Overdrive.txt
27-sep-2003/FK

Conceptually, the overdrive FX consists of:

Interpolation section
    - upsampler (zero injector)
    - a pre-emphasis filter
    - interpolation filter (anti-aliasing filter)

Transfer section
    - transfer function

Decimation section
    - decimation filter (anti-aliasing filter)
    - de-emphasis filter
    - downsampler

The interpolation and decimation sections can be integrated.


S-rate samples: ...   x-9 x-8 x-7 x-6 x-5 x-4 x-3 x-2 x-1 x0

After zero-injection
0  0  0 x-3  0  0  0 x-2  0  0  0 x-1  0  0  0 x0

Assume a FIR of 13 weights (which is much to small) for interpolation

       0  0  0 x-3  0  0  0 x-2  0  0  0 x-1  0  0  0 x0
y0=            w12           w8           w4          w0
y1=            w11           w7           w3
y2=            w10           w6           w2
y3=             w9           w5           w1

or

y0 = (w12 * x[-3]) + (w8 * x[-2]) + (w4 *x[-1]) + (w0 * x[0])
y1 = (w11 * x[-3]) + (w7 * x[-2]) + (w3 *x[-1])
y2 = (w10 * x[-3]) + (w6 * x[-2]) + (w2 *x[-1])
y3 = ( w9 * x[-3]) + (w5 * x[-2]) + (w1 *x[-1])

y4 = (w12 * x[-4]) + (w8 * x[-3]) + (w4 *x[-2]) + (w0 * x[-1])
y5 = (w11 * x[-4]) + (w7 * x[-3]) + (w3 *x[-2])
y6 = (w10 * x[-4]) + (w6 * x[-3]) + (w2 *x[-2])
y7 = ( w9 * x[-4]) + (w5 * x[-3]) + (w1 *x[-2])

Generalise this to interpolation rate R=4.

For each input sample x[i], generate R output samples:
y[i+0], y[i+1], ..., y[i+r]
where r = R-1.

y[i+0] = (w[0R-0] * x[i]) + (w[1R-0] * x[i-1]) + (w[2R-0] * x[i-2]) ... 
y[i+1] = (w[0R-1] * x[i]) + (w[1R-1] * x[i-1]) + (w[2R-1] * x[i-2]) ...
y[i+2] = (w[0R-2] * x[i]) + (w[1R-2] * x[i-1]) + (w[2R-2] * x[i-2]) ...
...
y[i+r] = (w[0R-r] * x[i]) + (w[1R-r] * x[i-1]) + (w[2R-r] * x[i-2]) ...

where w[] indices that are undefined are zero and therefore can be eliminated
from the expression.

These outputs are passed through the transfer section yielding:

z0 = t(y0)
z1 = t(y1)
...
z7 = t(y7)

Decimation then

      ... z7  z6  z5  z4  z3  z2  z1  z0
v0 = w0*z0 + w1*z1 + w2*z2 + w3*z3 + w4*z4 + ... + wn*zn
v1 = w0*z4 + w1*z5 + w2*z6 + w3*z7 + w4*z8 + ... + wn*z(n+4)

Generalizes to system output

v[i+0] = w[0] * z[i-0-0R] + w[1] * z[i-1-0R] + ... + w[n] * z[i-n-0R]
v[i+1] = w[0] * z[i-0-1R] + w[1] * z[i-1-1R] + ... + w[n] * z[i-n-1R]

Simply skip R samples before convolving the next output sample.

------------------------
Why can we not insert the zeros after the transfer function?
Because the harmonics will be present as aliases.