Got clicks? Listen to the voice:

"Many times clicks are caused by very fast attacks of 0.5 msec, setting attacks to 7 to 10 msec helps. Very fast decays of 0.5 msec can also cause a click when the note stops, decays of 50 msec are fast enough.
  

Big name for a basic rule. Most modules modulation input limit is at +/-64. If you reach the limit, the modulation signal will stay at 64. Mind you, there are exceptions.

The question on the mailing list was: why does my synth always sound different after loading a patch or after the G2 recalculates. Rob Hordijk gave the answer:

It seems like the G2 has a limit of 60 MIDI Send Modules per slot. But you can work around that limit by placing more MIDI modules in another slot and routing them to the first one.

MIDI can send about 1000 informations per second. Having about 250 MIDI modules in all four slots (4x60) would slow down each MIDI module to 4 informations per second.

Another nice thing you can do is to make a midi-controller (by placing a bunch of controller send modules) assign those to your favourite hard/software synth and.... assign morph-groups! ALL your synths (hard or software) now can have morph-groups!

If your synth doesn't have that already (like Cameleon 5000): all your synths can now have -via MIDI - modulations in the audio range, like cross-modulation.

You've got to discriminate between MIDI that sent out physically or internally. Channels 1-16 and "This" refer to the physical MIDI output of your G2. "This" is a shortcut to the channel the slot you're in is sending/receiving. If you try to send on "this" and receive on "this" in the same patch, nothing happens because "this" refers to the channel of the external midi connection.

"Slot A-D" are internal MIDI channels. You can compare this to the other internal routings.

The full modulation range spans from -64 to +64. A bipolar signal happens to span this range, from -64 to +64. But as a unipolar signal spans a range of 0 to +64 it is pretty logical that the modulation depth is not the full range but only half the range. So I guess that your real question is why does the LFO output not change to a range of 0 to +128 when set to unipolar and why some modulation inputs don't accept modulation signals out of the -64/+64 range. Which actually is rather a good question.
First the 'philosophical' question of why a unipolar signal has 'half the range' of a bipolar signal. It is a choice a designer of a system has to make to do it that way, and it is actually very common to do it that way. Basically it has to do with the fact that a modulation signal is always relative to some other value. E.g. a basic pitch or filter cutoff is set as a reference and then the effect modulation is applied in relation to that reference. Some things easily allow for bipolar modulations, like how a lesie type vibrato sweeps around a pitch. A leslie makes a steady pitch actually alternatingly lower and higher, a typical example of bipolar modulation. In contrast adding some vibrato to a guitar string by bending the string by a finger will only shift the pitch up, so this is a typical example of unipolar modulation. So, the basic question is does the modulation have to change some reference signal in only one direction or into two directions symmetrical around a reference signal, e.g. a pitch played by a key on the keyboard.
The reference signal is modulated between two extremes, the upper limit and the lower limit. Now in practice it is very often the case that when the lower limit is actually the reference signal (so unipolar modulation) the upper limit is still the same as when it would be bipolar modulation.
What you do in your patch is actually using the lowest pitch an osc can do as the reference pitch. But what you do is a 'special case'. Of course you can do it this way, but in many cases it is more practical to use e.g. the played note as the reference pitch to modulate around. In such a case you would find that if e.g. the modulation is set to a depth to sweep an octave (e.g. with a squarewave lfo to do 'octave popping') the changing from bipolar (one octave down <> one octave up) to unipolar (original pitch <> one octave up) is quite convenient, as it creates two different effects. If the unipolar setting would shift everything up, the effect would be that bipolar would be 'one octave down <> one octave up', but the unipolar 'original <> two octaves up', which would be the same as playing the bipolar modulation by a key just one octave higher on the keyboard. So in essence still be the same musical effect.
When a filter is modulated it is often modulated in a unipolar way, as a bipolar modulation can suppress the overall loudness when the lower limit of bipolar modulation would push the cutoff far below the played pitch. The modulation inputs on the filters are quite sensitive, twice as sensitive as on other modules (they go from 0% to 200%). With filters it is actually quite nice that the amplitude of an unipolar lfo signal is halved, as the filter modulation input sensitivity is double. They compensate each other nicely and make sure that the modulation is positive only in reference to the cutoff of the filter. It just works out nicely that way.
And now the question that touches the heart of your problem, why are there some modulation inputs that do not accept signals outside the -64/+64 range, in fact just clip the modulation at these boundaries. While there are other modulation inputs that can accept signals between -256 and +256. This has to do with how the modulation is handled internally within a module. Sometimes a modulation signal needs to be converted first, e.g. an osc is internally always linear and needs some exp/lin conversion of the pitch input signal. In some modules this is done by the calculation of a certain formula, in other modules by using a lookup tabel with values (and interpolate in the table if necessary). Tables have a fixed length and won't accept signals outside a certain modulation range that is fixed by the table length. Formulas can have intermediate calculation results that are outside a certain range, and so can set some limit to a modulation range as well. And on some modulation inputs the signals can be used straight away, so there is basically no range limit to the modulation input signal.
Now the pitch modulation input can basically modulate four types of scales, the scales that are set by the scale type button next to the coarse pitch knob. In the future there might be even more types of scales implemented, e.g. a Werckmeister scale. The limit on the modulation range might have to do with this, I don't know. Anyway, it might be the case that there is the eternal dilemma that facilitating for one option might put a limit to another option. Which means that on any system there must be compromises. The common sense is not to go for super elegant solutions, but rather use the 'when it works, it works' approach. Sometimes it might turn out to be elegant and other times not. Still, when it works, it works. In the end it all depends on how much time one wants to spend on improving the elegance of an already working solution.

The G2 has one single CV input, the Pedal input. Plus one input for a gate, the Sustain input. And for CV out there are of course the MIDI CC# out modules that will work with any proper Midi->CV box.

No CV to an audio input, they are AC coupled. But, I have a lot of fun with a Doepfer Theremin module connected to the G2 Expression Pedal input. Works like a charm! I use a simple 6.3mm mono jack with sleeve to earth and tip to the analog CV. The Pedal input accepts voltages between 0 and +5 V. But a proper Pedal input is protected to negative and overvoltages, as just by accident anything can be plugged in and it should survive. 220 is of course asking too much. But the Pedal inputs of both the NM and the G2 accept LFO and CV signals from ± 15V analog systems. I don't know if this is actually in the official specs, but I used these signals and the synths survived.

The Yamaha breathcontroller cannot be used directly, as the G2 Pedal input accepts a 0V to +5V signal. So, it would only read a 0 value. Also, the BC3 needs 12V power. A little bit of electronics with a 12V adapter would be required.

I have made me a sequenzer in the g2 for my E 64 Sampler. This is dead evil , you can send of course note on, note offs and the note numbers as all sequenzers do but the fun starts when sending CC to controll the Emi filters and the sampling start points, i use this on orchestral samples, a lot of fun!

There are some magic numbers for DSP percentage and memory percentage, these numbers are 100%, 50%, 33.3%, 25%, 20%, 16.6%, 14.2%, 12.5%, etc. The reason why these numbers are so important is as these are the points where the number of voices is changed. If you have 4 maximum voices at 51% you will immediately have 8 maximum voices at 49% and so on.

Sometimes tricks are necessary, e.g. when having to use the keyboard module, which now has five outputs, one must keep in mind that it uses no dsp but almost 4% of memory. If this is only to get a velocity value it is much wiser to use a constant module and assign the velocity morph to this constant module. Then dsp is also zero but memory is only .8%, a gain of 3.1% that can just pull the patch under e.g. the 33.3% memory and give three extra voices. I think that I can safely state that the rule of thumb will be to always use the velocity morph and never use the keyboard module for only
velocity. The keyboard gate is not necessary as the envelopes now have their keyboard gate signal built in. So, the keyboard module should only be used for special cases, when the note signal without the pitchbend or the pitchbend data are needed. And these are also made available by morphs, so morphs always have the priority.

An unexpanded G2 has got 4 (expanded 8) chips. If one voice uses 100% of one chip you have 4 (8) voices. When one voice uses 50% you have 8 voices. One voice uses 33% = 12 voices.

If the FX area uses 100% of one chip and one voice uses 100% of another one, you will have 3 voices (7). Same calculation with memory. Remember that one voice can not use more than one chip.
You know how many voices you've got by looking at the Patch Load and Memory Displays, dividing 100 with the higher number and multiplying the figure before the dot with 4 (8).

If the patch load is 26 and the memory 23 you'd calculate 100/26, that's 3.84, resulting in 12 voices.

To save memory, choose modules with fewer outputs. Use morphs when possible.