Event to Event offset time smaller then specs show

RE: Event to Event offset time smaller then specs show

Mathias — Mon, 11 Sep 2017 07:57:38 GMT

Hi guys, I'll close this question cause I'm working on other stuff at the moment and not really focusing on cases with multiple peripherals. I think I also understand the way the different BW configurations work better by developing more. So thanks for all your help! :)

RE: Event to Event offset time smaller then specs show

Petter Myhre — Mon, 28 Aug 2017 11:41:51 GMT

No worries, take your time :)

RE: Event to Event offset time smaller then specs show

Mathias — Fri, 25 Aug 2017 11:22:04 GMT

Okay, I will be able to provide you with this information after Tuesday next week, I have a deadline to get before that... :/ Thank you already for putting so much time and effort into it!

RE: Event to Event offset time smaller then specs show

Petter Myhre — Fri, 25 Aug 2017 09:28:30 GMT

We did some more testing, and now we actually get the same numbers as in the SDS. With a connection interval of 15 ms on both links. We would like to figure out why you get something different, but then it would be nice if you could explain a bit more about your test setup. And maybe you could also share some sniffer traces so we can have a look at them? (not pictures)

RE: Event to Event offset time smaller then specs show

Mathias — Tue, 22 Aug 2017 14:22:05 GMT

No problem, thanks for checking up on it :) Yes, I'm using that SoftDevice version.

RE: Event to Event offset time smaller then specs show

Petter Myhre — Tue, 22 Aug 2017 14:01:46 GMT

Sorry for the late reply. I have done some testing with our Ellisys sniffer. With medium bandwidth I get 3.995 ms between the events, and with high bandwidth I get 6.391 ms. Not the same as your measurements, but still less than the tEEO in the specification. I will contact the developers and get back to you. You are using S130 v2.0.1 right?

RE: Event to Event offset time smaller then specs show

Petter Myhre — Thu, 17 Aug 2017 14:31:57 GMT

I see. Then I need to look a bit more into it. I try to do it tomorrow morning, if not it might not be until Monday.

RE: Event to Event offset time smaller then specs show

Mathias — Thu, 17 Aug 2017 14:03:40 GMT

But you still have less then 4025 µs as well (3792 µs on the figure). And I've now tested it with HIGH BW for peripheral and central with still 15 ms connection interval on each, and I now get 4597 µs instead of 3792 µs as gap, which is still less then what it should be (6900 µs). :(

RE: Event to Event offset time smaller then specs show

Petter Myhre — Thu, 17 Aug 2017 13:25:06 GMT

Because the central device will schedule the connection events one after another, and if it can't receive more than 3 packets (medium bandwidth) it will not set aside 6900 us for every connection event, but only 4025 us. Please let me know if I have misunderstood what the problem is.

RE: Event to Event offset time smaller then specs show

Mathias — Thu, 17 Aug 2017 11:35:52 GMT

Why does it make sense then?

RE: Event to Event offset time smaller then specs show

Petter Myhre — Thu, 17 Aug 2017 11:10:28 GMT

What kind of bandwidth configuration do you have on the central device? If it is medium, doesn't this make sense? Since the tEEO of medium is 4025 us.

RE: Event to Event offset time smaller then specs show

Mathias — Wed, 16 Aug 2017 14:32:16 GMT

Okay, thanks for the input already. And also for those slides, some interesting stuff in there that I didn't know yet :) I also would like to share that I don't always get 3 packets per interval, sometimes only 1 (in the same measurement) and sometimes some packet loss. I tested it again with a bit higher connection interval then the theoretical minimum (50 ms instead of 15 ms) and then I get good timing (larger then 6.9 ms in every case) and almost no packet loss (maybe one or two in a few seconds) and always 3 packets per interval for both peripherals. So maybe it has something to do with other factors that don't allow smooth flow with the theoretical minimum connection interval for this case, don't know.

RE: Event to Event offset time smaller then specs show

endnode — Wed, 16 Aug 2017 14:28:04 GMT

I see it in the table and could come up with theories but the best will be to wait for Nordic support guys later this week;)

RE: Event to Event offset time smaller then specs show

Mathias — Wed, 16 Aug 2017 14:19:29 GMT

No, I'm asking why it doesn't start later. Because with the HIGH BW configuration, the specs say the start time of the next should be minimum 6900 µs after the first one.

RE: Event to Event offset time smaller then specs show

endnode — Wed, 16 Aug 2017 13:53:16 GMT

I start to understand, my fault;) i was confused by your sniffer logs but these obviously map only single connection at the time, I need to imagine second connection running in parallel... You are trying to map situation of two simultaneous Central connections on nRF51 with S130 and you ask why second "slot" of synchronous links (= both have 15ms connection intervals) start ~3.8ms after the first one while it could start sooner? Isn't that irrelevant because it doesn't matter what is their shift, they both run independently and with max throughput they can get (if SD spec says 3 PDUs per interval then it is 3)?

RE: Event to Event offset time smaller then specs show

Mathias — Wed, 16 Aug 2017 13:35:50 GMT

Hmm I think it's maybe less because only 3 packets are being send per interval then. Older Dongle firmware supported 6 packets per interval at pretty low connection intervals (7.5 - 20 ms) but the newer firmware doesn't apparently (max. 3 as you can see).

RE: Event to Event offset time smaller then specs show

endnode — Wed, 16 Aug 2017 13:24:34 GMT

Btw. value 3792us means 3.792ms but there is perfect silence at that time of each of your connection interval (because we counted all PDU lengths and gaps and arrived to ~2ms, right?). So what exactly this number means according to your tools?

RE: Event to Event offset time smaller then specs show

endnode — Wed, 16 Aug 2017 13:23:08 GMT

Then you probably misinterpret SD specification because as we both agreed above that the only static points in time of BLE link are starts of connection intervals, all the rest is relative to them based on PDU length and fixed air-speed or 150-microsecond gaps. And all these values perfectly match your trace... so?

RE: Event to Event offset time smaller then specs show

Mathias — Wed, 16 Aug 2017 13:15:41 GMT

Well, what I actually find strange is why the time between the start of the event for the first peripheral and the start of the event for the second peripheral, is only 3792 µs while the SoftDevice s130 specs show that this is minimally 6900 µs for both HIGH BW configuration?

RE: Event to Event offset time smaller then specs show

endnode — Wed, 16 Aug 2017 13:04:55 GMT

As I said I don't know how long your Slave "full" PDUs are but it should be easy to compute the length as on-air speed is 1MHz exactly and all the time. So you are now happy with the picture or anything unclear?

RE: Event to Event offset time smaller then specs show

Mathias — Wed, 16 Aug 2017 13:02:04 GMT

Thanks for the input. Answering your clarification questions:

Yes, indeed.
Time between each PDU (Master-Slave) beginning and PDU (Slave-Master) beginning, is 230 µs. Minus 80 µs for the Master-Slave PDU duration, this gives perfect 150 µs GAP. Time between each PDU (Slave-Master) beginning and PDU (Master-Slave) beginning, is 446 µs. Minus 330 µs, this gives 116 µs, which is a bit less then 150 µs ... But all in all the time between end of last PDU and start of first PDU of next connection interval, is indeed very close to 13 ms (12.972 ms to be precise). I'm using Peryton as Sniffer software, combined with a Sabrent hardware sniffer piece.

RE: Event to Event offset time smaller then specs show

endnode — Wed, 16 Aug 2017 12:44:42 GMT

3xEmpty PDU (Master->Slave) + 5xGAP (between each of 6 PDUs) + 3xFull PDU (Slave->Master) = 3x80 + 5x150 + 3x330 = 2ms meaning that final GAP between end of last PDU and start of first PDU of next connection interval should be very close to 13ms. Is it the case? Note that even some more expensive sniffers like Frontline do perform well in ~10us range but not very much below (they usually show almost precise 150us timing on gaps but then fail to be as accurate on other measurements), however still you should not get OFF by more then few hundreds of microseconds.

RE: Event to Event offset time smaller then specs show

endnode — Wed, 16 Aug 2017 12:38:51 GMT

Reading it again but not sure what are all the timings in your diagrams and what are corresponding numeric values:

Can we agree that basic connection interval of 15ms works in your case as it is delta between start events of first Master->Slave packet in each interval?
Then timing inside interval itself between Master Tx->GAP#1->Slave Tx->GAP#2 etc. isn't cut into slots but is dynamic depending on each PDU length. BLE spec just says how long after end of received PDU should opposite side start to transmit (150us to be specific). See slides 97, 124 and 125 here. Are your Empty PDU length + GAP#1 + full Slave PDU length in contradiction with this? If so then could you write it down more clearly?

Without knowing details about your payload length simplified numbers would be:

RE: Event to Event offset time smaller then specs show

Mathias — Wed, 16 Aug 2017 11:55:18 GMT

It's roughly the same, every snif.

RE: Event to Event offset time smaller then specs show

endnode — Wed, 16 Aug 2017 11:42:21 GMT

Have you considered inaccuracy of time display by your fancy sniffer HW&SW bundle?