Audio (video) - Kindle. Getting it up and running, Code level.

[geekmaster]

Originally Posted by knc1

...
I.E: Work done in the sense of a collaboration with the community at large, not in a competition with the community or any of its members.

Alternative methods and parallel development efforts are not necessarily competition anyway. It is just demonstrating another way to do something, which is a good thing. And ideas from the various methods can be combined later for a more robust solution. A diversity of good ideas is healthy and welcome.

[twobob]

Originally Posted by knc1

And an example of how to do it, with example C code?
http://processors.wiki.ti.com/index...._of_Real_Input

This is another great document, Currently this is just out of my skills price range (like so many other things) Here's a selection of the things I am stuck on:

To keep it simple, in precis, the points are about:

• *restrict
  
• Pragma: DATA_ALIGN and DATA_MEM_BANK ( and pragmas in general for the processors we have on the table)
  
• _nassert

Okay so the full blurb:

http://en.wikipedia.org/wiki/Restrict - for some reason I cant get my compiler to swallow this. no doubt some include or library somewhere is not being correctly invoked / linked by me - any thoughts? -lAwesomeRestrictLibrary? hehehe
I understand it is an assertion that values will only be accessed singly during operation to provide further optimisation options for the compiler. I'll keep reading on this one.
I'm sure it a noob mistake.

I am stuck on, data-aligning the values in memory, The Pragma: DATA_ALIGN (and also giving my hassle is: The Texas Instruments PRAGMA: DATA_MEM_BANK (which is explained in lovely detail here http://www.dsprelated.com/groups/c6x/show/4176.php),) can I find similar implementations of memory aligned / memory buffered value management for the complete range of K Processors or is this going to be one build for each processor type? (the implication of DATA_MEM_BANK is performance related to prevent mem read lock, If I am reading the docs right and could possibly be ignored to just get things going I suppose, I'll double check that thought)

But the data alignment is crucial for the maths - again assuming I understand the numbers right - so that's a required

Is this as simple as including the right Pragma.h file? http://processors.wiki.ti.com/index....Can_Understand seems to imply arm support of some nature - please be gentle with me, I'm trying to get my head up to speed.

Last thought _nassert ((int) pIn % 8 == 0);

an assertion to the preprocessor that the pln WILL be divisible by 8, if your value is NOT divisible by 8 after you say that it is - bad things will likely happen -

Again I just can't determine if I should be able to implement this. Again forgive me if I'm being nooby.

Finding suitable pragmas or implementing the http://processors.wiki.ti.com/index....Can_Understand ones, also _nassert

[knc1]

All compiler directives, particular to TI's special compiler for their DSP devices.
So those would need to be translated into something that behaves the same on gcc.

Aside from that, the algorithm displayed by the source code should be of value.
Plus, for things like operations in a loop on an array of values, gcc will "auto-vectorize" and generate the required VFP code.
Not so with the TI dsp compiler, there the programmer gets to write it all out themselves.

[twobob]

Originally Posted by knc1

All compiler directives, particular to TI's special compiler for their DSP devices.
So those would need to be translated into something that behaves the same on gcc.

Aside from that, the algorithm displayed by the source code should be of value.
Plus, for things like operations in a loop on an array of values, gcc will "auto-vectorize" and generate the required VFP code.
Not so with the TI dsp compiler, there the programmer gets to write it all out themselves.

So to summarise:

I can ditch the inline _nassertions,
I dont have to worry about the PRAGMAs as this is managed for me (it's the alignment requirement that was worrying me)

I am going to ditch the restrict as well for now then until I understand - how and whether I can get it going - seems to be core functionality - not sure what I'm doing wrong there : )

Yep: so basically - with enough knowledge to rewrite the DSPF_sp_fftSPxSP ASM sections this could be a goer of a project for the kindle.

Nice one, perhaps one day when my knowledge has expanded to encompass these skills I can make this a reality. In the interim I think I will have to go for something a little bit more "script-kiddie"

***The shame*** ; ) thanks very much

[knc1]

I would start with the first article because of the good description.
http://www.eetimes.com/design/embedd...low-power-MCUs

I have attached the source code (open source yeah!) - only one *.h and one *.c file.

[geekmaster]

Originally Posted by knc1

I would start with the first article because of the good description.
http://www.eetimes.com/design/embedd...low-power-MCUs

I have attached the source code (open source yeah!) - only one *.h and one *.c file.

I like how the .h file embedded comments even contain the commented-out source code for the programs used to generate the data tables.

There may be faster ways to do the bit-reversal though, but that should be plenty fast enough.

[twobob]

Originally Posted by knc1

I would start with the first article because of the good description.
http://www.eetimes.com/design/embedd...low-power-MCUs

I have attached the source code (open source yeah!) - only one *.h and one *.c file.

Excellent source again. I did give it the eye over. missed the source. doh.
Superb documentation. many thanks

[geekmaster]

This realtime FFT stuff is cool, but even back in the early apple-2 days, I did realtime spectrum plots (i.e. "waterfall display") by just measuring the zero-cross times to decide which bucket to increment and plot. The end result was much the same. We used it to analyze telephone signalling tones. No sines or cosines. No complex numbers. No bit reversals. But with no interrupts, every path through the code had to be padded to the same execution time for a constant time predictable run loop.

So, depending on your application, you may be able to "cheat" like I did and still have an interesting frequency spectrum visual effect. Of course, we relied on analyzing simple waveforms with a mixture of only a few pure tones. But interestingly, it did show voice plots that looked a lot like those published in the books at the time.

EDIT: We used zero-cross detection because the apple-2 built-in audio (cassette-tape) input only provided a 1-bit digital input value from an analog sound source.

[twobob]

Originally Posted by geekmaster

This realtime FFT stuff is cool, but even back in the early apple-2 days, I did realtime spectrum plots (i.e. "waterfall display") by just measuring the zero-cross times to decide which bucket to increment and plot. The end result was much the same. We used it to analyze telephone signalling tones. No sines or cosines. No complex numbers. No bit reversals. But with no interrupts, every path through the code had to be padded to the same execution time for a constant time predictable run loop.

So, depending on your application, you may be able to "cheat" like I did and still have an interesting frequency spectrum visual effect. Of course, we relied on analyzing simple waveforms with a mixture of only a few pure tones. But interestingly, it did show voice plots that looked a lot like those published in the books at the time.

Yes indeed estimation can transcend crunching everytime, given the right lateral.

Most recently Beat estimation was a project that I read through the code of, the same with amplitude estimation, but the languages were c# and csound respectively, Once I squirm past the - taking longer to read than think about - stage perhaps some shortcuts could be managed.

As an aside:

It's good to know there are strong signal chops people out there, between you, Knc and my friend Will, I seriously fear for any theoretically modeled signals "Theoretical" status. For my part I'm just muddling through what docs I can muster and a few decent books, at the end of the day if I can do this stuff then anyone can.
Like knc says - people who can follow instructions - I would like to think I at least make that list If I manage to get some data into Frequency space then happy days.

[geekmaster]

Originally Posted by twobob

...
It's good to know there are strong signal chops people out there, between you, Knc and my friend Will, I seriously fear for any theoretically modeled signals "Theoretical" status. For my part I'm just muddling through what docs I can muster and a few decent books, at the end of the day if I can do this stuff then anyone can. ...

It has been far too long since I messed with DSPs. You can do a LOT of that stuff these days with commodity SoCs like we have in our kindles.

Regarding writing code: It is best to start simple (like a "Hello World" program) and just keep adding features in small steps and testing them each time, until you have something resembling what you want. Or another way is to start with somebody else's program that does "almost" what you want, and bend it to your will. You learn more the first way, but you can move on to the next big thing faster the second way.

[twobob]

Originally Posted by geekmaster

It has been far too long since I messed with DSPs. You can do a LOT of that stuff these days with commodity SoCs like we have in our kindles.

Regarding code: it is best to start simple (like a "Hello World" program) and just keep adding features in small steps and testing them each time, until you have something resembling what you want. Or another way is to start with somebody elses program that does "almost" what you want, and bend it to your will.

Perhaps I should be clear. I am not a beginner programmer. Just foggy on the syntax and dictates of c.

I have been programming on and off for about a decade I suppose, professionally, longer as a hobby. I have some fairly complicated single-man projects - free to use - completed and out there, Utilising Silverlight, and web services tech so I am vaguely capable.

However cross-devving for an embedded device in a newish-to-me language is a little slow going. I will catch up the pace eventually, I usually do

[geekmaster]

Originally Posted by twobob

Perhaps I should be clear. I am not a beginner programmer. Just foggy on the syntax and dictates of c. ... I have some fairly complicated single-man projects - free to use - completed and out there, Utilising Silverlight, and web services tech so I am vaguely capable. ...

I know that. I was referring to how to learn the nuances of a "new-ish" language (C in this case). But I posted that here not specifically for you alone, but for others reading here as well. Sorry about any potential confusion about my intended audience.

P.S. My professional websites were either perl CGI scripts and javascript, or (later) PHP and javascript, developed in a LAMP (or WAMP) environment. Silverlight? *runs screaming in horror! *

[twobob]

Originally Posted by geekmaster

I know that. I was referring to how to learn a "new-ish" language (which for you is C in this case). But I posted that here not specifically for you along, but for others reading here as well. Sorry about any potential confusion about my intended audience.

Oh yeah. doh! Teeheehee. I 100% agree, test driven development and a solid idea of the results you expect is a good start.

Start little, cascade the changes through your current implementation and look for bugs or unintended effects. (particularly in c)

All sound advice

[knc1]

Originally Posted by twobob

However cross-devving for an embedded device in a newish-to-me language is a little slow going. I will catch up the pace eventually, I usually do

Maybe this will help (attached), the original Maxim publication.
At least a person stands a chance of reading fig. 2 in this copy.

[twobob]

Originally Posted by knc1

Maybe this will help (attached), the original Maxim publication.
At least a person stands a chance of reading fig. 2 in this copy.

Thanks . I have made some notes for when I have more c chops and reposted the orignal source (kinda) with the license (and notes) for my future perusal.

I'm giving the PDF the eye-over now.

EDIT: AH Excellent, much more effiecient than my current - highlight source - save as html - open with FF - save as html - import to calibre - wangle to get pictures embedded neatly in the web versions of stuff on the kindle many thanks!!!