Signals and systems class, why?

[prophoss]

I am currently a student in a computer engineering degree plan. As a part of my plan we have to take signals and systems class where we learn to ... well that is the problem. I really don't know what we are learning this stuff for or what I would be using it for after college. I sort of get the math side and how to use Laplace, Fourier, and Z transforms. So far I have done well and kept a good grade point average. This class though thoroughly kicked my butt. Partly, I think, because I don't see how it all applies. I was hoping someone out there might shed some light on the application of this subject on the job site. I look forward to hearing from y'all, and thanks in advance.
 

[ataradov]

Out of all things you can learn in the University this stuff is one of the most useful.

For anything electronics-related you need to develop good intuitive understanding of things like Fourier transforms. There is rarely a need for the actual math with integrals, but you need to understand the relationship between frequency and time domains.

This stuff shows up literally everywhere on a daily basis.

Same for general DSP and control systems concepts.

Also, the fact that is is hard to understand is good, it means you are actually about to learn something new, not just coast on the stuff you already know or can easily pick up.

I would say approaching this from the DSP side is easier and way more visual. You can actually do experiments that are visual in nature. It will be based around discreet transforms, not the continuous ones, but they are directly related to each other.

[tggzzz]

Listen to ataradov.

You may not understand the relevance now, but these are fundamental concepts that won't change during your career (unlike most stuff that has a half-life of a few years).

Even if you don't directly use it yourself, you will probably find it is a useful tool to cut through rubbish spouted by other people.

Plus, of course, the reason to take the course is because you don't understand it.

[Benta]

+++ to ataradov and  tggzzz.

This stuff will stay with you for life. There are plenty of "application note solderers" out there. But analytical thinking as an engineer requires mental tools.
Meaning complex maths, LaPlace, Fourier, Z-transforms and simply understanding orders of magnitude (especially relevant these days with the Corona panic) etc.

Stay with it. It'l give you a backbone you can't imagine yet.

[AntiProtonBoy]

Also, if you don't take the path of electronics engineering and stay inside the domain of computer science, signal processing will be a recurring theme across many disciplines, ranging from data compression, image processing, audio, scientific data processing, computer graphics, etc.

[emece67]

.

[coppice]

It puzzles me why you can't see the relevance of these techniques to computer engineering. Are you looking at computer engineering in a narrow logic function way? Try getting a signal from one place to another fast and reliably and you'll soon start seeing the relevance of Laplace, Fourier and Z transforms to computer engineering. Hint: even the humble gigabit ethernet chip uses DSP techniques, where these mathematical techniques are relevant.

[tszaboo]

IDK how is it for you, but for us signals and systems was used to weed out the weak and keep the rest for a few extra semester in the university. 80% of students failing tests and other similar bullshit.

[prophoss]

It puzzles me why you can't see the relevance of these techniques to computer engineering. Are you looking at computer engineering in a narrow logic function way?

You may actually be correct there. Most of what I have been looking toward is the programing not so much of the how it all works from a signal standpoint. Part of the reason I went the direction I did was to understand how things work from the bottom up. I have learned all about circuits from the electron up. When I got to this class though I guess I just didn't get the reason behind learning all this and was completely confused as to the need of it. I asked my teacher and he just said so that we can analyze circuits. That didn't exactly answer my question and that is why I asked the same question here.
Really good stuff y'all thanks.

[prophoss]

OK, I am trying to wrap my head around all this so be patient with me here. I got confused in this class and I am just trying to get it straight in my head.
DSP(digital signal processing) is basically like sampling a continuous wave form. The idea of that is just to do what then? Make sure there is no aliasing? Why do we need all the Z transform stuff?

[coppercone2]

because if you can do it in a program you are useful as a data processor

a job that pays might literary be someone draws a ladder filter and tells you that he wants that in labview working smooth

(probobly high end technology job that you need to sell yourself extremely well to get)

[ataradov]

DSP(digital signal processing) is basically like sampling a continuous wave form.

This is the most basic thing you can do. Most DSP courses just assume that you have correctly sampled digital representation of the signal. The fun starts from that.

The idea of that is just to do what then? Make sure there is no aliasing?

To be able to implement thing like mp3 compression. Or implement Zoom and make millions on the pandemic. Or understand how "Ok, google" or "Alexa, what time is that?" works. All of that requires deep understanding of the DSP.

And time-frequency transforms come in handy for a lot of reasons. Once of them is that not every part of the signal in the time domain carries useful information, and neural networks have much easier time dealing with the frequency domain information.

Or even more benign example that virtually any EE would have faced in their life time. You get a microcontroller and it reads ADC samples. Those samples are noisy. What is the best way to filer out the noise for a given scenario and a type of input signal?

And if you get into digital communications (Bluetooth, Wi-Fi, low power networks), you won't get anywhere without solid understanding of the DSP and transforms.

It is hard to understate just how important those concepts are in real life.

[bd139]

This stuff pops up in the business end of computing as well. We implement digital filters (usually IIR) regularly. 95% of stuff is putting and getting shit from databases. Without the additional knowledge things get boring.

Obviously if you go in the hard EE end, perhaps 90% of the interesting problems are signal processing.

[bd139]

Or you just do the comp sci thing and double-buffer it

(I spent the last few years writing high volume aggregators)

[tggzzz]

You may actually be correct there. Most of what I have been looking toward is the programing not so much of the how it all works from a signal standpoint.

The danger is you will correctly implement something that cannot work due to fatal high-level flaws.

Understand the difference between validation and verification!

Both top-down and bottom-up design and implementation are necessary.

Part of the reason I went the direction I did was to understand how things work from the bottom up. I have learned all about circuits from the electron up.

Good for you; such cross-disciplinary knowledge will stand you in good stead over the years.

But, of course, you haven't "learned all about circuits" Nobody can.

[T3sl4co1l]

Part of the reason I went the direction I did was to understand how things work from the bottom up. I have learned all about circuits from the electron up.

Have you?  Excuse my skepticism -- Fourier transforms show up frequently in physics, especially the statistical mechanics that underlies semiconductor theory!

Indeed, a transformation is incredibly useful when applied to repeating structures -- in this case, crystalline solids.  The physics application is to integrate over a periodic structure, instead of trying to add up each and every 10^26 or so particles' contributions, surely a hopeless endeavor.

So you see terms like "phase space" or "k space", which plot wavenumber -- that is, spacial frequency (units of 1/length).  The transform works just the same with respect to space, as it does to time (hertz == 1/time) -- anywhere you have a repeating structure with respect to some parameter, you can do a transform, and you'll probably find something meaningful.

Even more fundamental than that, matter itself is made of waves (quantum mechanics); FT is often used in QM solutions, though the boundary conditions are often more complex, so that an approach from differential equations is necessary.  (For example, the field in a hydrogen atom gives rise to discrete energy levels; but the orbital parameters (spin and angular momentum) give rise to spherical harmonics -- the FT of waves on a spherical shell.)

Understanding wave-particle duality, is simply understanding the Fourier transform -- and once you have an intuitive understanding of wave mechanics, my friend, you can understand literally 90%, maybe 98% or more, of all phenomena in the universe, from the smallest subatomic to the largest galactic-cluster scale!

I asked my teacher and he just said so that we can analyze circuits. That didn't exactly answer my question and that is why I asked the same question here.

Speaking of analysis, and periodic structures, an interesting application of FT is the classic nerd snipe: https://xkcd.com/356/

By symmetry, the nearest-neighbor pair of points is easy to solve; but the knights-move pair shown here is surprisingly challenging.  A typical solution is to consider the infinite array of nodes -- we're just doing nodal analysis like we would any other circuit -- and take the (2-dimensional) Fourier series of that entire system, with respect to position.  The two points in question are the boundary conditions, and, crank crank crank, out pops a... factor of pi?

Tim

[Tomorokoshi]

For a small view into this topic, and to spend some time on a cold winter day, watch this series:

[Electro Fan]

You might be too close to the trees to see the forest.

A huge part of the forest is what we refer to as Information Systems.  Information Systems are used to collect, distribute, and manage information. 

Information Systems are comprised of computer systems (that input, process, and output information) and networks (that move information, both with wires and wirelessly within and between machines).

Information comes in various forms including text, sound, images, and video. 

In order to design and build and maintain information systems it is necessary to move information back and forth among the analog and digital realms.  This requires ADCs, DACs, bandwidth, and much more to manage the signals that represent the information as it moves within and between machines and as the machines interface to the applications and humans they support.  The ability to reliably make all this happen requires the concepts, math, and physics that are being taught in the lessons and projects in your class.  You might think you are abstractly learning about frequency, amplitude, and phase but you are learning how to manage information comprised of many signals in the analog (continuously variable) and digital (binary) realms - all riding on the microscopic control of nature including electromagnetics.

Unfortunately it is a common approach to teach university students about lots of parts without teaching the big picture of the systems that are made from the parts but you should accept that what you are learning is a bunch of powerful building blocks that will help you design the vegetation and trees that make the forest of information technology.  Whether it’s signals in a circuit board like I2C or signals across the Internet with TCP/IP or any other type of signal it’s ultimately about the ability to reliably and ever more efficiently and cost-effectively transmit, process, and receive information.  Almost everything in your signals class should fit into this framework if you zoom out to see and assemble the big picture.  With signals machines and people get and give information and with information machines and people make decisions that lead to actions that drive productivity that drives profitability that creates the ROI that can be recycled back into R&D.  How we harness signals is fundamental to much of technology across all industries (energy, transportation, healthcare, you name it) and throughout the private and public sectors of human society.

Challenge your dynamic range to keep learning the details and the big picture at the same time.

[prophoss]

Excellent stuff, thanks.

[Benta]

Unfortunately it is a common approach to teach university students about lots of parts without teaching the big picture of the systems that are made from the parts but you should accept that what you are learning is a bunch of powerful building blocks that will help you design the vegetation and trees that make the forest of information technology.

Very true and well phrased. I remember the situation from my university days.
In retrospect I understand the issue: the lecturers on the basic stuff (maths, physics, transformations etc.) were all theorists from the maths/physics faculty and would present ridiculous examples using farads, henrys etc. in what they were lecturing. All highly knowledgeable people with a deep knowledge in their field, but only in that (there's a limit to how much one person can handle)
It wasn't until we got to the higher level courses that the lecturers were actually engineers with practical experience. From then on things just got better. But their math knowledge was not at the level of the maths/physics guys, just much more application oriented.

Normal for all studies, I guess.

[SiliconWizard]

I agree with what was said above.

I'm just sorry that (apparently) your teachers are not able to make you see what those concepts are useful for. That's part of teaching IMHO, especially teaching future engineers!

Or maybe they actually are showing students the applications, but *you* are not interested in those applications. For instance, If you're focusing on "computer engineering" and are not particularly interested in the hardware side of things, you may not see the point or even be interested in subjects like signal processing. I think you'd miss out on useful and interesting things in that case, but I can understand that not everyone is interested in those topics, and I also know that many students going for "computer" engineering these days are mostly interested in the pure software side of things. Of course those concepts could be useful even if you're just doing pure software, but I certainly know a lot of "software" people hating those topics and shying away from even signal processing as much as they can. If this is your case, I'd say: try and give it a chance, you might end up getting interested - do not hesitate to ask your teachers for practical examples if you can't see any. Other than that, just see this as part of your formal education, like what you had during high school.

[ataradov]

Normal for all studies, I guess.

Yes, this is very true. I noticed the exact same thing. Fortunately I learned early on to recognize this pattern.

One of the most helpful thing to get a better perspective is to actually watch educational YouTube videos. They will not teach you anything, but they are very useful in putting things into place. Like the video referenced above for the Fourier transform or videos from 3Blue1Brown and other popular creators.

Another useful source is the lectures from the universities (MIT, Stanford). What they publish is mostly 101 stuff, so it is not incredibly taxing, but can provide another view on things.

[ataradov]

Or maybe they actually are showing students the applications, but *you* are not interested in those applications.

I think part of it, they don't want to reveal the truth that a lot of stuff that is done in real life is not some glorious theory, but a lot of experimentation and trying things out (in an educated way, of course).   

In the control systems we looked at a lot of practical and would be useful examples of systems. But they all started something like this "CD-ROM head positioning over the track can be modeled as this combination of blocks, find the coefficients for optimal control". The answer why it is modeled this way is nowhere to be found.

After working with DSP and CS in practice on actually new systems for which there is no pre-made recipes, I realized that you just try things until the model behaves like the actual system and then add control elements. And hope that your model was good enough that the real system responds the same after adding the control stuff. If not, rinse and repeat until it does.

Knowing this from the beginning would have been very helpful.

[NiHaoMike]

I would say to learn it with SDR experiments. I learned more about it from videos about SDR than I did in school.

[tggzzz]

Or maybe they actually are showing students the applications, but *you* are not interested in those applications.

I think part of it, they don't want to reveal the truth that a lot of stuff that is done in real life is not some glorious theory, but a lot of experimentation and trying things out (in an educated way, of course).   

In the control systems we looked at a lot of practical and would be useful examples of systems. But they all started something like this "CD-ROM head positioning over the track can be modeled as this combination of blocks, find the coefficients for optimal control". The answer why it is modeled this way is nowhere to be found.

After working with DSP and CS in practice on actually new systems for which there is no pre-made recipes, I realized that you just try things until the model behaves like the actual system and then add control elements. And hope that your model was good enough that the real system responds the same after adding the control stuff. If not, rinse and repeat until it does.

Knowing this from the beginning would have been very helpful.

There is truth in that, but the "pure" and "simple" theoretical examples are a necessary base that enables guided experimentation and design.

Without that, all you have is blind fumbling.