FreeDATA/modem/demodulator.py
2024-02-22 15:39:18 +01:00

396 lines
15 KiB
Python

import numpy as np
import codec2
import ctypes
import structlog
import threading
import audio
import os
from modem_frametypes import FRAME_TYPE
import itertools
TESTMODE = False
class Demodulator():
MODE_DICT = {}
# Iterate over the FREEDV_MODE enum members
for mode in codec2.FREEDV_MODE:
MODE_DICT[mode.value] = {
'decode': False,
'bytes_per_frame': None,
'bytes_out': None,
'audio_buffer': None,
'nin': None,
'instance': None,
'state_buffer': [],
'name': mode.name.upper(),
'decoding_thread': None
}
def __init__(self, config, audio_rx_q, modem_rx_q, data_q_rx, states, event_manager, fft_queue):
self.log = structlog.get_logger("Demodulator")
self.tuning_range_fmin = config['MODEM']['tuning_range_fmin']
self.tuning_range_fmax = config['MODEM']['tuning_range_fmax']
self.rx_audio_level = config['AUDIO']['rx_audio_level']
self.AUDIO_FRAMES_PER_BUFFER_RX = 4800
self.buffer_overflow_counter = [0, 0, 0, 0, 0, 0, 0, 0]
self.is_codec2_traffic_counter = 0
self.is_codec2_traffic_cooldown = 5
self.audio_received_queue = audio_rx_q
self.modem_received_queue = modem_rx_q
self.data_queue_received = data_q_rx
self.states = states
self.event_manager = event_manager
self.fft_queue = fft_queue
# init codec2 resampler
self.resampler = codec2.resampler()
self.init_codec2()
# enable decoding of signalling modes
self.MODE_DICT[codec2.FREEDV_MODE.signalling.value]["decode"] = True
tci_rx_callback_thread = threading.Thread(
target=self.tci_rx_callback,
name="TCI RX CALLBACK THREAD",
daemon=True,
)
tci_rx_callback_thread.start()
def init_codec2(self):
# Open codec2 instances
for mode in codec2.FREEDV_MODE:
self.init_codec2_mode(mode.value)
def init_codec2_mode(self, mode):
"""
Init codec2 and return some important parameters
"""
# create codec2 instance
c2instance = ctypes.cast(
codec2.api.freedv_open(mode), ctypes.c_void_p
)
# set tuning range
codec2.api.freedv_set_tuning_range(
c2instance,
ctypes.c_float(float(self.tuning_range_fmin)),
ctypes.c_float(float(self.tuning_range_fmax)),
)
# get bytes per frame
bytes_per_frame = int(
codec2.api.freedv_get_bits_per_modem_frame(c2instance) / 8
)
# create byte out buffer
bytes_out = ctypes.create_string_buffer(bytes_per_frame)
# set initial frames per burst
codec2.api.freedv_set_frames_per_burst(c2instance, 1)
# init audio buffer
audio_buffer = codec2.audio_buffer(2 * self.AUDIO_FRAMES_PER_BUFFER_RX)
# get initial nin
nin = codec2.api.freedv_nin(c2instance)
# Additional Datac0-specific information - these are not referenced anywhere else.
# self.signalling_datac0_payload_per_frame = self.signalling_datac0_bytes_per_frame - 2
# self.signalling_datac0_n_nom_modem_samples = codec2.api.freedv_get_n_nom_modem_samples(
# self.signalling_datac0_freedv
# )
# self.signalling_datac0_n_tx_modem_samples = codec2.api.freedv_get_n_tx_modem_samples(
# self.signalling_datac0_freedv
# )
# self.signalling_datac0_n_tx_preamble_modem_samples = (
# codec2.api.freedv_get_n_tx_preamble_modem_samples(self.signalling_datac0_freedv)
# )
# self.signalling_datac0_n_tx_postamble_modem_samples = (
# codec2.api.freedv_get_n_tx_postamble_modem_samples(self.signalling_datac0_freedv)
# )
self.MODE_DICT[mode]["instance"] = c2instance
self.MODE_DICT[mode]["bytes_per_frame"] = bytes_per_frame
self.MODE_DICT[mode]["bytes_out"] = bytes_out
self.MODE_DICT[mode]["audio_buffer"] = audio_buffer
self.MODE_DICT[mode]["nin"] = nin
def start(self, stream):
self.stream = stream
for mode in self.MODE_DICT:
# Start decoder threads
self.MODE_DICT[mode]['decoding_thread'] = threading.Thread(
target=self.demodulate_audio,args=[mode], name=self.MODE_DICT[mode]['name'], daemon=True
)
self.MODE_DICT[mode]['decoding_thread'].start()
def sd_input_audio_callback(self, indata: np.ndarray, frames: int, time, status) -> None:
audio_48k = np.frombuffer(indata, dtype=np.int16)
audio_8k = self.resampler.resample48_to_8(audio_48k)
audio_8k_level_adjusted = audio.set_audio_volume(audio_8k, self.rx_audio_level)
audio.calculate_fft(audio_8k_level_adjusted, self.fft_queue, self.states)
length_audio_8k_level_adjusted = len(audio_8k_level_adjusted)
# Avoid buffer overflow by filling only if buffer for
# selected datachannel mode is not full
index = 0
for mode in self.MODE_DICT:
mode_data = self.MODE_DICT[mode]
audiobuffer = mode_data['audio_buffer']
decode = mode_data['decode']
index += 1
if audiobuffer:
if (audiobuffer.nbuffer + length_audio_8k_level_adjusted) > audiobuffer.size:
self.buffer_overflow_counter[index] += 1
self.event_manager.send_buffer_overflow(self.buffer_overflow_counter)
elif decode:
audiobuffer.push(audio_8k_level_adjusted)
def worker_received(self) -> None:
"""Worker for FIFO queue for processing received frames"""
while True:
data = self.modem_received_queue.get()
self.log.debug("[MDM] worker_received: received data!")
# data[0] = bytes_out
# data[1] = freedv session
# data[2] = bytes_per_frame
# data[3] = snr
item = {
'payload': data[0],
'freedv': data[1],
'bytes_per_frame': data[2],
'snr': data[3],
'frequency_offset': self.get_frequency_offset(data[1]),
}
self.data_queue_received.put(item)
self.modem_received_queue.task_done()
def get_frequency_offset(self, freedv: ctypes.c_void_p) -> float:
"""
Ask codec2 for the calculated (audio) frequency offset of the received signal.
:param freedv: codec2 instance to query
:type freedv: ctypes.c_void_p
:return: Offset of audio frequency in Hz
:rtype: float
"""
modemStats = codec2.MODEMSTATS()
codec2.api.freedv_get_modem_extended_stats(freedv, ctypes.byref(modemStats))
offset = round(modemStats.foff) * (-1)
return offset
def demodulate_audio(self, mode) -> int:
"""
De-modulate supplied audio stream with supplied codec2 instance.
Decoded audio is placed into `bytes_out`.
"""
audiobuffer = self.MODE_DICT[mode]["audio_buffer"]
nin = self.MODE_DICT[mode]["nin"]
freedv = self.MODE_DICT[mode]["instance"]
bytes_out = self.MODE_DICT[mode]["bytes_out"]
bytes_per_frame= self.MODE_DICT[mode]["bytes_per_frame"]
state_buffer = self.MODE_DICT[mode]["state_buffer"]
mode_name = self.MODE_DICT[mode]["name"]
try:
while self.stream.active:
threading.Event().wait(0.01)
while audiobuffer.nbuffer >= nin:
# demodulate audio
nbytes = codec2.api.freedv_rawdatarx(
freedv, bytes_out, audiobuffer.buffer.ctypes
)
# get current modem states and write to list
# 1 trial
# 2 sync
# 3 trial sync
# 6 decoded
# 10 error decoding == NACK
rx_status = codec2.api.freedv_get_rx_status(freedv)
if rx_status not in [0]:
self.is_codec2_traffic_counter = self.is_codec2_traffic_cooldown
self.log.debug(
"[MDM] [demod_audio] modem state", mode=mode_name, rx_status=rx_status,
sync_flag=codec2.api.rx_sync_flags_to_text[rx_status]
)
# decrement codec traffic counter for making state smoother
if self.is_codec2_traffic_counter > 0:
self.is_codec2_traffic_counter -= 1
self.states.set_channel_busy_condition_codec2(True)
else:
self.states.set_channel_busy_condition_codec2(False)
if rx_status == 10:
state_buffer.append(rx_status)
audiobuffer.pop(nin)
nin = codec2.api.freedv_nin(freedv)
if nbytes == bytes_per_frame:
self.log.debug(
"[MDM] [demod_audio] Pushing received data to received_queue", nbytes=nbytes
)
snr = self.calculate_snr(freedv)
self.get_scatter(freedv)
self.modem_received_queue.put([bytes_out, freedv, bytes_per_frame, snr])
state_buffer = []
except Exception as e:
error_message = str(e)
# we expect this error when shutdown
if "PortAudio not initialized" in error_message:
e = None
self.log.debug(
"[MDM] [demod_audio] demod loop ended", mode=mode_name, e=e
)
def tci_rx_callback(self) -> None:
"""
Callback for TCI RX
data_in48k must be filled with 48000Hz audio raw data
"""
while True:
audio_48k = self.audio_received_queue.get()
audio_48k = np.frombuffer(audio_48k, dtype=np.int16)
audio.calculate_fft(audio_48k, self.fft_queue, self.states)
length_audio_48k = len(audio_48k)
index = 0
for mode in self.MODE_DICT:
mode_data = self.MODE_DICT[mode]
audiobuffer = mode_data['audio_buffer']
decode = mode_data['decode']
index += 1
if audiobuffer:
if (audiobuffer.nbuffer + length_audio_48k) > audiobuffer.size:
self.buffer_overflow_counter[index] += 1
self.event_manager.send_buffer_overflow(self.buffer_overflow_counter)
elif decode:
audiobuffer.push(audio_48k)
def set_frames_per_burst(self, frames_per_burst: int) -> None:
"""
Configure codec2 to send the configured number of frames per burst.
:param frames_per_burst: Number of frames per burst requested
:type frames_per_burst: int
"""
# Limit frames per burst to acceptable values
frames_per_burst = min(frames_per_burst, 1)
frames_per_burst = max(frames_per_burst, 5)
frames_per_burst = 1
codec2.api.freedv_set_frames_per_burst(self.dat0_datac1_freedv, frames_per_burst)
codec2.api.freedv_set_frames_per_burst(self.dat0_datac3_freedv, frames_per_burst)
codec2.api.freedv_set_frames_per_burst(self.dat0_datac4_freedv, frames_per_burst)
def calculate_snr(self, freedv: ctypes.c_void_p) -> float:
"""
Ask codec2 for data about the received signal and calculate
the signal-to-noise ratio.
:param freedv: codec2 instance to query
:type freedv: ctypes.c_void_p
:return: Signal-to-noise ratio of the decoded data
:rtype: float
"""
try:
modem_stats_snr = ctypes.c_float()
modem_stats_sync = ctypes.c_int()
codec2.api.freedv_get_modem_stats(
freedv, ctypes.byref(modem_stats_sync), ctypes.byref(modem_stats_snr)
)
modem_stats_snr = modem_stats_snr.value
modem_stats_sync = modem_stats_sync.value
snr = round(modem_stats_snr, 1)
self.log.info("[MDM] calculate_snr: ", snr=snr)
# snr = np.clip(
# snr, -127, 127
# ) # limit to max value of -128/128 as a possible fix of #188
return int(snr)
except Exception as err:
self.log.error(f"[MDM] calculate_snr: Exception: {err}")
return 0
def get_scatter(self, freedv: ctypes.c_void_p) -> None:
"""
Ask codec2 for data about the received signal and calculate the scatter plot.
:param freedv: codec2 instance to query
:type freedv: ctypes.c_void_p
"""
modemStats = codec2.MODEMSTATS()
ctypes.cast(
codec2.api.freedv_get_modem_extended_stats(freedv, ctypes.byref(modemStats)),
ctypes.c_void_p,
)
scatterdata = []
# original function before itertool
# for i in range(codec2.MODEM_STATS_NC_MAX):
# for j in range(1, codec2.MODEM_STATS_NR_MAX, 2):
# # print(f"{modemStats.rx_symbols[i][j]} - {modemStats.rx_symbols[i][j]}")
# xsymbols = round(modemStats.rx_symbols[i][j - 1] // 1000)
# ysymbols = round(modemStats.rx_symbols[i][j] // 1000)
# if xsymbols != 0.0 and ysymbols != 0.0:
# scatterdata.append({"x": str(xsymbols), "y": str(ysymbols)})
for i, j in itertools.product(range(codec2.MODEM_STATS_NC_MAX), range(1, codec2.MODEM_STATS_NR_MAX, 2)):
# print(f"{modemStats.rx_symbols[i][j]} - {modemStats.rx_symbols[i][j]}")
xsymbols = round(modemStats.rx_symbols[i][j - 1] // 1000)
ysymbols = round(modemStats.rx_symbols[i][j] // 1000)
if xsymbols != 0.0 and ysymbols != 0.0:
scatterdata.append({"x": str(xsymbols), "y": str(ysymbols)})
# Send all the data if we have too-few samples, otherwise send a sampling
if 150 > len(scatterdata) > 0:
self.event_manager.send_scatter_change(scatterdata)
else:
# only take every tenth data point
self.event_manager.send_scatter_change(scatterdata[::10])
def reset_data_sync(self) -> None:
"""
reset sync state for modes
:param frames_per_burst: Number of frames per burst requested
:type frames_per_burst: int
"""
for mode in self.MODE_DICT:
codec2.api.freedv_set_sync(self.MODE_DICT[mode]["instance"], 0)
def set_decode_mode(self, modes_to_decode):
# Reset all modes to not decode
for m in self.MODE_DICT:
self.MODE_DICT[m]["decode"] = False
# signalling is always true
self.MODE_DICT[codec2.FREEDV_MODE.signalling.value]["decode"] = True
# Enable specified modes
for mode, decode in modes_to_decode.items():
if mode in self.MODE_DICT:
self.MODE_DICT[mode]["decode"] = decode