How much of the nRF5340 Shared RAM segment (0x2007000 - 0x207FFFF) does the Nordic BLE Soft Controller require?

[Kevin Doss]

We are using the nRF5340 in a multi-protocol application where we use the BLE Soft Controller in one mode and a proprietary protocol in another mode. Both protocols use the Shared RAM in region 0x20070000 - 0x2007FFFF for message and data exchange between their Hosts on the Application CPU and Controllers on the Network CPU.

We currently have 56 kBytes allocated to the Soft Controller (0x20070000 - 0x2007DFFF) and 8 kBytes allocated to our proprietary protocol (0x2007E000-0x2007FFFF). We have run out of space in this allocation for the proprietary protocol and would like to increase it by maybe 4 kBytes.

The question is what is the minimum Shared RAM allocation that the BLE soft controller needs? Our most intensive usage of the BLE soft controller configures it as a peripheral and GATT server. The Kconfig overlay for this configuration is attached.

Thank you,

Kevin

overlay-ble.conf

[Kevin Doss]

We can live with 14 kBytes of Shared SRAM in our current project. Longer term, though, we may have applications that require more than 14k. It would be helpful to have a more definitive answer to this question if/when that case arises. I find it a little hard to believe that the BLE soft controller can't live with less than 50 kBytes. If that's true trying to implement a data-intensive MPSL solution could be quite restrictive.

[Einar Thorsrud]

It is difficult to say what the minimum sizes of the shared SRAM is, as the amount that is needed will depend on the application (maximum packet sizes being used etc.). So reducing the size should be combined with proper testing to ensure you don't end up in a situation where it causes problems.

[Kevin Doss]

We understand that the answer to the question will vary depending on how hard the BLE stack is being driven (which is why we attached the configuration file above). Even if we knew the worst case (maximum packet size, minimum connection interval, moving as much data as possible in a 2Mbps connection) that would be helpful. It seems wrong to expect the customer to have to profile your application to figure out where it breaks.

[Einar Thorsrud]

I see your point. We have raised this issue internally, but I am not able to provide any numbers or way to estimate this at the moment though, unfortunately.

[Kevin Doss]

Just curious if there is anything new on this question.

Thanks,

Kevin