Hi, I'm new to audio processing but have done a good amount of image processing in the past so have a general background of DSP. Im currently trying to implement a Hamming window filter, but I don't know at what spacing to overlap the windows. Should sequential windows overlap when their height is 0.5, so the sum of both filters at any point is 1? Thanks for any advice Adam

# hamming window overlap

Started by ●November 12, 2008

Reply by ●November 12, 20082008-11-12

On Nov 12, 3:06�pm, Adam Chapman <adam.chap...@student.manchester.ac.uk> wrote:> Hi, > > I'm new to audio processing but have done a good amount of image > processing in the past so have a general background of DSP. > > Im currently trying to implement a Hamming window filter, but I don't > know at what spacing to overlap the windows. Should sequential windows > overlap when their height is 0.5, so the sum of both filters at any > point is 1? > > Thanks for any advice > AdamWhat do you mean by a Hamming window filter? Normally the overlap you describe is when there is block data processing and the Hamming window is used for block data weighting, not as a filter. If you are doing block processing, a 50% overlap would give a sum of 1.08, a 25% overlap would sum to twice that, ... . You will get a better answer if you explain what you are trying to do. Dirk

Reply by ●November 12, 20082008-11-12

On 12 Nov, 21:06, Adam Chapman <adam.chap...@student.manchester.ac.uk> wrote:> Hi, > > I'm new to audio processing but have done a good amount of image > processing in the past so have a general background of DSP. > > Im currently trying to implement a Hamming window filter, but I don't > know at what spacing to overlap the windows. Should sequential windows > overlap when their height is 0.5, so the sum of both filters at any > point is 1?It's not clear what you do, but I assume you have designed a FIR filter by the window method and used a Hamming window. If that's the case, you want to apply the filter as a direct convolution, that is, N-1 y[n] = sum x[n]w[n-k] n = 0,1,2,..... k=0 where w is the N-length window function. Rune

Reply by ●November 12, 20082008-11-12

On Nov 12, 10:06�pm, Rune Allnor <all...@tele.ntnu.no> wrote:> On 12 Nov, 21:06, Adam Chapman <adam.chap...@student.manchester.ac.uk> > wrote: > > > Hi, > > > I'm new to audio processing but have done a good amount of image > > processing in the past so have a general background of DSP. > > > Im currently trying to implement a Hamming window filter, but I don't > > know at what spacing to overlap the windows. Should sequential windows > > overlap when their height is 0.5, so the sum of both filters at any > > point is 1? > > It's not clear what you do, but I assume you have designed a > FIR filter by the window method and used a Hamming window. > > If that's the case, you want to apply the filter as a direct > convolution, that is, > > � � � �N-1 > y[n] = sum x[n]w[n-k] � � � � � � � n = 0,1,2,..... > � � � �k=0 > > where w is the N-length window function. > > RuneSorry I should have axplained better it is for block data weighting. Is there a rule -of-thumb that is used to work out a sensible overlap between sequential windows?

Reply by ●November 12, 20082008-11-12

Adam Chapman wrote:> On Nov 12, 10:06 pm, Rune Allnor <all...@tele.ntnu.no> wrote: >> On 12 Nov, 21:06, Adam Chapman <adam.chap...@student.manchester.ac.uk> >> wrote: >> >>> Hi, >>> I'm new to audio processing but have done a good amount of image >>> processing in the past so have a general background of DSP. >>> Im currently trying to implement a Hamming window filter, but I don't >>> know at what spacing to overlap the windows. Should sequential windows >>> overlap when their height is 0.5, so the sum of both filters at any >>> point is 1? >> It's not clear what you do, but I assume you have designed a >> FIR filter by the window method and used a Hamming window. >> >> If that's the case, you want to apply the filter as a direct >> convolution, that is, >> >> N-1 >> y[n] = sum x[n]w[n-k] n = 0,1,2,..... >> k=0 >> >> where w is the N-length window function. >> >> Rune > > Sorry I should have axplained better it is for block data weighting. > Is there a rule -of-thumb that is used to work out a sensible overlap > between sequential windows?Read up on the overlap-add or overlap-save methods of fast convolution. Fast convolution uses FFTs to replace traditional convolution, producing precisely the same result with fewer calculation steps. Somehow, a myth arose that the data for fast convolution needs to be windowed as with more conventional uses of FFT. That myth is untrue. Unfortunately, it is possible to window the data without much harm, so the myth lives on. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������

Reply by ●November 12, 20082008-11-12

On Nov 12, 2:31 pm, Adam Chapman <adam.chap...@student.manchester.ac.uk> wrote:> ... > Is there a rule -of-thumb that is used to work out a sensible overlap > between sequential windows?Adam One window-overlap criterion used in signal reconstruction is the "Princen-Bradley condition' which you may find in: http://research.microsoft.com/~malvar/papers/tsp498.pdf in equation (6) on page 6. It is used to minimize artifacts in the reconstruction of signals undergoing time varying modifications. Another resource considering applications and the windows appropriate to them is: SPECTRAL AUDIO SIGNAL PROCESSING JULIUS O. SMITH III Center for Computer Research in Music and Acoustics (CCRMA) Department of Music, Stanford University, Stanford, California 94305 USA at: http://ccrma.stanford.edu/~jos/sasp/ The people who generated these references may be practicing some signal processing closer to your interest than most of the responses here so far have been. Dale B. Dalrymple http://dbdimages.com