moved tx audio to class __init__

This commit is contained in:
DJ2LS 2021-01-06 18:01:54 +01:00 committed by GitHub
parent 508c01ffed
commit 0efc20defa
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23

View file

@ -32,8 +32,7 @@ class RF():
libname = pathlib.Path().absolute() / "codec2/build_linux/src/libcodec2.so"
self.c_lib = ctypes.CDLL(libname)
#--------------------------------------------OPEN AUDIO CHANNEL RX
self.p = pyaudio.PyAudio()
@ -48,13 +47,13 @@ class RF():
#--------------------------------------------OPEN AUDIO CHANNEL TX
#self.p = pyaudio.PyAudio()
#self.stream_tx = self.p.open(format=pyaudio.paInt16,
# channels=1,
# rate=static.AUDIO_SAMPLE_RATE,
# frames_per_buffer=2048, #n_nom_modem_samples
# output=True,
# output_device_index=static.AUDIO_OUTPUT_DEVICE, #static.AUDIO_OUTPUT_DEVICE
# )
self.stream_tx = self.p.open(format=pyaudio.paInt16,
channels=1,
rate=static.AUDIO_SAMPLE_RATE,
frames_per_buffer=static.AUDIO_FRAMES_PER_BUFFER, #n_nom_modem_samples
output=True,
output_device_index=static.AUDIO_OUTPUT_DEVICE, #static.AUDIO_OUTPUT_DEVICE
)
#--------------------------------------------START AUDIO THREAD
@ -72,22 +71,29 @@ class RF():
FREEDV_DATAC2_THREAD = threading.Thread(target=self.receive, args=[11], name="DATAC2 Decoder")
FREEDV_DATAC2_THREAD.start()
FREEDV_DATAC3_THREAD = threading.Thread(target=self.receive, args=[12], name="DATAC3 Decoder")
FREEDV_DATAC3_THREAD.start()
#self.transmit(7,b'12345')
#time.sleep(2)
#self.transmit(7,b'000000000000')
#self.transmit(7,b'ABCDEFGHIJKL')
#--------------------------------------------------------------------------------------------------------
def audio_listen(self):
print("STARTING AUDIO LISTENER")
while True:
time.sleep(0.05)
data = self.stream_rx.read(static.AUDIO_FRAMES_PER_BUFFER, exception_on_overflow = False)
static.AUDIO_BUFFER += data
#static.AUDIO_BUFFER += data.strip(b'\x00')
#static.AUDIO_BUFFER += data
static.AUDIO_BUFFER += data.strip(b'\x00')
rms = audioop.rms(data,2)
print(rms)
#--------------------------------------------------------------------------------------------------------
# GET DATA AND MODULATE IT
@ -105,13 +111,18 @@ class RF():
mod_out = ctypes.c_short * n_tx_modem_samples
mod_out = mod_out()
if mode < 10:
##preamble = bytes(payload_per_frame)
preamble = b'111111111111'
data_out = preamble + data_out
#data_out += data_out
data_list = [data_out[i:i+payload_per_frame] for i in range(0, len(data_out), payload_per_frame)] # split incomming bytes to size of 30bytes, create a list and loop through it
data_list_length = len(data_list)
for i in range(data_list_length): # LOOP THROUGH DATA LIST
if mode < 10: # don't generate CRC16 for modes 0 - 9
buffer = bytearray(bytes_per_frame) # use this if no CRC16 checksum is required
buffer[:len(data_list[i])] = data_list[i] # set buffersize to length of data which will be send
print("buffer for ACK: " + str(buffer))
@ -128,25 +139,25 @@ class RF():
data = (ctypes.c_ubyte * bytes_per_frame).from_buffer_copy(buffer)
self.c_lib.freedv_rawdatatx(freedv,mod_out,data) # modulate DATA and safe it into mod_out pointer
print(bytes(mod_out))
#print(bytes(mod_out).strip(b'\x00'))
p = pyaudio.PyAudio()
stream_tx = p.open(format=pyaudio.paInt16,
channels=static.AUDIO_CHANNELS,
rate=static.AUDIO_SAMPLE_RATE,
frames_per_buffer=n_nom_modem_samples,
output=True,
output_device_index=0, #static.AUDIO_OUTPUT_DEVICE
)
#p = pyaudio.PyAudio()
#stream_tx = p.open(format=pyaudio.paInt16,
# channels=static.AUDIO_CHANNELS,
# rate=static.AUDIO_SAMPLE_RATE,
# frames_per_buffer=n_nom_modem_samples,
# output=True,
# output_device_index=1, #static.AUDIO_OUTPUT_DEVICE
# )
audio = audioop.ratecv(mod_out,2,1,static.MODEM_SAMPLE_RATE, static.AUDIO_SAMPLE_RATE, static.TX_SAMPLE_STATE)
stream_tx.write(audio[0])
audio = audioop.ratecv(mod_out,2,1,static.MODEM_SAMPLE_RATE, static.AUDIO_SAMPLE_RATE, static.TX_SAMPLE_STATE)
self.stream_tx.write(audio[0])
print("KILL")
stream_tx.stop_stream()
stream_tx.close()
p.terminate()
#print("KILL")
#stream_tx.stop_stream()
#stream_tx.close()
#p.terminate()
return mod_out
@ -167,7 +178,7 @@ class RF():
bytes_out = bytes_out() #get pointer from bytes_out
i = 0
while static.MODEM_RECEIVE == True: # Listne to audio until data arrives
while True: # Listne to audio until data arrives
time.sleep(0.05) # here we reduce CPU load
nin = self.c_lib.freedv_nin(freedv)
@ -202,17 +213,15 @@ class RF():
# --------------- END DEBUGGING OUTPTUT -------------------------------------------
self.c_lib.freedv_set_sync(freedv, 0) #FORCE UNSYNC
# CHECK IF FRAMETYPE IS BETWEEN 10 and 50 ------------------------
frametype = int.from_bytes(bytes(bytes_out[:1]), "big")
if 50 >= frametype >= 10 and len(bytes_out) > 30: # --> The length check filters out random strings without CRC
static.AUDIO_BUFFER = bytearray()
print("MODE: " + str(mode) + " DATA: " + str(bytes(bytes_out[:-2])))
arq.data_received(bytes(bytes_out[:-2])) #send payload data to arq checker without CRC16
self.c_lib.freedv_set_sync(freedv, 0) #FORCE UNSYNC
else:
print("MODE: " + str(mode) + " DATA: " + str(bytes(bytes_out)))
@ -236,4 +245,4 @@ class RF():
if len(static.AUDIO_BUFFER) > i: # WE WILL LOOP THROUGH OUR DATA BUFFER WHILE OUR BUFFER IS BIGGER THAN THE CHUNK POSITION
i = (nin*2) + i
else:
i = 0
i = 0