Next: The lattice filter implementation
Up: Linear Prediction analysis
Previous: The LP spectrum
Figure 39 gives the amplitude response of a LP filter
for any single frequency. The sourcefilter excitations are flat, hence
the overall gain can be computed.
Durbin's algorithm gives the gain as .
This may also be calculated as:

Power in an impulse train separated by N samples is 1/N

Power is a Gaussian distributed variable with zero mean and unit
variance is 1.

Power in a random number uniformly distributed on [1,1] is:
Speech Vision Robotics group/Tony Robinson