MSP Sampling Tutorial 5: Waveshaping

From Cycling '74 Wiki
Revision as of 15:53, 28 June 2012 by gtaylorrtqenet (Talk | contribs)

Jump to: navigation, search

Click here to open the tutorial patch: Media:05sWaveshapingSynthesis.maxpat

In this tutorial, we'll look at a latent (but very useful) attribute of samples, which is that they can be used as lookup tables to transform the shape of other waveforms. This process is called waveshaping, and is used in synthesis to generate complex spectra from a sinusoidal input. It's also the basic signal processing technique behind many types of amplitude-dependent distortion, and can be used to model the non-linearities of different kinds of amplifier.


Using a stored wavetable as a transfer function

Take a look at the tutorial patcher. The basic sound-generating circuit in the upper-left of the patcher should look familiar, with one new object (lookup~) inserted into the chain. We have a cycle~ object going through an amplifier (*~) to the ezdac~.

  • Turn on the ezdac~ object and adjust the number box labeled amplitude. You should hear a sine wave at 220 Hz.

The new object in this signal chain at first seems to be doing nothing, and in terms of what we hear, it isn't... yet. The lookup~ object interprets a piece of sample memory stored in a buffer~ object as a transfer function, with the beginning of the sample used representing input values of -1 and the end of the sample used representing values of 1. Incoming values are scaled across this X axis, and the resulting audio comes from the corresponding values along the Y axis.

In our patch, the buffer~ named waveform is serving as a lookup table (or transfer function) for the incoming sine wave. When the cycle~ object generates a -1, for example, whatever sample value is at the beginning of the buffer~ comes out of the lookup~ object. When our cycle~ hits 1, the lookup~ object reads from the end of the buffer~ to find its outgoing sample.

  • Double-click the buffer~ object at the bottom of the tutorial patcher. Notice that the waveform loaded in is a simple ramp. Because our waveshape (the sample in the buffer~ is a linear ramp with the beginning of the sample at -1 and the end at 1, our cycle~ object sounds unchanged.

The lookup~ object takes three possible arguments: the first is the name of the buffer~ to use as its waveshape; the second and third are the start and end points (in samples to use within the buffer~ as the boundaries of the transfer function. Because we want our buffer~ to be exactly 512 samples long in our patcher, we created it with the argument of 10.66667 milliseconds. How did we get this number. At the right of the patcher, you can see the sampstoms~ object, which allows us to convert from samples to milliseconds.

The waveform~ object

Look at the graphical object at the top of the tutorial patcher. Notice that it contains the same shape that is loaded into our buffer~. The waveform~ object allows us to view, select regions of, and directly modify the contents of a buffer~ with a drawing tool.

  • Using your mouse, doodle in the waveform~ object's display. Notice how different shapes affect the output sound. Try drawing smooth curves, then jagged ones, then ones with lots of plateaus (straight horizontal lines). Notice that even slight variations in the shape have tremendous impact on the spectrum generated by the lookup~ object.

The waveform~ object is operating in draw mode, where you can literally modify a sample loaded into a buffer~ with your mouse. Other modes allow you to select regions, the boundaries of which can be used as Max messages for other objects.

Setting sample values with Max messages: peek~

  • Click on the button labeled A in the tutorial patcher. Our waveshape (and our sound) should return to normal.

The way we got the default waveshape in our tutorial patcher is through the logic controlled by the uzi below the button labeled A. The uzi object, you may recall, generates a lot of data instantly depending on its argument... the right outlet of the object generates a numeric ramp from 1 to its argument. We've subtracted 1 from that value to generate a stream of Max numbers from 0 to 511 when you click the button. These numbers are then sent to the middle and left inlets of a peek~ object... the left inlet receives the numbers unchanged; the middle inlet receives them after they have been scaled into the range of -1 to 1 (via the scale object).

The peek~ object allows us to manually set the sample values within a buffer~ via Max messages. The left inlet (which is "hot", and actually performs the operation) sets which sample we're changing; the middle inlet sets the value to change that sample to. In our case, the uzi object generates a stream of numbers that our entire buffer~ to an ascending ramp, e.g.

Sample Value

0 -1.

1 -0.996086

2 -0.992172

3 -0.988258

4 -0.984344

5 -0.980431


508 0.988258

509 0.992172

510 0.996086

511 1.

For the math geeks in the room

  • Click on the other button objects in the patcher (labeled B, C, and D). Notice how the waveshape in the buffer~ changes, and note how it affects the sound. Each waveshape seems to multiply the frequency of the sound generated by the cycle~ object. The logic at B makes our 220 Hz wave sound at 440, C at 660, and D at 880.

The equations that the expr objects are doing in these parts of the patch generate special types of transfer functions called Chebyshev polynomials. These functions are interesting in that they have the ability to transform sinusoidal input to different harmonic multiples. The four Chebyshev polynomials in our tutorial are:

y = x (uzi object A, leaves the input unchanged)

y = 2*x^2-1 (uzi object B, doubles the frequency)

y = 4*x^3-3*x (uzi object C, triples the frequency)

y = 8*x^4-8*x^3+1 (uzi object D, quadruples the frequency)

In practice, what they do looks like this:

Samplingchapter05a.png Our four Chebyshev polynomials and their effect on a cosine input

  • Click on the read messages in patcher area E. These will load 512-sample audio files into our waveshape buffer~. Notice their effect on the sound:

Samplingchapter05b.png Waveshaping through gtr512.aiff and blp512.aiff, respectively.

  • Now that you know a bit more about the expected behavior of the waveshaping distortion, return to drawing in the waveform~ object to see what kind of results you can achieve.


The lookup~ object allows you to use a buffer~ as a transfer function to perform a process called waveshaping on an input sound. Different shapes cause different distortions of the spectra and can create very complex timbres. The waveform~ object allows you to directly view and modify the contents of an MSP buffer~ using your mouse, and the peek~ object allows you to set sample values programmatically with Max messages. Some transfer functions (such as Chebyshev polynomials) have special properties when used as a waveshaping function on a sinusoidal input.

See Also

lookup~ - Transfer function lookup table

waveform~ - buffer~ viewer and editor

peek~ - Read and write sample values

sampstoms - Convert time from samples to milliseconds