search

xmos / sw_usb_audio

XMOS USB Audio reference design
Other
32 stars 17 forks source link

S/PDIF transmit discontinuities #159

Closed danielpieczko closed 1 year ago

danielpieczko commented 1 year ago

Using the new S/PDIF transmit code from lib_spdif, output tests sometimes report discontinuities:

Channel 8: discontinuity (samples -5121286, -5121272 do not differ by 7 but by 14)
Channel 9: discontinuity (samples -3532169, -3532184 do not differ by -5 but by -15)

In this example, one sample went missing on channel 8, and two samples on channel 9. In the reproductions I've seen, it is usually one or two samples on each channel that are missing. Sometimes there is an invalid sample: Channel 9: discontinuity (samples -1243255, -1243268 do not differ by -5 but by -13) (expected samples: -1243265, -1243270)

So far, this has only been seen at sample rates of 176400 and 192000.

Running output tests with a three minute duration, this error is seen 10% of the time. Often it occurs 10-30s into the stream.

On the 316 board, it was most commonly seen on the 2AMi8o10xxsxxx config, but on others as well.

On the 216 board, it was seen only on the 2AMi8o10xxsxxx_mix8 config.

xross commented 1 year ago

Since this feels like a timing issue I've pushed some basic optimisations to the S/PDIF transmitted on the head of develop

xross commented 1 year ago

Looking at the new implementation, I'm wondering if this static int lastbit should be reset to 0 on SR change @XMOS-JoeG

https://github.com/xmos/lib_spdif/blob/30f0039ae9d6adad16700b612ed2cea32995b40d/lib_spdif/src/SpdifTransmit.xc#L102

xross commented 1 year ago

Looking at the new implementation, I'm wondering if this static int lastbit should be reset to 0 on SR change @XMOS-JoeG

https://github.com/xmos/lib_spdif/blob/30f0039ae9d6adad16700b612ed2cea32995b40d/lib_spdif/src/SpdifTransmit.xc#L102

This would only be important if we wanted to make the start of stream predicable, for testing etc.

danielpieczko commented 1 year ago

Since this feels like a timing issue I've pushed some basic optimisations to the S/PDIF transmitted on the head of develop

Following these changes in lib_spdif, we no longer see the discontinuities reported in this issue where one or two samples are missing (across hundreds of ouput test attempts).