Systems Laboratory

New model-based approaches to halftoning are being developed. They use well-known models of visual perception along with models of printing that we have developed. One approach minimizes the mean-squared error between the perceived intensity of the continuous-tone image and the perceived intensity of the printed halftoned image. Another is an adaptation of the well-known error diffusion method to include the printer model. Traditional approaches, for example, ordered clustered dither, obtain robustness to printer distortions, such as ink spreading, at the expense of spatial resolution and the visibility of graininess. In contrast, our new methods exploit the printer distortions to produce higher quality images than would be obtained with RperfectS printers. Improvements due to model-based halftoning are expected to reduce the resolution requirements for laser printers used in high-quality printing (e.g., 400 dots/inch instead of 600). Model-based halftoning can be especially useful in transmission of high-quality documents using high-fidelity, gray-scale image encoders. In such cases, halftoning is performed at the receiver, just before printing. Apart from coding efficiency, this approach permits the halftoner to be tuned to the individual printer, whose characteristics may vary considerably from those of other printers, for example, write-black vs. write-white laser printers.



Professor: D. L. Neuhoff

Gracuate Students: M. Horowitz, M. Slyz

Image coding is the process of creating binary image representations with the dual goals of efficiency (as few bits as possible in the representation) and accuracy (the reproduced images shall be as similar as possible to the original). Two approaches are being pursued. The first involves the use of a detailed model of the intermediate level human visual sensors to construct transform codes that hide quantization noise. The second involves the design of lossless image codes based on adaptive prediction, with new kinds of predictors and adaptation strategies. These lossless image codes are intended for applications, such as medical imaging, where an exact reproduction of the image is required. On the other hand, the first project is intended for more everyday applications where exact reproduction is not necessary, but good quality and high efficiency are needed.



Professor: W. E. Stark

Graduate Student: S. Lee

A very noisy channel is a channel whose capacity is close to zero. Very noisy channels (VNCs) can be used to model many physical channels operating at low signal-to-noise ratios. More importantly, a large class of physical channels, operating at arbitrary signal-to-noise ratios, can be modeled as repeated uses of a VNC. In particular, this is true for the infinite bandwidth additive white Gaussian noise channel and the direct detection optical Poisson channel. The error exponent indicates the best achievable performance of any block codes used over a communications channel. A code which achieves this best performance is said to be exponentially optimum. For most channels, the error exponent is not known and can only be bounded. In this research, the error exponent is computed exactly for a large class of VNCs, and exponentially optimum codes are explicitly constructed for these channels. These ideas are applied to derive both the error exponent and exponentially optimum codes for the direct detection, polarization-switched, optical channel.



Professor: K. A. Winick

Distance Bounds for Runlength-Constrained Codes

Graduate Student: S-H. Yang

One of the most basic problems in coding theory is to find the largest code of a given length and minimum distance. There are several known upper and lower bounds when the codewords are unconstrained. In many digital transmission and recording systems, considerations such as spectral shaping, self-clocking, and reduction of intersymbol interference require that the recorded sequences satisfy special run-length constraints. In this research distance bounds and the construction of runlength-constrained error-correcting codes are investigated. Upper bounds are derived for the minimum achievable distance of runlength-constrained sequences, and lower bounds are also derived which include cost constraints.



Professor: K. A. Winick

Graduate Student: S-H. Yang

Runlength-Constrained Write-Once Memories

Sponsors: Office of Naval Research; Office of Naval Technology

A write-once memory (WOM) is a storage medium where the value in each bit location can only be changed from the virgin 0-state to the permanent 1-state irreversibly. Data can be recorded by marking blank (i.e., 0-state) bits. Those marked locations are stuck in the 1-state and hence limit to some degree further use of the memory. Examples of WOMs in the electronic and computer industry are punch cards, paper tapes, PROMs and optical disks. Current laser-optics technology produces the "write-once" CD-ROMs that are especially sutiable for storing archival data. Usually this data must be periodically updated after it has been initially recorded. If we can re-use the write-once disk by implementing an efficient coding scheme, then the expense of replacing the whole disk may be saved. In this research, the ultimate capacity of runlength-constrained write-once memories is investigated using techniques from information theory.



Professor: K. A. Winick

Corrugated Waveguide Filters

Graduate Students: C. Brooks and G. Vossler

Sponsor: National Science Foundation

Corrugated thin film waveguides play a major role in lightwave devices. Applications include distributed feedback lasing, bistable switching, phase matching in nonlinear materials, pulse compression, grating coupling, and optical filtering. In many of these applications, the corrugation is periodic. In an aperiodically corrugated thin film waveguide, however, the frequency dependent coupling between waveguide modes can be used to produce a filter which has a specified spectral response. Inverse scattering techniques have been developed for designing such filters, and efforts are currently underway to fabricate these devices. Several new fabrication techniques are being pursued. These include an optical direct write method, based on photobleaching gamma ray-induced defect centers in ion-exchangeable glasses, and a Rbent waveguideS approach. The first filter to be demonstrated will compensate for dispersion-induced pulse spreading in optical fibers.



Professor: K. A. Winick

Rare Earth-Doped Waveguide Lasers

Graduate Students: G. Vossler and C. Brooks

Sponsors: National Science Foundation;
NSF Center for Ultrafast Optical Science;
Smith Industries; IMRA America, Inc.

Recently, the development of rare earth-doped fiber lasers has received considerable attention. These fiber lasers exhibit a host of desirable properties. First, they permit wide tuning ranges and short pulse generation because of their broad emission lines. Second the pump powers required for lasing are low, since the pump beam is strongly confined to a small volume. Finally, rare earth-doped lasers offer better frequency stability, longer lifetimes, and less temperature sensitivity than semiconductor devices. These traits make them promising devices for telecommunications, sensing, and spectroscopic applications. Glass waveguide lasers on planar substrates are a natural extension of the fiber technology. As opposed to a fiber, it should be possible to integrate monolithically multiple components onto a single glass substrate. These components could include distributed feedback laser mirrors, grating couplers, mode-lockers, and nonlinear elements. We have fabricated neodymium-doped, channel, waveguide lasers in special glass melts and have demonstrated the first glass integrated optic distributed Bragg reflector laser. Efforts are currently under way to passively mode-lock these lasers and to extend theses results to rare earth-doped lithium niobate hosts. Novel sensors, based on this technology, are also under development.