward the target to be investigated. The energy reflected from the target is then measured by the sensor.