Windows 7's Midi OUT jitter disaster

[sensing]

Midi is basicaly two wires transporting square waves over a bandwith equivalent to an oldschool modem.
Me thinks that should be possible with spdif\audio if
Proof is in the eating though.

[Haven]

Originally Posted by phutureboy

I have read all this already but as far as I can understand all this is about monophonic MIDI. My question is about achieving polyphonic MIDI + CC commands, etc...
bil_g says he "play chords, twist knobs, and wank on the wheels", "I'm using my synths just like I would with any USB, Firewire, or PCI MIDI interface". I assume this is simultaneusouly otherwise there is no sense in saying what he says. And I doubt you can transmit all polyphonic note on/note off, CC commands, pitch bend and wheel messages at the same time via just 1 cable coming from the ES-4 soldiered to a MIDI plug like they are here : CV to MIDI & DINsync connectors - a set on Flickr
So I guess there is something else that he did that is missing in his description. Reason why I am asking to describe the full path and connectors to achieve this. That'd be very helpful...

MIDI is a serial protocol. That means all the messages get sent one after another. Even the bits that make up each byte (or word) get sent one after another. It only uses two connectors between devices. One is ground, the other is a binary signal. Polyphony is just multiple note ons and offs lined up down a single wire. Same with all MIDI messages, just a string of 1's and 0's down a single wire in order.

If the audio system of you DAW could generate that stream of serial data and send it to an audio interface that would then convert it to the voltages MIDI expects it would be sample accurate up to the point of entering your external MIDI device.

THAT is how Expert Sleepers does audio to MIDI. Polyphonic MIDI too of course, since there is no difference in the transmitting hardware between playing three notes and playing one.

Expert Sleepers uses a plug in within your DAW that has its audio routed to the ES interface. An ES interface that supports MIDI will use either ADAT or SPDIF to connect to your audio interface. The ES interface then sends out the MIDI via a ES Gate expander. You will need to build an adapter like that shown in this thread to patch in a MIDI DIN connector.

The ES interfaces and plugins originally were designed to output sample accurate CV, gate and clocks. MIDI was added later. Aside from misunderstanding MIDI communication this is why some people are confused as they were not aware of the recent developments in the hardware and software.

[bil_g]

phutureboy,

What haven and sensing said. Damn, I type slow. Haven gave more detail than what I was about to post, anyway.

Just want to add that it looks like Expert Sleepers will be offering pre-made MIDI/DINsync adapter cables, soon. You can see them here.

Also worth mentioning...I actually use it at 44.1khz all the time. Others have had success, too. You can see my results here.

[phutureboy]

Well, thanks for the replies you all. I am really amazed all the data flow of a polyphonic play can travel via 1 mono jack only !
Now I wonder why they decided to use 5-pin plugs for the MIDI standard instead of mono jacks when MIDI was invented...

And what is the difference of use between the '2 jacks for 1 MIDI plug' cable and the '1 jack for 1 MIDI plug' cable ?

[nms]

Originally Posted by Resonant Serpent

It is a complete myth that changing services to background with give you better performance. One of the programmers on the Sonar forum gave a detailed account a few years back on how this actually harms the timing of your DAW, so don't do it. Check that your not experiencing

Anyone else care to weigh in here? This is the first I ever heard of this.

[valis]

The tweak was to set processing priority to Background processes (instead of foreground application) to insure that your ASIO driver stack was still being served, and this dates back to the single core P3 era. Completely irrelevant in modern computing and the Sonar programmer is correct, background process priority will harm the foreground app (your DAW). Fwiw I had dual P3's, Dual P4's (Xeons) and never had to use this tweak on those machines, while I did on single socket machines of that era...but this was 10+ years ago! For any machine from the Core1 Duo era on I would just ignore this setting and leave it to foreground apps.

[Jauqq]

Originally Posted by bil_g

Also worth mentioning...I actually use it at 44.1khz all the time. Others have had success, too. You can see my results here.

Really exciting results!!
Can I ask how did you test and check each of your synths jitter values? I'm guessing you recorded each synth to audio, then zoomed in on the transients in a sample editor?

Also which audio interface are you using to send SPDIF to the ES4?

Thanks

[bil_g]

Originally Posted by Jauqq

Really exciting results!!
Can I ask how did you test and check each of your synths jitter values? I'm guessing you recorded each synth to audio, then zoomed in on the transients in a sample editor?

Correct, but I was able to do everything within Sonar. I used two sequences of 16ths(50% length) separated by a few measures at 120bpm.

Also which audio interface are you using to send SPDIF to the ES4?

Thanks

I was using a Fireface 400. I've used it with a Motu 2408mk2, as well.

Originally Posted by Dogboy73

Good old Black Viper Black Viper is to Windows what Jesus is to Christianity! They guy is bloody God send. I've used his tweaks for years in XP to tune up & strip down all the crap that MS lumps in there. Good stuff Of course the Apple crowd are probably standing on the sidelines laughing their arses off at all this stuff. Okay, you got us. We know Mac is thought to be better at this stuff. But we just like getting under the hood I think. Like having a car & being able to work on it yourself. Very rewarding. Apple users have to take theirs into the Apple service centre if they get any problems (I should stress I in NO WAY intend to start a Mac V PC debate. IMO these are always utterly, utterly pointless. Black & white. Chalk & cheese. Different but both have equal merrit).

haha. I like this post. Also it interests me because I remember when folks mocked me and said that Windows 7 didn't need any tweaking or optimizing for DAW use. I had a feeling Windows 7 tweaks and optimizations would come about, and here they are (and have been around for a while!).

I do kind of enjoy tweaking sometimes, just for aesthetic reasons. But it is kind of a waste of time... like spending 30 minutes to choose a desktop background haha...

[remote337]

USB on XP for me with no issues whatsoever using an MOTU 8x8.

Not sure about 7 though which is why I stayed on XP.

[TS-12]

Switching priority from "Background Services" doesn't fix this either, as someone recommended to try.

[BaconTastesGood]

After dealing with so much frustration with MIDI timing I decided to run one final test with my ESI M8U XL on Win7/64 and using Reaper 64. I didn't do any of the crazy black viper or whatever configuration, but I did shut down all background tasks that were non-essential (basically anything in my tray). This is a Mac Pro (quad core w/ HT, 16GB RAM, 2.8GHz) booted into OS X (not using Boot Camp or Parallels).

I programmed in 50% gate, 1/32nd notes, triggering my TX81Z FM Hi Hat patch, and recorded audio (128 sample buffers, 24-bit, 48KHz, Mackie 1640i ASIO).

I then measured sample deltas between consecutive 32nd note hits -- at 120bpm there should be 3000 samples between each hit.

One measure resulted in some pretty grim data. Up to 204 samples off between hits and up to 127 samples off the grid. Average 66 samples off the grid.

Note-to-note variance: up to 4.25ms
Grid variance: up to 2.65ms (averaging around 1.3ms)

I was hoping my RS7K would be more accurate. And it IS in general, but it has an alarming accumulated drift between notes. Less than 14 samples of hit-to-hit variance UNTIL the the measure repeats, at which time it had a 135 sample delta as it (seemingly) resynced at the start of the the measure. The off-grid variance started very small (0-3 samples) and slowly climbed until by the end of the measure it was at 133 samples, then it would resync and drop back to being on the grid perfectly.

This indicates to me that it's using a fairly inaccurate note to note timer which accumulates error/drift until it resets at the start of the measure, at least that's my best guess.

[Catabolic]

Forgive my ignorance but from reading this thread it seems the Innerclock does not do anything about computer jitter, it just prevents the synth being triggered from jittering further. Do I have that right?

[TS-12]

I'm debating right now, Either to buy Expert Sleepers system or downgrade to Windows XP 64bit

[Marco1]

Is there only a big problem with Windows 7's Midi OUT, or is there also a problem with its Midi IN? (I mean a more serious problem than any modern OS seems to have anyway.)

Even after having read about it at Experts Sleepers website, I have not yet fully understood what the ES-4 (plus additional Midi accessories) does: Is its Midi out multi-channel? Does it only offer Midi out or also Midi in?

@TS-12: Have you tried Window 7 Service Pack 1 by now?

[bil_g]

ES is just MIDI out. It is multi-channel. It's easier to understand once you use it. Worth it.

[TS-12]

Originally Posted by Marco1

@TS-12: Have you tried Window 7 Service Pack 1 by now?

Finished backing up OS drive.
Now Installing updates and service pack 1...

[spaceacademy]

Originally Posted by nms

Anyone else care to weigh in here? This is the first I ever heard of this.

Yes its irrelevant these days as the poster below you said. We all lost our chance to have purpose build Daws with built in tight midi when Atari went down and just became game designers for other companies. C Lab bought the remaining Falcons and did a lot of work on them such as built in audio, faster processors, better monitor suppoprt (you can use your new fancy monitor today with one of those) built in Scsi drives etc (still a great midi sequencer if you had one of those machines). If that had carried on who knows what would have happened. But they were expensive and by that time the first cheapish PCs had arrived developers like fruity loops abd Steinberg started making software for PCs so C Lab who the became Emagic started to concentrate on logic for the PC. But even today there are many remnants of the midi side from C Labs creator in Logic. Perhaps if Steinberg and C Lab had joined forces back then things may have been different today and we wouldnt have these midi slop problems.

The thing about the Atariu/Clab MMX machines is the midi is not a background, irrelevant service. Its built right in to the motherboard taking instructions directly from the cpu. Its like that with all ataris, not just the CLab one

We will never really get tight midi from our Daws unless the fundamental design of motherboards change and thats not going to happen.

I get through the problem by using an MPC400 slaved via an inneclock systems sync gen which is slaved to the audio clock of my Daw (sounds more complex than it is). I get spot on very tight midi. You dont need an MPC 400. Even an Atari would do the job better than the PC or Mac.

The worst way to wire up your midi is by using a USB interface but what choice do you have if you have an Imac. And even if you have a tower machine its getting difficult to find dedicated PCI mid interfaces. But even their not tight enough.

I was trying out an EMUX2 the other day attached to my moog and sequencing from my Daw. It really is a joke. Notes that were half a second out and so on. Ridiculous state of affairs in this day and age. That's why most people record a few bars, find a bit thats ok, maybe edit a note or two and leave it at that. They cannot record entire tracks from an instrument using USB midi without a lot of tidying up---and remember its a cumulative effect. The more tracks recorded with USB midi the sloppier the whole thing is.

But as I always say some people actually dont hear midi slop. This is obvious to me from some of the demos Ive heard posted on GS. Each to their own though

[Jauqq]

Originally Posted by spaceacademy

We will never really get tight midi from our Daws unless the fundamental design of motherboards change and thats not going to happen.

Even an Atari would do the job better than the PC or Mac.

The worst way to wire up your midi is by using a USB interface but what choice do you have if you have an Imac.

This is not correct.
A Mac with it's midi time stamping drivers built into the OS (OSX) and a USB midi interface with firmware that supports the OS time stamping (Motu, Edirol are the only ones i know of), surpasses an Atari.

I tested it extensively. I had an Atari/C-Lab/ Unitor set-up against my Mac and Motu 128 express. Long story short, midi was recorded direct as audio, using a midi cable to audio jack system. I checked the results in a sample editor. My Motu averaged jitter 0.3ms, the Atari was around 1.5ms. Also the Motu sent midi at the same time across all of it's eight ports. The Atari delayed midi data when using Log3, Unitor 2, so that midi coming out of the last ports on the Unitor 2 were around 2ms -3ms lagging midi coming from the Atari main port.

My Atari rig was later Ebayed....

[Everett Paul]

Another Mac Pro here, with ESI M8U-XL, dual booting between Lion and Windows 7 64 (and another spare XP install).

I have not measured scientifically, but I experience no subjective increase in midi jitter with the ESI on Windows 7. In my experience it is identical to the unit's behaviour under OSX. I bought the ESI to replace the AMT8 which a) didn't support Windows 7 and b) had atrocious lag in both Windows and OSX.

I note that some of the posters above are running midi tests with Reaper driving the midi interface. Don't get me wrong I have been with Reaper since about v2, although around about v4 I walked away from Reaper due to the overwhelming focus on adding minor actions/features for whoever shouted the loudest on the Reaper forum rather than fixing longstanding and fundamental 'amateur' design flaws....but that's another story.

Anyway, In my experience Reaper's midi timing is atrocious, on every interface, at both input and output. Before anyone jumps down my throat, I've used Reaper on self build Core2's, under clean Windows XP installs, with both firewire and pci audio interfaces, with RME midi, TC midi, AMT8, ESI, under Windows 7 32bit, Windows 7 64bit, Macbook, Macbook Pro, Mac Pro, Snow Leopard, Lion.... In every case the behaviour is the same, midi plays second fiddle to audio and there's an overall level of audible jitter both at input recording, and at output even playing quantized 16ths, similar to the good old days of Cubase on the PC :(

By the way I am baffled to this day why people refer to Atari timing, that's another machine where the timing was appalling, I used one for years and often had to pre-bounce chunks of songs down to DAT tape for reimport as a single sample just so that the atari had significantly less midi to perform in realtime so that I wouldn't get the dreaded serial midi note shuffling and glitches between sequence joins. I distinctly remember working on a remix for someone where I had to deliver the remix pretty much the next day and I was up all night finishing the track, not because I hadn't completed the remix, but because I had to keep printing 'simpler' parts that would play in time until the final song midi was as basic as possible just triggering long phrases. The next day the atari got packed away forever.

If you want midi tighter than a female fruit fly's privates you can only really do it with something like an MPC or Cirklon, or at a push using an Innerclock with your choice of daw. If you must use computer based midi it's been clear for years that there are daws and then there are daws, I am yet to see any daw capture and output midi as tightly as Logic, and I don't even use Logic any more so you can know I'm giving an honest opinion......

[Everett Paul]

by the way, when I said I don't experience worse midi lag on Windows 7, that's not me giving a glowing report to usb midi on computers: The ESI is the best interface I've found for playing in time, but overall I still find midi input and output timing on computers in general to be like having your teeth pulled without anaesthetic, through your asshole.

[Catabolic]

Still running Vista but what about the Virtual XP machine inside Windows 7 Professional?

[nms]

I'm on windows 7 x64, Ableton, & a Cakewalk UM-3G and would dread going back to XP personally. I wasn't happy before but since the UM-3G I've had no complaints. Not sure if it's been mentioned, but it's wise to run tests with both MME & DirectMusic midi drivers in ableton to see which works best for you too.

[TS-12]

Finally had time to do MIDI OUT tests in Windows 7 64bit SP1: Same Jitter Problem,
I didn't do "mathematical" tests to be able to tell the jitters in ms, because it was audible to the ear anyway.

[nms]

I just did the test outlined on the innerclock page. 2 bars of 50% length 16th notes at 120bpm and 48khz.

Interface: UM-3G midi out to kenton midi to DCB converter
Synth: Juno 60.

In 32 hits recorded & measured against the 6000 sample grid I got average jitter of 20 samples, 80 samples max. I tried again and got 18 samples avg & 43 samples max.

Ableton Live 8.2, MME midi driver type, Win 7x64.

Edit: To illustrate the importance of having your shit configured properly... here are the results I get when I choose the DirectSound midi driver in ableton's preferences..

In 32 hits recorded & measured note to note I got average jitter of 66 samples with a wopping 258 samples max!

As a side test, I recorded the notes 1 octave down at 60bpm then changed the file's sample rate to 96khz and got 13 samples average jitter (after dividing by 2 to compensate for the double sample rate).

I'm curious to test this technique more particularly with chords as it will definitely give you tighter timing. I work at 88khz normally, so recording at half tempo an octave down then changing the sample rate (no resample) to 88.2 would be workable.

[Jauqq]

Originally Posted by nms

I just did the test outlined on the innerclock page. 2 bars of 50% length 16th notes at 120bpm and 48khz.

UM-3G doing midi out to a kenton midi to DCB converter controlling my Juno 60.

In 32 hits recorded & measured note to note I got average jitter of 20 samples, 80 samples max.

Ableton Live 8.2, MME port type, Win 7x64.

How much of that jitter is down to the Kenton? A better way to test, and the method I use, is to take pure midi recorded direct to audio. You can butcher a midi cable, and wire a jack plug the other end. I've a custom made unit which serves the same purpose.

[nms]

Couldn't tell ya. Also couldn't tell you how much of that jitter is from the Juno.

It's measurements that you see in actual use that matter to me. Considering that 20 samples avg equals 0.42ms, I don't care much about it. The kenton's doing its job.

[nms]

wow this is really strange. I'm testing my Virus TI2 via midi from the UM3G and here's what I get..

the notes are ALL either 5,990 or 5,991 with the occasional 6,060.

Interesting because whatever's happening here is obviously not random. it's happening at those values with 1 sample consistency.

When I disconnect the midi cable and plug a usb cable into the Virus to control it directly (bypassing the UM3G) the values stay the same. So it appears the UM3G introduces no jitter itself.

[Acid Mitch]

Originally Posted by Catabolic

Forgive my ignorance but from reading this thread it seems the Innerclock does not do anything about computer jitter, it just prevents the synth being triggered from jittering further. Do I have that right?

It does not trigger synths. It gives a jitter free clock stream to sync sequencers/drum machines.

[Marco1]

MIDI out jitter may also be machine-dependent.

I had the opportunity to test my Midisport 2x2 on a small consumer notebook (Sony VAIO) and a much stronger workstation notebook (Dell Precision M6600). Both have Windows 7 SP1 64 Bit with the same Windows updates installed, and both have the same Midisport driver installed. On both I used Cubase 6.5 Trial to play and rerecord eight bars of repetitive 8th notes at 200BPM (connecting the Midisport outout to the input). On the Dell, I tried both USB 2.0 and 3.0 ports, and various options in Cubase's device configuration.

The results on the Dell were disastrous. Note-off events of some notes sometimes arrived later than note-on events off succeeding notes, thus choking some notes. On the Sony, that did not happen.

I do not know what the cause of the problem is on the Dell. Since it is a brand-new machine (which also has other issues), I am probably going to return it.