Any DSP professionals among Thinkpad comunity fellows ?

[alexzabr]

I was thinking we have some percentage of hi-tech professionals here among fellows, perhaps there are DSP (digital signal processing) gurus that may be willing to help a DSP newcomer...
I know, there are several known online DSP-related communities, I participate in two of such (DSPrelated and the community at Mathwork.com), both of these aren't very responsive forums once it deals with a DSP theoretical issues.
Anyway, if there are any DSP experienced professionals willing to help me - I'll be happy, just let me know.
I'm just finishing my B.Sc.EE degree majoring in DSP (among Control and Microelectronics) and is trying to widen my theoretical and practical horizons in certain DSP fields.
The most interesting fields are a non-stationary signal spectrum analyzing and processing by regular and wavelet means...

Thanks in advance, Alex

[christopher_wolf]

I have some experience in this area, but I don't quite understand what you would want help with.

[alexzabr]

Thanks Christopher.
I'm not sure whether it is good idea to start the dedicated correspondance here in forum, perhaps you would prefer to your PM or email, just let me know. I will anyway provide the general direction here.
I'm undertaking my final project now which is in the field of DSP. As part of it I need to establish a reliable spectrum analysis of certain audio-frequency non-stationary signal by means of STFT and wavelet decomposition. So far I managed to rpoceed with wavelet decomposition and further STFT within the decomposed bands and obtain some useful info, but besides of it, at further stages, once the entire algorithm will be developed and stabilized, I'll have to try to implement one in real-time processing system. For that reason, I'll have to estimate the computational complexity of the algorithm to assess whether it is realizable in real life under particular real-time system constraints. Since wavelet (and discrete wavelet transform in particular) is quite new field to me (aside of theory which is readily available), so far I'm having difficulties trying to realize the complexity of discrete wavelet transform implementation, i.e. how much complex/real elementary operations are necessary for dwt at single level and for given data vector length.

Besides of that, there is additional issue I'm trying to clarify for me:
as an alternative approach to wavelet decompostion and band analysis, I'll be willing to run STFT (spectrogram in MATLAB) on teh entire sequence within its entire frequency range (nyquist limited of course) but rather then using regular FFT in it I wuold like to use chirp Z to attain the higher spectrum resolution within pre-defined narrow frequency bands. Now, spectrogram MATLAB (specgram) implementation uses FFT in it, I would like to replace it with chirp Z whithin user-defined band. So this is what I stumbled upon: is it possible to replace the FFT within specgram by chirp Z easily or I have no choice by to create my own spectrogram function in which chirp Z is utilized instead of FFT ?

Any clue ?

Regards, Alex

[BillMorrow]

I have no problem entertaining the creation of a private group to discuss esoteric hardware issues..

[alexzabr]

Thank you Bill, this will be appreciated.
How it can be done ?

[NS]

Thank you Bill, this will be appreciated.
How it can be done ?

I am waiting anxiously for it. Look forward to it...

I am waiting...

Regards:

[christopher_wolf]

Thank you Bill, this will be appreciated.
How it can be done ?

I am waiting anxiously for it. Look forward to it...

I am waiting...

Regards:

Wow, you know about DSP theory, NS?

Tell me then, what is the purpose and form of the Chirp Z transform? What are the advantages? What about the special cases? What is a CTA and how does it represent the DFT?

Answer those first before you say anything else.


Alex; sure...Although I am not sure if you can easily use a CZT in place of the FFT that the MATLAB implementation of specgram uses. Rather, why not use the CZT function in MATLAB itself (may have to because I don't think there is a way to get it through the spectogram)?
If I recall correctly, it is CZT(X,M,W,A); X is the sequence, M is the length, W is the complex ratio between the endpoints given by the transform, and A is the complex starting point.

[alexzabr]

First of all thanks Cristopher for your response.
Well, just plain chirp Z (its MATLAb help elaborates czt() usage) isn't much of help for me from the similar reason as a regular FFT. The reason is that the analyzed signal is non-stationary and is actually featured by a several instances of short-time high-energy noise appearance. Chirp Z as it is cannot correlate the spectral characteristics with time bases, just like DFT isn't capable of that.
Contrary, by means of STFT (spectrogram), I would be willing to apply chirp Z within the sliding time window instead of FFT and that chrp Z would be defined to run at high enough resolution within several relatively narrow ferquency bands to locate the noisy disturbance energy within these bands (I suspect the most of this noise energy resides within these bands). This is why I think spectrogram+chirp Z combination may help me.
BTW, yesterday I opened up specgram() MATLAb source code - was surpized to figure their usage of chirp Z inside the function, aside of FFT usage. Specgram() help also hints about possibility of usage chirp Z insteda of FFT within the function, but unfortunately only for
spectrum analysis at the set of pre-defined frequencies, not within the pre-defined bands as I need.
I'll have to try to figure the possibility of modifying the specgram() source in order to be able to utilize its chirp Z for frequency bands instead of single frequencies...may not be easy task though...

In fact this is an alternatiev analysing approach to the wavelet decomposition I do also...