Alignment of external MIDI events and audio

Question

Hi,
I've been doing some careful measurements of the timing of MIDI events under Logic 8.0.2. I record a lot of external MIDI hardware, so the timing of external MIDI events and its relation to recorded audio is very important to me.
My setup is OS X 10.5.5, Logic 8.0.2., all relevant patches and updates. I'm using an RME fireface 800 as both the audio and MIDI interface in these examples.
I've created a project in which there is no plug-in of any kind: neither instrument nor effect. Even the klopgeist is gone.
My preferences settings:
Audio/Devices:
buffer=1024 samples
sw monitoring=off
indep mon level=off
I/O safety buffer=off
Audio/General:
PDC=Off (I also measured with PDC=all, to verify that it has no effect)
low latency mode=off
I create an external MIDI track of quantized whole notes: C3, playing on the first tick of each bar. BPM is 120.
I then create another external MIDI track, cable MIDI OUT to MIDI IN on the Fireface, and record the first MIDI track onto the second (i.e., loopback test). The note data is right on the money; within one tick of the original values. I set PDC to all, repeat the test, same result.
I then cable MIDI out on the Fireface to a hardware synth, and call up a patch with a sharp attack. I route the audio output of the synth to an audio input of the Fireface. I create an audio track with this input selected, and record the output of the synth. The recorded audio is early by 20ms or so -- roughly equivalent to the buffer size.
I repeat the test with PDC set to All, same result. PDC has no effect, as you would expect, as there are no plug-ins in this project.
I then set Preferences/MIDI/Sync/All MIDI Output delay to 21ms. I repeat the audio recording test, and now the recorded events are on the money.
Here are my questions. First, am I doing something wrong, or does Logic have a fundamental bug concerning the linkage between MIDI timing and audio timing?
It seems clear to me that this timing error has nothing to do with PDC. There is no "P". Right?
I thought that the RME has timestamped MIDI. The RME manual speaks of very accurate MIDI timing, and I can verify that the MIDI jitter of the Fireface is indeed low, in the 1ms neighborhood. This is borne out by the MIDI loopback test I did above. But MIDI timestamps don't mean anything if they are not correlated to the time code of the audio, haha! What am I missing here?
If timestamping the MIDI data can't work for some reason, could this problem not be solved by having Logic automatically delay the transmission of external MIDI data by an amount equal to the audio buffer size? Put differently, my solution to this problem is to set a MIDI output delay in preferences equal to the audio buffer size, i.e., 21.3ms. And this seems to work. Is that a proper solution, and if so, why does Logic not do such a thing automatically?
Thanks,
-Luddy

Answer 1

OK, so I just retested and here is the best part... I recorded a loop from my drum machine, cut the first note nicely to the 1 on the first bar... using a count in etc.
I played it back and looped outs 3-4 back into 1-2 on my Ensemble. I also played back the same kick on my drum machine just to check this reverse latency issue. I was not concerned with hit for hit layering accuracy. What I got back astounds me!
The "loopback" recording was spot on to the original being played out... totally locked. BUT, the one that was midi clocked by Logic and play from my drum machine came in ahead of the 1??? How is this possible when I am hearing it when monitoring this recording and it is slightly behind? RIDICULOUS! What does this mean?
Logic is NOT recording what I am hearing, but rather recording some timewarped version of it in which it places events ahead of when they happened, but ONLY on midi clocked or midi sent notes!
I have lurked here a long time and I know that there is not much known here about this issue, so before anyone tells me the typical solution I will say as I did in my last post:
I have done a ton of tests like this, I also followed the loopback testing as described on Logicprohelp.com.
Yes I tried it with all PDC modes, SW monitoring on AND off.
I am truly beginning to think that this is a major bug of some sort.
Anyone else out there with hardware that reports latency to Logic (RME, Metric Halo, and Apogee stuff are ones I know that do this) experiencing this abberation?
thanks