On 20-12-02 17:04:47, Glenn Schmottlach wrote: > On Tue, Dec 1, 2020 at 4:43 PM Glenn Schmottlach <gschmottlach@xxxxxxxxx> wrote: > > Hi Ruslan - > > > > Thanks for the feedback but unfortunately I've experienced mixed > > results with the gadget UAC2 driver on both Windows/Linux. Let me > > describe my environment. My host platform is either a Linux Ubuntu > > 18.04 or Windows 10 laptop while the target environment is a > > BeagleBone Black (Linux beaglebone 5.4.74-g9574bba32a #1 PREEMPT). I'm > > testing two different scenarios: > > > > Scenario #1: > > BeagleBone Black (BBB) runs speaker-test generating a single channel > > (S32_LE) audio stream containing a 1KHz tone with a 48K sample rate, > > e.g. > > > > > speaker-test -D hw:1,0 -r 48000 -c 1 -f 1000 -F S32_LE -t sine > > > > The host laptop is running Audacity and recording the tone over the > > UAC2 adapter. On the Linux host the capture is correct and the tone is > > bit-perfect. On the Windows 10 the capture contains numerous missing > > samples which translates into a lot of audible pops and clicks. > > > > Scenario #2: > > The Linux/Windows host plays a single channel, 48K, S32_LE 1K sine > > tone to the target using either Audacity (on Windows) or 'aplay' (on > > Linux), e.g. > > > > > aplay -D hw:4,0 -c 1 -r 48000 -t wav tone_1k.wav (Linux) > > > > On the BBB target I use 'arecord' to record the tone to a RAM disk and > > then copy the recorded file back to the host where I can verify the > > quality of the recording. In both instances (e.g. using either Windows > > or Linux for playback) the recording on the target results in a > > captured file with missing samples and audible pops/clicks. In this > > scenario the UAC2 gadget is configured with c_sync == asynchronous. I > > wouldn't expect things to improve with c_sync == adaptive since you > > mentioned in your patch that it always reports back the nominal > > frequency to the host from the feedback endpoint. > > > > Do you have any suggestions that might explain (the above) behavior. > > Can you describe your test environment in more detail so that I can > > perhaps re-create it? What Linux target are you using with your tests? > > You mentioned you tested an 8x8 playback/capture scenario. Can you > > provide any details of how you performed this test and the method you > > used to confirm the audio quality for the capture/playback? > > > > Thanks for any insights you might be able to offer . . . > > > > Glenn > > Hi Ruslan - > > This is a follow-up from my post yesterday. I recompiled my kernel > *WITHOUT* your UAC2 patches and repeated Scenario #2 where the Linux > PC plays a single channel tone to the BeagleBone Black where it's > recorded with 'arecord'. Yesterday, I recorded garbled audio on the > target but today, without any UAC2 kernel patches, the recorded audio > on the target is glitch-free and appears to be bit-perfect. > > This experiment leads me to believe your patches may be inadvertently > corrupting the data-path. Have you been able to repeat my experiment > and either confirm or refute my findings? I am interested to learn > more how you tested your patches and whether it's something I can > recreate here. > > Assuming we can sort out this data corruption issue, what are your > thoughts on how the Linux target device can properly provide the > Windows feedback endpoint with real frequency updates rather than the > constant nominal frequency. If I understood your patch notes correctly > it seems this is an outstanding issue that requires additional > attention. I'm a bit of a noob when it comes to how this might be > addressed. > > Thanks for your continued insights and support . . . > > Glenn Hi Glenn & Ruslan, Do you know why WIN10 can't recognized UAC2 device if I configure the sample rate as 48000HZ? Configuring sample rate as 44100HZ, the playback function would work well at my platforms (chipidea IP), no glitch is heard. At WIN10, I use Windows Media Player, at board side I use command: arecord -f cd -t wav -D hw:4,0 | aplay -f cd -Dplughw:3,0 & >From the USB Bus analyzer: Feedback EP is scheduled every 1ms, there are nine 176-byte packets and one 180-byte packet among 10ms transfers. -- Thanks, Peter Chen
