[pulseaudio-discuss] Proposal for a new API and usage of Bluetooth HSP and HFP profiles on Linux

[Arun Raghavan]

+Wim and George from PipeWire

On Tue, 3 Dec 2019, at 12:15 AM, Pali Rohár wrote:
> On Monday 02 December 2019 19:01:11 Tanu Kaskinen wrote:
> > On Sun, 2019-12-01 at 19:57 +0100, Pali Rohár wrote:
> > > Hello!
> > > 
> > > I'm sending this email to relevant mailing lists and other people who
> > > could be interested in it. (I'm not subscribed to all of ML, so please
> > > CC me when replying).
> > > 
> > > 
> > > I would like to open a discussion about a completely new way of handling
> > > Bluetooth HSP and HFP profiles on Linux. These two profiles are the only
> > > standard way how to access microphone data from Bluetooth Headsets.
> > > 
> > > 
> > > Previously in bluez4, HFP profile was implemented by bluez daemon and
> > > telephony HFP functionality was provided by either dummy modem, ofono
> > > modem or by Nokia's CSD Maemo modem.
> > > 
> > > In bluez5 version was modem code together with implementation of HFP
> > > profile removed. And let implementation of HSP and HFP profiles to
> > > external application.
> > > 
> > > Currently HSP profile is implemented in pulseaudio daemon to handle
> > > microphone and Bluetooth speakers. HFP profile is not implemented yet.
> > > 
> > > 
> > > HSP and HFP profiles use AT modem commands, so its implementation needs
> > > to parse and generates AT commands, plus implement needed state machine
> > > for it.
> > > 
> > > And now problem is that last version of HFP profile specification is too
> > > complicated, plus Bluetooth headsets vendors started to inventing and
> > > using of own custom extensions to HFP profile and AT commands.
> > > 
> > > Main problem of this "external" implementation outside of bluez is that
> > > only one application can communicate with remote Bluetooth device. It
> > > is application which received needed socket from bluez.
> > > 
> > > So in this design if audio daemon (pulseaudio) implements HFP profile
> > > for processing audio, and e.g. power supply application wants to
> > > retrieve battery level from Bluetooth device, it means that audio daemon
> > > needs to implement also battery related functionality.
> > > 
> > > It does not make sense to force power supply daemon (upower) to
> > > implement audio routing/encoding/decoding or audio daemon (power supply)
> > > to force implementing battery related operations.
> > > 
> > > 
> > > For handle this problem I would like to propose a new way how to use and
> > > implement HSP and HFP profiles on Linux.
> > > 
> > > Implement a new HSP/HFP daemon (I called it hsphfpd) which register HSP
> > > and HFP profiles in bluez and then exports functionality for all other
> > > specific applications via DBus API (API for audio, power supply, input
> > > layer, telephony functions, vendor extensions, etc...). So it would acts
> > > as proxy daemon between bluez and target applications (pulseaudio,
> > > upower, ofono, ...)
> > > 
> > > This would simplify whole HFP usage as applications would not need to
> > > re-implement parsing and processing of AT commands and it would allow
> > > more applications to use HFP profile at one time. And also it means that
> > > audio software does not have to implement telephony stack or power
> > > supply operations.
> > > 
> > > 
> > > I wrote a document how such DBus API could look like, see here:
> > > 
> > >   https://github.com/pali/hsphfpd-prototype/raw/prototype/hsphfpd.txt
> > > 
> > > 
> > > And also I implemented "prototype" implementation to verify that
> > > designed API make sense and can be really implemented. Prototype fully
> > > supports HSP profile in both HS and AG role, plus HFP profile in HF
> > > role. This prototype implementation is available here:
> > > 
> > >   https://github.com/pali/hsphfpd-prototype
> > > 
> > > Some other details are written in README:
> > > 
> > >   https://github.com/pali/hsphfpd-prototype/raw/prototype/README
> > > 
> > > 
> > > What do you think about it? Does it make sense to have such design?
> > > Would you accept usage of such hsphfpd daemon, implemented according to
> > > specification, on Linux desktop?
> > > 
> > > I would like to hear your opinion if I should continue with this hsphfpd
> > > design, or not.
> > > 
> > > 
> > > With this design and implementation of hsphfpd is possible to easily fix
> > > pulseaudio issue about power supply properties:
> > > 
> > >   https://gitlab.freedesktop.org/pulseaudio/pulseaudio/issues/722
> > 
> > I quite like this proposal. Avoiding the need to implement the button
> > press, battery level etc. handling separately in PulseAudio, oFono and
> > PipeWire seems like a pretty good justification to me. I assume you're
> > volunteering to maintain this new daemon?
> 
> Yes, I can maintain this new daemon.
> 
> > It's not clear to me how this would affect the PulseAudio <-> oFono
> > interaction, if at all.
> 
> Only one application can register to bluez and "own" bluez socket for
> communication with bluetooth headset. Currently it can be either
> pulseaudio (with native backend in pulseaudio for HSP profile) or ofono
> (with ofono backend in pulseaudio for HFP profile).
> 
> So if hsphfpd would be in use, then ofono would not be able to get
> socket from bluez. And it would be needed to add a new "plugin" into
> ofono to use hsphpfd for telephony operations. See hsphfpd.txt file
> (there is diagram of usage).
> 
> > When oFono is used, would PulseAudio get the
> > audio socket from oFono or hsphfpd?
> 
> Always from hsphfpd.
> 
> > What about volume commands, would they go through oFono or would
> > PulseAudio interact with hsphfpd directly?
> 
> Internally they are handled by hsphpfd directly and appropriate API is
> exported via DBus for audio application daemon (pulseaudio). For details
> see hsphfpd.txt file, interface org.hsphfpd.AudioTransport.
> 
> > I think hsphfpd should be part of bluetoothd, but if that's not
> > possible, then that's not possible.
> 
> I do not know if bluez developers are interested in having this code as
> part of bluez project, specially when in bluez4 HFP profile was there
> and in bluez5 was HFP code completely removed.
> 
> But it should not matter where hsphpfd is implemented.
> 
> > (I don't want to get into a lengthy discussion about programming
> > languages, but I'll just note here that I don't like Perl.)
> 
> It is just a prototype. I used Perl just because it is faster for me to
> implement parsing of AT commands needed for prototype purposes.

Broadly, I think this is a good thing to do. My main concern is having this be adequately maintained over a longer period of time.

The other thing to note is that in PipeWire, it is possible for external entities to implement the equivalent of a "sink" or "source" (it's just a node in PipeWire terminology), so in the long run, it might be good to have a single BT audio daemon that manages A2DP and HSP/HFP..

This would also have the benefit of moving codec dependencies out of PulseAudio (which I'll help mitigate in other ways within PulseAudio anyway).

Cheers,
Arun