My trip to China and what I learned at the Universities: DSP Theory or DSP Design?

Gene:

               Nice blog.  I have four comments:

1. I too noticed the heavy use to DSP’ors that the Chinese researchers reported to us.  

       However, could the sample set be biased?  After all they were called to a min-conference

       by a producer of DSP’ors.  If our sample set is biased, maybe most China Universities are more like US schools.

2. Whether the samples are biased or not, the challenge is to get the right balance between theory

       and implementation.  I think Wayne Padgett and Dave Anderson are on track with their fixed-point stuff.  

       I also like your idea of a 16-bit floating-point.  Can you really get the speed of fixed-point without all the pain?

3. If you are really serious about learning Chinese, check out:  http://chinesepod.com/

       You can subscribe to it via iTunes.  Just a few minutes a day and you’ll be recognizing common phrases.

4. I know 6 young ladies that might not like your labeling Susannah as my favorite. :->

It is really very simple. Here is a new perspective on China...

1. Most distinguished degree in China is engineering - even for woman. Most CEOs in high-tech industries in China are engineers by education. They drive the need for application developments towards universities.

2. The Chinese education system, just like the way you learn Chinese, is one of memorizing and absorbing as much as you can very fast and then repeating it a lot. Not one of creating new ideas.

3. The mindset of Chinese is very pragmatic. If we want to improve, let first make something based on whatever is existing and build from that. The Chinese market is so vast and of very different income levels that you can always sell cheaper products to someone, somewhere.

4. As Chinese students have been traumatized for 16 years (from age of  2 onwards) to study 90 hours per week to become the best students in order to enter the best university, after they enter the university, the last thing on their mind is to force themselves to study more stuff. So let's make something real.

5. Professors in the university get not so well paid as in US. But making projects that can be used in industry can earn a professor some nice extra cash, or shares in a start-up or at the minimum a reputation as it provides visibity when put in the market.

6. The Chinese government is heavily promoting security and communication field, and so universities are getting large budgets to do basic R&D on DSP projects, or at least the D when they can get R from elsewhere (such as students returning from US).

7. DSPs are everywhere in China. Ti has really done a wonderfull job of making it pervasive technologies in every corner of China. I know of at least 100 universities working daily on DSPs.

8. The quantity of DSP engineers is so vast, and they are mostly connected (in China called Guanxi), so if we all work on a little piece of the theoretical puzzle, we'll get it done quickly and we'll all benefit.

9. Chinese are very top-down thinkers. Let's make money first, then we can afford to invest in growth by theory, imagination and passion.

10. Chinese students will prefer staying in China because there are more opportunities here than in US these days, especially now with bad economy China has gained lots of confidence and the American Dream has become a Chinese Dream for many. This said, many parents, who are paying for this eduction have not been informed of that midset change, so they will still send their kids to US - whether they like it or not!

So in my view, theory will gain importance again once China has the luxury to do so, and that is when the Chinese market is gettng saturated in some domestic industries, or the low value solution is no longer in demand. An example is digital video recorders, where exports are getting the best way to grow the supplier's market. That is the right moment to go back to the theory and improve the designs.

Finally, one personal advice. Don't learn Chinese (beyond the polite 1 month course) unless you want to come and live in China. By the time westerners master it, most Chinese DSP engineers will speak better English than we would have been able to learn Chinese in that timeframe. Thousands of English native speakers are doing quite well in China by teaching 2 year olds to learn english on Saturday and Sundays. Just a thought...

Hi, Pascal Coppens

Nice analyses! you seem to know China very well.

I am Chinese. I think the teaching way in US Universities is better. The Universities in china just teach the students how to use the DSPors, but don't teach how to design the DSPors, so there will not a company like TI in China.

Haha, interesting analysis.

Great subject Gene!

Having been both a professor and a developer of DSP technology I have to say "Finally people who get it!". It is clear to me nowadays that most breakthroughs come not from new theory, or new clever implementations, but from the smart co-design of principle and implementation. It takes people who understand how to present the problem and formulate the algorithms to take the best advantage of efficient hardware architectures. This is what leads to new technologies.

It has always seemed strange to me that at the same time where mechanical engineers established the concept of "mechatronics" Electrical engineers were busy specializing and compartmentalizing the sub-fields of their trade, and the teaching of these sub-fields further.

It does not help that (in North-America anyway - I do not know elsewhere) society generally attaches great value to extreme specialization. So engineers who do want to get recognized in their field are pressured to specialize. In doing so, they miss opportunities to contribute great new ideas.

I will even go one step further and say that high-performance analog design should also be a part of any DSP course. A design engineer who understands well al the facets of a particular design challenge will be best equipped to optimize the compromise between analog, high-speed design, Signal processing and even software design, and hopefully reorganize the problem to yield a novel, efficient solution. Now would such an engineer if he/she exists be able to break the barriers of conventional wisdom. This is another challenge!

Gene, Thank you for a great eye-opener!

An interesting view; and written to trigger reactions. So here are a few:

1. Can it be that your sample of US professors (three) is too small to conclude that US schools teach DSP vs. DSPs (processors)? In Canada, many of us teach both (and frequently with TI parts thanks to TI’s university program ;-)).  I personally have gone from TMS320C30-based designs in the nineties to TMS320C5402-based designs a few years back, to TMS320C6711 and13 currently, and environments such as Matlab are only used for filter coefficients determination and algorithmic simulations prior to actual implementation.

2. My impression of the Chinese higher educational system – based solely on observing some major universities I should mention, but confirmed by many Chinese students and colleagues - is that its quality is driven by probability rather than thoroughly planed  pedagogical policy, and that its apparent current success is mostly due to Gaussian statistics (Mainland China produces several million engineers par year!). Still, there appears to be a slight upward trend to achieve excellence, as in many other Asian countries for that matter.

3. Finally, I share your prospect that our children may end up going to China for quality education in the not distant future. As before in human history, it won’t be as much due to the new leader being better as to the previous one having gone to sleep. In my view, North-America’s ongoing fascination with immediate material gain is doing just that: preparing itself for a long slumber to come. On the positive side of things, eating Dim Sum and stir-fried veggies is not so bad, and we’ll get to use a language that does not conjugate verbs, a relief from the arcane rules of western languages! :-)

I studied Electronic Engineering at the University of Southampton, UK. In that course EVERYONE had to learn how to program microprocessors (Motorola 68000) using machine code (assembler) and BCPL. We were also taught the basics of software design.

Regarding Digital Signal Processing, we were really only taught the concepts (time and frequency domains, Fourier transform, etc) and the basic building blocks, such as FFTs, convolution, auto and cross correlation, but not how to design a program to perform DSPing on a DSPor, or any advice on writing embedded applications. This I have learnt since graduation. I even did a Third Year Project in signal processing but never really programmed a DSPor, instead I used a tool to do the DSPor work that imported and exported files, which I then processed using my own C++ PC applications. The emphasis was still on the DSPing and not the DSPor.

That was 15 years ago and I hope that things have changed and students program DSPors in the same way I programmed microprocessors when I was a student.