From: Tim Wescott on
On 07/08/2010 02:26 PM, Nasser M. Abbasi wrote:
> On 7/8/2010 2:07 PM, HardySpicer wrote:
>
>> ahhh diddums...you should try some advanced control engineering and
>> see how you get on.
>> No sympathy.
>>
>> Hardy
>
> But DSP and control in a way are interrelated?
>
> A filter is just a system. IIR has feedback. Feedback is used in DSP.
> Using Costas loop (phase-locked loop) in demodulation sues feedback loop
> to detect carrier frequency, and I am sure there many other examples.
>
> Matlab uses state space approach in converting analog filter to digital
> filter. Modern control theory is all state space.
>
> For me, control/ linear system theory/ signal processing are all very
> much interrelated. Advanced control theory goes a little more crazy with
> advanced math and matrix theory than DSP, but at the end of the day, it
> is all just a system, with input/output and feedback and fancy
> disturbances thrown in to make it real.
>
> I love to study control theory also, and I also found it very hard. I
> think control engineers and DSP engineers have the same genetics.

Real control (forget theory) is about attempting to fit some tractable
mathematical model to a plant that is -- at root -- viciously nonlinear
with unknowable dynamics. Many control engineers are so used to doing
this that they don't even consciously do so -- they use integrator
anti-windup because "things won't work if I don't", they use
conservative plant models, etc., -- because at root, it's a nasty, nasty
problem to solve.

What makes control theory hard is all the bits and bobs that have been
attached to it to make it work with the real world, and those are
different from the bits and bobs that have been attached to systems
theory for communications.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html
From: Fred Marshall on
Nasser M. Abbasi wrote:
>
> I find DSP the hardest subject to become good at. Other students like me
> at school also complain how hard the DSP courses are compared to the
> other EE courses and other engineering courses in general.
>
> I think some of the reasons are:
>
> 1. DSP courses Has the most math. (including complex variables).

***Well, if you don't learn the math in school then you will likely be
working as if you had one hand tied behind your back.

> 2. Two domains to worry about, time and frequency. Jumping from one to
> the other can get confusing.

***It should not be so much confusing but dealing with it might be
disorienting - like flying upside down. Use lots of cartoons.

> 3. Two other domains to worry about, continuous time vs. discrete time.

***Just crank up the sample rate and they start looking the same.
Then sprinkle in how sampling perturbs things from there....
Then learn to think about periodic things....

> 4. Many relations between many concepts to get right.

***Think of a tool box. Hammer/nail. Screwdriver/screw. etc.

> 5. One has to also be good in programming.

***Likely more so all the time.

> 6. Demodulation is just hard.

???

Filter design is hard.

Well, I learned analog filter synthesis in undergrad school. Never knew
where those darned transfer functions came from as if they were handed
down by some gurus or something. Eventually I came to learn about
approximation theory - which is where they came from!! Yay. It would
have been so much nicer if the approximation theory stuff had come first
- even if only briefly. Then it wouldn't have been so abstract.

> 7. Need to also be good in probability and statistics to do random
> signals (real life).

***I might say one needs to be good in random signals to apply some bit
of probability and statistics.

> 8. Has to know how to do lab work also. Hard stuff.

***Well, the reports were always a real chore.

> And many more. May be you can add more items to the list.

***It used to be "getting the computer time" to finish my work!!! :-)

***Coupling what you need with what you have to work with - including
the physics.
***Getting to understand the difference between what's possible and what
isn't.
>
> Do many of the DSP experts here also found DSP hard at school? It seems
> only the very smart can become good at DSP.
>
> I think one is either born to do DSP or not. I think it is genetics.
>
> --Nasser
>
>
>

You forced me to remember just *how* I learned it!! It wasn't in school
really. More a blend of grad school interests and on-the-job needs.
So, I sorta eased into it by writing a program very similar to the P-M
program (during the same time frame). Then Tom Stockham gave some
lectures at our workplace and I applied what I learned there to some
real world problems. And a couple of other lecture series. Then I
designed some interesting (analog then digital) tapped delay line
filters. Then I learned about half-band filters and wrote a program to
design them.
School was about linear systems - Laplace transforms and I think Z
transforms were self-taught. But I have never been to drawn into the
math - in my view the math follows the situation. If it's a necessity
then fine - but much is to be understood with cartoons really. I
learned that method in an RCA lecture series on DSP. It's just a
variation on the "graphical convolution" method - which is something
worth understanding.
So, it took a long time and was needs-driven or just interest.

Why should it be hard in school?
- If you don't have fundamental understanding of the physics and the
applications then the math can leave you a bit cold. Not that the math
doesn't help. Somehow I had a very hard time relating the "taught math"
to the real world. I finally "got" some of it in grad school when we
learned what was called "computer applications" which was really about
using computers to solve differential equations - ordinary and partial
(i.e. field problems). Not that this was DSP so much but it's very
closely related.
- Practical application of what you might be asked to learn are
sometimes elusive. So the motivation is lessened perhaps.
- It's a bit like learning to swim. If someone throws you in the deep
end of the pool, you may not like it but you may learn to swim all that
much faster!!

I don't think that any of us are "good at it" at least in my way of
thinking. Each of us are good at some part of it and maybe OK at most
of it. But really being good at something requires that you've done it.
And, real life only gives us a limited number of opportunities to *do*
things. Take a look at some of the very smart folks in comp.dsp who
still ask questions!! So, don't be disheartened by "not being good at
it". Find something to be good at. Then something else. Then
again.... If you're an academic then you may have more time and
motivation. If you're in industry then the motivations are driven by
needs beyond learning.

Fred
From: steveu on
>On 07/08/2010 02:26 PM, Nasser M. Abbasi wrote:
>> On 7/8/2010 2:07 PM, HardySpicer wrote:
>>
>>> ahhh diddums...you should try some advanced control engineering and
>>> see how you get on.
>>> No sympathy.
>>>
>>> Hardy
>>
>> But DSP and control in a way are interrelated?
>>
>> A filter is just a system. IIR has feedback. Feedback is used in DSP.
>> Using Costas loop (phase-locked loop) in demodulation sues feedback
loop
>> to detect carrier frequency, and I am sure there many other examples.
>>
>> Matlab uses state space approach in converting analog filter to digital
>> filter. Modern control theory is all state space.
>>
>> For me, control/ linear system theory/ signal processing are all very
>> much interrelated. Advanced control theory goes a little more crazy
with
>> advanced math and matrix theory than DSP, but at the end of the day, it
>> is all just a system, with input/output and feedback and fancy
>> disturbances thrown in to make it real.
>>
>> I love to study control theory also, and I also found it very hard. I
>> think control engineers and DSP engineers have the same genetics.
>
>Real control (forget theory) is about attempting to fit some tractable
>mathematical model to a plant that is -- at root -- viciously nonlinear
>with unknowable dynamics. Many control engineers are so used to doing
>this that they don't even consciously do so -- they use integrator
>anti-windup because "things won't work if I don't", they use
>conservative plant models, etc., -- because at root, it's a nasty, nasty
>problem to solve.

Isn't a more common case that the plant is pretty well known, and pretty
well linear, but only over a certain range. The tricky stuff is typically
ensuring things don't do wacky if you step outside the well characterised
areas.

>What makes control theory hard is all the bits and bobs that have been
>attached to it to make it work with the real world, and those are
>different from the bits and bobs that have been attached to systems
>theory for communications.

Steve


From: steveu on
Hi,

>I find DSP the hardest subject to become good at. Other students like me
>at school also complain how hard the DSP courses are compared to the
>other EE courses and other engineering courses in general.

I wonder if they've just dumbed down the other courses. Courses are all
supposed to be set at a level of depth where they are pretty demanding for
the target audience.

>I think some of the reasons are:
>
>1. DSP courses Has the most math. (including complex variables).

Huh? Did you skip the EM theory courses?

>2. Two domains to worry about, time and frequency. Jumping from one to
>the other can get confusing.

Isn't that true of a number of courses? Doesn't all the analogue
processing, RF and optical comms stuff have the same requirements?

>3. Two other domains to worry about, continuous time vs. discrete time.

If anything, the discrete stuff is easier to work with than the continuous
time processing in most other non-digital electronics.

>4. Many relations between many concepts to get right.

Any system means handling many concepts to get it right. Maybe in the other
courses you are just partitioning things into smaller packages.

>5. One has to also be good in programming.

In most college courses now, it seems the only practical work is
programming. Sadly, its mostly of the mind rotting Matlab variety for DSP
and everything else.

>6. Demodulation is just hard. Filter design is hard.

Going beyond a core idea to make a successful practical system is always
hard. Little piecemeal projects at college often seem simple. However, in
the real world nothing is easy. Even for relatively simple systems there
are other smart people out there trying hard to be just a little better or
more cost effective than you. People choose a winner. There is a very
limited market for anything which is not world class in some key
parameter.

>7. Need to also be good in probability and statistics to do random
>signals (real life).

The real world is noisy above zero K, which means most electronics, beyond
simple logic design, is a mass of statistics work.

>8. Has to know how to do lab work also. Hard stuff.

You mean you actually need to make stuff work? :-\

>And many more. May be you can add more items to the list.

>Do many of the DSP experts here also found DSP hard at school? It seems
>only the very smart can become good at DSP.

You'll find anything hard if it doesn't interest you. If it does interest
you, then probably you will persevere and not concern yourself about
difficulty too much. Do I find DSP hard? In general terms, no. However,
there are specific areas that perplex me to this day, and I've been doing
DSP since the mid 70's - a time when DSP was one algorithm designer to
100 logic designers trying to implement his stuff.

>I think one is either born to do DSP or not. I think it is genetics.

If you are intent on looking for a pile of reasons its a horrible topic,
its probably the wrong topic for you.

Steve

From: Jeff Cunningham on
On 7/8/10 5:12 PM, Vladimir Vassilevsky wrote:
>
>
> Nasser M. Abbasi wrote:
>
>>
>> I find DSP the hardest subject to become good at.
>
> If you know any subject easy to become good at, let me know.
>
> VLV

Indeed. I heard an interview of a journalist who went undercover to
investigate the jobs of migrant farm workers. After several weeks of
lettuce picking he mentioned to a co-worker that he was disappointed he
wasn't getting any better at it. The worker said "Oh, it takes a good
five years to get really good at lettuce picking".

-Jeff