openwebrx-clone/csdr/csdr.py
2021-01-17 18:11:10 +01:00

888 lines
36 KiB
Python

"""
OpenWebRX csdr plugin: do the signal processing with csdr
This file is part of OpenWebRX,
an open-source SDR receiver software with a web UI.
Copyright (c) 2013-2015 by Andras Retzler <randras@sdr.hu>
Copyright (c) 2019-2020 by Jakob Ketterl <dd5jfk@darc.de>
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as
published by the Free Software Foundation, either version 3 of the
License, or (at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Affero General Public License for more details.
You should have received a copy of the GNU Affero General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
"""
import subprocess
import os
import signal
import threading
import math
from functools import partial
from owrx.kiss import KissClient, DirewolfConfig
from owrx.wsjt import Ft8Profile, WsprProfile, Jt9Profile, Jt65Profile, Ft4Profile, Fst4Profile, Fst4wProfile
from owrx.js8 import Js8Profiles
from owrx.audio import AudioChopper
from csdr.pipe import Pipe
import logging
logger = logging.getLogger(__name__)
class output(object):
def send_output(self, t, read_fn):
if not self.supports_type(t):
# TODO rewrite the output mechanism in a way that avoids producing unnecessary data
logger.warning("dumping output of type %s since it is not supported.", t)
threading.Thread(target=self.pump(read_fn, lambda x: None), name="csdr_pump_thread").start()
return
self.receive_output(t, read_fn)
def receive_output(self, t, read_fn):
pass
def pump(self, read, write):
def copy():
run = True
while run:
data = None
try:
data = read()
except ValueError:
pass
if data is None or (isinstance(data, bytes) and len(data) == 0):
run = False
else:
write(data)
return copy
def supports_type(self, t):
return True
class dsp(object):
def __init__(self, output):
self.samp_rate = 250000
self.output_rate = 11025
self.hd_output_rate = 44100
self.fft_size = 1024
self.fft_fps = 5
self.center_freq = 0
self.offset_freq = 0
self.low_cut = -4000
self.high_cut = 4000
self.bpf_transition_bw = 320 # Hz, and this is a constant
self.ddc_transition_bw_rate = 0.15 # of the IF sample rate
self.running = False
self.secondary_processes_running = False
self.audio_compression = "none"
self.fft_compression = "none"
self.demodulator = "nfm"
self.name = "csdr"
self.base_bufsize = 512
self.decimation = None
self.last_decimation = None
self.nc_port = None
self.csdr_dynamic_bufsize = False
self.csdr_print_bufsizes = False
self.csdr_through = False
self.squelch_level = -150
self.fft_averages = 50
self.wfm_deemphasis_tau = 50e-6
self.iqtee = False
self.iqtee2 = False
self.secondary_demodulator = None
self.secondary_fft_size = 1024
self.secondary_process_fft = None
self.secondary_process_demod = None
self.pipe_names = {
"bpf_pipe": Pipe.WRITE,
"shift_pipe": Pipe.WRITE,
"squelch_pipe": Pipe.WRITE,
"smeter_pipe": Pipe.READ,
"meta_pipe": Pipe.READ,
"iqtee_pipe": Pipe.NONE,
"iqtee2_pipe": Pipe.NONE,
"dmr_control_pipe": Pipe.WRITE,
}
self.pipes = {}
self.secondary_pipe_names = {"secondary_shift_pipe": Pipe.WRITE}
self.secondary_offset_freq = 1000
self.unvoiced_quality = 1
self.modification_lock = threading.Lock()
self.output = output
self.temporary_directory = None
self.pipe_base_path = None
self.set_temporary_directory("/tmp")
self.is_service = False
self.direwolf_config = None
self.direwolf_port = None
self.process = None
def set_service(self, flag=True):
self.is_service = flag
def set_temporary_directory(self, what):
self.temporary_directory = what
self.pipe_base_path = "{tmp_dir}/openwebrx_pipe_".format(tmp_dir=self.temporary_directory)
def chain(self, which):
chain = ["nc -v 127.0.0.1 {nc_port}"]
if self.csdr_dynamic_bufsize:
chain += ["csdr setbuf {start_bufsize}"]
if self.csdr_through:
chain += ["csdr through"]
if which == "fft":
chain += [
"csdr fft_cc {fft_size} {fft_block_size}",
"csdr logpower_cf -70"
if self.fft_averages == 0
else "csdr logaveragepower_cf -70 {fft_size} {fft_averages}",
"csdr fft_exchange_sides_ff {fft_size}",
]
if self.fft_compression == "adpcm":
chain += ["csdr compress_fft_adpcm_f_u8 {fft_size}"]
return chain
chain += ["csdr shift_addfast_cc --fifo {shift_pipe}"]
if self.decimation > 1:
chain += ["csdr fir_decimate_cc {decimation} {ddc_transition_bw} HAMMING"]
chain += ["csdr bandpass_fir_fft_cc --fifo {bpf_pipe} {bpf_transition_bw} HAMMING"]
if self.output.supports_type("smeter"):
chain += [
"csdr squelch_and_smeter_cc --fifo {squelch_pipe} --outfifo {smeter_pipe} 5 {smeter_report_every}"
]
if self.secondary_demodulator:
if self.output.supports_type("secondary_fft"):
chain += ["csdr tee {iqtee_pipe}"]
chain += ["csdr tee {iqtee2_pipe}"]
# early exit if we don't want audio
if not self.output.supports_type("audio"):
return chain
# safe some cpu cycles... no need to decimate if decimation factor is 1
last_decimation_block = []
if self.last_decimation >= 2.0:
# activate prefilter if signal has been oversampled, e.g. WFM
last_decimation_block = ["csdr fractional_decimator_ff {last_decimation} 12 --prefilter"]
elif self.last_decimation != 1.0:
last_decimation_block = ["csdr fractional_decimator_ff {last_decimation}"]
if which == "nfm":
chain += ["csdr fmdemod_quadri_cf", "csdr limit_ff"]
chain += last_decimation_block
chain += [
"csdr deemphasis_nfm_ff {audio_rate}",
"csdr agc_ff --profile slow --max 3",
]
if self.get_audio_rate() != self.get_output_rate():
chain += [
"sox -t raw -r {audio_rate} -e floating-point -b 32 -c 1 --buffer 32 - -t raw -r {output_rate} -e signed-integer -b 16 -c 1 - "
]
else:
chain += ["csdr convert_f_s16"]
elif which == "wfm":
chain += [
"csdr fmdemod_quadri_cf",
"csdr limit_ff",
]
chain += last_decimation_block
chain += [
"csdr deemphasis_wfm_ff {audio_rate} {wfm_deemphasis_tau}",
"csdr convert_f_s16"
]
elif self.isDigitalVoice(which):
chain += ["csdr fmdemod_quadri_cf"]
chain += last_decimation_block
# dsd modes
if which in ["dstar", "nxdn"]:
chain += ["dc_block", "csdr limit_ff", "csdr convert_f_s16"]
if which == "dstar":
chain += ["dsd -fd -i - -o - -u {unvoiced_quality} -g -1 "]
elif which == "nxdn":
chain += ["dsd -fi -i - -o - -u {unvoiced_quality} -g -1 "]
chain += [
"digitalvoice_filter",
"CSDR_FIXED_BUFSIZE=32 csdr agc_s16 --max 30 --initial 3",
"sox -t raw -r 8000 -e signed-integer -b 16 -c 1 --buffer 32 - -t raw -r {output_rate} -e signed-integer -b 16 -c 1 - ",
]
# m17
elif which == "m17":
chain += [
"csdr limit_ff",
"csdr convert_f_s16",
"m17-demod",
"CSDR_FIXED_BUFSIZE=32 csdr agc_s16 --max 30 --initial 3",
"sox -t raw -r 8000 -e signed-integer -b 16 -c 1 --buffer 32 - -t raw -r {output_rate} -e signed-integer -b 16 -c 1 - ",
]
# digiham modes
else:
chain += ["dc_block", "rrc_filter", "gfsk_demodulator"]
if which == "dmr":
chain += [
"dmr_decoder --fifo {meta_pipe} --control-fifo {dmr_control_pipe}",
"mbe_synthesizer -f -u {unvoiced_quality}",
]
elif which == "ysf":
chain += ["ysf_decoder --fifo {meta_pipe}", "mbe_synthesizer -y -f -u {unvoiced_quality}"]
max_gain = 0.005
chain += [
"digitalvoice_filter -f",
"CSDR_FIXED_BUFSIZE=32 csdr agc_ff --max 0.005 --initial 0.0005",
"sox -t raw -r 8000 -e floating-point -b 32 -c 1 --buffer 32 - -t raw -r {output_rate} -e signed-integer -b 16 -c 1 - ",
]
elif which == "am":
chain += ["csdr amdemod_cf", "csdr fastdcblock_ff"]
chain += last_decimation_block
chain += [
"csdr agc_ff --profile slow --initial 200",
"csdr convert_f_s16",
]
elif self.isFreeDV(which):
chain += ["csdr realpart_cf"]
chain += last_decimation_block
chain += [
"csdr agc_ff",
"csdr convert_f_s16",
"freedv_rx 1600 - -",
"csdr agc_s16 --max 30 --initial 3",
"sox -t raw -r 8000 -e signed-integer -b 16 -c 1 --buffer 32 - -t raw -r {output_rate} -e signed-integer -b 16 -c 1 - ",
]
elif self.isDrm(which):
if self.last_decimation != 1.0:
# we are still dealing with complex samples here, so the regular last_decimation_block doesn't fit
chain += ["csdr fractional_decimator_cc {last_decimation}"]
chain += [
"csdr convert_f_s16",
"dream -c 6 --sigsrate 48000 --audsrate 48000 -I - -O -",
"sox -t raw -r 48000 -e signed-integer -b 16 -c 2 - -t raw -r {output_rate} -e signed-integer -b 16 -c 1 - ",
]
elif which == "ssb":
chain += ["csdr realpart_cf"]
chain += last_decimation_block
chain += ["csdr agc_ff"]
# fixed sample rate necessary for the wsjt-x tools. fix with sox...
if self.get_audio_rate() != self.get_output_rate():
chain += [
"sox -t raw -r {audio_rate} -e floating-point -b 32 -c 1 --buffer 32 - -t raw -r {output_rate} -e signed-integer -b 16 -c 1 - "
]
else:
chain += ["csdr convert_f_s16"]
if self.audio_compression == "adpcm":
chain += ["csdr encode_ima_adpcm_i16_u8"]
return chain
def secondary_chain(self, which):
chain = ["cat {input_pipe}"]
if which == "fft":
chain += [
"csdr fft_cc {secondary_fft_input_size} {secondary_fft_block_size}",
"csdr logpower_cf -70"
if self.fft_averages == 0
else "csdr logaveragepower_cf -70 {secondary_fft_size} {fft_averages}",
"csdr fft_exchange_sides_ff {secondary_fft_input_size}",
]
if self.fft_compression == "adpcm":
chain += ["csdr compress_fft_adpcm_f_u8 {secondary_fft_size}"]
return chain
elif which == "bpsk31" or which == "bpsk63":
return chain + [
"csdr shift_addfast_cc --fifo {secondary_shift_pipe}",
"csdr bandpass_fir_fft_cc -{secondary_bpf_cutoff} {secondary_bpf_cutoff} {secondary_bpf_cutoff}",
"csdr simple_agc_cc 0.001 0.5",
"csdr timing_recovery_cc GARDNER {secondary_samples_per_bits} 0.5 2 --add_q",
"CSDR_FIXED_BUFSIZE=1 csdr dbpsk_decoder_c_u8",
"CSDR_FIXED_BUFSIZE=1 csdr psk31_varicode_decoder_u8_u8",
]
elif self.isWsjtMode(which) or self.isJs8(which):
chain += ["csdr realpart_cf"]
if self.last_decimation != 1.0:
chain += ["csdr fractional_decimator_ff {last_decimation}"]
return chain + ["csdr agc_ff", "csdr convert_f_s16"]
elif which == "packet":
chain += ["csdr fmdemod_quadri_cf"]
if self.last_decimation != 1.0:
chain += ["csdr fractional_decimator_ff {last_decimation}"]
return chain + ["csdr convert_f_s16", "direwolf -c {direwolf_config} -r {audio_rate} -t 0 -q d -q h 1>&2"]
elif which == "pocsag":
chain += ["csdr fmdemod_quadri_cf"]
if self.last_decimation != 1.0:
chain += ["csdr fractional_decimator_ff {last_decimation}"]
return chain + ["fsk_demodulator -i", "pocsag_decoder"]
def set_secondary_demodulator(self, what):
if self.get_secondary_demodulator() == what:
return
self.secondary_demodulator = what
self.calculate_decimation()
self.restart()
def secondary_fft_block_size(self):
base = (self.samp_rate / self.decimation) / (self.fft_fps * 2)
if self.fft_averages == 0:
return base
return base / self.fft_averages
def secondary_decimation(self):
return 1 # currently unused
def secondary_bpf_cutoff(self):
if self.secondary_demodulator == "bpsk31":
return 31.25 / self.if_samp_rate()
elif self.secondary_demodulator == "bpsk63":
return 62.5 / self.if_samp_rate()
return 0
def secondary_bpf_transition_bw(self):
if self.secondary_demodulator == "bpsk31":
return 31.25 / self.if_samp_rate()
elif self.secondary_demodulator == "bpsk63":
return 62.5 / self.if_samp_rate()
return 0
def secondary_samples_per_bits(self):
if self.secondary_demodulator == "bpsk31":
return int(round(self.if_samp_rate() / 31.25)) & ~3
elif self.secondary_demodulator == "bpsk63":
return int(round(self.if_samp_rate() / 62.5)) & ~3
return 0
def secondary_bw(self):
if self.secondary_demodulator == "bpsk31":
return 31.25
elif self.secondary_demodulator == "bpsk63":
return 62.5
def start_secondary_demodulator(self):
if not self.secondary_demodulator:
return
logger.debug("starting secondary demodulator from IF input sampled at %d" % self.if_samp_rate())
secondary_command_demod = " | ".join(self.secondary_chain(self.secondary_demodulator))
self.try_create_pipes(self.secondary_pipe_names, secondary_command_demod)
self.try_create_configs(secondary_command_demod)
secondary_command_demod = secondary_command_demod.format(
input_pipe=self.pipes["iqtee2_pipe"],
secondary_shift_pipe=self.pipes["secondary_shift_pipe"],
secondary_decimation=self.secondary_decimation(),
secondary_samples_per_bits=self.secondary_samples_per_bits(),
secondary_bpf_cutoff=self.secondary_bpf_cutoff(),
secondary_bpf_transition_bw=self.secondary_bpf_transition_bw(),
if_samp_rate=self.if_samp_rate(),
last_decimation=self.last_decimation,
audio_rate=self.get_audio_rate(),
direwolf_config=self.direwolf_config,
)
logger.debug("secondary command (demod) = %s", secondary_command_demod)
my_env = os.environ.copy()
# if self.csdr_dynamic_bufsize: my_env["CSDR_DYNAMIC_BUFSIZE_ON"]="1";
if self.csdr_print_bufsizes:
my_env["CSDR_PRINT_BUFSIZES"] = "1"
if self.output.supports_type("secondary_fft"):
secondary_command_fft = " | ".join(self.secondary_chain("fft"))
secondary_command_fft = secondary_command_fft.format(
input_pipe=self.pipes["iqtee_pipe"],
secondary_fft_input_size=self.secondary_fft_size,
secondary_fft_size=self.secondary_fft_size,
secondary_fft_block_size=self.secondary_fft_block_size(),
fft_averages=self.fft_averages,
)
logger.debug("secondary command (fft) = %s", secondary_command_fft)
self.secondary_process_fft = subprocess.Popen(
secondary_command_fft, stdout=subprocess.PIPE, shell=True, start_new_session=True, env=my_env
)
self.output.send_output(
"secondary_fft",
partial(self.secondary_process_fft.stdout.read, int(self.get_secondary_fft_bytes_to_read())),
)
# direwolf does not provide any meaningful data on stdout
# more specifically, it doesn't provide any data. if however, for any strange reason, it would start to do so,
# it would block if not read. by piping it to devnull, we avoid a potential pitfall here.
secondary_output = subprocess.DEVNULL if self.isPacket() else subprocess.PIPE
self.secondary_process_demod = subprocess.Popen(
secondary_command_demod, stdout=secondary_output, shell=True, start_new_session=True, env=my_env
)
self.secondary_processes_running = True
if self.isWsjtMode():
smd = self.get_secondary_demodulator()
chopper_profiles = None
if smd == "ft8":
chopper_profiles = [Ft8Profile()]
elif smd == "wspr":
chopper_profiles = [WsprProfile()]
elif smd == "jt65":
chopper_profiles = [Jt65Profile()]
elif smd == "jt9":
chopper_profiles = [Jt9Profile()]
elif smd == "ft4":
chopper_profiles = [Ft4Profile()]
elif smd == "fst4":
chopper_profiles = Fst4Profile.getEnabledProfiles()
elif smd == "fst4w":
chopper_profiles = Fst4wProfile.getEnabledProfiles()
if chopper_profiles is not None and len(chopper_profiles):
chopper = AudioChopper(self, self.secondary_process_demod.stdout, *chopper_profiles)
chopper.start()
self.output.send_output("wsjt_demod", chopper.read)
elif self.isJs8():
chopper = AudioChopper(self, self.secondary_process_demod.stdout, *Js8Profiles.getEnabledProfiles())
chopper.start()
self.output.send_output("js8_demod", chopper.read)
elif self.isPacket():
# we best get the ax25 packets from the kiss socket
kiss = KissClient(self.direwolf_port)
self.output.send_output("packet_demod", kiss.read)
elif self.isPocsag():
self.output.send_output("pocsag_demod", self.secondary_process_demod.stdout.readline)
else:
self.output.send_output("secondary_demod", partial(self.secondary_process_demod.stdout.read, 1))
# open control pipes for csdr and send initialization data
if self.has_pipe("secondary_shift_pipe"): # TODO digimodes
self.set_secondary_offset_freq(self.secondary_offset_freq) # TODO digimodes
def set_secondary_offset_freq(self, value):
self.secondary_offset_freq = value
if self.secondary_processes_running and self.has_pipe("secondary_shift_pipe"):
self.pipes["secondary_shift_pipe"].write("%g\n" % (-float(self.secondary_offset_freq) / self.if_samp_rate()))
def stop_secondary_demodulator(self):
if not self.secondary_processes_running:
return
self.try_delete_pipes(self.secondary_pipe_names)
self.try_delete_configs()
if self.secondary_process_fft:
try:
os.killpg(os.getpgid(self.secondary_process_fft.pid), signal.SIGTERM)
# drain any leftover data to free file descriptors
self.secondary_process_fft.communicate()
self.secondary_process_fft = None
except ProcessLookupError:
# been killed by something else, ignore
pass
if self.secondary_process_demod:
try:
os.killpg(os.getpgid(self.secondary_process_demod.pid), signal.SIGTERM)
# drain any leftover data to free file descriptors
self.secondary_process_demod.communicate()
self.secondary_process_demod = None
except ProcessLookupError:
# been killed by something else, ignore
pass
self.secondary_processes_running = False
def get_secondary_demodulator(self):
return self.secondary_demodulator
def set_secondary_fft_size(self, secondary_fft_size):
# to change this, restart is required
self.secondary_fft_size = secondary_fft_size
def set_audio_compression(self, what):
self.audio_compression = what
def get_audio_bytes_to_read(self):
# desired latency: 5ms
# uncompressed audio has 16 bits = 2 bytes per sample
base = self.output_rate * 0.005 * 2
# adpcm compresses the bitstream by 4
if self.audio_compression == "adpcm":
base = base / 4
return int(base)
def set_fft_compression(self, what):
self.fft_compression = what
def get_fft_bytes_to_read(self):
if self.fft_compression == "none":
return self.fft_size * 4
if self.fft_compression == "adpcm":
return int((self.fft_size / 2) + (10 / 2))
def get_secondary_fft_bytes_to_read(self):
if self.fft_compression == "none":
return self.secondary_fft_size * 4
if self.fft_compression == "adpcm":
return (self.secondary_fft_size / 2) + (10 / 2)
def set_samp_rate(self, samp_rate):
self.samp_rate = samp_rate
self.calculate_decimation()
if self.running:
self.restart()
def calculate_decimation(self):
(self.decimation, self.last_decimation, _) = self.get_decimation(self.samp_rate, self.get_audio_rate())
def get_decimation(self, input_rate, output_rate):
decimation = 1
correction = 1
# wideband fm has a much higher frequency deviation (75kHz).
# we cannot cover this if we immediately decimate to the sample rate the audio will have later on, so we need
# to compensate here.
# the factor of 6 is by experimentation only, with a minimum audio rate of 36kHz (enforced by the client)
# this allows us to cover at least +/- 108kHz of frequency spectrum (may be higher, but that's the worst case).
# the correction factor is automatically compensated for by the secondary decimation stage, which comes
# after the demodulator.
if self.get_demodulator() == "wfm":
correction = 6
while input_rate / (decimation + 1) >= output_rate * correction:
decimation += 1
fraction = float(input_rate / decimation) / output_rate
intermediate_rate = input_rate / decimation
return decimation, fraction, intermediate_rate
def if_samp_rate(self):
return self.samp_rate / self.decimation
def get_name(self):
return self.name
def get_output_rate(self):
return self.output_rate
def get_hd_output_rate(self):
return self.hd_output_rate
def get_audio_rate(self):
if self.isDigitalVoice() or self.isPacket() or self.isPocsag() or self.isDrm():
return 48000
elif self.isWsjtMode() or self.isJs8():
return 12000
elif self.isFreeDV():
return 8000
elif self.isHdAudio():
return self.get_hd_output_rate()
return self.get_output_rate()
def isDigitalVoice(self, demodulator=None):
if demodulator is None:
demodulator = self.get_demodulator()
return demodulator in ["dmr", "dstar", "nxdn", "ysf", "m17"]
def isWsjtMode(self, demodulator=None):
if demodulator is None:
demodulator = self.get_secondary_demodulator()
return demodulator in ["ft8", "wspr", "jt65", "jt9", "ft4", "fst4", "fst4w"]
def isJs8(self, demodulator = None):
if demodulator is None:
demodulator = self.get_secondary_demodulator()
return demodulator == "js8"
def isPacket(self, demodulator=None):
if demodulator is None:
demodulator = self.get_secondary_demodulator()
return demodulator == "packet"
def isPocsag(self, demodulator=None):
if demodulator is None:
demodulator = self.get_secondary_demodulator()
return demodulator == "pocsag"
def isFreeDV(self, demodulator=None):
if demodulator is None:
demodulator = self.get_demodulator()
return demodulator == "freedv"
def isHdAudio(self, demodulator=None):
if demodulator is None:
demodulator = self.get_demodulator()
return demodulator == "wfm"
def isDrm(self, demodulator=None):
if demodulator is None:
demodulator = self.get_demodulator()
return demodulator == "drm"
def set_output_rate(self, output_rate):
if self.output_rate == output_rate:
return
self.output_rate = output_rate
self.calculate_decimation()
self.restart()
def set_hd_output_rate(self, hd_output_rate):
if self.hd_output_rate == hd_output_rate:
return
self.hd_output_rate = hd_output_rate
self.calculate_decimation()
self.restart()
def set_demodulator(self, demodulator):
if demodulator in ["usb", "lsb", "cw"]:
demodulator = "ssb"
if self.demodulator == demodulator:
return
self.demodulator = demodulator
self.calculate_decimation()
self.restart()
def get_demodulator(self):
return self.demodulator
def set_fft_size(self, fft_size):
self.fft_size = fft_size
self.restart()
def set_fft_fps(self, fft_fps):
self.fft_fps = fft_fps
self.restart()
def set_fft_averages(self, fft_averages):
self.fft_averages = fft_averages
self.restart()
def fft_block_size(self):
if self.fft_averages == 0:
return self.samp_rate / self.fft_fps
else:
return self.samp_rate / self.fft_fps / self.fft_averages
def set_offset_freq(self, offset_freq):
if offset_freq is None:
return
self.offset_freq = offset_freq
if self.running:
self.pipes["shift_pipe"].write("%g\n" % (-float(self.offset_freq) / self.samp_rate))
def set_center_freq(self, center_freq):
# dsp only needs to know this to be able to pass it to decoders in the form of get_operating_freq()
self.center_freq = center_freq
def get_operating_freq(self):
return self.center_freq + self.offset_freq
def set_bandpass(self, bandpass):
self.set_bpf(bandpass.low_cut, bandpass.high_cut)
def set_bpf(self, low_cut, high_cut):
self.low_cut = low_cut
self.high_cut = high_cut
if self.running:
self.pipes["bpf_pipe"].write(
"%g %g\n" % (float(self.low_cut) / self.if_samp_rate(), float(self.high_cut) / self.if_samp_rate())
)
def get_bpf(self):
return [self.low_cut, self.high_cut]
def convertToLinear(self, db):
return float(math.pow(10, db / 10))
def set_squelch_level(self, squelch_level):
self.squelch_level = squelch_level
# no squelch required on digital voice modes
actual_squelch = -150 if self.isDigitalVoice() or self.isPacket() or self.isPocsag() or self.isFreeDV() else self.squelch_level
if self.running:
self.pipes["squelch_pipe"].write("%g\n" % (self.convertToLinear(actual_squelch)))
def set_unvoiced_quality(self, q):
self.unvoiced_quality = q
self.restart()
def get_unvoiced_quality(self):
return self.unvoiced_quality
def set_dmr_filter(self, filter):
if self.has_pipe("dmr_control_pipe"):
self.pipes["dmr_control_pipe"].write("{0}\n".format(filter))
def set_wfm_deemphasis_tau(self, tau):
if self.wfm_deemphasis_tau == tau:
return
self.wfm_deemphasis_tau = tau
self.restart()
def ddc_transition_bw(self):
return self.ddc_transition_bw_rate * (self.if_samp_rate() / float(self.samp_rate))
def try_create_pipes(self, pipe_names, command_base):
for pipe_name, pipe_type in pipe_names.items():
if self.has_pipe(pipe_name):
logger.warning("{pipe_name} is still in use", pipe_name=pipe_name)
self.pipes[pipe_name].close()
if "{" + pipe_name + "}" in command_base:
p = self.pipe_base_path + pipe_name
encoding = None
# TODO make digiham output unicode and then change this here
# the whole pipe enoding feature onlye exists because of this
if pipe_name == "meta_pipe":
encoding = "cp437"
self.pipes[pipe_name] = Pipe.create(p, pipe_type, encoding=encoding)
else:
self.pipes[pipe_name] = None
def has_pipe(self, name):
return name in self.pipes and self.pipes[name] is not None
def try_delete_pipes(self, pipe_names):
for pipe_name in pipe_names:
if self.has_pipe(pipe_name):
self.pipes[pipe_name].close()
self.pipes[pipe_name] = None
def try_create_configs(self, command):
if "{direwolf_config}" in command:
self.direwolf_config = "{tmp_dir}/openwebrx_direwolf_{myid}.conf".format(
tmp_dir=self.temporary_directory, myid=id(self)
)
self.direwolf_port = KissClient.getFreePort()
file = open(self.direwolf_config, "w")
file.write(DirewolfConfig().getConfig(self.direwolf_port, self.is_service))
file.close()
else:
self.direwolf_config = None
self.direwolf_port = None
def try_delete_configs(self):
if self.direwolf_config:
try:
os.unlink(self.direwolf_config)
except FileNotFoundError:
# result suits our expectations. fine :)
pass
except Exception:
logger.exception("try_delete_configs()")
self.direwolf_config = None
def start(self):
with self.modification_lock:
if self.running:
return
self.running = True
command_base = " | ".join(self.chain(self.demodulator))
# create control pipes for csdr
self.try_create_pipes(self.pipe_names, command_base)
# send initial config through the pipes
if self.has_pipe("bpf_pipe"):
self.set_bpf(self.low_cut, self.high_cut)
if self.has_pipe("shift_pipe"):
self.set_offset_freq(self.offset_freq)
if self.has_pipe("squelch_pipe"):
self.set_squelch_level(self.squelch_level)
if self.has_pipe("dmr_control_pipe"):
self.set_dmr_filter(3)
# run the command
command = command_base.format(
bpf_pipe=self.pipes["bpf_pipe"],
shift_pipe=self.pipes["shift_pipe"],
squelch_pipe=self.pipes["squelch_pipe"],
smeter_pipe=self.pipes["smeter_pipe"],
meta_pipe=self.pipes["meta_pipe"],
iqtee_pipe=self.pipes["iqtee_pipe"],
iqtee2_pipe=self.pipes["iqtee2_pipe"],
dmr_control_pipe=self.pipes["dmr_control_pipe"],
decimation=self.decimation,
last_decimation=self.last_decimation,
fft_size=self.fft_size,
fft_block_size=self.fft_block_size(),
fft_averages=self.fft_averages,
bpf_transition_bw=float(self.bpf_transition_bw) / self.if_samp_rate(),
ddc_transition_bw=self.ddc_transition_bw(),
flowcontrol=int(self.samp_rate * 2),
start_bufsize=self.base_bufsize * self.decimation,
nc_port=self.nc_port,
output_rate=self.get_output_rate(),
smeter_report_every=int(self.if_samp_rate() / 6000),
unvoiced_quality=self.get_unvoiced_quality(),
audio_rate=self.get_audio_rate(),
wfm_deemphasis_tau=self.wfm_deemphasis_tau,
)
logger.debug("Command = %s", command)
my_env = os.environ.copy()
if self.csdr_dynamic_bufsize:
my_env["CSDR_DYNAMIC_BUFSIZE_ON"] = "1"
if self.csdr_print_bufsizes:
my_env["CSDR_PRINT_BUFSIZES"] = "1"
out = subprocess.PIPE if self.output.supports_type("audio") else subprocess.DEVNULL
self.process = subprocess.Popen(command, stdout=out, shell=True, start_new_session=True, env=my_env)
def watch_thread():
rc = self.process.wait()
logger.debug("dsp thread ended with rc=%d", rc)
if rc == 0 and self.running and not self.modification_lock.locked():
logger.debug("restarting since rc = 0, self.running = true, and no modification")
self.restart()
threading.Thread(target=watch_thread, name="csdr_watch_thread").start()
audio_type = "hd_audio" if self.isHdAudio() else "audio"
if self.output.supports_type(audio_type):
self.output.send_output(
audio_type,
partial(
self.process.stdout.read,
self.get_fft_bytes_to_read() if self.demodulator == "fft" else self.get_audio_bytes_to_read(),
),
)
self.start_secondary_demodulator()
if self.has_pipe("smeter_pipe"):
def read_smeter():
raw = self.pipes["smeter_pipe"].readline()
if len(raw) == 0:
return None
else:
return float(raw.rstrip("\n"))
self.output.send_output("smeter", read_smeter)
if self.has_pipe("meta_pipe"):
def read_meta():
raw = self.pipes["meta_pipe"].readline()
if len(raw) == 0:
return None
else:
return raw.rstrip("\n")
self.output.send_output("meta", read_meta)
if self.csdr_dynamic_bufsize:
self.process.stdout.read(8) # dummy read to skip bufsize & preamble
logger.debug("Note: CSDR_DYNAMIC_BUFSIZE_ON = 1")
def stop(self):
with self.modification_lock:
self.running = False
if self.process is not None:
try:
os.killpg(os.getpgid(self.process.pid), signal.SIGTERM)
# drain any leftover data to free file descriptors
self.process.communicate()
self.process = None
except ProcessLookupError:
# been killed by something else, ignore
pass
self.stop_secondary_demodulator()
self.try_delete_pipes(self.pipe_names)
def restart(self):
if not self.running:
return
self.stop()
self.start()