Diva Vs. Real Analog

[aciddose]

Are you serious?

the typical accuracy for a typical pot is going to be more in the range of 12-14 bits, depending on the pot.

sometimes even that isn't enough - see coarse vs. fine tune for an example. (or, less likely you know what i mean but also see multi-turn pots.)

floating point parameters in the 1.0 normalized range have an accuracy of 24 bits. that is definitely enough for any ordinary parameter.

the interesting thing though is that if i were designing a spec i would find it very difficult to settle on a range between those two. i would rather let the limitations imposed by the protocol dictate that than attempt to justify any choice myself. there will be a case where someone can justify a reason for having more.

that's why modern specs don't have limitations. see OSC for example.

[xh3rv]

Are you actually claiming you can tell the difference? I wasn't even aware that there was some kind of discussion over the quality of the smoothness of moving midi parameters.

I'd agree that ModWheel -> Filter isn't completely broken or anything, but there are lots of cases where more than 7 bits is critical.

For example, oscillator detuning. You can definitely hear changes in the rate of beating with very small tweaks here. In Diva it's easy to hear extreme beating effects with two triangles mixed, detuned 20 cents throughout the middle-ish octaves. It changes a whole lot with even just a few cents up or down.

It gets more complicated down the line. Carrier/modulator arrangements can be extremely sensitive to subtle detunings. Something like adding a little extra keyfollow to one oscillator will really change the way things sound up and down the keyboard, but it's a little trickier to dial this in. Using envelopes to drive large pitch modulations for percussion sounds can lead to nice, snappy transients but setting the depth such that they spike at the right frequencies is definitely fine-tunable.

In all these cases a mouse is capable. I think we all believe knobs really give a different perspective to be having this conversation - I really think it's the finer motor skills in finger muscles rather than shoulder/elbow/wrist for the mouse. So I'm really just trying to say, there are cases where I think the intuition of a mouse vs. 7bits vs. high-precision are all completely different, and it's nice to have options and different ways of exploring sound.

I've been using a Novation Nocturn with some Uhbik maps for fine-tuning controls, hope to have a good set of presets up from that soonish, and can honestly say there's absolutely no way some of the sounds would emerge from mouse or 7-bit tweaking (and plenty of them were completely or partially edited by mouse too - the Uhbik trick of clicking below a knob for fine-tuning is fantastic). It's a joyous thing to have a really fine-tunable knob, and technically productive.

[Gonga]

I've brought this up many times, but because of the discussion I'll say it again. There's no way to pitch bend a Moog Voyager Rack Mount Edition without very noticeable zippering at higher frequencies, and it doesn't matter what the resolution of the midi controller is. The synth just sees 127 steps, and with a wide bend the stepping sounds awful. I sold my Moog because of it. It is a major issue. All the U-He synths provide much smoother, more analog bends than my Voyager did. It's "all analog signal path" is worthless when the control signal is 127 steps.

Diva sounds more like a Model D than my Moog Voyager RME.

[bmrzycki]

With encoders, it's not a big deal, there are encoders that do more than 10000 pulses per revolution but you wouldn't want that, your hand isn't capable of getting reliable results from that kind of resolution. The issue with pots is that 16-bit ADC is too expensive to use on a MIDI controller.

I see you and I have similar obsessions -- I have done a lot of research to find high-quality encoders and came across the same results. To be honest I think 8-9 bit resolution is fine enough. Most people forget that adding that one extra bit doubles the data-size

A big factor to consider here is that most MIDI hardware can't handle large amounts of 14-bit data in real time. Not to mention that MIDI over USB or LAN is transferred at the same rate as MIDI via DIN cable (31.25 kBps).

That's only true if you convert the signal back to the 5 pin DIN plug. The USB-MIDI spec doesn't specify anywhere a maximum speed limit -- most implementations of it simply assume 38400 is good enough. Same is true for RTP MIDI over ethernet, as long as it stays in a medium that expects very, very fast MIDI messages. You're absolutely right though, if you talk to hardware via the MIDI din plug you have to clock and buffer the hell out of the bytestream, a PITA.

I plan to investigate this in more detail further with the Rasperry Pi boards. I'll keep you informed of my progress.

[bmrzycki]

the typical accuracy for a typical pot is going to be more in the range of 12-14 bits, depending on the pot.

I would very much like to see the technical specifications for devices you are referencing to have 12-14 bits of resolution on average. The only devices I found were that came close at 10-bit are magnetic encoders such as:
http://www.sherbornesensors.com/Rotary_ ... ncoders/73

Justin's comments are spot-on IMO, esp the one about the human hand's native resolution. You'd need a very large knob to accurately convey such a fine granularity to the encoder anyway.

[bmrzycki]

14-bit resolution MIDI messages exist, it's a shame so few programs understand them. They're called LSB/MSB in the updated midi spec released years ago.

ftfy.

No, I meant exactly what I said. They're still listed on the midi.org site quite clearly:
http://www.midi.org/techspecs/midimessages.php
Table 3 - Control Change Messages (Data Bytes), bytes 98-101 decimal.
LSB/MSB CC pairings are nothing like NRPNS/RPNs. One big difference is there are only 32 of them (CC 0-31 is the MSB, CC 32-63 is the corresponding LSB). On top of that most have predefined meanings like Modwheel, Pan, Foot Pedal, etc. There are 16,384 NRPNs available to every manufacturer -- by definition they are not pre-set (Non-Registered Parameter Number). They also look completely different from a bytestream perspective.

The only softsynth I've ever seen use them is Vember Audio's Surge.

I am very interested to see what Urs does with midi in his synths and effects though.

[kmonkey]

72 pages...something about midi bit resolution..Someone is beating dead horse...isn't it..

[SadPuppyBlues]

isn't it..

Someone's from Britain, innit?

[aciddose]

the typical accuracy for a typical pot is going to be more in the range of 12-14 bits, depending on the pot.

I would very much like to see the technical specifications for devices you are referencing to have 12-14 bits of resolution on average.

you don't need an unusually large knob, but it definitely has a limit to it's lower bound for a given accuracy. 12 = 4096 of course.

in order to test this you'd want to set up a ADC with enough bits and accuracy and sample from the pot in some way designed to minimize noise.

the usual method is to sample the rate of charge/discharge of a capacitor with current fed through the pot. there are two problems with this. or three maybe. first you have temperature sensitivity. you also have noise generated by the resistor, although that is filtered by the capacitor. then you have non-linearity of the whole system which can interact with every component and that all gets awfully complicated.

i know for a fact i can select any bit in the 9-bit (edit, it's actually an 8-bit range. cheap POS.) range of the pitch bender on my alpha juno right here...

that thing is not more accurate than a typical 1" knob into a 10k carbon pot with a low noise adc.

i'll tell you what i'll do. i'm interested in verifying this now.

i plugged my cheap meter into my function generator and turned off the waveform. i turned on the dc offset. using a 1cm (smaller than your pinky!) knob i can easily adjust from -13.00 to +13.00 on my digital meter with exact selection of any millivolt value. what does this tell us? well, we know that there is a range of 2600 there. that's ~11.3443 bits. it also tells us that my dinky $40 meter probably has 11-bit accuracy as i'm in "20v" range mode. so the most we can test with this method is 12-bit (+1 thanks to sign.)

now what could my 40 dollar mil-spec carbon pot with a 1" knob do?

accuracy of the human hand? have you ever tuned the speed on a turn-table? that's at least twice if not four times more accuracy required than what i was doing to produce 11 bit with the same size knob control. that would mean that the human hand is in fact capable of tuning to at least ~13 bits of angular precision with a 1-cm knob.

an average rmse for a surgeon's hand is 0.08mm. say we have a slide-potentiometer 5cm long. that's 500 / 0.08 = 6250. so linear position is fairly accurate to ~12.6 bits real-time for these guys over a 5cm area.

you have to take into account though that the major component of that error signal is slow dc drift/slew, jerks and a ~10hz sinusoid wobble.

we can control drift when we're fine-tuning a rotation. jerks are the major issue. wobble or "physiological tremor" is the easiest to control for as you simply have to time when you let go of the knob. once you let go, assuming there is no coincidental jerk motion occurring, it doesn't move from where you set it.

[Urs]

Well, getting practical here, we're messing with Atmel microcontrollers that have built-in 10 bit ADCs. So we can probably get like 8 usable bits from pots, unless we go higher quality converters.

Rotary encoders with too high resolution are as unsuitable as ones with too low resolution.

I think the ideal solution are high res pots for each parameter and an encoder to fine control the last selected parameter. Much like the value encoder of the Little Phatty. Knobs for coarse settings, encoder for the small details.

On a positive note, we pretty much researched all technologies and suppliers for building a Diva controller. The microcontroller is working for most of what we want to do, so we might be seeing something soon. Can't wait to have a hands-on Diva in front of me

Cheers,

Urs

[Howard]

What you're saying here is that you can actually hear the difference between a filter's cutoff set at say 86, and 87, and that even when these changes occur in quick succession (i.e. 30 times a second), you can hear the jumps from 86->87->88 etc. Are you really saying this?

I'm sure I'm not alone in claiming "of course I can!"

...would it have been possible to control a parameter on two old analog synths with one value inverted in the past? No way!

Way.

[bmrzycki]

I plugged my cheap meter into my function generator and turned off the waveform. i turned on the dc offset. using a 1cm (smaller than your pinky!) knob i can easily adjust from -13.00 to +13.00 on my digital meter with exact selection of any millivolt value. what does this tell us? well, we know that there is a range of 2600 there. that's ~11.3443 bits. it also tells us that my dinky $40 meter probably has 11-bit accuracy as i'm in "20v" range mode. so the most we can test with this method is 12-bit (+1 thanks to sign.)

An interesting test, thank you for trying it and posting your results.

Part tolerances between parts matter considerably and is why manufacturers list them on spec sheets. This is why I asked you for spec sheets for parts that led you to believe your original statement about average pots having 12-14 bits of resolution. Potentiometers are essentially variable resistors and it's clear the ADC (voltage meter) in your tester is a decent good one, likely much too expensive to attach to an entire bank of potentiometers.

As Urs mentioned above, practical engineering of sourced parts from the usual players (Atmel and others) will more likely yield 8-9 bits of usable resolution.

[Gonga]

72 pages...something about midi bit resolution..Someone is beating dead horse...isn't it..

I would ask why you're beating this poor horse? Why not contribute something pertaining directly to your craft instead? Isn't that the purpose of the forum?

[xh3rv]

I think the ideal solution are high res pots for each parameter and an encoder to fine control the last selected parameter. Much like the value encoder of the Little Phatty. Knobs for coarse settings, encoder for the small details.

Random thought (well, inspired by a calculator's mem assign and recall) - would it make sense to have a toggle-able locked assignment on the encoder? It might be nice to have that assignment stay static while tuning against other tweaks, or for a quick recall, or for a quick way to flip between two fine-tuning parameters.

[el-bo (formerly ebow)]

you don't need an unusually large knob

.... ......... .......... .................... .................