MediaMatrix Latency Times

MediaMatrix can run at three sample rates, 48 kHz, 44.1 kHz, or 32 kHz. These have delays per sample of 0.020833 mS, 0.022676 mS, or 0.031250 mS respectively.

Note : Processing delay if I/O and all processing is on the same chip = 40 samples.

The minimum possible delay of an analog I/O MediaMatrix system is therefore 118 (37+40+41) samples using MM8802’s or 94 (20+40+34) samples if using the old BoB’s.  The delay that is reported by the MediaMatrix compiler is always 22 samples shorter than reality for the delay in a system using MM8802’s, or 2 samples longer than reality for the delay in a system using old BoBs, or in this case 96 samples.  Therefore at 48 kHz a minimum delay system will be reported to delay 2.00 ms but actually delay 2.46 ms using MM8802’s or 1.96 ms using old BoB’s.  At 44.1 kHz a minimum delay system will be reported to delay 2.18 ms but actually delay 2.68 ms using MM8802’s or 2.13 ms using old BoB’s.  At 32 kHz a minimum delay system will be reported to delay 3.00 ms but actually delay 3.69 ms using MM8802’s or 2.94 ms using old BoB’s.

The minimum possible delay of an AES/EBU I/O MediaMatrix system using the new AES/EBU card with Sample Rate Converting is therefore 82 (40+40+2) samples.  The delay that is reported by the MediaMatrix compiler is always 14 samples longer than reality for the delay in a system using new AES/EBU cards, or in this case 96 samples.  Therefore at 48 kHz a minimum delay system will be reported to delay 2.00 ms but actually delay 1.71 ms or 0.29 ms shorter than reported.  At 44.1 kHz a minimum delay system will be reported to delay 2.18 ms but actually delay 1.86 ms or 0.32 ms shorter than reported.  At 32 kHz a minimum delay system will be reported to delay 3.00 ms but actually delay 2.56 ms or 0.44 ms shorter than reported.

The minimum possible delay of a CobraNet I/O MediaMatrix system is therefore 552 (256+40+256) samples, plus the delay in the converters or AES interfaces used in the CobraNet devices used for input and output.  You will need to check with the manufacturer of the device to determine what delay is added over the CobraNet network delay of 256 samples.  The delay that is reported by the MediaMatrix compiler is always 456 samples shorter than reality for the delay in a system using CobraNet cards, or in this case 96 samples.  Therefore at 48 kHz (the only sample rate CobraNet runs at) a minimum delay system will be reported to delay 2.00 ms but actually delay 11.5 ms or 9.5 ms longer than reported.  Remember there is also additional delay in the converters at each end.

Though it is not simple, it is possible to use the CCW and CCR commands in the Terminal window to determine the placement into the chips of your algorithms.  Once you know the placement, you can use the information above to determine delays for each path through MediaMatrix.  In many cases a simpler course of action is to hard assign the algorithms in a time critical path to the chip you desire.  This is accomplished by adding to the end of the algorithm string in the Algorithm Object Properties window the command “ chip=xy” without the quotation marks but with the leading space, where x = the DSP board number, and y = the chip letter (ie. a, b, c, or d).

You can also check MediaMatrix delays using the Frequency Response device.  In the lower right hand corner it reports the system induced delay, which is the delay due to inter chip transfers (both DAB and LDAB).  When you clip onto a wire, the point referenced is the output of the algorithm driving that wire.  Therefore the delay measured includes the delay (if any) associated with the wire the source clip is connected to, but not the delay (if any) associated with the wire the destination clip is connected to.  Therefore you can’t be certain if there is a transfer delay associated with a wire connected to an audio output since there is no output of an audio output device in the schematic for you to clip onto.

One way to actually determine the delay out a specific output is to use the I/O Probe device.  Put an I/O Probe in your schematic and wire it to the input of something (a signal presence indicator will work).  Clip one of the Frequency Response probes to the source you wish to measure from.  Connect the other Frequency Response probe to the output of the I/O Probe.  Switch the I/O Probe to the output you wish to check.  Look at the delay shown in the Frequency Response device, and subtract one transfer delay from what is shown.  That is subtract 1.33 ms if running at 48 kHz, 1.45 ms for 44.1 kHz, or 2.00 ms for 32 kHz.  This is because the I/O Probe adds an extra DAB transfer to the true system induced delay of that output.

If you wish to know the total delay you then must add the minimum delay calculated for a MediaMatrix with the associated I/O device you are using to the system induced delay measured.

The internal data paths in MediaMatrix are 24 bit.  In the input algorithms, the incoming audio is shifted 2 bits away from full scale to provide internal headroom.  In the output algorithms, the 2 bit shift is reversed.  When using an original 24 bit AES input or output, where you have actual 24 bits of valid data (instead of 22 bits or less of valid data), you may want to almost perfectly compensate for these bit shifts by dialing in 12 dB of gain on the input, and attenuating the output by 12 dB.  The reason the compensation is “almost” perfect is because 2 bits are a little more than 12 dB.  If you do this, remember you do not have the extra 12 dB of internal headroom.  When using the new AES input with sample rate conversion, the least significant 2 or 3 bits are mostly noise due to the sample rate conversion process, and therefore it makes little sense to try to compensate for the bit shift.