multiple subscriptions slows down characteristic throughput

[chrishunt93]

TLDR: Does anyone have an idea how to subscribe to multiple notifications using PyGATT without significant packet loss?

I have been having an issue with some BLE interfacing code for some while now. I am using the PyGATT module in Python 3 on a Linux machine to receive data from a Bluetooth LE device (MyoArmband from Thalmic Labs). For some reason, I am not able to get the advertised sampling rate for any of the data streams (EMG and IMU). I can connect and set up the device fine, however, my computed sampling rates are much slower than they should be.

For example, I should be getting a data rate of 200 Hz for EMG and 50 Hz for IMU, however, with this code I am only getting ~70 Hz from EMG and ~17 Hz IMU. I've tested a few configurations and determined that if I only subscribe to a single characteristic at a time, I can get the correct sampling rate for that characteristic. This suggests that my code must be dropping packets only when multiple characteristics are subscribed to. Is this a known issue with PyGATT / is there any way around this? If not with PyGATT, does anyone have a suggestion for another module that I can use in Python for interfacing with BLE devices (preferably cross-platform)?

Note: I've attached my code for reference.

import pygatt
import time
import numpy as np

emgcount = 0
imucount = 0

def emg_handler(handle, value): 
    global emgcount 
    emgcount += 2

def imu_handler(handle, value): 
    global imucount 
    imucount += 1

def main():
   btle = pygatt.BGAPIBackend()

    btle.start()
    myo = btle.connect( 'e0:f2:99:e7:60:40' )

    try:
        # set parameters
        myo.char_write('d5060401-a904-deb9-4748-2c7f4a124842', b'\x01\x03\x00\x00\x00')   # deregister all streaming
        myo.char_write('d5060401-a904-deb9-4748-2c7f4a124842', b'\x0a\x01\x02')           # lock myo
        myo.char_write('d5060401-a904-deb9-4748-2c7f4a124842', b'\x09\x01\x01')           # don't sleep
        myo.char_write('d5060401-a904-deb9-4748-2c7f4a124842', b'\x01\x03\x02\x01\x00')   # stream filtered emg and imu

        # subscribe to everything
        myo.subscribe('d5060402-a904-deb9-4748-2c7f4a124842', callback = imu_handler)
        myo.subscribe('d5060105-a904-deb9-4748-2c7f4a124842', callback = emg_handler)
        myo.subscribe('d5060205-a904-deb9-4748-2c7f4a124842', callback = emg_handler)
        myo.subscribe('d5060305-a904-deb9-4748-2c7f4a124842', callback = emg_handler)
        myo.subscribe('d5060405-a904-deb9-4748-2c7f4a124842', callback = emg_handler)

        print( 'Streaming...' )
        t0 = time.time()
        while ( time.time() - t0 ) < 10.0: 
            time.sleep( 1.0 )

    finally:
        tf = time.time()
        btle.stop()

        print('')
        print( 'EMG SAMPLING RATE:', emgcount / ( tf - t0 ) ) # should be ~200, is returning ~70
        print( 'IMU SAMPLING RATE:', imucount / ( tf - t0 ) ) # should be ~50, is returning ~17

if __name__ == '__main__':
    main()

[bem22]

I have the exact problem when trying to acquire data from 3 characteristics from multiple BLE devices. I believe it's because of the threading mechanism behind the adapters, that is trying to acquire the hci interface. I have subscribed successfully to characteristics that almost reach the maximum BLE bandwidth, however, when having multiple devices at the same time, the communication is badly slowed down.

[peplin]

I expect the two backends (GATTTool and BGAPI) to have very different performance. The former is (yuck) parsing string output from the gatttool CLI and I wouldn't be surprised if the performance isn't very good. This is the backend using hci devices; @bem22 your issue is likely different than @chrishunt93 's since the example in this ticket is using BGAPI.

I unfortunately don't have any high rate BLE devices like this, so I haven't done and am not planning on doing any performance testing, but I would welcome PRs if you made any progress.