Acoustical measurement "windowing'

tonio · 19th May 2010

With measurement software there seems to be various GUI " windowing" algo's.
I Apologize if I used any terms incorrectly.
I know that the following is available to me:
Rectangular
Hamming
1/2 hamming
Hanning
1/2 hanning
Bingham
1/2 Bingham

So ... what is the difference?

Is there a purpose for each type vs. intended use?

T

DrFrankencopter · 22nd May 2010

In an spectral analysis windowing refers to the method by which the time sequence is broken up and analyzed.

You can perform an FFT based spectrum calculation on your entire test sequence, but you'll find the the high frequencies will tend to be noisy. To improve signal to noise ratio in the high frequencies, the signal is broken into discrete windows. These windows are defined in terms of the #of points in them. This is a convenient for the FFT algorithm, but is hard for us to understand as users. In my opinion it's better to see the window as a length of time (i.e. divide the window length by the sample rate). So, if you're using windows of length 8192 (for pure FFT's they need to be powers of 2) at a sample rate of 44.1kHz this is 0.185 seconds worth of data. The longest frequency visible in this window is about 5Hz. So, longer windows will improve your low frequency resolution at the expense of a bit more high frequency noise.

The windows of the data are typically overlapped to some extent, and this will have a big influence on the speed/cpu use as well as the apparent noise. These can really smooth out a spectrum, but it can also prevent you from resolving narrow poorly damped modes.

The window type (Cosine, Hamming, Blackman-Harris, etc) simply determine how the data is weighted across the window. Usually you want to de-emphasize the data at the start and end of the window. Hamming window is probably the most popular window type.

I hope that helped...

Cheers

Kris

Dange · 23rd May 2010

Just to add that Hann and Hamming are two different types of windows. Unfortunately similar in name and type!

As SAC said window functions do have a specific use depending on the signal being analysed and are not arbitary. The window functions are used to reduce errors.

For periodic signals, like sine tones, the uniform or rectangular window can be used, but the window must contain exactly one period.

So if you were analysing a 500Hz sine wave, then the window should be 0.002 secs long.

If you want to delve into the maths of FFT then the need for windowing becomes apparent..........eventually

Why not try different windows and see what they actually do? You should be able to see some differences.

What are you measuring?

19th May 2010	#1
tonio Gear addict Joined: Sep 2002 Location: San Diego Posts: 339 Thread Starter	Acoustical measurement "windowing' With measurement software there seems to be various GUI " windowing" algo's. I Apologize if I used any terms incorrectly. I know that the following is available to me: Rectangular Hamming 1/2 hamming Hanning 1/2 hanning Bingham 1/2 Bingham So ... what is the difference? Is there a purpose for each type vs. intended use? T

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search

Similar Threads
Thread	Thread starter	Forum	Replies	Last Post
Acoustical Curtains?	nafspark	Studio building / acoustics	11	27th June 2009 02:02 PM
Acoustical Paint	JBaptista68	Bass traps, acoustic panels, foam etc	14	6th April 2009 07:31 AM
Acoustical Caulk...	AnalogBrain	Geekslutz forum	4	5th January 2008 05:15 AM
Acoustical Treatment?	Clintr89@yahoo.c	Remote Possibilities in Acoustic Music & Location Recording	1	11th September 2007 10:16 AM

22nd May 2010	#2
DrFrankencopter Lives for gear Joined: Sep 2003 Location: Ottawa Posts: 917	In an spectral analysis windowing refers to the method by which the time sequence is broken up and analyzed. You can perform an FFT based spectrum calculation on your entire test sequence, but you'll find the the high frequencies will tend to be noisy. To improve signal to noise ratio in the high frequencies, the signal is broken into discrete windows. These windows are defined in terms of the #of points in them. This is a convenient for the FFT algorithm, but is hard for us to understand as users. In my opinion it's better to see the window as a length of time (i.e. divide the window length by the sample rate). So, if you're using windows of length 8192 (for pure FFT's they need to be powers of 2) at a sample rate of 44.1kHz this is 0.185 seconds worth of data. The longest frequency visible in this window is about 5Hz. So, longer windows will improve your low frequency resolution at the expense of a bit more high frequency noise. The windows of the data are typically overlapped to some extent, and this will have a big influence on the speed/cpu use as well as the apparent noise. These can really smooth out a spectrum, but it can also prevent you from resolving narrow poorly damped modes. The window type (Cosine, Hamming, Blackman-Harris, etc) simply determine how the data is weighted across the window. Usually you want to de-emphasize the data at the start and end of the window. Hamming window is probably the most popular window type. I hope that helped... Cheers Kris

23rd May 2010	#3
Dange Lives for gear Joined: Feb 2008 Location: London, UK Posts: 857	Just to add that Hann and Hamming are two different types of windows. Unfortunately similar in name and type! As SAC said window functions do have a specific use depending on the signal being analysed and are not arbitary. The window functions are used to reduce errors. For periodic signals, like sine tones, the uniform or rectangular window can be used, but the window must contain exactly one period. So if you were analysing a 500Hz sine wave, then the window should be 0.002 secs long. If you want to delve into the maths of FFT then the need for windowing becomes apparent..........eventually Why not try different windows and see what they actually do? You should be able to see some differences. What are you measuring?