BLE data transfer speed issues with nRF52840

[masterLee]

Hi, I want to measure the data transfer speed of BLE. I  use the nRF52840 DK with the ble_app_uart example

I created a dummy string of 20 bytes (default length of one packet) and I send the string again and again in an endless loop. 

Every time the string is  transferred to the android smartphone, I invert the state of an output pin on the DK so that I can measure

the frequency (speed) with the oscilloscope.

After connecting the DK to the smartphone, data transmission starts and I can observe the data coming on my smartphone BLE uart application.

At the same time I take measurements from the inverting output pin with the oscilloscope and I have noticed a few things that I cannot explain.

BLE Settings are : MAX_CONN_INTERVAL = 20, MIN_CONN_INTERVAL = 10, PHY= 2Mbps,

1. For the period of one connection interval only a  few transmissions occur at the beginning and for the rest of the connection interval no transmission takes place until the next interval.

2. When I remove the  do{}while (err_code == NRF_ERROR_RESOURCES) check which includes the ble_nus_data_send() , then the transmissions occur for the whole length of the connection interval.

 Why this do{}while (err_code == NRF_ERROR_RESOURCES) check  occupies such a big chunk of the connection interval period?

3.The theoretical speed is 250kBps (PHY2) but I can transfer only 30kBps. Why is that? I expected  it to be much higher...

4. I read that in order to transmit more data I must enable the DLE function of BLE so that I can send up to 251 bytes in one transfer packet. How can I do that in the example?

5. Is there something else that I should take care of in order to come closer to the speed  of 250kBps?

[run_ar]

1. There is a limitation to connection event length, set in the sdk_config.h Try increasing NRF_SDH_BLE_GAP_EVENT_LENGTH

2. The NRF_ERROR_RESOURCES error means there are no available tx buffers. If you don't wait for the buffer to be freed before you move on to the next data bulk you will not transfer all the data.

3. You need to optimize the connection settings. More on that in other devzone cases e.g.https://devzone.nordicsemi.com/f/nordic-q-a/62363/nrf52832-throughput-test-with-nordic-uart-service

4. This seems to be enabled by default in sdk 17.0.2. but look at sdk_config.h NRF_SDH_BLE_GATT_MAX_MTU_SIZE and also the defines in ble_nus.c

#define BLE_NUS_MAX_RX_CHAR_LEN        BLE_NUS_MAX_DATA_LEN /**< Maximum length of the RX Characteristic (in bytes). */
#define BLE_NUS_MAX_TX_CHAR_LEN        BLE_NUS_MAX_DATA_LEN /**< Maximum length of the TX Characteristic (in bytes). *

5. 250 kBytes is not possible. Due to protocol timing restrictions and packet overhead the max is about 1.3 mbps. See SD guide on throughput: infocenter.nordicsemi.com/.../ble_data_throughput.html

[masterLee]

Thank you for your reply.

In image 1 as you can see, the time that data are being transferred, occupy a very small portion of the entire connection interval(each small pulse represents a transfer). The rest of the connection interval is not used for data transfer as I would expect until the buffer is free again and a new connection interval begins. This takes 90% of the connection interval so the way I see it, it is impossible to achieve higher transfer speed without removing the ( do{}while (err_code == NRF_ERROR_RESOURCES)  check.

[run_ar]

NRF_ERROR_RESOURCES means there are no available TX buffers. As long as you get that error message, data is not placed in the buffers and transmitted.

[masterLee]

Ok so what can I do to fix it?

[run_ar]

Did you try to increase the gap event length?

You could possibly test at the Experimental: ATT_MTU Throughput Example to see how changing various parameters affect the throughput you get.

[masterLee]

If I get it right the problem here is that  there are no available TX buffers.

So how is this related to the gap event length?

No I have not changed ot, because it is already set to 6 which is the max number I can setsince I use an android central device(smartphone). Right?

But since you mention it again I will try it.

The NRF_SDH_BLE_GATT_MAX_MTU_SIZE is set already to 247

Is there any way that I can make the do{}(while) much faster? I mean the Tx Buffer.

[run_ar]

Hi,
The gap event length is used by the scheduler. So I think this is limiting the length of the event in this case. This is also used to calculate the number of buffers the sd will allocate up to a maximum of 7. As en alternative you can use sd_ble_opt_set and enable ble_common_opt_conn_evt_ext_t. This will allow the  connection event to extend past the allocated time. But I still think you need to increase the length a little to allocate the maximum number of tx buffers.

I the central device can handle a maximum of 6 packets per event the only way to increase the throughput is by using shorter intervals. In addition to the long MTU's and 2mbits. But I don't think that is limited by Android. But might be an issue with certain (older?) chipsets.