M3 Maximum Polyphony?

FRETPICK · Post by **FRETPICK** » Mon Jun 15, 2009 12:42 am

120 voices. Is there any tech reason why it's not 128?

Is there any loss in fidelity?

Just asking....

Stefanus Wayanartha · Post by **Stefanus Wayanartha** » Mon Jun 15, 2009 7:41 am

maybe because the processing speed of the M3 main processor (chip)... if m3 has 128 voices polyphony, maybe it will cause some latency ... am i correct?? because i think the M3's system is similiar than a PC, and it's ancestor (OASYS) has a PC architecture inside it..

Rob Sherratt · Post by **Rob Sherratt** » Mon Jun 15, 2009 7:52 am

The EXB-RADIAS produces 24 voce polyphony. Given that both 120 (for EDS) and 24 are divisible by 12, the number of semitones in an octave, I would guess there is some relationship between the circuitry for polyphony and the circuitry for note generation.

X-Trade · Post by **X-Trade** » Mon Jun 15, 2009 1:15 pm

from what I can gather, the M3 isn't based on a PC system, however all microchip-based devices, even the tritons, can have comparisons drawn to a PC. because they are all technically 'computers'.

Particularly from the error messages reported in the past, it sounds like the M3 is based on an ARM CPU, like that in smartphones. ARM processors are good because they are relatively low-power devices.
The ARM probably handles the front end, voice loading, MIDI event processing, and the voice allocation etc, and sends instructions in a form technically similar to MIDI, then the sound is produced by DSP chips. this so far conforms to the standard model for most DSP based synthesizers.

I imagine the voice limit is based on two factors:
- maximum processing power
- memory allocation and efficiency issues

if the number of voices were an odd number, it may be more difficult for the allocation system to allocate voices when there is only one left, for example. also, if you are keeping an array or other datastructure with information on the state of each voice, it is always going to be more efficient for the system to work on a number which is or close to a power of two (i.e. a binary number close to its max limits). so pick the number closest to the most that you can squeze out of the M3, that isn't going to result in dropped or delayed notes i.e. latency.
also, its always best to leave a little headroom so that you don't get dropouts if the processor has a little hiccup

Rob Sherratt · Post by **Rob Sherratt** » Mon Jun 15, 2009 1:37 pm

The M3 and the Pa2x use the same Freescale Semiconductor ARM processor chip that was originally developed many high end cellphones. There is a daughter board coming out soon that allows you to receive mobile calls through your M3 and Pa2x when you are on stage, so all the audience can share in your conversations

FRETPICK · Post by **FRETPICK** » Mon Jun 15, 2009 4:51 pm

So what you guys are saying is that the polyphony is at an even number so it doesn't get confused within it's own allocation system.

I still don't understand why the polyphony isn't 128. EDS?

Sorry...I'm not that tech savy. However....

If what you are saying is that there is room left in the chip for errors & clips, that still to me doesn't make sense. The reason being is that other keyboards have a full polyphony range with 0 hanging or a 99.5% success rate.

X-Trade · Post by **X-Trade** » Mon Jun 15, 2009 6:09 pm

well, what do you mean by a 'full range'?
M3's full range is 120 notes. when I said it needs to leave some headroom, that is maybe two or even less than one voice's worth of processing time.

Korg calculated that they can optimally get 120 notes poly out of the chips, its not so simple as 'this chip does this many voices'. chips have a clock speed, so the number of voices depends on the number of processes (or processor cycles) you're doing for a voice. the more complex your synth, the less voices you'll get out of the same chip because you're doing more processing, so you can't meet that deadline to 'push' data out to the DACs otherwise. you have to increase the output buffer which requires more memory and results in a more sluggish response - latency.

lets say for example you have an 800mhz (so 800 thousand cycles a second) chip, and each voice is taking up roughly 150 thousand cycles (a bit high, I think). you're going for an output buffer of roughly a second (again, this is very poor latency), but then you can get a poly of 5 voices, with a bit of idle time left over...

anyway, its probably not 128 because it can not do 128, but 120 was probably the best next lowest option (lets say for example that you can get 122 voices out of it).

If what you are saying is that there is room left in the chip for errors & clips, that still to me doesn't make sense. The reason being is that other keyboards have a full polyphony range with 0 hanging or a 99.5% success rate.

also, how do you know that the other keyboards have a 'full polyphony range'? they have however much the manufacturer programmed into them.....

I'm not going to go any further with this because its a bit of a complicated area....

FRETPICK · Post by **FRETPICK** » Tue Jun 16, 2009 9:06 pm

^
Thanks.

If anyone else can shed some light please go right ahead.

Rob Sherratt · Post by **Rob Sherratt** » Wed Jun 17, 2009 12:27 am

FRETPICK wrote:120 voices. Is there any tech reason why it's not 128?

Yes the EDS chip inside the M3 was built with a hardware limit of 120 independent oscillators. If you need more polyphony, buy an M3-88 or an M3-73 and fit an extra M3M to it so you will then have 240 oscillator capability. You can play both M3M's simultaneously from the M3 keybed using the same controllers. No other manufacturer can do this.

FRETPICK wrote:Is there any loss in fidelity?

If you run out of polyphony because you have a very complex multi-track arrangement with more than 120 EDS samples required at any one time, or if you like playing glissandos with the sustain pedal held down so that more than 120 EDS samples are playing, then you will notice that some samples will drop out.

If you compare the M3 with 120 simultaneous EDS sample capability with the Tyros with 128 simultaneous sample capability, then the M3 performs better than the Tyros. This is because the M3 uses a more sophisticated algorithm to decide which samples to stop playing, for example it will cut fret noises etc before it cuts the main EDS sample. The Tyros just cuts the oldest note played.

FRETPICK · Post by **FRETPICK** » Wed Jun 17, 2009 5:42 pm

Thanks. Very educational.

Are keyboard prices very high in comparison to say a leading VSTi?

For example: Camel Audio's Alchemy.

rfoshaug · Post by **rfoshaug** » Thu Jun 18, 2009 7:27 am

This is an old Korg "thing".

The Triton family had 60 (or was it 62?) voices instead of the "logical" 64. I believe it was said at the time that the "missing" voices had something to do with processing sound input and sampling. Even my Korg Karma, which didn't have audio inputs, had the same circuitry and also had the same number of voices as the other Tritons, for commonality.

So the electronics still had 64 "slots" of processing power, but some of these were reserved for sound input instead of sound output. Maybe this is the case for M3 as well, so that it actually has 128 "slots" of computing power, but 8 of these are reserved for other duties?

kanthos · Post by **kanthos** » Thu Jun 18, 2009 12:34 pm

FRETPICK wrote:Thanks. Very educational.

Are keyboard prices very high in comparison to say a leading VSTi?

For example: Camel Audio's Alchemy.

VSTs are generally cheaper, unless you're getting a highly-sampled symphonic orchestra. This shouldn't be a surprise; keyboards cost more to produce because of the physical parts, while software is written once and then put onto cheap media.