UE gets disconnected in a few seconds

[smunaut]

UE gets disconnected in a few seconds

[ismagom]

The issue was introduced in commit 065ac91

[ismagom]

Hi @smunaut , we were able to reproduce the issue and provided a fix. Could you test if:

https://github.com/srsran/srsRAN_Project/tree/fix_disconnect

fixes it for you?

FYI, the problem seemed to be we were using the extended size bit in the MAC header to indicate the size of a PDU which didn't need it.

[smunaut]

@ismagom 🎉 It works perfectly. Been connected for 10 min without issues. It also goes IDLE after 5s inactivity and comes back directly no issues there either.

Thanks for the fix !

Is that behavior phone dependent that it wasn't noticed before ?

[ismagom]

The behaviour is phone dependant yes. Even depends on the release, for instance the Quectel RM500 has the issue but the RM520 doesn't...

[smunaut]

Release 23_10_1 works fine too FWIW.
I'll try to bisect but not sure how that will work out.

[smunaut]

Around 66e7ef966adb5b08634b897483db882514857d06 is my best guess from the bisecting ... not super easy to tell because although in recent commits the bug happens very quickly, in some older commits, it happens after a longer time, and some commits around there just plain crash/segfault, but it should be close I think.

bcbc26eec2c5935bf7af5dc09216b397015f4ddf a few commits before seems to be fine.

[frankist]

Thanks for reporting this @smunaut. The git commit hashs you mentioned as stable are quite old, so it is weird that this would remain undetected for so long.

Can you share new logs with the gnb mac log set to debug level? I want to observe the SINR of the PUCCH that is reporting ack=222 (which means inability to detect the ACKs)

Edit: Sorry, can you also set the phy log to debug level? I hope it won't be too heavy

[smunaut]

Hi @frankist

Sure, here's the requested log attached.
gnb_log.zip
Once with just MAC as debug and once with both PHY and MAC. (I didn't see the "edit" at first 😅 )

And yeah ,those commits are quite old so I'm not sure why I would be the only one to hit that, I find it weird too.
Although at those commit, although the same issue seems to happen, it's not as "immediate", it can take a couple of minutes.

But with 23_10 for sure it works fine, I've been connected all morning without issue and when I go out of RF coverage, I do get lots of errors and then loose connection, but when I come back, it reconnects immediately.

With latest master when the issue occur, getting the phone to reconnect is a hassle, I need to go in/out of airplane mode several time, change APN, use the 5G forcer on/off and eventually it will retry but it takes a while ...

[ismagom]

Hi @smunaut , could you repeat the same log with the release 23_10_1 that works? thanks

[smunaut]

@ismagom Sure. Here's the same log, using the same config on 23_10 _1 and running a speedtest while connected.

gnb-working.zip

[achkonsa]

sorry to disturb...but my version is 23_10_1-1542-g4d9f22329, it still gets released exactly after 5 lines gnb output. Always 5 lines...

[herlesupreeth]

@achkonsa Your issue maybe different than what @smunaut is facing. In your case the behavior maybe due to UE inactivity timer. You could use the below configuration in your gNB config file so that UE is not released after 5 seconds of inactivity (inactivity time is by default set to 5 seconds)

cu_cp:
  inactivity_timer: 7200

[achkonsa]

thank you so much!!! It helps me a loooooot. And I want to ask that does 7200 mean the gnb will end after 7200 seconds? Or it will keep connection all the time? Thanks a loooooot!!!!! By the way, my partner does not meet this issue. I wonder why...His version is 23_10_1-821-g0b2702cca...:)

[herlesupreeth]

And I want to ask that does 7200 mean the gnb will end after 7200 seconds?

inactivity_timer represents the time in seconds after which UE gets release if there is no activity in UE (e.g. no data flow from/to UE etc). Its not the time after which gNB will end.

[smunaut]

@achkonsa Please open a discussion if you want to discuss this idle behavior but this is distinct from the actual problem discussed in this issue.

[achkonsa]

OK, Im sorry.

[smunaut]

And as a side question : was I anywhere close with the bisect as to where that was introduced ? 😅