Receptor Latency Tests - WAY TO GO MUSE!!

[jeamsler]

I've had a Receptor for a while now. I got it because it was advertised to make software feel like hardware. I've played hardware for a long time and now software instruments. Software instruments have versatility but playing was not always satisfying because I always had the feeling the keys were running away underneath my fingers. In other words I could never have as sure a sense of rhythm while playing software as playing good hardware synths or a real piano. For a long time I have tested hardware synths for latency precisely because of this sense sometimes that some synths were not on as sure a footing as others. Generally my tests have correlated with what I have felt. So when I tried Receptor and now after owning and playing one for a while I also feel a sure sense with it though I never have tested it until now. Let me say the Receptor has tested as well as any hardware synth and better than every computer softsynth setup I have tested including gigastudio/gsif instruments and good pci sound card setups (Lynx, Echo, etc). The latency in absolute terms is not lower but what is tremendously better is the consistency of latency. Hardware tests out as being generally more consistent than computer/softsynth setups. Hardware synths have their own latency issues usually having to do with not enough processor power to keep up with demand but usually that does not affect the consistency of response. Cheaper hardware is worse of course and you ususally get what you pay for. With computers while the latency isn't bad the variance is and that generally correlates with my playing experiences. High variance translates for me as that rubbery rhythm that I have a hard time getting under control. The Receptor is unbelievably consistent with an extraordinarily low variance. Now latency values given by soundcard control panels, etc. are essentially useless. In testing they don't correlate at all. For most computers a buffer setting of 128 will get you about 10-12 ms of average latency with variances as high as 1.5ms and min to max of about 3ms. These are values I have found by experiment. Take the buffers down to 64 and you get about 6-8ms with variance of about 1ms and min to max of 2ms. Now that does not tell the whole story. When you look at the data you see with computer/softsynths that the data has a general ramp pattern similar to a sawtooth wave. The latency values gradually go down to a min then jump up abruptly to a max value then start desending again and repeat the pattern. This happens at any buffer setting I've ever tried and I imagine results from buffers filling then emptying. Hardware synths don't have this characteristic. They may have terrible latencies but their variances but don't have this regular pattern. To me the pattern is important because random fluctuations I feel I can absorb to a degree in playing but regular ramping latency is going to be like playing one rhythm against another. Hardware synths can go as low as 2-3ms and as high as 10-15ms depending on the synth. If you spend money for a good synth you will probably get the lower end these days (wasn't always true in the past though). And with 2-5ms of initial latency you also get a low variance. Now where does the Receptor stack up? The latency is generally the same as a computer but the variance is much better. At a buffer of 128 you get about 12ms with a variance of about .04ms! In other words it is dead consistent. Pretty amazing for me to see that. At 64 buffers you get about 7.5ms with the same variance. And at 32 buffers you get 5.5 ms with the same very very small variance. Extraordinary results really. The latency may not be what some people are used to seeing in specs and soundcard panels but these are real world figures and I have yet to see any computer setup get better initial latency figures than these at the same buffer settings. More importantly no computer can come anywhere near the consistency of response that the Receptor has shown to me. The Receptor has many issues to resolve: some big some small. However, the platform and implementation of the audio and midi drivers is really superb in giving a truly hardware level of performance. In this respect the specs and advertisements are not hype they are true from what I can see. Well this has been a lengthy post and I am not going to claim that I have the last word in measurements. I'm only saying what I have found in my own tests over 20 years and what I have come to believe based on those tests. Feel free to comment.
Way to go MUSE

Jon

[Acrobat]

Really interesting, thank you for sharing your knowledge. The "Ramp shape" issue I'm particularly interested as it's something I always noticed but couldn't give a reason for.
Your explanation is pretty convincing but opens to a 1000 more questions about,

I.E. -> To WHAT is the period of that phenomenon related most?

[jeamsler]

I'm thinking it is related to the buffers filling and emptying. I think the buffers hold a certain number of samples and when full they fire. Maybe someone with more knowledge about how audio is fired in a computer can say. An interesting calculation is that if you take one period of the ramp and subtract the Last (min) from the First (max) you get a number of samples roughly equal to the buffer size. I don't know enough about audio and how it is done in computers to say what that result means but perhaps it does connect with how the audio process works in computers. On the other hand Receptor is a computer with real life os that can do other things (not simple one optimized entirely for midi and audio as on hardware synth) and yet it does not display the ramping or the variance so I don't know how it all fits.

Jon

[jeamsler]

I neglected to include that all the tests were at 44.1 khz. I used a simple patch from the ZR-3 organ. More complex synths/sounds may change things since unlike hardware synths computer synths will either output all simultaneous events as fast as midi permits or simply fail. So the thing that tends to give in this is the initial latency and the polyphony. This is what I have observed at least up to now. Hardware synths on the other hand will generally keep a fairly constant initial latency (within 2-3ms for a decent synth) but other simultaneous notes will be progressively delayed sometimes by quite a bit. I can't test every patch or synth obviously but my experience is that even in extreme circumstances the change in initial latency is similar to good hardware (1-2ms).

Jon

[khubbard]

Thanks so much for the interesting discussion. I'm having some trouble with latency running BFD on my Receptor and triggering the MIDI with a Roland TD 10. I run the sound through an M-Audio Project Mix I/O to a pair of reference monitors. The MIDI out of the Roland is plugged directly into the back of the Receptor.

I still have noticeable latency when striking the pads. It's happening with mesh and rubber pads, as well as the rubber cymbal pads. I've turned off the sound source in the TD 10 to minimize processing from outside the Receptor. So the TD 10 is acting as a trigger-to-MIDI interface essentially.

Any thoughts? It's hard to play the drums with this much latency. I'm estimating (no hard core test here) that it's probably about 25ms. It's like bad slapback that you'd get from playing with a back wall close to the stage. Really annoying, and it would drive other musicians nuts if I took it to a live gig.

Thanks for any thoughts or suggestions!

Kevin

[Acoutistic]

Hello,

FWIW, I trigger BFD 1.5 on Receptor with a Zendrum with very low latency and very very satisfying results. You might want to check the sample buffer size on your Receptor (on the setup tab of the GUI) and be sure its set as low as possible. I usually use 128; sometimes I can get away with 64.

M