What are some programming-centric ideas for EE statistics labs?

[TheKracken5]

Hi everyone, I recently got a job through my instructor to design labs for a new course we are offering. It is Engineering Statistics for Electrical and Computer engineering. We will be using Yates text. We previously had to take a engineering stats class in the math department that is more geared toward our much larger Mechanical Engineering department.

Our universities Engineering program uses the “Learn by doing” model and as such, every engineering course has a lab component. My instructor is VERY busy this semester and has made this basically a independent project for me.

I was hoping some of you might have some ideas on what sort of labs would be useful for both industry as well as in a academic setting. Other than Calculus, no other courses are prereqs for this course, so no EE background is assumed.Our department uses Matlab for all of our courses, and I figured maybe our labs should be geared more toward this. I don’t know if there will be any plans to use actual equipment for the labs, all up in the air right now.Thanks for any help!

 

[berkeman]

The main place I've seen statistics and probability used in EE is in communication theory. You can certainly design some interesting labs around testing the limits of communication theory. Have you looked into that?

 

[TheKracken5]

The main place I've seen statistics and probability used in EE is in communication theory. You can certainly design some interesting labs around testing the limits of communication theory. Have you looked into that?

Yes I have been looking into that. It certainly is easy enough to have them calculate the probability that there will be exactly 20 1’s in a 64 bit random sequence (one of the ideas I am working on now to illustrate and introduce basic communications along with probability mass functions). It just feels really weird to call this a lab when I can replace bits and communications with m&m’s or skittles….I am also looking into signal processing/ filtering, and target detection.But, I am trying to find the fine line between too advanced and too easy. No EE classes are required for this course and calc 1 is the math requirement.This is a very exciting job for me because I am planning on going to grad school and studying something like information theory, signal processing or controls.

 

[StoneTemplePython]

... Our universities Engineering program uses the “Learn by doing” model and as such, every engineering course has a lab component. My instructor is VERY busy this semester and has made this basically a independent project for me.

I was hoping some of you might have some ideas on what sort of labs would be useful for both industry as well as in a academic setting. Other than Calculus, no other courses are prereqs for this course, so no EE background is assumed.Our department uses Matlab for all of our courses, and I figured maybe our labs should be geared more toward this

Given this, and what you said in your follow up post

I am planning on going to grad school and studying something like information theory, signal processing or controls.

Have you picked up Cover and Thomas's Elements of Information Theory? There is a lot of interesting stuff in there. For instance, in the chapter called "Gambling and Data Compression" you could probably do 1 or 2 labs on a well chosen problem or two in there. E.g. consider 6.8

The following analysis is a crude approximation to the
games of Lotto conducted by various states...

[after setup, the questions follow]

(a) What is the optimal strategy to divide your money among
the various possible tickets so as to maximize your long-term
growth rate? (Ignore the fact that you cannot buy fractional
tickets.)

(b) What is the optimal growth rate that you can achieve in this
game?
1 1 1 1 1

(c) If (f 1 , f 2 , . . . , f 8 ) = ( 18 , 18 , 14 , 16
, 16 , 16 , 4 , 16 ), and you start
with $1, how long will it be before you become a millionaire?

You could have a lab where people conjecture various strategies, and work through coding them up in Matlab. Then as the lead, you could give a sketch of the optimal strategy at the end, and run a simulation showing how great it is...

(If needed you can put in the extra work at the very end to tie this explicitly into data compression...)

goals of said lab: 1.) programming practice, 2.) unexpected uses of things related to EE, 3.) motivate additional topics like information theory

- - - -

I'm kind of wondering about the use of Matlab, though, if there is no programming course pre-req and people haven't taken Linear Algebra yet.

 

[jasonRF]

What is the syllabus? Just knowing the title and the book doesn't tell us a whole lot, unless that course just marches linearly through the book. Even then we need to know how far into the book you get.

jason

 

[Stephen Tashi]

. We will be using Yates text.

Presumably https://www.amazon.com/dp/B00I86TB5S/?tag=pfamazon01-20

When I searched in that book on Amazon for "shot noise", I didn't find anything. Is that topic omitted?

 

[NTL2009]

Well, it's been a while, but would this be applicable?

Back when cell phones were going from analog to digital, I got exposed to a statistical measure of receiver sensitivity for these digital systems, the Bit Error Rate (BER). For analog you use a low level RF input signal (modulated at a standard level), and measured the analog signal-to-noise ratio in the recovered (de-modulated) signal. Too much noise, and it was a failure. Done.

But that didn't work well for digital, since it either recovered the signal, or it didn't, and the 'borderline' was narrow, and... the digital receiver used error correction and there was redundancy in the transmitted signal. And in production, you want fast go-no-go tests, you don't want to have to 'hunt' for the value at that borderline edge (those sorts of comparative tests were done in the lab).

So the BER was a more 'raw' measure of how many bits were dropped from a known pattern (I may be fuzzy on details here, it has been a long time).

But the interesting thing to me at the time was how statistics were applied. If there were very few errors, or very many errors, the test could be very short - you had a high probability that this unit was good/bad. If there were some more errors, the test would run a little longer to collect enough data before deciding good/bad.

 

[berkeman]

I am also looking into signal processing/ filtering, and target detection.

Depending on the level of your students, you can put together a pretty interesting lab using Excel to implement DSP filter functions. The graphing functions in Excel can be used to illustrate the filtering effects (LPF, HPF, BPF), and you can also use the FFT function in Excel to demonstrate the juxtaposition of the time and frequency domains for DSP filters...

 

[TheKracken5]

Hi everyone, have been very busy. I appreciate all the replies. Here is some more detail.

Due to this course being a brand new development, there has not been a syllabus made. I am currently taking this course with about 10 other students as a test.

The instructor had plans for this course, but it seems he got very busy and basically we are supposed to read the text at home and then when we get to class we just ask questions, he hasn’t had the time to really learn the material himself. A little disappointing, but this give me a great creative outlet for my ADHD and my passion for math and teaching to help this course be developed correctly.

Currently the requirement to only have Calc 1 as a pre req seems to be based on the fact we are piggy backing off the mechanical engineering course in the math department. Their focus is a lot more statistics based than probability and stochastic processes like this course.

My hope is that this course can be changed to be included in the actual EE curriculum. Currently you can take the engineering stats course anytime you want, so if you want it to be a senior class you can. It would be nice if this was a pre req for courses in DSP, communication, ect.I am meeting with the instructor for the first time since starting, tomorrow. Hopefully I will be able to get some more clarification on a lot of things. He has apologized for kinda just throwing me into this with no direction.The topics planned to be covered in the text still seems to be up in the air, but for now it is ch1 – 8 in our Yates Probability and stochastic processes textbook. I am hoping to get this changed a bit.

 

[TheKracken5]

Given this, and what you said in your follow up post
Have you picked up Cover and Thomas's Elements of Information Theory? There is a lot of interesting stuff in there. For instance, in the chapter called "Gambling and Data Compression" you could probably do 1 or 2 labs on a well chosen problem or two in there. E.g. consider 6.8
You could have a lab where people conjecture various strategies, and work through coding them up in Matlab. Then as the lead, you could give a sketch of the optimal strategy at the end, and run a simulation showing how great it is...

(If needed you can put in the extra work at the very end to tie this explicitly into data compression...)

goals of said lab: 1.) programming practice, 2.) unexpected uses of things related to EE, 3.) motivate additional topics like information theory

- - - -

I'm kind of wondering about the use of Matlab, though, if there is no programming course pre-req and people haven't taken Linear Algebra yet.

I am very confused about the Matlab and pre req math as well. But our department introduces MATLAB in our very first circuits course, I think it is a good idea, even if I am terrible at programming.

Also thank you for the book recommendation! The more and more I have been reading up on labs, the more interested I am in information theory and communications.

 

[StoneTemplePython]

I am very confused about the Matlab and pre req math as well. But our department introduces MATLAB in our very first circuits course, I think it is a good idea, even if I am terrible at programming.

Also thank you for the book recommendation! The more and more I have been reading up on labs, the more interested I am in information theory and communications.

To me it seems obvious that you should try to make these labs as programming centric as possible. It's win-win because running things like simulations are a great way to try to replicate the experiments that occur at the sample space level in probability and reinforce intuition for learners, and I think doing so will force you to get up the programming learning curve (i.e. you wouldn't recommend a lab project you haven't fully vetted). EE or otherwise, getting up the curve in programming should be immensely valuable. You won't be terrible for too long.

Another idea, which might be too advanced given the minimal pre-reqs but maybe not if you cherry pick the right stuff, is too comb through the iconic text from Snell and Doyle that basically shows Random Walks and Electric Networks are 2 sides of the same coin. Link here:

https://math.dartmouth.edu/~doyle/docs/walks/walks.pdf

(e.g. look at pages 12 - 14 and you should recognize a circuit and see discussion of a Monte Carlo approach, both of which seem relevant. There's a lot more of course.)