Elektron-Users

Thanks Niall - been busy but great to know they have it in focus.....

Regards - David.

Didn't really have any doubts about the testing methods but glad to know all the same.

I got the go-ahead from Elektron to post this.

I have concatenated two emails. The paragraph enclosed with brackets came from the second email.


Hello Niall,

We are very sorry you are not fully satisfied with our product.

Of course we take these matters seriously and are looking into improving matters. We doubt we can ever make it as tight as some people may like it because of technical difficulties (it is a matter of synchronising 2 DSPs and a CPU, a multi sample resolution is necessary for us to keep communication and audio rendering efficient), so a latency and a small "jitter" will always be necessary.

There is nothing wrong with your method of measuring the jitter, any fully working sound card will do. But remember that the difference in time between two trigs from different recordings can be approximately twice as big as the difference from the ideal timing.

[Say a sound should have started at sample position 22050, but in the first measurement it begins at 22060 and in the second measurement it begins at position 22040, when you look at them you see a difference of 20 positions, but none of the sounds were off by more than 10 positions from the ideal startpoint. I have not read the posts on elektron-users and therefore I don't know what analysis have been made from the measurements.]


Could you please send examples of patterns you feels are lagging behind to us for testing, preferably with information on what to listen for and how to approximate your setup?

best regards

Anders G?rder, developer at Elektron Music Machines.

I am glad to hear their reply.

If they look at the problem in detail, it seems at least possible they will spot a do-able way to reduce the maximum jitter.

Thanks for your efforts, everyone who is working on this...

What i would really like is for elektron to look at the jitter pattern people have measured, and give an explanation of whether it could be
-improved overall
or at least
-changed to reduce the 2ms gaps;
or if not, why it's impossible or a bad idea.

Niall have you proceeded to give them examples of the distinctive jitter pattern people have measured?

Hmmm - thanks for posting the Elektron Support response - personally the bit about DSP/CPU clocking points to a problem that may not be easily resolved I fear. Sounds like a compromise was made early on in the design that traded internal timing/event precision for DSP/Voice processing/rendering which I still feel is not really cool.

Niall - if you are communicating with Elektron directly - maybe they could provide an option similar to how Alesis dealt with these issues in the Andromeda A6 - they had CPU/Processing options in the system menu that let the user choose between (a) ultra hi-resolution audio rendering with slower VCA/Envelope attack precision or (b) lower resolution audio rendering with super fast VCA/Envelope attack precision for timing critical sounds/patches.

They could offer a similar option/trade off at the System Level - maybe in Classic/Pattern Mode (no param. locks) and by disabling internal effects as well they could prioritise timing precision.

Regards - David.

innerclock wrote:
Sounds like a compromise was made early on in the design that traded internal timing/event precision for DSP/Voice processing/rendering which I still feel is not really cool.
Regards - David.


I dont know what kind of DSP?s are used in the MD... but the machine is doing quite a lot of sounds with just 2 dsp?s...

We have very precise timing in the nord modulars but even with an expanded 8 dsp model i see problems to do all what the MD is doing...

I also heared that the MD is using some kind of compressed audioformat inside... ensuring a relatv higher sound quality for given samplerates and bandwidth... also the 12 bit thing in theire advertizing might be related to that...

I have no idea if that is reality or just gossip...but the MD sounds indeed very strong and the filters very smooth for a digital syntheziser...

realtime en and decoding procedures might be a reason for a certain unstability in the output of the soundengine... just a phenomen you only would be able to organize somehow but probably cant get rid of it totaly...

Elektrons mail is pretty clear about that the wobbeling is caused by something they just cant switch off...

whatever the real reasons are...it must be a structural thing and nothing that is related to only dsp load..other wise a singel track with a few events would behave better than 16 tracks..

But there seems to be no relation between the wobbeling and the dsp load..

So just switching down dsp use probably wouldnt change anything...

rudebop wrote:

But there seems to be no relation between the wobbling and the dsp load..

So just switching down dsp use probably wouldn?t change anything...

That is my fear also. If it is not DSP related - I wonder if a voice/step buffering option might be able to be added in the OS? That way - step/voice events could all be held and then fired at precise clock/tempo intervals rather than with the push/pull as it is now.

This would probably mean a sacrifice of 3 ms overall midi Clock Sync I/O lag in order to absorb the current event slop into a buffer. But this would be inconsequential if using the MD on its own and could be very simply compensated for when using the SPS-1 in a sync situation by using some sort of Master/Slave offset.

I would much rather a sequencer stay in tight time with a fixed offset than have a sloppy sequencer with no offset at all.

You can work with the first case - the second is hard work.

Maybe it could be an option in the SPS-1 OS?

Regards - David.

a 3 ms late machine? a electrontribe or what?...no thanks

I agree with David.

If I had to choose I'd rather have a drum-machine that has a rock solid internal timing, even if it's 3 ms late. You can find ways to compensate for that.

In any case it would be great if the user could have the option to decide which timing response to use.

That said I REALLY hope Elektron will be able to fix the issue without any kind of compromise.

PS. Hey David, check your Inbox when you have time.

rudebop wrote:
I dont know what kind of DSP?s are used in the MD... but the machine is doing quite a lot of sounds with just 2 dsp?s...


They're bog-standard, 56k Motorolas, IIRC. Not that I've looked inside, or anything!

rudebop wrote:

whatever the real reasons are...it must be a structural thing and nothing that is related to only dsp load..other wise a singel track with a few events would behave better than 16 tracks..



But it does, if my test was indicative.
Left channel is impulse with all other channels muted. Right channel is impulse with all other channels active, but mixed to zero:

http://syncretism.net/snd/niall_-_impulse_120_bpm.wav