Where is novel DSP developed (in math or in practical DSP or ...?)?

[soundmodel]

Where is novel DSP developed (in math or in practical DSP or ...?)?

Particularly, what got me interested in this topic was that I've perceived that a lot of audio DSP innovations actually came elsewhere. E.g. the FFT was not developed for audio applications, but it could be applied to audio signals. So the research about FFT (and FT) was done elsewhere.

Similarly, some algorithms such as linear predictive coding have been developed for other purposes than what they were applied to in audio signal processing.

So,

If one wishes to innovate in audio DSP, then should one work on pure mathematics or communications engineering or something else?

Of course nowadays one could as well work directly on audio DSP, but I wonder if it's possible to come up with novel techniques in such "applied" domain. Rather, cutting edge research is probably done on other fields, such as in physics.

[FabienTDR]

IMHO you are looking at the wrong place. And innovation is a big empty word anyway.

Audio Tools in general are far more than a DSP algorithm. I'd even say that it represents the most primitive aspect of audio product design. It really starts with the purchase and installation process and ends with customer support and maintenance.

What you are mentioned above are math tools. Theorems, proven methods. They don't make a product. Otherwise, nobody would spend 200$ on an EQ plugin and would simply directly set biquads coeffs, direct IR drawing, Fourier transforms by hand and such. The problem is, they are horribly unusable for the average person. And it's exactly here where one can innovate and optimize whatever workflows.

[mystran]

Where is novel DSP developed (in math or in practical DSP or ...?)?

In the minds of people who genuinely care about DSP.

Seriously, you don't research just by deciding you want to do research and you certainly don't come up with something novel by deciding you want to come up with something novel.

You do it by figuring out some problem you want to look into and then you use whatever methods you can think of to try to solve that problem and find whatever prior research you can find, whatever the prior field happens to be. Sometimes you find that a good solution is known either in your field of interest or some other field. Sometimes you find that a related solution used in another domain can be adapted. Sometimes you find you can combine a few known methods to produce something greater than the parts and then sometimes when you think hard enough about something you accidentally come up with something novel if you're lucky. Sometimes you think you came up with something novel only to realise that it's already known, possibly in another field where it might or might no be used, because it might not be suitable for that other field the same way it's suitable for your chosen problem.

But like.. none of this happens because you decide that you want to do novel research.

PS. Also what FabienTDR said about products being a lot more than just DSP.

[Smashed Transistors]

"Creation" is not "problem solving", but "problem solving" is necessary to practical creation.

To have a chance to have innovative and practical ideas in a technical field, you need to have a broad and deep (i.e. down to the details) knowledge of the field along with an open mind.

The problem is that "experts" tend to have closed minds.

To get expertise while preserving open-mindedness, I think that a good habit is to study more than one solution for a technical problem.

Sometimes some ideas cannot be combined because of incompatible problems. Knowledge of various solutions is really helpful.

In the case of todays DSP, accuracy and CPU charge are not big issues, but aliasing and stability are often the big problems you will have to deal with.

I think that it is also a good idea to read old classics. Personally, I love cybernetics science books of the 50s or 60s.

[nonnaci]

Known problems are often useful for narrowing the search-light but the trade-off necessarily restricts your search space unless you perform your own abstractions. This is where innovation lies. Recognizing contradictions, paradoxes, before others do and establishing a language / new paradigm for it is invention (at least on paper).

Figure out if your knack for abstraction is greater than pragmatics and pursue the field accordingly.

[Oden]

Try this one:

Trial and error.

That's how the most significant break-throughs are made. The wheel etc.

Mathematical theorems? Avoid like plague, unless you actually need to do something deductive.

[Vertion]

Novelty comes from a number of directions, but most of all from your own perception. FFT for example is about transformation of data from one perspective to another, and yet both perspectives are useful. They also can contain the same amount of information, which is a notable point.

Look at the concept of morphing from A to B. To morph between, what are you moving across to get from A to B? What is at the midpoint? How to break down in to smaller parts (method of division) will be tantamount to how to can put things back together (method of construction). Find new ways to look at things, new ways to divide and build up, keep going further into meta-cognition and higher in order. However, every thought should follow with an action if you are to create with your mind and learn the real results of your thoughts. Experimentation, improvisation, necessity, and simplicity are keys, the only way to have no success is to do
nothing. We are all looking for answers and to increase order for the good of our lives.

Quest!

[Michael L]

nvm

[stratum]

My understanding so far is that you need to understand musicians, recording engineers and the way they work, and find a way of making them know how your product meets their needs. If you are lucky they might discover that without your help, but then you may not know what you did right

[soundmodel]

My understanding so far is that you need to understand musicians, recording engineers and the way they work, and find a way of making them know how your product meets their needs. If you are lucky they might discover that without your help, but then you may not know what you did right

Yeah well obviously developing audio DSP "for use" is different from developing DSP algorithms "purely".

The original question concerns mainly the latter one, i.e. "basic research" in DSP.

[stratum]

The original question concerns mainly the latter one, i.e. "basic research" in DSP.

I remember a day I had seen the following words printed in a university professors office door : "Think about why we call it re-search"

[camsr]

It all has roots in mathematics, and mostly discretized from there.

[Smashed Transistors]

In this respect,
mathematics is not an abstract way of theorising,
it is a practical set of tools for problem solving.


A way to know you are on the road for novelty:

- you try to combine two (or more) ideas
- you theorize (on paper) down to the details
- you implement
- you play and test
- the results are good but problems arise
- you google the problem...
- no google solution to the problem => you are on the road for novelty
- Here, the real work starts...

search and re-search as said @stratum

[stratum]

If you compare the discretization chapter of Vadim's book to the classical dsp literature you'll see a good example: People ask questions and find a solution that fits their needs. Other people ask different questions and existing answers do not fit. Then they theorize in different ways (i.e. they have different models because they have different needs).

This is implied here https://www.quora.com/Why-is-it-called- ... first-time in a different way, quoting:

research (v.)

1590s, from Middle French recercher, from Old French recercher "seek out, search closely," from re-, intensive prefix (see re-), + cercher "to seek for" (see search (v.)). Related: Researched; researching.

re- (prefix)

word-forming element meaning "back to the original place; again, anew, once more," also with a sense of "undoing," c. 1200, from Old French and directly from Latin re- "again, back, anew, against," "Latin combining form conceivably from Indo-European *wret-, metathetical variant of *wert- "to turn" [Watkins].

The sense "back to the original place; again, anew, once more," is interesting. It implies doing something again in a different way. You go to the original place to see something different, perhaps something you have already seen but didn't notice the importance of or discarded as being irrelevant (for that time or purpose). The sense "undoing" is also interesting: It has been done before, but done in a way that now looks wrong from a different perspective.

[camsr]

- no google solution to the problem => you are on the road for novelty

Thanks, that made my day.