Lapas attēli
PDF
ePub

other languages, especially Japanese and English. There are projects on handwritten address recognition for the post office and form filling for numerous organizations. There is also machine translation research.

engaged in training as well as pilot (2) Networking, both local area and
production. In 1991 more than 8,500 wide area networks (LAN and
companies obtained some kind oftech- WAN), both to take advantage of
nical service from ITRI and almost the trend toward integrated ser-
40,000 people attended various semi- vices and data network (ISDN) and
nars, workshops, or training courses. for more heterogeneous comput-
The staff generated about 350 patent ing environments that are seen to
applications, half outside of Taiwan. be common in the 1990s.
R&D activity is in

(3) Consumer electronics, including • Electronics

improvements to analog devices • Computers & Communications

using digital technology, high def• Opto-electronics

inition TV (HDTV), and related • Measurement & Standards

audio-visual products. • Materials • Chemicals

CCL has almost 1,000 staff mem• Energy & Resources

bers, including 40 with Ph.D. degrees, • Machinery

440 with M.S. degrees, and 300 with • Aerospace

B.S. degrees. The 1991 total budget • Industrial Safety & Health was about $61M. Some specific proj• Industrial Pollution Control ects that I saw include the following.

• I was shown several demos of multi

media applications. These mostly involved building easy-to-use tools on top of commercial (U.S.) multimedia software.

The aerospace activity is new, begun in
1990 when ROC launched the National
Initiative for Aerospace Industry
Development.

The Computer & Communications Research Laboratories (CCL), where I spent most of my time, run by my host Mr. Wang, is also new, although it was spun off from an older Electronics Research and Services Organization whose job was to foster the IC and PC industries in Taiwan. (Acer is Taiwan's largest PC maker and offers a range of clone products including various high end machines.) Current IC production is just beginning to include the technology to make central processing units (CPUs). CCL and some local private firms are negotiating with MIPS for manufacturing MIPS's R4000 CPUs in Taiwan. There are also discussions between Hewlett-Packard (HP) and several local manufacturers about possible cooperation. CCL's focus is in three general areas.

• A 40-MHz Sparc VME-based work

station, based on a RISC CPU and
client/server architecture. A chip-
set for peripheral devices was also
designed. A three-dimensional (3D)
color graphics module, an interac-
tive 3D graphics interface, etc. Quite
a bit of this chip design was com-
pleted but it does not seem as if the
actual packaging was completed. In
any case the experience is described
as the design of a super-mini. Based
on this CCL is working on a file
server and also, I was told, a multi-
processor and vector processor. I
am somewhat skeptical about prog-
ress on these later projects, how-
ever, as my discussions with Wang
did not suggest much interest in
really high performance systems.
There is substantial work on vari-
ous open systems software tools.

• There are projects to develop sev

eral new products, including at least two notebook PCs, a hand-held (palmtop) PC, an X-terminal, IC card, and a 900-MHz cordless radio frequency (RF) module. (Some of these may have already been completed.) These involve consortia with local companies. Given the state of current technology in these areas there is no reason to think that the projects will not be successful. A few of these projects involve liquid crystal display (LCD) technology, but I had no opportunity to learn about progress on this topic. One of the most interesting and practical of technology dissemination projects was an automated ticketing system for the Taipei railroad system. This has been installed and is in daily operation.

• There is a collection of research

projects related to communications, including ISDN applications such as a PC S interface card, a digital telephone for simultaneous voice and data transmission, and ISDN software. Plans are to work on a G4 fax, video phone, etc. There is a great deal of work on digital video processing associated with a 5-year program to develop key HDTV technologies by 1996, i.e., digital video processing and recording, image acquisition, and LCD projection. Prototype products include

(1) Workstation and peripherals,

multimedia systems, and system software.

• Research on Chinese character

processing has been in progress since
1985 and includes printed Chinese
character recognition, on-line char-
acter recognition, etc. CCL wants
to expand its recognition work to

an HD-camcorder and an HD-video Systems Workshop that was to be held upgrade, with NCHC staff being allowed cassette recorder. The HDTV proj- in the capital the day after my visit.

in the capital the day after my visit to make quarterly visits to the factory ects are intended to revive what is Nevertheless, CCL seemed like an inter- to check that progress on this machine seen as a declining consumer prod- esting place to work. Concerning coop- is on schedule. Cray and NEC also ucts industry, as well as to promote eration with Western laboratories, Wang made proposals but these did not qualdefense, medical, and other pointed out to me that because much ify. Chang told me that Cray's proposal Taiwanese industries.

of their work is tied to improving indus- missed because the company made some

trial activities, some projects might be errors in the paperwork they submitted. My general sense of work at CCL off limits, but given that this could be NEC's four-processor SX-3 did not run was that projects are not so different handled, I think that there would be NCHC's seven benchmark programs from those at some other Asian indus- opportunities for Western researchers. as well as expected after tuning, and trial research institutes. The research On the other hand, young CCL staff their upgrade, an SX-4, was only to be is applied and the emphasis is on excel- could provide a welcome additional available in 1995. In fact, NCHC has lent liaison with industry rather than pair of hands in Western laboratories repeatedly told me that having a strong excellent academic output. With so much as medium-term visitors.

Taiwan-based support team was an emphasis on industry it is not surpris

important part of their requirements, ing that staff turnover is high, with NATIONAL CENTER FOR and IBM already has a very significant young scientists and engineers moving HIGH PERFORMANCE

presence there. from CCL to industrial organizations COMPUTING (NCHC)

While I was visiting NCHC's Taipei with which they already have relation

office I was introduced to Dr. Jean H. Su ships. ITRI management is aware of I have described plans for NCHC in [E-mail: ajin@nchc.ntu.edu.tw). Su is the issue and is trying to address it by earlier reports and so will not repeat a former student of Prof. Steven Pruess, making the work and environment highly those here. My primary reason for a a well known numerical analyst at the attractive. There is a great deal of enthu- repeat visit was to see how the con- University of New Mexico. During my siasm and energy about working on struction of the center was progressing travels in Asia I have met relatively few interesting prototype projects by what and to learn if a decision had been scientists with formal training in numeriappeared to me to be a relatively young made about a vendor. NCHC is being cal mathematics and application to high staff. CCL seems to be engaged in exactly built just adjacent to the Hsinchu Sci- performance computation. Su's presthe kind of R&D that its charter and ence Park (see above), but surprisingly,

ence Park (see above), but surprisingly, ence at NCHC reaffirms my sense that mission specify. The research activities my park administration hosts didn't this organization is going to focus on are not too “far out,” so it is possible know about it initially, and even later applications and provide professional for the staff to sense where they are could not locate the construction activ- support to their user community. Also going. I did not see (or was not shown) ity amidst a great deal of other building while I was there I observed a course any really unusual directions. As in projects except to say that it was

projects except to say that it was being taught by an American molecuother laboratories in this part of the "somewhere over there." However, I lar biologist on the use of canned softworld, there is a heavy emphasis on was assured by Prof. San-Cheng Chang, ware for molecular modelling. This was building things and less emphasis on the NCHC Director, that construction tied in with a new visualization laborasoftware. Equipment is good, but not is mostly on schedule and that a machine tory that had several SGI workstations as good as at first class research insti- will be installed later this year. and had just been opened. tutes in Japan or the United States; At the moment the sole acceptable there were many PCs and fewer work- vendor is IBM, who proposed a four- INTELLIGENT SIGNAL stations than I expected. I did not see processor ES9000/A20 high end PROCESSING AND anywork in algorithms or other aspects machine, each processor equipped with COMMUNICATION SYSTEMS of computer science; work of this kind a vector unit. Permission has to be WORKSHOP (ISPACS'92) would be more likely to occur at a obtained from the Taiwanese university or the Academia Sinica. CCL Government to allow a sole source This 1992 IEEE International did not appear to be much interested in

procurement and then NCHCstaff will Workshop, held from 19-21 March 1922 supercomputing either. Also, surpris- begin negotiations about details. The in Taipei, brought together about 150 ingly, my host was unaware of the Sig- plan is to upgrade this in 1994, and people, including 35 from outside of nal Processing and Communication IBM has proposed an SS1 for the Taiwan (half from Japan) and 45 stu

dents. The meeting organizers were

Prof. Lin-shan Lee
Prof. David G. Messerschmitt

Nishitani emphasized that the Prof. of Electrical Engineering Department of Electrical

community should concentrate not on Department of Computer Science Engineering & Computer Science abstraction but rather on building and Information Engineering University of California

domain-specific digital signal processors National Taiwan University

517 Cory Hall

and that this would enable DSPs to be Taipei, Taiwan, ROC

Berkeley, CA 94720

developed that were small, required Tel: +886-2-351-0231 x3212/3232 Tel: (415) 642-1090

low power, and were low cost. (The Fax: (415) 642-2739

speaker is a recognized expert in the and

E-mail: messer@janus.berkeley.edu field of programmable DSP and was

responsible for the very early developDr. Tomonori Aoyama and

ment of a single chip DSP for NEC.) Executive Manager, Transport

Build it, try it, and learn from it might Processing Laboratory Dr. Takao Nishitani

be his motto. Nishitani repeatedly NTT Transmission Systems

Senior Manager, Terminal Systems described the possibilities for realizLaboratories

Research Laboratory

able hardware such as a domain spe1-2356, Take

C&C Systems Research Laboratories cific video signal processor that can Yokosuka, Kanagawa 238-03, Japan NEC Corporation

realize motion compensation in conTel: +81-468-59-3020

1-1, Miyazaka 4-chome

trast to a mesh-parallel computer that Fax: +81-468-59-3014

Miyamae-ku, Kawasaki

he felt was too general and hence both Kanagawa 216, Japan

too expensive and not powerful enough. The motivation for this workshop Tel: +81-44-856-2118

Other examples included real-time was that the organizers (mostly coming Fax: +81-44-856-2232

computer graphics, high-resolution from the hardware side of signal pro- E-mail: takao@tsl.cl.nec.co.jp video, 3D video, a 128-module video cessing) felt that more computer sci

signal processor for HDTV, and finally ence ideas need to be injected into the The titles of their talks speak volumes: a video compression chipset that is field of signal processing and that bring- Messerschmitt, “Telecommunications

Messerschmitt, “Telecommunications available now, implementable because ing members of both communities and Signal Processing: The Evolving it was specific rather than general. together would be a small step in that Relationship"; Nishitani, “Hardware

Relationship”; Nishitani, “Hardware Nishitani is also very interested in picdirection. Nevertheless, most of the Approaches to Signal Processing on ture coding algorithms using wavelet participants were still heavily focused Communication Systems."

transforms but gave no specific details on hardware, so it was clear that a great Messerschmitt emphasized that of his work in this area. deal more needs to be done to move telecommunications, signal processing, To me, these papers were perfect these groups together.

and computing are merging, even though metaphors for the different approaches Typically, this field has been heavily he saw a polarization into wireless and to research as practiced in Japan and in oriented toward digital signal process- optical transmission technologies with the United States. ing (DSP) applications and especially almost opposite bandwidth and reli- In addition to invited and presented techniques to increase the information ability problems. He felt that in the talks, there was a full afternoon session content of signals by increasing the bit future a proliferation of services with devoted to posters. An English lanrate by using new physical medium, smaller interest groups will mean that guage Proceedings contains all the reducing the bits needed to be trans- abstraction, complexity management, presented papers, not including the mitted by compressing techniques, system design, and structured design keynotes, unfortunately. There were improvement of the output signal by methodologies will become increasingly several papers on the “hot” topic of echo cancellation, adaptive filters, sig- important. He concluded that he was digital communication. These centered nal transmission/equalization and finally anxious to learn more about object on methods to code speech signals. speech, text, and character recognition/ oriented programming, which he felt Traditionally speech coders (the consynthesis, etc. (Signals can be speech, will be a key to future developments. version of speech signals into digital data, or video.) More interesting to me Everyone agreed that his presentation data streams) have been judged on three were the very different world views was brilliant and exactly what was called major performance criteria: speech expressed by the two keynote speakers, for by the meeting organizers. quality, rate (number of bits per second necessary to transmit after compres- there are many algorithms that attempt "Application of High-Speed Analog sion), and complexity (equivalent to to do this. Jehng has studied one par

Neural Networks to Optical cost). In efforts to bring coding rate ticular method, the three-step hierarchi- Communication Systems" down to the required international cal search algorithm (3HSA), which is Takao Matsumoto, Masafumi Koga, standard of 8 kbit/s (Consultative very efficient in terms of operation count Hiroshi Miyao, Yoshihito Committee on International Telephony but is difficult to implement in hard- Amemiya and Telegraphy (CCITT) standard), ware. In his paper he considers a tech- NTT Transmission Systems coding algorithms become more com- nique due to H.T. Kung for a systolic Laboratories plex and hence take more time. Conse- array implementation of the algorithm. 1-2356, Take quently a fourth criterion, communi- It is a nice piece of work but needs Yokosuka, Kanagawa 238-03, Japan cations delay, has become increasingly hardware implementation and testing Tel: +81-468-59-3160 important for speech encoding. In a to determine if it will be practical.

Fax: +81-462-40-2162 complex network, the delays of many encoders add together, transforming "Determination of IFS Codes Using The main idea is to develop a neural the delay into a significant impairment Scale-Space Correlation net, optical WDM (wavelength demuleven after echo cancellation has been

Functions"

tiplexer). Typical optical WDM are performed. The international CCITT Masayuki Kawamata, Hiroaki difficult to fabricate and are not flexistandard specifies an algorithmic delay Kanbara, Tatsuo Higuchi ble. A single-mode transmission fiber lower than 16 ms (with an objective of Department of Electronic

allows multichannel signals to propa5 ms) and a total coder-decoder delay Engineering

gate through a multimode waveguide of 32 ms (with an objective of 10 ms). Tohoku University

and couple to a detector array. When The CCITT 16-kbit/s standard speci- Aramaki, Aoba, Sendai 980, Japan the optical signals are coherent the fies a total delay lower than 2 ms. In E-mail: kambara@higuchi.ecei. result is an inhomogeneous optical mobile telephony, there are related

tohoku.ac.jp

power distribution (speckle pattern) at objectives to get a frequency use effi

the output caused by the interference ciency in the range of 2-3 bps/Hz. IFS (iterated function systems) have among propagating modes. If this is I found three other papers particu- been very heavily studied in the context

been very heavily studied in the context input to a neural net (through a deteclarly interesting.

of image compression (M. Barnsley is tor array) it might be possible to sepa

one of its earliest and most active pro- rate the signals as the output of the net. “Realization of Array Architectures ponents, and of course work by The advantages of this approach are for Hierarchical Block Matching B. Mandelbrot was seminal) for natural

B. Mandelbrot was seminal) for natural that the optical components are easy to Algorithms”

images. Kanbara's work involves the build and the demultiplexing characYeu-Shen Jehng, Liang-Gee Chen, systemic analysis of fractal images and teristics can be changed by changing

Tzi-Dar Chiueh, Thou-Ho Chen the development of an algorithm to the neural net parameters, i.e., without Room 416

determine the IFS codes using digital changing the optical components. Department of Electrical

signal processing techniques. What Matsumoto says it better, “In a fabricaEngineering

makes this approach interesting is that tion sense, the burden which is imposed National Taiwan University while many clever people have been on the optics side in a conventional Taipei, Taiwan 10764, ROC working on the topic of IFS, Kanbara's WDM demultiplexer is moved to the

group is unique in their use of DSP electronics side.” His paper describes The basic problem being looked at thinking. In particular he proves an the experiments and an analog learnin this paper (Jehng's thesis) is how to important theorem relating the maxi- ing method (called multifrequency extract information from a two- mum value (with respect to a param- oscillation (MFO)). The main probdimensional sequence of images in order eter) of a one-parameter spatial corre- lem with the approach is that more to predict movement. For digital trans- lation function to the IFS codes (when than 1K synapses/chip and a speed of mission of image data, it is much more applied to binary images). Finally, his over 16 input patterns per second are efficient to transmit "motion vectors” paper provides an algorithm implement- beyond the ability of current large scale rather than the primitive images, and ing how the theorem can be applied. integration (LSI) fabrication technol

ogy. Nevertheless, the technique is a very unusual one.

« iepriekšējāTurpināt »