power spectral density via LPC

Hello,

I have question about LPC spectrum. I would like to compute LPC of given signal and compare it with PSD

x=loadbin('ma004009.ils');
fs=16e3;
tt=(0:slen-1)./fs ;
 slen=length(x);
 NFFT=slen; 
y=fft(x3,NFFT);
Pyy=y.*conj(y);      
Pyy=10*log10(Pyy); % PSD of signal

LPC spectrum

[en,a]=lpc(x,10);
 N2=fft(en,NFFT);
 Puu=N2.*conj(N2)/NFFT; 
 Puu=10*log10(Puu);

this gives me PSD of signal via LPC but the curve is upside up. why? And how to fix it? Thanks for any ideas.

Tomas

[BAPPA MUKHERJEE]

what do you mean by slen

Answer 1

[Wayne King]

You want to multiply by -1 because its 1/Puu. For example:

load mtlb;
x = mtlb.*hamming(length(mtlb));
a = lpc(x,12);
NFFT = 2^nextpow2(length(mtlb));
lpcDft = fft(a,NFFT);
lpcDft = lpcDft(1:NFFT/2+1);
plot(-20*log10(abs(lpcDft)))