mirror of
https://github.com/DJ2LS/FreeDATA
synced 2024-05-14 08:04:33 +00:00
190 lines
7.8 KiB
Python
190 lines
7.8 KiB
Python
#!/usr/bin/env python3
|
|
# -*- coding: utf-8 -*-
|
|
"""
|
|
Created on Wed Dec 23 07:04:24 2020
|
|
|
|
@author: DJ2LS
|
|
"""
|
|
|
|
import ctypes
|
|
from ctypes import *
|
|
import pathlib
|
|
import pyaudio
|
|
import audioop
|
|
import sys
|
|
import logging
|
|
|
|
import static
|
|
|
|
|
|
|
|
class RF():
|
|
|
|
def __init__(self):
|
|
|
|
|
|
self.p = pyaudio.PyAudio()
|
|
self.defaultFrames = static.DEFAULT_FRAMES
|
|
self.audio_input_device = static.AUDIO_INPUT_DEVICE
|
|
self.audio_output_device = static.AUDIO_OUTPUT_DEVICE
|
|
self.tx_sample_state = static.TX_SAMPLE_STATE
|
|
self.rx_sample_state = static.RX_SAMPLE_STATE
|
|
self.audio_sample_rate = static.AUDIO_SAMPLE_RATE
|
|
self.modem_sample_rate = static.MODEM_SAMPLE_RATE
|
|
self.audio_frames_per_buffer = static.AUDIO_FRAMES_PER_BUFFER
|
|
self.audio_channels = static.AUDIO_CHANNELS
|
|
self.format = pyaudio.paInt16
|
|
self.stream = None
|
|
|
|
|
|
#self.data_input = "stdin"
|
|
self.data_input = "audio"
|
|
#self.data_output = "stdout"
|
|
self.data_output = "audio"
|
|
|
|
|
|
libname = pathlib.Path().absolute() / "codec2/build_linux/src/libcodec2.so"
|
|
self.c_lib = ctypes.CDLL(libname)
|
|
|
|
|
|
self.mode = static.FREEDV_MODE # define mode
|
|
|
|
|
|
self.freedv = self.c_lib.freedv_open(self.mode)
|
|
self.bytes_per_frame = int(self.c_lib.freedv_get_bits_per_modem_frame(self.freedv)/8)
|
|
self.payload_per_frame = self.bytes_per_frame -2
|
|
self.n_tx_modem_samples = self.c_lib.freedv_get_n_tx_modem_samples(self.freedv)*2 #get n_tx_modem_samples which defines the size of the modulation object
|
|
self.n_max_modem_samples = self.c_lib.freedv_get_n_max_modem_samples(self.freedv)
|
|
self.n_nom_modem_samples = self.c_lib.freedv_get_n_nom_modem_samples(self.freedv)
|
|
self.nin = self.c_lib.freedv_nin(self.freedv)
|
|
|
|
|
|
|
|
|
|
# MODULATION-OUT OBJECT
|
|
def ModulationOut(self):
|
|
return (c_short * self.n_tx_modem_samples)
|
|
|
|
# MODULATION-IN OBJECT
|
|
def ModulationIn(self):
|
|
return (c_short * (self.n_max_modem_samples))
|
|
|
|
# FRAME BYTES
|
|
# Pointer for changing buffer data type
|
|
def FrameBytes(self):
|
|
return (c_ubyte * self.bytes_per_frame)
|
|
|
|
# GET DATA AND MODULATE IT
|
|
def Transmit(self,data_out):
|
|
|
|
mod_out = self.ModulationOut()() # new modulation object and get pointer to it
|
|
|
|
data_list = [data_out[i:i+self.payload_per_frame] for i in range(0, len(data_out), self.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 self.mode < 10: # don't generate CRC16 for modes 0 - 9
|
|
|
|
buffer = bytearray(self.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
|
|
|
|
if self.mode >= 10: #generate CRC16 for modes 10-12..
|
|
|
|
buffer = bytearray(self.payload_per_frame) # use this if CRC16 checksum is required ( DATA1-3)
|
|
buffer[:len(data_list[i])] = data_list[i] # set buffersize to length of data which will be send
|
|
|
|
crc = c_ushort(self.c_lib.freedv_gen_crc16(bytes(buffer), self.payload_per_frame)) # generate CRC16
|
|
crc = crc.value.to_bytes(2, byteorder='big') # convert crc to 2 byte hex string
|
|
buffer += crc # append crc16 to buffer
|
|
|
|
data = self.FrameBytes().from_buffer_copy(buffer) #change data format from bytearray to ctypes.u_byte and copy from buffer to data
|
|
|
|
self.c_lib.freedv_rawdatatx(self.freedv,mod_out,data) # modulate DATA and safe it into mod_out pointer
|
|
|
|
if self.data_output == "stdout":
|
|
sys.stdout.buffer.write(mod_out) # print data to terminal for piping the output to other programs
|
|
sys.stdout.flush() # flushing stdout
|
|
|
|
if self.data_output == "audio":
|
|
#print(self.audio_channels)
|
|
stream_tx = self.p.open(format=self.format,
|
|
channels=self.audio_channels,
|
|
rate=self.audio_sample_rate,
|
|
frames_per_buffer=self.n_nom_modem_samples,
|
|
output=True,
|
|
output_device_index=self.audio_output_device,
|
|
)
|
|
|
|
audio = audioop.ratecv(mod_out,2,1,self.modem_sample_rate, self.audio_sample_rate, self.tx_sample_state)
|
|
stream_tx.write(audio[0])
|
|
stream_tx.close()
|
|
|
|
return mod_out
|
|
|
|
|
|
# DEMODULATE DATA AND RETURN IT
|
|
def Receive(self):
|
|
|
|
# Open Audio Channel once
|
|
stream_rx = self.p.open(format=self.format,
|
|
channels=self.audio_channels,
|
|
rate=self.audio_sample_rate,
|
|
frames_per_buffer=self.n_max_modem_samples,
|
|
input=True,
|
|
input_device_index=self.audio_input_device,
|
|
)
|
|
|
|
|
|
while static.MODEM_RECEIVE == True: # Listne to audio until data arrives
|
|
|
|
|
|
#if self.data_input == "stdin":
|
|
# samples = self.c_lib.freedv_nin(self.freedv)*2 ### MIT DER *2 funktioniert das irgendwie recht zuverlässig bei mode 5! Bei Mode 12 auch
|
|
# data_in = sys.stdin.buffer.read(samples)
|
|
if self.data_input == "audio":
|
|
|
|
data_in = stream_rx.read(self.c_lib.freedv_nin(self.freedv), exception_on_overflow = False)
|
|
#print(bytes(data_in))
|
|
buffer = bytearray(self.n_max_modem_samples*2) # N MAX SAMPLES * 2
|
|
buffer[:len(data_in)] = data_in # copy across what we have
|
|
|
|
|
|
self.ModulationIn()() #Create new ModulationIn Object
|
|
modulation = self.ModulationIn()# get an empty modulation array
|
|
modulation = modulation.from_buffer_copy(buffer) # copy buffer across and get a pointer to it.
|
|
|
|
bytes_out = self.FrameBytes()() # initilize a pointer to where bytes will be outputed
|
|
|
|
nbytes = self.c_lib.freedv_rawdatarx(self.freedv, bytes_out, data_in) # Demodulated data and get number of demodulated bytes
|
|
|
|
if nbytes == self.bytes_per_frame: # make sure, we receive a full frame
|
|
|
|
|
|
print(bytes(bytes_out[:-2]))
|
|
self.c_lib.freedv_set_sync(self.freedv, 0)
|
|
|
|
#print(bytes(bytes_out))
|
|
|
|
# CHECK IF FRAME CONTAINS ACK------------------------
|
|
if bytes(bytes_out[:6]) == b'REQACK':
|
|
|
|
logging.info("RX | ACK REQUESTED!")
|
|
time.sleep(5)
|
|
logging.info("TX | SENDING ACK FRAME")
|
|
self.Transmit(b'ACK')
|
|
#----------------------------------------------------
|
|
|
|
# CHECK IF FRAME CONTAINS ACK------------------------
|
|
if bytes(bytes_out[:3]) == b'ACK':
|
|
|
|
logging.info("TX | ACK RCVD!")
|
|
static.ACK_TIMEOUT = 1 #Force timer to stop waiting
|
|
static.ACK_RECEIVED = 1 #Force data loops of TNC to stop and continue with next frame
|
|
|
|
#----------------------------------------------------
|
|
|
|
#return bytes(bytes_out[:-2])
|
|
|
|
|
|
|
|
|