On Nov 3, 2004, at 11:10 AM, Fernando wrote:
definitive application survey, I think. I didn't find them until after
I had done my experimentation.
I do have a few things to add to his reports, more about functionality
than applications. Some of this I mentioned before:
1. The Stepped section is really just like the Smooth section (which is
a VC lag processor), with some extra input circuitry. The differences
are that (1) the slew rate for a given RATE knob setting and VC input
is much faster, and (2) the SAMPLE input works as follows: it detects
rising edges (but see below), turns them into fixed (very short)
duration pulses, then inverts that signal to effectively feed into an
internal HOLD input (as on the smooth section). So the Stepped section
is normally in a hold state; you release it and let it slew very
quickly for a very short duration with a trigger pulse - effectively,
this steps it by an amount depending on the RATE knob and VC. The
slewing is clear looking at the signals on an oscilloscope.
2. John noted that if you patch the Smooth CYCLE output to Stepped
SAMPLE, you get two sample steps per Smooth oscillation period. I also
don't know exactly why this is, but there is nothing special about the
fact that the signal is coming from CYCLE. Evidently Stepped SAMPLE
detects not only rising edges, but also falling edges with a ~20v drop,
which is what CYCLE generates when it goes from +10v to -10v. It is
rather tricky to generate such a signal externally (processors are
nonlinear when the voltages get large, and don't fully add their
inputs), but it can be done, and Stepped SAMPLE again triggers on the
falling edges as well as the rising ones. This doesn't happen with
normal, 5v to 0v falling edges. Also, it's something peculiar to the
Stepped module; no other rising edge detectors I've tried (e.g. DSG,
sequencer clock) trigger on CYCLE falling edges. Presumably the
behavior has to do with the fact that the edges are being turned into
short pulses.
3. John also noted that CYCLE is normally low (on both Smooth and
Stepped). This is simply because it only goes high when OUTPUT goes
below about -1v (on my system, anyway). It goes low again when OUTPUT
goes above about 4v.
4. Beware the SSG diagrams in the Gold book, which mistakenly draw
CYCLE as an input - it is an output.
5. I thought I'd seen other posts puzzling over some CYCLE weirdness,
but I can't seem to dig them up at the moment. One thing that puzzled
me for a while was this: if you feed a square wave into Smooth INPUT,
and crank the RATE up enough to get triangle waves, CYCLE just sits at
-10v. But if you patch CYCLE to INPUT, it alternates between -10v and
10v. It's like CYCLE somehow alters its behavior based on the fact that
it's patched within the SSG. But the answer is simply given by #3 above
- an ordinary square wave does not go below 0v, but CYCLE does. There's
really nothing weird about CYCLE.
6. You can make a Random Voltage Generator even without a Noise Source!
John described how to turn the SSG into an RVG, with smooth, stepped,
and pulse random voltage outputs. This is also described in the Gold
book (page 5-39). This is an incredibly cool patch - if you fully
understand it, you have got the SSG down. However, one thing is that
that patch requires the S/H source from a Noise Source module (or a
Random Source, which includes a Noise Source internally). That signal
is a randomly modulated sawtooth wave. The idea is that every time you
sample it (unless two sequential times are very close together), you
get a random voltage. (Actually the Gold book mistakenly says to use
the S/H output, not S/H source - that will not work.) Well, it turns
out you can make an RVG even without a Noise Source! Just use a regular
sawtooth wave instead of the S/H source, say from a PCO. (A DSG would
also work.) Set the PCO to high range, with the frequency knob at 12
'clock. Now, this signal is not randomly modulated, so you might think
it would not work - you won't get a random voltage when you sample it.
But in fact, the overall system is chaotic; the samples you get are
actually random, because they are sampled at random intervals - because
the patch in fact works! It's circular - the samples are random,
because the patch output voltages are random, because the samples are
random... but it works. I haven't done the hairy math to ensure that
the system will never get locked into a repeating state, but the output
looks the same to me as an actual RVG. In dynamical terms, I believe
the cyclic states are unstable.
Note that if the sawtooth PCO rate is much faster than 12 o'clock, the
patch doesn't work - I think because the signal is changing too fast
for the Stepped section to lock onto. This makes sense given #1 above;
the Stepped module is actually slewing for a brief period, and not
actually sampling. Setting the PCO rate to 12 o'clock roughly matches
the S/H source signal's mean frequency, btw. Of course, this is really
only useful if you happen to have an SSG but no Noise Source... still,
I think it's cool that you can get your randomness for free, as it
were.
Bob Hearn
---------------------------------------------
Robert A. Hearn
http://www.swiss.ai.mit.edu/~bob/
> > A month or so ago I spent some time really grokking the SSG. IJohn's SSG posts that you mentioned pretty much constitute the
> think I
> > was able to completely satisfy myself on every point about its
> > behavior, even answering some unresolved questions from SMOG. The
> > module really does make sense; all of its apparent strangeness can
> be
> > explained.
>
> Please Bob, it would be a pleasure to read an extensive report!
>
> Fernando
definitive application survey, I think. I didn't find them until after
I had done my experimentation.
I do have a few things to add to his reports, more about functionality
than applications. Some of this I mentioned before:
1. The Stepped section is really just like the Smooth section (which is
a VC lag processor), with some extra input circuitry. The differences
are that (1) the slew rate for a given RATE knob setting and VC input
is much faster, and (2) the SAMPLE input works as follows: it detects
rising edges (but see below), turns them into fixed (very short)
duration pulses, then inverts that signal to effectively feed into an
internal HOLD input (as on the smooth section). So the Stepped section
is normally in a hold state; you release it and let it slew very
quickly for a very short duration with a trigger pulse - effectively,
this steps it by an amount depending on the RATE knob and VC. The
slewing is clear looking at the signals on an oscilloscope.
2. John noted that if you patch the Smooth CYCLE output to Stepped
SAMPLE, you get two sample steps per Smooth oscillation period. I also
don't know exactly why this is, but there is nothing special about the
fact that the signal is coming from CYCLE. Evidently Stepped SAMPLE
detects not only rising edges, but also falling edges with a ~20v drop,
which is what CYCLE generates when it goes from +10v to -10v. It is
rather tricky to generate such a signal externally (processors are
nonlinear when the voltages get large, and don't fully add their
inputs), but it can be done, and Stepped SAMPLE again triggers on the
falling edges as well as the rising ones. This doesn't happen with
normal, 5v to 0v falling edges. Also, it's something peculiar to the
Stepped module; no other rising edge detectors I've tried (e.g. DSG,
sequencer clock) trigger on CYCLE falling edges. Presumably the
behavior has to do with the fact that the edges are being turned into
short pulses.
3. John also noted that CYCLE is normally low (on both Smooth and
Stepped). This is simply because it only goes high when OUTPUT goes
below about -1v (on my system, anyway). It goes low again when OUTPUT
goes above about 4v.
4. Beware the SSG diagrams in the Gold book, which mistakenly draw
CYCLE as an input - it is an output.
5. I thought I'd seen other posts puzzling over some CYCLE weirdness,
but I can't seem to dig them up at the moment. One thing that puzzled
me for a while was this: if you feed a square wave into Smooth INPUT,
and crank the RATE up enough to get triangle waves, CYCLE just sits at
-10v. But if you patch CYCLE to INPUT, it alternates between -10v and
10v. It's like CYCLE somehow alters its behavior based on the fact that
it's patched within the SSG. But the answer is simply given by #3 above
- an ordinary square wave does not go below 0v, but CYCLE does. There's
really nothing weird about CYCLE.
6. You can make a Random Voltage Generator even without a Noise Source!
John described how to turn the SSG into an RVG, with smooth, stepped,
and pulse random voltage outputs. This is also described in the Gold
book (page 5-39). This is an incredibly cool patch - if you fully
understand it, you have got the SSG down. However, one thing is that
that patch requires the S/H source from a Noise Source module (or a
Random Source, which includes a Noise Source internally). That signal
is a randomly modulated sawtooth wave. The idea is that every time you
sample it (unless two sequential times are very close together), you
get a random voltage. (Actually the Gold book mistakenly says to use
the S/H output, not S/H source - that will not work.) Well, it turns
out you can make an RVG even without a Noise Source! Just use a regular
sawtooth wave instead of the S/H source, say from a PCO. (A DSG would
also work.) Set the PCO to high range, with the frequency knob at 12
'clock. Now, this signal is not randomly modulated, so you might think
it would not work - you won't get a random voltage when you sample it.
But in fact, the overall system is chaotic; the samples you get are
actually random, because they are sampled at random intervals - because
the patch in fact works! It's circular - the samples are random,
because the patch output voltages are random, because the samples are
random... but it works. I haven't done the hairy math to ensure that
the system will never get locked into a repeating state, but the output
looks the same to me as an actual RVG. In dynamical terms, I believe
the cyclic states are unstable.
Note that if the sawtooth PCO rate is much faster than 12 o'clock, the
patch doesn't work - I think because the signal is changing too fast
for the Stepped section to lock onto. This makes sense given #1 above;
the Stepped module is actually slewing for a brief period, and not
actually sampling. Setting the PCO rate to 12 o'clock roughly matches
the S/H source signal's mean frequency, btw. Of course, this is really
only useful if you happen to have an SSG but no Noise Source... still,
I think it's cool that you can get your randomness for free, as it
were.
Bob Hearn
---------------------------------------------
Robert A. Hearn
http://www.swiss.ai.mit.edu/~bob/