I'm finding consecutive runs of REW - with no changes to the room, levels or mic position - do not produce the same waterfall plot results (though the SPL and phase plots are always the same). REW is running on Mac OS 10.13.6 with a 4 core i7 chip at 3.2 GHZ base speed using all SSD storage directly off the PCIe buss.
REW is using its own dedicated Metric Halo (MH) audio interface via audio over ethernet - which is quite fast. The digital output from the MH interface goes into a second (apogee) interface running on the same machine connected thru a PCIe card - also a fast audio connection. Some latency is introduced by a DAW program processing the REW output on the Apogee interface. The DAW program implements a crossover for the monitors via filter plugins that rolloff to the sub. The filters are setup to be minimum phase with the impulse response customized to be low latency, but the audio buffer for the DAW needs to be 512 samples to run these filters. Then the analog out from the second interface goes to a 2.1 monitor system. I assume the round trip latency REW sees would be = latency of interface one + 512 samples (DAW program) + latency of interface 2.
REW 5.18 seems to mostly work with this setup, e.g., checking the main and sub levels via the preferences menu. Similarly, in the measurement window, REW 5.18 can check the levels which always come in at the same +/- 0.1 dB. You can see there is some latency between when REW generated the test signal and it is picked up by the mic. But it makes not difference if I enable acoustic timing reference or not - waterfall plots are not the same 2X in a row.
With acoustic timing reference enabled the delay REW calculates is not always the same. At 48K doing 20 second sweeps, for 2 consecutive measurements the delay reported came in at: (1) -0.066 +/- 0.01 msec; (2) -0.085+/- 0.01 msec. At 96K the sweeps are 10 seconds and for 3 consecutive runs the delays came in at: (1) 0.101 +/- 0.005 msec; (2) -0.043 +/- 0.005 msec; (3) -0.044 +/- 0.005 msec. But it does not matter if I use 48K or 96K sampling rate, e.g., if on the first run if I see some ringing on the waterfall plot at ~200 Hz, on the next run that can be gone and I see ringing somewhere else were there was none.
This machine has no disks to go to sleep - and those settings are turned off anyway in MacOS. But after REW has been running for a bit, the check for the levels starts to fail. The latency visibly increases between the time when REW generates the tone and when it gets sent out thru the system and picked up by the mic. This happens even if I only use 1 interface w/o any crossover filtering. But unlike REW 5.18, REW 5.19 release and the current 5.20 beta can fail the first time I try to check the levels because of too much delay between the input and output. But if I quit REW and re-start it, then at least on the first try checking the level with 5.18 works. So I've ended up using 5.18 and not 5.19 release or 5.20 beta.
It seems like I'm dealing with some latency in REW that accumiulates after sweeps and the level checks have been run a few times; but does not happen when the program has first been launched. It's not clear to me if this is contributing to some kind of time skew that causes waterfall plots to never be the same 2 times in a row. In REW I tried changing the time to end the impulse response from 1.1 to 4.4 to keep all - but it made no difference. The windowing for the IR left, right, and decay is Tukey 0.25. Happy to hear any pointers about how to get more consistent waterfall results.
REW is using its own dedicated Metric Halo (MH) audio interface via audio over ethernet - which is quite fast. The digital output from the MH interface goes into a second (apogee) interface running on the same machine connected thru a PCIe card - also a fast audio connection. Some latency is introduced by a DAW program processing the REW output on the Apogee interface. The DAW program implements a crossover for the monitors via filter plugins that rolloff to the sub. The filters are setup to be minimum phase with the impulse response customized to be low latency, but the audio buffer for the DAW needs to be 512 samples to run these filters. Then the analog out from the second interface goes to a 2.1 monitor system. I assume the round trip latency REW sees would be = latency of interface one + 512 samples (DAW program) + latency of interface 2.
REW 5.18 seems to mostly work with this setup, e.g., checking the main and sub levels via the preferences menu. Similarly, in the measurement window, REW 5.18 can check the levels which always come in at the same +/- 0.1 dB. You can see there is some latency between when REW generated the test signal and it is picked up by the mic. But it makes not difference if I enable acoustic timing reference or not - waterfall plots are not the same 2X in a row.
With acoustic timing reference enabled the delay REW calculates is not always the same. At 48K doing 20 second sweeps, for 2 consecutive measurements the delay reported came in at: (1) -0.066 +/- 0.01 msec; (2) -0.085+/- 0.01 msec. At 96K the sweeps are 10 seconds and for 3 consecutive runs the delays came in at: (1) 0.101 +/- 0.005 msec; (2) -0.043 +/- 0.005 msec; (3) -0.044 +/- 0.005 msec. But it does not matter if I use 48K or 96K sampling rate, e.g., if on the first run if I see some ringing on the waterfall plot at ~200 Hz, on the next run that can be gone and I see ringing somewhere else were there was none.
This machine has no disks to go to sleep - and those settings are turned off anyway in MacOS. But after REW has been running for a bit, the check for the levels starts to fail. The latency visibly increases between the time when REW generates the tone and when it gets sent out thru the system and picked up by the mic. This happens even if I only use 1 interface w/o any crossover filtering. But unlike REW 5.18, REW 5.19 release and the current 5.20 beta can fail the first time I try to check the levels because of too much delay between the input and output. But if I quit REW and re-start it, then at least on the first try checking the level with 5.18 works. So I've ended up using 5.18 and not 5.19 release or 5.20 beta.
It seems like I'm dealing with some latency in REW that accumiulates after sweeps and the level checks have been run a few times; but does not happen when the program has first been launched. It's not clear to me if this is contributing to some kind of time skew that causes waterfall plots to never be the same 2 times in a row. In REW I tried changing the time to end the impulse response from 1.1 to 4.4 to keep all - but it made no difference. The windowing for the IR left, right, and decay is Tukey 0.25. Happy to hear any pointers about how to get more consistent waterfall results.