Continue to Site

Welcome to EDAboard.com

Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.

Register Log in

FFT convolution - FFT resolution vs length

Status
Not open for further replies.
E

Eugen_E

Full Member level 6
Joined
Nov 29, 2004
Messages
383
Helped
44
Reputation
86
Reaction score
11
Trophy points
1,298
Location
Romania
Activity points
2,862
fft resolution

Hello,

I need to implement an FFT convolution for a speech signal, sampled at 8kHz. The questions are:
1. what FFT length is enough to obtain good frequency resolution?
2. if i use a long FFT, egg 1024, the time needed to fill the input buffer will be 128ms, higher than the maximum interval the speech signal is quasi-stationary (<20ms), and this will lead to a spectral "averaging" and distorsion.

To satisfy all requirements, a long FFT could be used, taking a short interval of the input signal, and filling with zeros to the required length to get good resolution? If this is correct, please guide me in choosing appropriate values for FFT length, and number of input samples taken for this typical application.

Thanks
 

convolution fft

Would it make sense to do a "sliding FFT". Means you collect 1024 samples every time and do an FFT but you start a new FFT every 20 ms. This would cause it to have samples that are common to two or more sequential FFTs.
-b
 

    E

    Eugen_E

    Points: 2
    Helpful Answer Positive Rating
fft length resolution

Mathematically (I underscore - mathematically, i.e. equal) FFT convolution is nothing more than the fast algorithm to do linear convolution. You don't need to worry about stationarity interval, while you're performing linear convolution.

If you process audio signal in adaptive way (i.e. use different filter coefficients) for each stationarity interval, then of course you should worry about it. But in your task that is not the case. Just linear convolution as I understood.

Filter resolution is defined by filter length (reverse proportionality), and FFT length should be at least twice greater than filter length.
 

Status
Not open for further replies.

Similar threads

Part and Inventory Search

Welcome to EDABoard.com

Sponsor

Back
Top