Maximizing NRF52 & iOS BLE throughput in both directions

[Juju]

Hello

I'm trying to maximize the throughput between a NRF52832 and iOS devices in both directions.

Currently, I seem to achieve about 400 Kbps from an iPhone 8 to the NRF52 using:

• CBCharacteristicWriteWithoutResponse
• maximumWriteValueLengthForType:CBCharacteristicWriteWithoutResponse
• canSendWriteWithoutResponse on iOS side.

This seem to be what I can expect considering the Apple WWDC 2017 talk about Core Bluetooth.

However, when I send data from the NRF52 to the iPhone, I cannot go beyond 128 Kbps.

I do this using:

• sd_ble_gatts_hvx()
• BLE_GATT_HVX_NOTIFICATION
• multiple calls to sd_ble_gatts_hvx() until the SoftDevice queue is filled up
• calling again sd_ble_gatts_hvx() as soon as I get a BLE_GATTS_EVT_HVN_TX_COMPLETE

In both cases, the interval is set to 15ms, PHY=2Mbps, ATT MTU set to 185 bytes, Gatt data length set to 251 bytes.

1. How can I improve the throughput from NRF52 to iOS?
2. Isn't it supposed to be the same speed in both directions?

Thanks! 

[Vidar Berg]

Hello,

I expected the throughput to be roughly the same in both directions.  To verify this I modified the ble_app_uart example in sdk 15.2.0 to transmit dummy packets and request 2mb phy. With this I was able to achieve  ~400kbps+ with an iphone 7 running ios 12.1 (15ms CI, PHY=2, 185 byte MTU). However, the results seems to have been impacted by the fact that I had been running the ANCS example on the same DK prior to this test. After rebooting the phone (clears device cache) I was not able to get more than 128 kbps. In other words, it looks like the iphone is capping throughput at 128 kbps in some cases. I haven't found any relevant Apple documentation on this so not sure if it there's a proper way to override this.  

Here are some throughput numbers with Nordic on both sides in case you haven't seen it:

Bluetooth Low Energy data throughput

[Juju]

Thanks for your reply so it seems to match what I am currently getting.

Did you ever check the possible throughput improvement using a L2CAP channel between iOS and NRF52?

[Vidar Berg]

We have not tested a L2CAP channel with iOS as far as I know, but we have the OTS example if you want to try it. 

[Juju]

Thanks, I will put some figures when I have them.

[Juju]

I was able to achieve around 380 kbps using a L2CAP channel with the stream going from the NRF52 to the iPhone which is much better than the original 128 kbps.

However, this is still below below the 400 kbps obtained in the other directions and without the lightweight protocol advantage of L2CAP.

I still need to try the L2CAP channel from the iPhone to the NRF52.

[Vidar Berg]

Thanks for sharing the results. Are you using a custom app to establish the L2CAP channel? 

[Juju]

Yes I have my own app and needed to experiment with different values for:

• tx_mps / tx_queue_size for l2cap config
• ble_data.len for sd_ble_l2cap_ch_tx()

I used:

• tx_mps=512
• tx_queue_size=6
• ble_data.len = 255 (in my case data comes directly from SPI easyDMA)

Not sure how I can optimize further since I don't know about low level L2CAP transmission.

[Vidar Berg]

Based on your result (380 kbps) and the settings you use I'm pretty sure the throughput is again limited by the iPhone. We have measured 1447.2 kbps between two nordic devices.

[aaronxu]

Great experiment! Any update since iOS13 will be released soon.