|
Hello everyone! I have a few doubts in the ACELP part of G.723.1 encoder. In G.723.1 ,ACELP is used for secondary residual quantisation in 5.3 kbps rate mode.The algebraic stucture of the codebook allows for fast search procedures, since the excitation vector contains only 4 non zero pulses. DOUBT 1. As per the recommendation, the Correlation C, to be maximised is given by (eq 30): C = (alpha0)d[m0] + (alpha1)d[m1] + (alpha2)d[m2] + (alpha3)d[m3] Now, why are the correlation terms ie d[] multiplied by alpha terms (signs of the pulses) ? if C happens to be the net correlation due to the 4 pulses,it should have been C= d[m0] + d[m1] + d[m2] + d[m3]. Then what is the significance of the alpha terms(signs of pulses)? DOUBT 2. Next, there is a simplification made prior to codebook search...(eq 32) A signal s[j] is defined as : s[j] = s[2j+1] = sign(d[2j]) if |d[2j]| > |d[2j+1]| s[j] = s[2j+1] = sign(d[2j+1]) otherwise This signal is used to redefine d[j], leading to simplification in further computations. Now, my question is what is the reasoning behind this eq ? surely there has to be a mathematical derivation for this. I would really appreciate if anyone can suggest a solution. Thank you Sameer Kibey, Pune. |
ACELP doubts
Started by ●April 4, 2002
