SO_RAI_NO_DATA still does not cause RRC release

[Michal Mühlpachr]

Issue is related to  RE: NRF_SO_RAI_NO_DATA does not cause RRC release in SLM/nrf9160 without send 

Just after AT#XSOCKETOPT=1,50 modem should negotiate RRC Release, but it is not the case.

NCS v2.3 nrf9160_1.3.4

Serial LTE Modem, prj.conf changes:
CONFIG_SLM_START_SLEEP=y
CONFIG_SLM_WAKEUP_PIN=6
CONFIG_SLM_INDICATE_PIN=2
CONFIG_SLM_DATAMODE_URC=y
CONFIG_SLM_EXTERNAL_XTAL=y
CONFIG_NRF_MODEM_LIB_TRACE=y

Modem Trace: trace-2023-03-15T22-56-29.105Z.bin

AT dialogue and PPKII measurements:

[PPK2] DUT Powered, Measurement started
   0.13 <RESET> Wed Mar 15 23:56:33 2023
   1.01 [PPK2] 100k/s AVG   4.02 mA    Max  28.6  mA
   1.25 <INDICATE transition> -> High
   2.03 [PPK2] 100k/s AVG   2.21 mA    Max  29.8  mA
   3.01 [PPK2] 100k/s AVG   0.002mA    Max   0.009mA
   4.02 [PPK2] 100k/s AVG   0.002mA    Max   0.008mA
   4.26 <WAKEUP>
   5.02 [PPK2] 100k/s AVG   1.11 mA    Max  24.5  mA
   5.60     [RX] Ready
   5.60 [TX] AT
   5.61     [RX] OK
   5.61 [TX] AT+CFUN=0
   5.77     [RX] OK
   5.77 [TX] AT%MDMEV=1
   5.78     [RX] OK
   5.78 [TX] AT%XPOFWARN=1,30
   5.80     [RX] OK
   5.80 [TX] AT%XTEMPHIGHLVL=60
   5.81     [RX] OK
   5.81 [TX] AT%HWVERSION
   5.81     [RX] %HWVERSION: nRF9160 SICA BQA
   5.81     [RX] OK
   5.81 [TX] AT%SHORTSWVER
   5.83     [RX] %SHORTSWVER: nrf9160_1.3.4
   5.83     [RX] OK
   5.83 [TX] AT#XSLMVER
   5.83     [RX] #XSLMVER: "2.3.0","2.3.1-8622ee1c632e"
   5.83     [RX] OK
   5.83 [TX] AT%XSYSTEMMODE=0,1,0,0
   5.85     [RX] OK
   5.85 [TX] AT%REL14FEAT=1,1,1,1,0
   5.86     [RX] OK
   5.86 [TX] AT%RAI=1
   5.86     [RX] OK
   5.87 [TX] AT%REDMOB=1
   5.88     [RX] OK
   5.88 [TX] AT%XDATAPRFL=0
   5.89     [RX] OK
   5.89 [TX] AT+CEPPI=1
   5.90     [RX] OK
   5.90 [TX] AT%XNETTIME=1
   5.91     [RX] OK
   5.91 [TX] AT+CNEC=24
   5.92     [RX] OK
   5.92 [TX] AT+CMEE=1
   5.93     [RX] OK
   5.93 [TX] AT%XSIM=1
   5.94     [RX] OK
   5.94 [TX] AT+CEREG=5
   5.95     [RX] OK
   5.95 [TX] AT+CSCON=1
   5.97     [RX] OK
   5.97 [TX] AT+CGEREP=1
   5.97     [RX] OK
   5.97 [TX] AT%XTIME=1
   5.99     [RX] OK
   5.99 [TX] AT%XMODEMSLEEP=1,500,10240
   6.00     [RX] OK
   6.00 [TX] AT+CPSMS=1,,,"00111000","00000000"
   6.00     [RX] %XMODEMSLEEP: 4
   6.00     [RX] OK
   6.01 [TX] AT%XBANDLOCK=1,"0000000000000000000000000000000000000000000000000000000000001000000010000000100010011010"
   6.02 [PPK2] 100k/s AVG   4.64 mA    Max  37.1  mA
   6.02     [RX] OK
   6.03 [TX] AT+COPS=1,2,"23003"
   6.05     [RX] OK
   6.05 [TX] AT+CGDCONT=1,"IP","hardwario.com"
   6.05     [RX] OK
   6.05 [TX] AT+CFUN=1
   6.10     [RX] OK
   6.11     [RX] %XMODEMSLEEP: 4,0
   7.01 [PPK2] 100k/s AVG  27.2  mA    Max  48.0  mA
   7.02     [RX] %XSIM: 1
   7.02 [TX] AT+CGSN
   7.02     [RX] 352656106109476
   7.02     [RX] OK
   7.02 [TX] AT+CIMI
   7.03     [RX] 901288000012723
   7.03     [RX] OK
   8.01 [PPK2] 100k/s AVG  35.5  mA    Max  65.8  mA
   9.00 [PPK2] 100k/s AVG  39.3  mA    Max  66.5  mA
  10.01 [PPK2] 100k/s AVG  19.2  mA    Max  62.9  mA
  11.02 [PPK2] 100k/s AVG   1.89 mA    Max   2.53 mA
  12.03 [PPK2] 100k/s AVG   1.89 mA    Max   2.50 mA
  13.01 [PPK2] 100k/s AVG   1.89 mA    Max   2.51 mA
  13.46     [RX] +CEREG: 2,"AE38","000AC51F",9
  14.01 [PPK2] 100k/s AVG   4.55 mA    Max  55.8  mA
  14.16     [RX] +CSCON: 1
  15.02 [PPK2] 100k/s AVG  33.2  mA    Max 290mA
  16.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.3  mA
  17.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.3  mA
  18.02 [PPK2] 100k/s AVG  33.6  mA    Max  46.1  mA
  19.01 [PPK2] 100k/s AVG  33.1  mA    Max  47.8  mA
  20.01 [PPK2] 100k/s AVG  33.2  mA    Max  46.5  mA
  21.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.3  mA
  22.02 [PPK2] 100k/s AVG  33.6  mA    Max  46.2  mA
  23.02 [PPK2] 100k/s AVG  33.6  mA    Max  46.3  mA
  24.02 [PPK2] 100k/s AVG  33.3  mA    Max  46.1  mA
  24.43     [RX] +CGEV: ME PDN ACT 0,0
  24.43     [RX] +CNEC_ESM: 50,0
  24.43     [RX] %MDMEV: SEARCH STATUS 2
  24.44     [RX] +CEREG: 5,"AE38","000AC51F",9,,,"00000000","00111000"
  24.44 [TX] AT+COPS?
  24.45     [RX] +COPS: 1,2,"23003",9
  24.45     [RX] OK
  24.45 [TX] AT%XCBAND
  24.46     [RX] %XCBAND: 20
  24.46     [RX] OK
  24.46 [TX] AT+CEINFO?
  24.46     [RX] +CEINFO: 0,1,C,8,1,-87,18
  24.46     [RX] OK
  24.46 [TX] AT+CGDCONT?
  24.47     [RX] +CGDCONT: 0,"IP","hardwario.com","10.0.0.157",0,0
  24.48     [RX] +CGDCONT: 1,"IP","hardwario.com","",0,0
  24.48     [RX] OK
  24.48 [TX] AT#XDATACTRL=40
  24.48     [RX] %XTIME: ,"32305122655540",
  24.49     [RX] OK
  24.49 [TX] AT#XSOCKET=1,2,0
  24.49     [RX] #XSOCKET: 0,2,17
  24.50     [RX] OK
  24.50 [TX] AT%CONEVAL
  24.51     [RX] %CONEVAL: 0,1,7,54,24,42,"000AC51F","23003",493,6447,20,0,0,16,2,1,114
  24.51     [RX] OK
  25.00 [PPK2] 100k/s AVG  33.0  mA    Max 148mA
  25.65 [PPK2] ====== SUM: ATTACH   0.157uA/h during  25.6  s ======
  26.73 [TX] AT#XCONNECT="192.168.168.1",20000
  26.73     [RX] #XCONNECT: 1
  26.73     [RX] OK
  26.73 [TX] AT#XSOCKETOPT=1,50
  26.74     [RX] OK
  27.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.2  mA
  28.02 [PPK2] 100k/s AVG  33.6  mA    Max  46.3  mA
  29.00 [PPK2] 100k/s AVG  33.4  mA    Max  46.3  mA
  30.01 [PPK2] 100k/s AVG  33.0  mA    Max  46.2  mA
  31.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.3  mA
  32.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.1  mA
  33.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.1  mA
  34.01 [PPK2] 100k/s AVG  33.5  mA    Max  46.4  mA
  35.02 [PPK2] 100k/s AVG  32.8  mA    Max  46.3  mA
  36.00 [PPK2] 100k/s AVG  33.6  mA    Max  46.2  mA
  37.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.2  mA
  38.02 [PPK2] 100k/s AVG  33.6  mA    Max  46.1  mA
  39.01 [PPK2] 100k/s AVG  33.6  mA    Max  46.5  mA
  40.00 [PPK2] 100k/s AVG  32.6  mA    Max  46.0  mA
  41.01 [PPK2] 100k/s AVG  33.7  mA    Max  46.1  mA
  42.02 [PPK2] 100k/s AVG  33.6  mA    Max  46.2  mA
  43.02 [PPK2] 100k/s AVG  33.6  mA    Max  46.6  mA
  44.02 [PPK2] 100k/s AVG  33.7  mA    Max  48.2  mA
  45.02 [PPK2] 100k/s AVG  32.7  mA    Max  45.9  mA
  45.49     [RX] +CSCON: 0
  45.49     [RX] %XMODEMSLEEP: 1,86399999
  45.49 [TX] AT#XSLEEP=2
  45.49     [RX] OK
  46.00 [PPK2] 100k/s AVG  16.5  mA    Max 194mA
  47.02 [PPK2] 100k/s AVG   0.017mA    Max   0.023mA
  48.01 [PPK2] 100k/s AVG   0.017mA    Max   0.023mA
  49.03 [PPK2] 100k/s AVG   0.017mA    Max   0.023mA
  50.02 [PPK2] 100k/s AVG   0.017mA    Max   0.023mA

[Achim Kraus]

Maybe, it's easier to use SO_RAI_LAST and a own "empty message".

At least in the past versions, the idea with SO_RAI_NO_DATA seems to be the same, sending a last empty message with NO_RESPONSE. Therefore you have to "connect" the udp socket in order to tell the stack, where to send this injected empty message.

Just because you use a 90128 SIM:

Do you know the HPPLMN search interval of your SIM card? Not that you get too surprised by some additional power consumption from that function (see HPPLMN search - reason unknown , it explains, why Nordic will always search for such a 9xxxx network.) By the way, did your modem ever report that 90128 network (see nRF9160 - who's device has seen/reports a network with a global PLMN starting with 9 ).

[Michal Mühlpachr]

SO_RAI_LAST and an own "empty message" works as workaround.
But SO_RAI_NO_DATA should still be fixed.

As you can see from AT dialogue, UDP socket is "connected" (without any IP traffic).
As you can see from modem trace, stack is not sending any empty message after SO_RAI_NO_DATA.

HPPLMN search does not apply because Vodafone CZ uses 90128 for home NB IoT (all Vodafone CZ NB IoT SIMs are 90128 - i.e. "roaming"), so it is the highest priority PLMN already.

Since SO_RAI_NO_DATA  and SO_RAI_ONE_RESP work reliably like a charm, I see no reason why SO_RAI_LAST should struggle with  HPPLMN search  or another corner strange causes. You can see from LTE RRC/NAS signaling (Modem Trace) that it is not the case also.

[dejans]

Hi,

If you use SO_RAI_NO_DATA with SO_RAI_ONE_RESP, you would not need to know when UDP packet has been sent.

Best regards,
Dejan

[Michal Mühlpachr]

dejans said:

ONE_RESP + UDP does not deliver the expected result with R14 AS-RAI, since the R14 AS-RAI has no capability to signal such combination. With AS-RAI, the UE can only inform the network "more data expected". It is in R16 when 3GPP defines AS-RAI to be at the same expression level as CP-RAI.

Do I understand well, that ONE_RESP with R14 AS-RAI is not useful at all for power consumption optimization?

[Michal Mühlpachr]

I am interested in that also, because there is not ANY signaling to the application, that send buffers are free to obtain another datagram. In case of bulk transfer through UDP, the application has to estimate the actual link transfer rate only based on lost datagrams. That is an extremely unfriendly technique for power saving. The application should be aware, whether the next datagram to be sent will or will not be dropped inside the device.

[Michal Mühlpachr]

Achim Kraus said:

I log a couple of modem events and measure some times. If the time from "RRC idle" to "enter sleep" is significant larger than the actual PSM RAT, then something prevents the modem from "enter sleep" in time. One of such "something" is a HPPLMN search.

Clever idea, I'll use that, thanks.

[dejans]

Hi,

UDP is non-reliable protocol. Therefore, the information that UDP message left the modem would not guarantee anything or give any additional information. Typically, the assumption could be made that single UL UDP has left the modem when modem switches to RRC_IDLE.

Best regards,
Dejan

[Michal Mühlpachr]

dejans said:

UDP is non-reliable protocol. Therefore, the information that UDP message left the modem would not guarantee anything or give any additional information. Typically, the assumption could be made that single UL UDP has left the modem when modem switches to RRC_IDLE.

Yes, UDP is non-reliable as any datagram, but available information for data transfer timings has to be passed to the application.
Assumption can not be made on RRC_IDLE, because in such case link speed could not be utilized, i.e. such an approach wastes energy.

For good energy management, it is essential to know whether packet dropping occurs/does not occur in the device, as this information is inherently available to the device.
Another possibility is a statistical estimation, which needs to be based on the current line rate, which is known by the modem but not by the application, which needs to do the transmission timing, so the modem has to pass the information about the current line rate.

How does the application know that the packet is not dropped in the modem?

How does the application know what the current line rate is?

[Achim Kraus]

> UDP is non-reliable protocol. 

UDP is the MOST reliable base for an upper protocol running over cellular modems using NB-IoT! The most serious upper protocols based on UDP do a much better job with NB-IoT and are much more reliable than TCP when running over NB-IoT.

TCP handles "congestion control" and "message drops" internally, but that depends a lot on additional messages and the timings. The generally assumed times are far away from the times of NB-IoT and that causes then a lot of retransmissions. So the only thing you get reliable using TCP over NB-IoT is a lot more data ;-)

> Typically, the assumption could be made that single UL UDP has left the modem when modem switches to RRC_IDLE.

For a useful congestion control of the radio layer, the event, that a message is sent by the radio layer helps a lot. Other modems offer that.

Using the RRC Idle event doesn't really help in that case. RRC IDLE is reached, if no more data is to be exchanged. That's either indicated by RAI (therefore all this discussion) or when the network's RRC Active Timer expires. It's not usable for the mentioned bulk transfer.

I guess, if you ask the modem developers, there is a number of messages, which are queued and send one after the other. But the exact implementation is not documented. And for many applications it's better to have a feedback, how fast actually the messages are sent in order to do the queuing in a more specific way in the application. Maybe using POSIX poll with POLLOUT works, but AFAIK, that is missing in the documentation.