ΣSigmaStudio Ask AI Русский Log in

FFT and Analysis Window blocks documentation

SigmaDSP & SigmaStudio pedromartin3z Dec 15, 2017 00:00 5 replies Original on forum ↗
Subscribe
pedromartin3z Question
Dec 15, 2017 00:00

Hi folks and analog devices employees,

I recently was recommended (See this topic) SigmaDSP processors ADAU145x and ADAU146x for an specific application which requires frequency domain analysis.

Now, I'm considering purchasing the EVAL-ADAU1466 Evaluation Kit, but first I do require further documentation about FFT and Analysis Window blocks to ascertain I will be able to develop my application.

1. As per this topic, maximum FFT size was 1024 on 2014. Has a larger FFT size (4096) been implemented?
2. Which kind of window is used on the Analysis Window? (i.e Blackman)

3. Is there a way of modifying the overlap? (i.e 25% instead of 50%)

4. How to interprete complex FFT Data? (i.e DC Frequency component, positive frequencies, negative frequencies)
5. Is data normalized? (i.e Normalization factor = 2*sum(window))

Application schematics example:

Thank you in advance,
Pedro

Audio Signal Processors

frequency domain

fft

sigmastudio

SigmaDSP Audio Processors

adau1466

Show More

MikeSmith12 Reply
Dec 19, 2017 00:00

I wouldn't expect much help with the FFT side of the the 145X devices. I would hardly even classify it as being in the alpha stages of development, as things just don't work. I have been trying for a long time to get the FFT working, and there is little to no support.

Sadly to be blunt, if you truly need to do frequency domain operations, I would look elsewhere.

pedromartin3z Reply
Dec 19, 2017 00:00

I've been assuming this since I made the post. Thank you for confirming.

I'm currently looking at SHARC processors, which I find overkill right now but at least will allow me to develop more complex algorithms as the project progresses in the future.

ADIApproved Reply
Dec 20, 2017 00:00

HI Pedro:

Sorry about the disappointment. We are certainly making progress on the availability of frequency domain algorithms, although we also would want them to go faster. Certainly Sharc is a great floating point alternative. A limitation that the 145x has is the memory when using long FFT, for this reason and because of the fixed point nature of the DSP, we don't recommend FFT sizes larger than 1024.

The current built in window is modified hann that allows zero sum overlap. You could inspect the coefficients of the window by playing around with SigmaStudio.

Currently there is no real way to modify the 50% overlap.

The current FFT is complex, we are working on a Real only FFT. Complex FFTs are symmetrical with real signals and therefore in most applications, the imaginary part can be ignored.

The FFT does normalization on every one of it's intermediate steps.

Best luck, we certainly appreciate your feedback and look forward to meet your requirements in the future.

Regards

Miguel

1. As per this topic, maximum FFT size was 1024 on 2014. Has a larger FFT size (4096) been implemented?
2. Which kind of window is used on the Analysis Window? (i.e Blackman)

3. Is there a way of modifying the overlap? (i.e 25% instead of 50%)

4. How to interprete complex FFT Data? (i.e DC Frequency component, positive frequencies, negative frequencies)
5. Is data normalized? (i.e Normalization factor = 2*sum(window))

pedromartin3z Reply
Dec 20, 2017 00:00

Hola Miguel,

thank you for the answer, you clearly answered all of my questions and helped me evaluating your platform. I'll move to SHARC.

Looking forward to work with you in the future,
Pedro

maurizio Reply
Jun 17, 2018 00:00

Hello,
it is possible to find documentation.
How can I read the fft values?
Thank you