We have an internet app that uses BLE COC to synchronize the call audio of Android devices to the iPhone side microphone. All of this functions normally, but if two conditions are met: upgrading to IOS27; Connect iPhone to headphones. So the up and down latency of our BLE COC data will be significant and increase proportionally with time. Through the analysis of the air interface logs, it can be seen that the number of retries for COC data on the air interface has significantly increased, and the interval between the other two packets is very large (far greater than the 30 millisecond conn interval). I would like to ask if Apple has adjusted the scheduling strategy of the Bluetooth controller on iOS 27, and how can we adapt to this situation?
The attachment is a screenshot of the air interface log. The packet circled in the picture is sent from iPhone to Android, with an interval of 300+ms. It can be inferred that the iPhone protocol stack and Bluetooth controller have already squeezed a large amount of data, and our business bitrate is not high, only about 10K per second
BLE IOS27. When the device is connected to headphones, the Bluetooth controller does not tend to perform normal scheduling on BLE
I would like to ask if Apple has adjusted the scheduling strategy of the Bluetooth controller on iOS 27, and how can we adapt to this situation?
Have you filed a bug on this and, if so, what's the bug number?
__
Kevin Elliott
DTS Engineer, CoreOS/Hardware
Yes, I have submitted feedback which includes the issue time and detailed iPhone logs, but they are not the same sample as the screenshot provided above. Please do not confuse them.
Feedbacks:
https://feedbackassistant.apple.com/feedback/25527230
I would like to ask if Apple has adjusted the scheduling strategy of the Bluetooth controller on iOS 27,
Yes, particularly in terms of prioritizing A2DP/HFP traffic over BLE traffic. More specifically, when Bluetooth headphones are connected to an iPhone, the phone's single Bluetooth radio must give strict, real-time priority to the headphone audio stream (eSCO for calls, or A2DP for playback).
Because BLE L2CAP CoC guarantees in-order delivery, any BLE packet that collides with headphone audio time slots is repeatedly retransmitted. If packets are generated faster than the radio can clear them, data builds up in the Bluetooth controller queues, causing latency to grow continuously over time.
and how can we adapt to this situation?
I have a few different suggestions that might help:
(1) Request High Connection Priority (Shorter Interval)
When audio streaming begins, have the Android device request high connection priority.
Why it helps:
This prompts iOS to shorten the connection interval from 30 ms down to 11.25 to 15 ms. A shorter interval provides more frequent transmission windows between headphone audio slots, preventing multi-packet delays.
(2) Enable LE 2M PHY and Maximum Data Length (DLE)
-
Request LE 2M PHY (PHY_LE_2M) on the BLE connection.
-
Negotiate maximum Data Length Extension (DLE = 251 bytes).
Why it helps:
Transmitting at 2 Mbps cuts the physical airtime of each packet in half (~250 microseconds vs. over 500 microseconds). Shorter packets fit much more easily into the brief idle gaps between headphone audio frames without colliding.
(3) Implement Application-Level Frame Dropping (Prevent Buffer Bloat)
-
Do not continuously push audio frames into the socket or stream if data is backing up.
-
Attach a timestamp or sequence number to each audio frame.
-
If the transmit buffer depth or measured latency exceeds an acceptable ceiling (for example, greater than 80 to 100 ms), drop older frames and send only the most recent audio frame.
Why it helps:
For real-time voice, stale audio data is useless to the listener. Dropping delayed frames at the sender prevents the controller's transmit queue from growing continuously.
(4) (Optional) Consider Unacknowledged Transmissions for Live Audio
If you only stream real-time voice, consider using GATT Write Without Response (CBCharacteristicWriteType.withoutResponse) paired with standard audio Packet Loss Concealment (PLC), rather than L2CAP CoC.
Why it helps:
Unacknowledged packets are never retried at the radio level. If a packet misses an air slot, it is simply discarded rather than stalling subsequent packets, ensuring that stream latency remains near real-time.
__
Kevin Elliott
DTS Engineer, CoreOS/Hardware
Thank you for your detailed response. I will consider buffer optimization strategies for the application layer.
For the optimization of the BLE link, we have already negotiated a 2M PHY, while the interval can only be adjusted to multiples of 15 (not 1.25). I wonder if this is also one of the modified features after IOS27? And I'm also thinking if we can adjust the interval to a multiple other than 7.5, can we avoid conflicts with the SCO link? If this idea doesn't work, I want to shorten it to 11.25 as you mentioned, but it seems impossible at the moment. Is there any way to do it?