As an addendum to the previous workshop, we're going to implement the lowpass filter indicator that we've used. A second order lowpass filter is not a very complex indicator, but it's still a bit more serious than its simple minded colleagues SMA, EMA, Ichimoku & Co. So it's a good exercise for indicator implementation.

When we're only interested in implementing an indicator from a ready
formula, we can skip all math and
start directly with the code. If
you're interested in the theory, get the books by **John Ehlers** (book
list) who introduced signal processing for traders. The Gaussian highpass filter formula below
is from his book. For avoiding math, skip the following section.

A lowpass filter suppresses the high frequency components in a data stream, a highpass filter suppresses the low frequency components. The higher the order of the filter, the sharper is the frequency cutoff, which is normally desired for indicators. The filter's formula usually expresses the filter gain - output divided by input - in Z-transform notation. Here's the gain of a second order Gaussian highpass filter that you can get from a digital filter design book:

*GainHP = Out / In = (1 - a/2)*^{2}* · (1 - 2 z*^{-1}* + z*^{-2}*)
/ (1 - (2 - 2a) z*^{-1}*
+ (1 - a)*^{2}* z*^{-2}*)
*

where ** a** is the filter coefficient and the operator

*GainLP = Out / In = 1 - (1 - a/2)*^{2}* · (1 - 2 z*^{-1}* + z*^{-2}*)
/ (1 - (2 - 2a) z*^{-1}*
+ (1 - a)*^{2}* z*^{-2}*)
*

Converting to nominator / denominator form for separating input and output:

*Out / In = (a - a*^{2}*/4 + a*^{2}*/2 z*^{-1}*
- (a - 3a*^{2}*/4) z*^{-2}*) /
(1 - (2 - 2a) z*^{-1}* + (1 - a)*^{2}*
z*^{-2}*) *

Cross multiplication:

*(1 - (2 - 2a) z ^{-1} + (1 - a)^{2} z^{-2}) ·
Out = (a - a*

Applying the **z** operators to the input and output data (1 ·
Out = Out[0], z^{-1} · Out = Out[1], etc):

*1 · Out[0] - (2 - 2a) · Out[1] + (1 - a) ^{2} · Out[2] = (a - a^{2}/4)
· In[0] + a^{2}/2 · In[1] - (a - 3a^{2}/4) · In[2] *

where *Out[0]* is the current output value, *Out[1]* is the
output delayed by one bar, and so on. The final result:

**Out[0] = (a - a ^{2}/4) · In[0] + a^{2}/2 · In[1] -
(a - 3a^{2}/4) · In[2] + (2 - 2a) · Out[1] - (1 - a)^{2} ·
Out[2]**

This is the formula that we can now implement as an indicator function.

With the series function that we already used in
the previous workshop, it's easy to implement the delayed data from the above
formula. The filter coefficient **a** is converted to a
more convenient time period: **a = 1/(1+Period)**. The higher the
**Period**, the smaller is **a** and the stronger is
the high frequency attenuation effect of the filter.

That's the implementation (**Workshop4a**):

var lowpass(vars In,int Period) { var a = 1./(1+Period); vars Out = series(In[0],3); return Out[0] = (a-0.25*a*a)*In[0] + 0.5*a*a*In[1] - (a-0.75*a*a)*In[2] + (2-2*a)*Out[1] - (1-a)*(1-a)*Out[2]; }

The first line converts the time period to the filter coefficient **a**. Mind the
dot of the '**1.**'. The decimal point tells the compiler that it's a real number. A
plain '**1**' would be an **int**. **1+Period** is also an **int**, so we
had an integer calculation that yields an integer result. Which will be always
**0** since the integer **1** divided by a higher integer is **0**. This
is one of the programming traps. If in doubt, put a decimal point to all numbers in any floating point formula
in your scripts.

For the output we're creating a new **series**, but since we only need delay by of up to 2 bars,
the series only needs a length of 3 elements. If we omitted the length parameter, the series had the length of the
lookback period, which would be a waste of resources. We also initialize the series to the input data for preventing
that it starts with 0.

Note that **Period** hs no effect on the amount of data. The
filter uses only the last 3 bars. The plot of the **lowpass** function is the same as in the
previous workshop, because the internal LowPass filter
has the same formula:

You can now look into the file **indicators.c** in the **
Source** folder that contains the code of many indicators, and check out how those indicators work.

- The
**Z-transform**is a conventient way to define and calculate digital filters. - Applying the Z transform to data is equivalent to taking a delayed element
from a
**series**. - The length of a series can be given in the second argument.
- Mind the difference of integer and floating point calculations. Use decimal points when needed.
- Even complex filters need only a few lines of code.