hi, I am new to these group. I want to know why white noise floor is added to the autocorrelation function in the lpc analysis. what is the difference between lsp and lsf thanks in advance venugopal

# regarding white noise

Started by ●December 21, 2005

Reply by ●December 21, 20052005-12-21

venugopal wrote:> I want to know why white noise floor is added to the autocorrelation > function in the lpc analysis.This is done to avoid the ill behavior of the LPC analysis when the input is close to combination of the pure tones.> > what is the difference between lsp and lsfBasically the same thing. Just the matter of naming. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com

Reply by ●December 22, 20052005-12-22

> I want to know why white noise floor is added to the autocorrelation > function in the lpc analysis.I'm not an expert with LPC, but I know with adaptive dsp, you need to keep the inverse of the autocorrelation matrix stable. This is especially true with highly self-correlated signals. Injecting white noise into a signal before taking the autocorrelation is similar to adding a constant to the diagonals of the matrix. This increases the singular values of the matrix which decreases the condition number. A matrix with a lower condition number will invert more accurately (and quicker). However, this comes at the cost of biasing the solution. http://www.signalsguru.net

Reply by ●January 21, 20072007-01-21

>> I want to know why white noise floor is added to the autocorrelation >> function in the lpc analysis. > >I'm not an expert with LPC, but I know with adaptive dsp, you need to >keep the inverse of the autocorrelation matrix stable. This is >especially true with highly self-correlated signals. Injecting white >noise into a signal before taking the autocorrelation is similar to >adding a constant to the diagonals of the matrix. This increases the >singular values of the matrix which decreases the condition number. A >matrix with a lower condition number will invert more accurately (and >quicker). However, this comes at the cost of biasing the solution. > >http://www.signalsguru.net > >Hey, hope you still remember this conversation, since it is quite old. I need to know what exactly is the ill behaviour of LPC with pure tones. I understand that the autocorrelation matrix has a high condition number if dealing with pure tones ... still it doesn�t seem to produce a problem with LPC analysis and synthesize. I use sin(x) function, which is (should be) highly autocorrelated, but the condition number is "only" around 4*10^3. I think that this simply isn�t big enough to cause trouble during inversion. Do I need another function ?? Thank you in advanve