[etherlab-users] Sync problems and DC mode

[Graeme Foot]

The following function get the time from the slave at which the EtherCAT frame went through it
ecrt_master_reference_clock_time(master->master, &master->reference_time);

The following function will update the masters time with the last slaves time plus the app time period
ecrt_master_application_time(master->master, master->reference_time+master->app_time_period);


One problem is that the ecrt_master_application_time() call may not be exactly one master->app_time_period value beyond the slaves reference time.  There are many sources of jitter in your application, from the realtime wakeup time to various amounts of processing time since the wakeup to the above function call.  So you need to relate the ecrt_master_application_time() to your app's clock time.

Another thing you need to do for option B is relate the previous master->reference_time to the ecrt_master_reference_clock_time() and adjust your apps's time base to account for drift.  You can then sleep and wake up your rt thread in keeping with your slaves time period.


Graeme.

From: Ignacio Rosales Gonzalez [mailto:narogon at gmail.com]
Sent: Wednesday, 24 January 2018 12:57 a.m.
To: Graeme Foot <Graeme.Foot at touchcut.com>
Cc: Boris Skegin <boris.skegin.de at googlemail.com>; etherlab-users at etherlab.org
Subject: Re: [etherlab-users] Sync problems and DC mode

Thanks for your reply Graeme!
I would try your changes for the A method (a EtherCAT master is the master clock:) putting domain_queue the first.
By the way for the B method (Slave DC master is the master clock) what is wrong in this approach?:

  // send process data
  rtapi_mutex_get(&master->mutex);

  // queue domain data
ecrt_domain_queue(master->domain);


//get the DC slave reference clock time and save in master->reference_time
ecrt_master_reference_clock_time(master->master, &master->reference_time);

// sync slaves to ref clock
ecrt_master_sync_slave_clocks(master->master);

//update master time with time got from dc ref slave
  ecrt_master_application_time(master->master, master->reference_time+master->app_time_period);

// send domain data

  ecrt_master_send(master->master);
  rtapi_mutex_give(&master->mutex);

Best regards,
Ignacio Rosales


2018-01-23 2:18 GMT+01:00 Graeme Foot <Graeme.Foot at touchcut.com<mailto:Graeme.Foot at touchcut.com>>:
As Boris says, master->app_time needs to actually match your PC's time (converted to your app's timeframe).

Another change you could try is as below.  From line 1115:

  // send process data
  rtapi_mutex_get(&master->mutex);

  // queue domain data
  ecrt_domain_queue(master->domain);

  // update application time
  master->app_time = rt_get_app_time();
  ecrt_master_application_time(master->master, master->app_time);

  // sync ref clock to master
  ecrt_master_sync_reference_clock(master->master);

  // sync slaves to ref clock
  ecrt_master_sync_slave_clocks(master->master);

  // send domain data
  ecrt_master_send(master->master);
  rtapi_mutex_give(&master->mutex);


You want as constant a time between calling ecrt_master_application_time() and ecrt_master_send() as possible.  The calls to ecrt_domain_queue() call takes variable amounts of processing time, so do that first.  Also call ecrt_master_sync_reference_clock() every cycle to ensure the reference slave gets lot of small updates rather than a few big updates.  This helps all slaves keep sync better.  It also avoids another source of variable processing time between ecrt_master_application_time() and ecrt_master_send().


Graeme.


-----Original Message-----
From: etherlab-users [mailto:etherlab-users-bounces at etherlab.org<mailto:etherlab-users-bounces at etherlab.org>] On Behalf Of Boris Skegin
Sent: Tuesday, 23 January 2018 2:07 p.m.
To: etherlab-users at etherlab.org<mailto:etherlab-users at etherlab.org>; Ignacio Rosales Gonzalez <narogon at gmail.com<mailto:narogon at gmail.com>>
Subject: Re: [etherlab-users] Sync problems and DC mode

Hello.

> With this version
> <https://github.com/narogon/linuxcnc-ethercat/commit/e4ab86ba6167ced53
> 2e49904059df580062b2d97#diff-059a684a933530837771b5a249433ff3>
> (also as attachment lcec_main.c) I get the servos sync and OP but it
> seems that the PDO doesn't arrive for some of the slaves (no idea why).

master->app_time += master->app_time_period;  means that you just sum
up constant cycle times of  the LinuxCNC thread. So any latency information gets lost here.

Rather  make  master->app_time  be equal to something like
rt_get_time() transferred to EtherCAT time.

I also think that a proper value for sync0Shift  can help a lot.

If however nothing of that helps, then proceed  to  Graeme Foot's option b) .

BTW, what exactly is your kernel and network card and do you really use an adopted ( non-generic) network driver?

Regards,
boris
_______________________________________________
etherlab-users mailing list
etherlab-users at etherlab.org<mailto:etherlab-users at etherlab.org>
http://lists.etherlab.org/mailman/listinfo/etherlab-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.etherlab.org/pipermail/etherlab-users/attachments/20180124/b905dcb9/attachment-0004.htm>