nRF51822 BLE_ERROR_NO_TX_BUFFERS issue

[Tosa]

I'm using the nRF51822 DK + SDK10.0.0 + s110. I modified the app_ble_hrs example to send 2x 20B packets of data (via a modified heart_rate_meas_send() function) upon a m_battery_timer_id timeout (the other timers are disabled). My goal is to reach 40B (2x 20B packets) every 6ms. The client is a 2013 Asus Nexus 7 which allows 3-4 packets per connection event @ 7.5ms connection intervals. I'm using sd_ble_tx_buffer_count_get() and BLE_EVT_TX_COMPLETE evt.count to keep track of the availble TX buffers per the doc.

I'm seeing really poor performance in that at data intervals of <15ms, I quickly reach 0 available buffers). 15ms for 40B is only 21kbps. Even at 15ms data intervals or higher I see the number of TX buffers available slowly drop over time (starting off at 7 buffers). I'm running it now at 20ms data intervals and the number of TX buffers was 4 after 20 minutes and has now actually started missing data after 30 minutes and has only 2 TX buffers available :( . I also noticed that this calculation of available buffers does not necessarily match the number of buffers really available, since the counter that I decrement in heart_rate_meas_send() may decrement more than the total number of TX buffers without BLE_ERROR_NO_TX_BUFFERS being triggered.

Any help is appreciated! Would an nRF52 help with this issue?

[Stefan Birnir Sverrisson]

Hi Tosa

What is your actual connection interval?

According to my calculations, 40 bytes every 6ms would require 53,4 kpbs throughput. Transferring 3 x20 byte packets per 7.5ms connection interval should be sufficient. Not sure why you only get 21kbps.

This is possibly lack of CPU time, but I doubt it can limit the througput down to 21kbps, unless you have loads of other activity on the nRF51 at the same time draining CPU time.

Are you using nRF51 revision 2 or softdevice older than S110 8.0.0? If so, the softdevice will block the CPU during the whole BLE radio event, which results in less CPU time for the application.