Throughput (Speed) Test

[CheMax]

Hello Sirs,

We implemented a joint bandwidth test (or speed test) for a radio channel by transmitting notifications.

Parameters set in SDK config:

• #define NRF_SDH_BLE_GATT_MAX_MTU_SIZE  247;
• #define NRF_SDH_BLE_GAP_DATA_LENGTH      251;
• #define NRF_SDH_BLE_GAP_EVENT_LENGTH   320;

Parameters of the radio channel:

• PHY 1M;
• Connection interval 50 ms (40 units);
• Transmiter power 0 dbm;

After the connection is established and the exchange of parameters, the host allows the transfer of notification characteristics of the test characteristic(size 244 bytes). Next, the device in the loop begins to send data.

According to the data plotted. I attach his part below.

The average speed was 73 kB / s. But the range is from 35 kB / s to 80 kB / s.

If I understand correctly, this is due to the peculiarity of the protocol. namely, transmission within the connection interval to the first error. As a result, during one connection interval, data can be transmitted, for example, 50 ms, and in the next only 10.

The question is: is it possible to somehow stabilize the data flow?

Thanks, 

Max

[Kenneth]

s a result, during one connection interval, data can be transmitted, for example, 50 ms, and in the next only 10.

That is in essence the reason yes, and also why I recommend to always use the shortest possible connection interval.

[CheMax]

I tried it, but the speed only decreases.

I noticed that:

• with the interval> = 50 ms there is no difference in traffic.
• With an interval of 20-50 ms, the traffic decreases (the average speed is approximately 35-40 kB / s).
• With an interval of 8-20 traffic is even lower.

Is there a rule for choosing the connection interval based on the traffic volume? For example,
1. For large packets, the connection interval should be small
2. Or for large packets, the connection interval should be large

Also interested, failures in transmission speeds can be caused by non-optimal application code (in particular, the use of the operating system)?

Also in your example there was an option to "expand the connection interval event". But we can not find it in the specification. Is this a feature of your stack implementation or is this feature described in the standard?

[CheMax]

Thanks, about the intervals and the length of the packets, I understood.

What about other influencing factors:
Power level set upon connection

case BLE_GAP_EVT_CONNECTED:
    NRF_LOG_INFO("Connected");
    m_conn_handle = p_ble_evt->evt.gap_evt.conn_handle;
    err_code = nrf_ble_qwr_conn_handle_assign(&m_qwr, m_conn_handle);
    APP_ERROR_CHECK(err_code);

    /*-40dBm, -20dBm, -16dBm, -12dBm, -8dBm, -4dBm, 0dBm, +2dBm, +3dBm, +4dBm, +5dBm, +6dBm, +7dBm and +8dBm.*/
    uint32_t err_code = sd_ble_gap_tx_power_set(BLE_GAP_TX_POWER_ROLE_CONN, p_ble_evt->evt.common_evt.conn_handle, 4); 
	APP_ERROR_CHECK(err_code); 

	NotifyEvent.flags.fConnect = true;
    xTaskNotify(hDevCtrl, NotifyEvent.Value, eSetValueWithOverwrite);
break;

Will it be fixed (both for a distance of 0.1m and 10m) or will the chip reduce the signal power at a short range?

[Kenneth]

It's fixed yes, as configured by sd_ble_gap_tx_power_set().

[CheMax]

With the power level also understood, but for general development: how the receiver will behave (nrf52840) with a high transmitter signal power (+8 dbm) and a small distance:
1. will the signal exceed the permissible level and is not always perceived correctly?
2. Will the automatic tuning of the amplifier work and the signal will be "partially suppressed"?

[Kenneth]

Never heard this to be an issue, there is just so much loss through air because the antenna radiate freely in all directions that it's impossible to saturate the input in any way.

Best regards,
Kenneth

[CheMax]

Ok, thanks.

I currently have no more questions)

[Jimmy Wong]

Hi,

I did some measurement on difference connection interval / throughput / distance / Packet error rate on between 2 NRF52840 DK.

The example code is 

https://github.com/jimmywong2003/nrf5-packet-error-rate-measurement-on-ble-connection.

from the experiment result, smallest connection interval is the best because it would be lower packet error rate.

But at the same time, you need to tune the payload (DLE).

[CheMax]

Hi Jimmy,

Many thanks for your addition. I didn't even have a thought in my head that one could try to count radio events.

Good luck in your projects!

[Jimmy Wong]

HI Chemax,

I will share the result to you by email if you are interested.  Here are part of the result between 2 NRF52840 DK (0dBm), very close ~ 20 cm.  Event length = 400ms.