We developed an algorithm for real-time classification of mental states such as for the detection of engagement, mental workload, and mental fatigue in pilots or vehicle operators. The algorithm uses kernel partial-least squares (KPLS) to decompose multi-sensor EEG spectra into a small set of components. In this way, the algorithm can process practically unlimited input channels and spectral resolutions. No a priori information about the spatial or spectral distributions of the sources is required. We tested the algorithm with EEG data from Air Force pilots who performed simulated flight maneuvers over a 37-hour period. Over a range of six distinct signal-to-noise ratios ranging from -10 dB to +10 dB, classifier performance increased smoothly, with test proportions correct ranging from 84% to 94% (Wallerius, J., Trejo, L. J, Matthew, R., Rosipal,R., and Caldwell, J. A., 2005).

Wireless telephones are increasingly popular, but with that popularity there are increased problems. There are not enough radio frequencies for everyone to have a clear connection to the base station. So when many users are on the network at the same time, their calls can interfere with each other. When this happens, the quality of a call may be reduced, or even worse, the call be may lost. To reduce the annoying problem of interference and poor call quality, we are working the RadioCosm Inc., to develop mathematical models for wireless telephone networks. We are also developing powerful algorithms to tune these models, so that existing networks can maximize the quality of service they provide to the customer. Once the model is tuned, the parameters of the model can be applied to an existing network. Thus, without installing any new hardware, an existing network can handle increased call volumes while retaining a high level of call quality.

Severely disabled patients with diseases such as amyotrophic lateral sclerosis (ALS) may still be conscious and aware of their surroundings, but they may be unable to speak or even move a muscle. To help such patients communicate, we worked with the Bio-Logics and the University of Illinois on a brain-computer interface project for the severely disabled. Brain electrical signals for interface control are very weak and noisy. We improved the quality of the signal measurement and interpretation analyzed the signals generated by developing a custom wavelet preprocessor, which greatly improved the quality of the brain signals and the overall performance of the interface. A paper containing our method was published in the IEEE transactions on Rehabilitation Engineering: Donchin, E., Spencer, K. M. , and Wijesinghe, R. (2000). The Mental Prosthesis: Assessing The Speed Of A P300-Based Brain-Computer Interface, IEEE Transactions on Rehabilitation Engineering, 8, 174-179.

Awarded five US Patents on algorithm development work in the field of communications signal processing. All of the patents relate to an algorithm for extracting confidence metric information from maximum likelihood sequence estimation (MLSE) equalizer and combining confidence metric information from multiple base-stations to improve signal quality. This procedure dramatically improves system quality by minimizing the negative effect of shadow fading and multipath fading. Invented, developed and simulated the basic algorithms, and implemented them in real time DSP software in a fielded system.

Initiated and led a project which culminated in the deployment of an advanced acoustic echo cancellation system as part of a high-end commercial video conferencing system. The adaptive acoustic echo canceller is capable of fast convergence on echoes with tails as long as a quarter second. A filter bank is used to separate the signal into multiple sub-bands, which are each adapted independently. This approach minimizes computational load and makes fast convergence possible.