A friend is keen to buy a DAC with I²S in over HDMI as LVDS. He already has a USB to I²S over HDMI via LVDS, but the pin polarities are flipped for Data and Word Clock. So the DAC will get the Data and Left-Right signal flipped. There's clearly no industry standard 
I understand the I²S word he will use is 32bits with padding of 16bit and 24bit data.
Im keen to use this conundrum to understand a bit about digital encoding if anyone can kindly help me please?
If he plugs his USB-I²S to that DAC with his standard HDMI cable, what will happen?
Will the data + & - polarity flipped matter? Is it just going to cause a one bit delay? A delay that doesn't matter, as the LSB and MSB will be just a padded "0" and will be evenly matched across left and right channels?
Or is it going to cause a loss of one bit depth and a loss of 6dB SNR? If it does, is that a good thing to prevent intersample overs clipping?
Will there be the same music, but just with left and right stereo channels reversed? Easy corrected by just plugging the DAC output cables across channels.
Or do I need to make him a bespoke HDMI cable
Thanks!
I understand the I²S word he will use is 32bits with padding of 16bit and 24bit data.
Im keen to use this conundrum to understand a bit about digital encoding if anyone can kindly help me please?
If he plugs his USB-I²S to that DAC with his standard HDMI cable, what will happen?
Will the data + & - polarity flipped matter? Is it just going to cause a one bit delay? A delay that doesn't matter, as the LSB and MSB will be just a padded "0" and will be evenly matched across left and right channels?
Or is it going to cause a loss of one bit depth and a loss of 6dB SNR? If it does, is that a good thing to prevent intersample overs clipping?
Will there be the same music, but just with left and right stereo channels reversed? Easy corrected by just plugging the DAC output cables across channels.
Or do I need to make him a bespoke HDMI cable
Thanks!
The inverted WS will just flip channels.
The inverted SCK is a problem, IMO. The I2S transmitter (by default) writes the data bit with the falling edge, expecting the receiver to read it with the rising edge. If the receiver gets the signal inverted, it will read the data at the same edge the transmitter on the other side is writing a new data bit. Depending on overall transmission delay of the data line, it may or may not work reliably.
Some I2S transmitters can flip their SCK polarity , e.g. in RPi Linux it can be configured in software (and it does work, tested). But it takes some non-trivial configuration or even recompilation, IMO easier to flip the wires in HW.
The inverted SCK is a problem, IMO. The I2S transmitter (by default) writes the data bit with the falling edge, expecting the receiver to read it with the rising edge. If the receiver gets the signal inverted, it will read the data at the same edge the transmitter on the other side is writing a new data bit. Depending on overall transmission delay of the data line, it may or may not work reliably.
Some I2S transmitters can flip their SCK polarity , e.g. in RPi Linux it can be configured in software (and it does work, tested). But it takes some non-trivial configuration or even recompilation, IMO easier to flip the wires in HW.
It will do that, but it also does a little more - no longer are L and R samples co-incident in time. Normally this isn't an issue though.
Right, I forgot about that.
There were even commercially manufactured soundcards with L/R samples shifted by wrong design, probably a similar mistake.
This is an interesting post https://www.diyaudio.com/community/...mi-i2s-over-rj45-devices.282985/#post-5100325 . The twist length are by specs different https://en.wikipedia.org/wiki/Category_5_cable#Individual_twist_lengths
The question is how long in ns that skew is. https://www.flukenetworks.com/knowledge-base/dtx-cableanalyzer/propagation-delay and https://www.flukenetworks.com/blog/cabling-chronicles/101-series-getting-picture-delay-skew say maximum allowed skew for CAT5E is 50ns for 100m legth, with most cables down at 25ns, or 2-3ns for special CAT5 cables for analog video distribution apps (e.g. https://www.prysmiangroup.com/sites...kets/markets/downloads/datasheets/ste1e_0.pdf )
At 768kHz samplerate 32bit sample size the bitclock is 49MHz, e.g. 20ns period isch. Taking the skew to be no more than 10% it makes a 2ns max skew. A standard CAT5E cable should have no more than 0.5ns of inter-pair skew per 1m, most likely less.
IMO based on this it can be concluded that CAT5E is OK in single-meter lengths for LVDS I2S at standard samplerates.
The question is how long in ns that skew is. https://www.flukenetworks.com/knowledge-base/dtx-cableanalyzer/propagation-delay and https://www.flukenetworks.com/blog/cabling-chronicles/101-series-getting-picture-delay-skew say maximum allowed skew for CAT5E is 50ns for 100m legth, with most cables down at 25ns, or 2-3ns for special CAT5 cables for analog video distribution apps (e.g. https://www.prysmiangroup.com/sites...kets/markets/downloads/datasheets/ste1e_0.pdf )
At 768kHz samplerate 32bit sample size the bitclock is 49MHz, e.g. 20ns period isch. Taking the skew to be no more than 10% it makes a 2ns max skew. A standard CAT5E cable should have no more than 0.5ns of inter-pair skew per 1m, most likely less.
IMO based on this it can be concluded that CAT5E is OK in single-meter lengths for LVDS I2S at standard samplerates.
That's enlightening. I never even thought of the time delay with differential twisting in a Cat cable.
Thanks very much for all your help. Ive learnt heaps. I hope this helps the next DIY'er with LVDS I2S issues.