FreeDATA/test/util_rx.py
2022-06-19 10:04:46 -04:00

259 lines
8.1 KiB
Python

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Receive-side station emulator for test frame tests over a high quality audio channel
using a physical sound card or STDIO.
Legacy test for sending / receiving connection test frames through the codec2 and
back through on the other station. Data injection initiates directly through
the codec2 API.
Invoked from CMake, test_highsnr_stdio_{P_C, P_P}_datacx.py, and many test_virtual[1-3]*.sh.
@author: DJ2LS
"""
import argparse
import ctypes
import sys
import time
import numpy as np
import sounddevice as sd
# pylint: disable=wrong-import-position
sys.path.insert(0, "..")
sys.path.insert(0, "../tnc")
from tnc import codec2
def util_rx():
args = parse_arguments()
if args.LIST:
devices = sd.query_devices(device=None, kind=None)
for index, device in enumerate(devices):
print(f"{index} {device['name']}")
index += 1
# pylint: disable=protected-access
sd._terminate()
sys.exit()
N_BURSTS = args.N_BURSTS
N_FRAMES_PER_BURST = args.N_FRAMES_PER_BURST
AUDIO_INPUT_DEVICE = args.AUDIO_INPUT_DEVICE
MODE = codec2.FREEDV_MODE[args.FREEDV_MODE].value
DEBUGGING_MODE = args.DEBUGGING_MODE
MAX_TIME = args.TIMEOUT
# AUDIO PARAMETERS
# v-- consider increasing if you get nread_exceptions > 0
AUDIO_FRAMES_PER_BUFFER = 2400 * 2
MODEM_SAMPLE_RATE = codec2.api.FREEDV_FS_8000
AUDIO_SAMPLE_RATE_RX = 48000
# make sure our resampler will work
assert (AUDIO_SAMPLE_RATE_RX / MODEM_SAMPLE_RATE) == codec2.api.FDMDV_OS_48 # type: ignore
# check if we want to use an audio device then do a pyaudio init
if AUDIO_INPUT_DEVICE != -1:
# auto search for loopback devices
if AUDIO_INPUT_DEVICE == -2:
loopback_list = []
devices = sd.query_devices(device=None, kind=None)
for index, device in enumerate(devices):
if "Loopback: PCM" in device["name"]:
print(index)
loopback_list.append(index)
if loopback_list:
# 0 = RX 1 = TX
AUDIO_INPUT_DEVICE = loopback_list[0]
print(f"loopback_list tx: {loopback_list}", file=sys.stderr)
else:
print("not enough audio loopback devices ready...")
print("you should wait about 30 seconds...")
sd._terminate()
sys.exit()
print(f"AUDIO INPUT DEVICE: {AUDIO_INPUT_DEVICE}", file=sys.stderr)
# audio stream init
stream_rx = sd.RawStream(
channels=1,
dtype="int16",
device=AUDIO_INPUT_DEVICE,
samplerate=AUDIO_SAMPLE_RATE_RX,
blocksize=4800,
)
stream_rx.start()
# ----------------------------------------------------------------
# DATA CHANNEL INITIALISATION
# open codec2 instance
freedv = ctypes.cast(codec2.api.freedv_open(MODE), ctypes.c_void_p)
# get number of bytes per frame for mode
bytes_per_frame = int(codec2.api.freedv_get_bits_per_modem_frame(freedv) / 8)
payload_bytes_per_frame = bytes_per_frame - 2
n_max_modem_samples = codec2.api.freedv_get_n_max_modem_samples(freedv)
bytes_out = ctypes.create_string_buffer(bytes_per_frame)
codec2.api.freedv_set_frames_per_burst(freedv, N_FRAMES_PER_BURST)
total_n_bytes = 0
rx_total_frames = 0
rx_frames = 0
rx_bursts = 0
rx_errors = 0
nread_exceptions = 0
timeout = time.time() + MAX_TIME
receive = True
audio_buffer = codec2.audio_buffer(AUDIO_FRAMES_PER_BUFFER * 2)
resampler = codec2.resampler()
# time meassurement
time_start = 0
time_end = 0
# Copy received 48 kHz to a file. Listen to this file with:
# aplay -r 48000 -f S16_LE rx48.raw
# Corruption of this file is a good way to detect audio card issues
frx = open("rx48.raw", mode="wb")
# initial number of samples we need
nin = codec2.api.freedv_nin(freedv)
while receive and time.time() < timeout:
if AUDIO_INPUT_DEVICE != -1:
try:
# data_in48k = stream_rx.read(AUDIO_FRAMES_PER_BUFFER, exception_on_overflow = True)
data_in48k, overflowed = stream_rx.read(AUDIO_FRAMES_PER_BUFFER) # type: ignore
except OSError as err:
print(err, file=sys.stderr)
# if str(err).find("Input overflowed") != -1:
# nread_exceptions += 1
# if str(err).find("Stream closed") != -1:
# print("Ending...")
# receive = False
else:
data_in48k = sys.stdin.buffer.read(AUDIO_FRAMES_PER_BUFFER * 2)
# insert samples in buffer
x = np.frombuffer(data_in48k, dtype=np.int16) # type: ignore
# print(x)
# x = data_in48k
x.tofile(frx)
if len(x) != AUDIO_FRAMES_PER_BUFFER:
receive = False
x = resampler.resample48_to_8(x)
audio_buffer.push(x)
# when we have enough samples call FreeDV Rx
while audio_buffer.nbuffer >= nin:
# start time measurement
time_start = time.time()
# demodulate audio
nbytes = codec2.api.freedv_rawdatarx(
freedv, bytes_out, audio_buffer.buffer.ctypes
)
time_end = time.time()
audio_buffer.pop(nin)
# call me on every loop!
nin = codec2.api.freedv_nin(freedv)
rx_status = codec2.api.freedv_get_rx_status(freedv)
if rx_status & codec2.api.FREEDV_RX_BIT_ERRORS:
rx_errors = rx_errors + 1
if DEBUGGING_MODE:
rx_status = codec2.api.rx_sync_flags_to_text[rx_status] # type: ignore
time_needed = time_end - time_start
print(
f"nin: {nin:5d} rx_status: {rx_status:4s} "
f"naudio_buffer: {audio_buffer.nbuffer:4d} time: {time_needed:4f}",
file=sys.stderr,
)
if nbytes:
total_n_bytes += nbytes
if nbytes == bytes_per_frame:
rx_total_frames += 1
rx_frames += 1
if rx_frames == N_FRAMES_PER_BURST:
rx_frames = 0
rx_bursts += 1
if rx_bursts == N_BURSTS:
receive = False
if time.time() >= timeout:
print("TIMEOUT REACHED")
time.sleep(0.01)
if nread_exceptions:
print(
f"nread_exceptions {nread_exceptions:d} - receive audio lost! "
"Consider increasing Pyaudio frames_per_buffer...",
file=sys.stderr,
)
print(
f"RECEIVED BURSTS: {rx_bursts} "
f"RECEIVED FRAMES: {rx_total_frames} "
f"RX_ERRORS: {rx_errors}",
file=sys.stderr,
)
frx.close()
# and at last check if we had an opened audio instance and close it
if AUDIO_INPUT_DEVICE != -1:
sd._terminate()
def parse_arguments():
# --------------------------------------------GET PARAMETER INPUTS
parser = argparse.ArgumentParser(description="Simons TEST TNC")
parser.add_argument("--bursts", dest="N_BURSTS", default=1, type=int)
parser.add_argument(
"--framesperburst", dest="N_FRAMES_PER_BURST", default=1, type=int
)
parser.add_argument(
"--mode", dest="FREEDV_MODE", type=str, choices=["datac0", "datac1", "datac3"]
)
parser.add_argument(
"--audiodev",
dest="AUDIO_INPUT_DEVICE",
default=-1,
type=int,
help="audio device number to use, use -2 to automatically select a loopback device",
)
parser.add_argument("--debug", dest="DEBUGGING_MODE", action="store_true")
parser.add_argument(
"--timeout",
dest="TIMEOUT",
default=10,
type=int,
help="Timeout (seconds) before test ends",
)
parser.add_argument(
"--list",
dest="LIST",
action="store_true",
help="list audio devices by number and exit",
)
args, _ = parser.parse_known_args()
return args
if __name__ == "__main__":
util_rx()