1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178

[The following note originally appeared on the emusicl mailing list. It is
reprinted here with the author's permission]
From xrjdm@FARSIDE.GSFC.NASA.GOV Wed Nov 23 11:26:39 1994
Date: Tue, 4 Oct 1994 15:09:23 0500
From: Joe McMahon <xrjdm@FARSIDE.GSFC.NASA.GOV>
Reply to: Electronic Music Discussion List <EMUSICL@AMERICAN.EDU>
To: Multiple recipients of list EMUSICL <EMUSICL@AMERICAN.EDU>
Subject: Automata: the longawaited summary
Back in August, I think, I promised to post a quick intro to cellular
automata and how they can be used as a soundgeneration tool. Since I'm
going to take a couple of different sources and sum them up with little or
no direct attribution, combined with my own opinions, I'll give everybody
my references *first* so they can delete the article and draw their own
conclusions if they so prefer.
The primary reference that got me started on all this is one in the CMJ:
Vol 14, No. 4, Winter 1990: "Digital Synthesis of Selfmodifying Waveforms
by Means of Cellular Automata" (Jacques Chareyon). Those who are already
familiar with automata may just skip to that article and forget about the
rest of this one.
Note: the article gives a mail address for M. Chareyon, but he did not
answer an inquiry about any available recordings using this technique in
1990.
So. Anyone still here? Good.
Cellular automata are a mathematical concept first introduced in the late
1940's. Generally speaking, a cellular automaton consists of a grid of
cells. Each cell may take on any of a number of values  binary automata
(cell on or cell off) are the most commonly studied. Each cell has a
neighborhood, defined more simply as other cells which influence its state.
The exact nature of this influence is defined by what are called transition
rules. The cellular automaton starts off with some cells in any of the
allowable states. for each "step" in the automaton's history, the
neighborhood of every cell is checked, and the state of the cell is
updated. All updates occur simultaneously.
The transition rule must describe the resulting state of a cell for every
possible configuration of other cells in the neighborhood. For large
numbers of states, the amount of memory required to hold the transition
rule becomes increasingly large, Therefore, some automata use what is known
as a "totalistic" rule. These rules simply sum the values of the cells in
the neighborhood and then assign a result on this basis. The resulting
tables are far smaller.
Many readers may already be familiar with John Horton Conway's game of
"Life". This is a twodimensional binary automaton with a totalistic rule.
This makes for a very small rule set:
i) If fewer than two filled cells (cells with value 1) surround a cell,
it becomes empty next generation.
ii) If more than three filled cells surround a cell, it becomes empty
next generation.
iii) If exactly three cells filled cells surround a cell, it becomes
filled on the next generation.
This corresponds to a totalistic rule set with a total of 8(21)+1 or 9
rules (one each for the sum values of 0 (no cells with a value) through 9
(all cells with a value) ).If the transition rule were represented as a
nontotalistic one, the rule set would need 2**8 or 256 entries. There are
many interesting totalistic automata, so giving up detailed description of
every nuance of the transitions to save memory space isn't a big sacrifice.
Interesting as two dimensional automata are, they really aren't terribly
useful for music making. There have been some experiments which have
attempted to use a twodimensional automaton to generate MIDI events 
synthesis at the note level, using :
Battista, T. and M. Giri, 1988. "Composizione Tramite Automi Cellulari."
Atti del VII Cooloquio di Informatica Musicale. Rome, Italy: Edizione Arti
Grafiche Ambrosini, pp. 181182.
Edgar, R. and J. Ryan, 1986. "LINA" Exhibition of the 1986 International
Computer Music Conference, San Francisco: Computer Music Association.
I have not heard any of the music from these efforts, so I certainly can't
pass any judgement on them. For the purposes of this summary, we'll just
look at onedimensional automata. These use a linear array of cells, with
the neighborhood generally being one or two cells on either side of each
cell.
(This is the type of automaton dealt with in M. Chareyon's article, which I
will be paraphrasing broadly hereafter).
M. Chareyon's automata are wavetables. A digitized signal is stored as a
linear array of numbers in memory. A totalistic rule is used to determine a
lookup value which indexes into an array containing the resulting value;
this is saved into a second array. After the first array is completely
processed, the roles of the two are swapped and the process is repeated.
The limiting factor in this process is the number of bits of resolution
being used to generate the sound. For a totalistic rule using a twocell
neighborhood and 12bit individual samples, we have 3*(2*12) = 12288
entries in the rule table. At 2 bytes each, this is 24K of storage. If we
go to 16bit sample resolution, we have 196608 entries at 2 bytes each for
a total of 393216 bytes, or 384K.
The key point of M. Charyeon's method is the use of small neighborhoods
with large numbers of cellular states. Since the computation of the new
wavetable is all table lookup, very complex transition rules can be
precomputed and loaded into the tables, allowing the synthesis to
essentially be a fast sumandlookup loop to calculate each new wavesample.
>From the article, it appears that M. Chareyon was able to produce 2 or 3
voices in realtime on a Mac II with a Digidesign Sound Accelerator board.
It seems that it would probably be possible to use an AV Mac to do it
without the board.
This LASy (Linear Automaton Synthesis) method is closely related to the
KarplusStrong pluckedstring algorithm, in that a wavesample is run
through an algorithm which recirculates the samples to "selfmodify" the
wave. In fact, a judicious choice of table entries allows one to very
simply simulate the KS algoritm directly.
So what are the sounds like? Some automata produce waveforms which quickly
"rampup" to complex spectra and then drop off quickly. Others move to a
steady state and then remain there. Yet others produce neverending and
unpredictable waveforms, whose harmonic content is constantly changing.
Obviously enough, the original wavesample can be obtained mathematically,
or by actual sampling and using LASy as a waveshaper. As M. Chareyon notes,
a quick estimate of the number of possible automata for a 2neighbor
totalistic rule using a 256entry wavetable with 12bit entries is
(2**12)**256 * (2**12)**(3*2**12) or about 10**4500 possible automata. Of
course, many, many of these would not be suitable for music (e.g., the 4096
automata in which all values go to one vlaue in one step, etc.); however,
the number of musically useful automata is still likely to be an immense
number.
M. Chareyon provides a number of examples of ways to fill out the rule
tables and a number of hints on creating wave tables  generally speaking,
one can create a function which is used to compute the values to be placed
into the table and then fill it so it can simply be loaded and used by the
basic algorithm. His experience in using LASy is that he manages
approximately 50% of the time to produce sounds with the desired
characteristics, and that about 10% of the remaining time he gets
unexpected but useful results which can be used as starting points for
further exploration.
Again, the important point is that the basic automaton uses wavesamples at
full resolution, calculating a new wavesample for each step of the
automaton; the next wavesample can be played while the new one is being
calculated. Because of the large number of states, mathematical tools for
the analysis of automata and the construction of automata with specifically
desired qualities require too much storage and compute time to make them
useful for LASy purposes.
Again, much of this article is paraphrased from M. Chareyon's article; I
take no credit for any of the work in this note. I'm just summarizing.
The following other articles were referenced by M. Chareyon's article:
Burks, A., ed. 1970. Essays on Cellular Automata. Champaign/Urbana, IL:
University of Illinois Press.
Chareyon, J. 1988a. "Sound Synthesis and Processing by Means of Linear
Cellular Automata." Proceedings of the 1988 Internation Computer Music
Conference. San Francisco: Computer Music Association.
Chareyon, J. 1988b. "Wavetable come Automa Cellulare: una Nuova Tecnica di
Sintesi." Atti del VII Colloquio di Informatica Musicale, Rome, Italy:
Edizioni Arti Grafiche Ambrosini, pp. 174177.
Farmer, D., T. Toffoli, and S. Wolfram, eds. 1984. Cellular Automata.
NorthHolland Physics Publishing. [One of the definitive works on cellular
automata  fairly heavy math, not a popular presentation  JM]
Gardner, M. 1970. "The Fantastic Combinations of John Conway's New Solitare
Game 'Life'". Scientific American 223(4) 120123. [A good introduction to
cellular automata, focusing on 'life' in specific. Useful intro if my
1paragraph summary of automata was confusing :)  JM]
 Joe M.

"At the end of the hour, we'll have information on the sedatives used by
the artists,,," (MST3K)
