Project

General

Profile

Bug #3527 » btech.py

Chris Bates, 03/28/2016 05:52 PM

 
# Copyright 2016:
# * Pavel Milanes CO7WT, <co7wt@frcuba.co.cu> <pavelmc@gmail.com>
# * Jim Unroe KC9HI, <rock.unroe@gmail.com>
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 2 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 General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <http://www.gnu.org/licenses/>.

import time
import struct
import logging

LOG = logging.getLogger(__name__)

from chirp import chirp_common, directory, memmap
from chirp import bitwise, errors, util
from chirp.settings import RadioSettingGroup, RadioSetting, \
RadioSettingValueBoolean, RadioSettingValueList, \
RadioSettingValueString, RadioSettingValueInteger, \
RadioSettings
from textwrap import dedent

MEM_FORMAT = """
#seekto 0x0000;
struct {
lbcd rxfreq[4];
lbcd txfreq[4];
ul16 rxtone;
ul16 txtone;
u8 unknown0:4,
scode:4;
u8 unknown1:2,
spmute:1,
unknown2:3,
optsig:2;
u8 unknown3:3,
scramble:1,
unknown4:3,
power:1;
u8 unknown5:1,
wide:1,
unknown6:2,
bcl:1,
add:1,
pttid:2;
} memory[200];

#seekto 0x1000;
struct {
char name[6];
u8 unknown1[10];
} names[200];

#seekto 0x3C90;
struct {
u8 vhf_low[3];
u8 vhf_high[3];
u8 uhf_low[3];
u8 uhf_high[3];
} ranges;

// the 2501+220 has a different zone for storing ranges

#seekto 0x3CD0;
struct {
u8 vhf_low[3];
u8 vhf_high[3];
u8 unknown1[4];
u8 unknown2[6];
u8 vhf2_low[3];
u8 vhf2_high[3];
u8 unknown3[4];
u8 unknown4[6];
u8 uhf_low[3];
u8 uhf_high[3];
} ranges220;

"""

# A note about the memmory in these radios
#
# The real memory of these radios extends to 0x4000
# On read the factory software only uses up to 0x3200
# On write it just uploads the contents up to 0x3100
#
# The mem beyond 0x3200 holds the ID data

MEM_SIZE = 0x4000
BLOCK_SIZE = 0x40
TX_BLOCK_SIZE = 0x10
ACK_CMD = "\x06"
MODES = ["FM", "NFM"]
SKIP_VALUES = ["S", ""]
TONES = chirp_common.TONES
DTCS = sorted(chirp_common.DTCS_CODES + [645])
NAME_LENGTH = 6
PTTID_LIST = ["OFF", "BOT", "EOT", "BOTH"]
PTTIDCODE_LIST = ["%s" % x for x in range(1, 16)]
OPTSIG_LIST = ["OFF", "DTMF", "2TONE", "5TONE"]

# this var controls the verbosity in the debug and by default it's low (False)
# make it True and you will to get a very verbose debug.log
debug = False

# Power Levels
NORMAL_POWER_LEVELS = [chirp_common.PowerLevel("High", watts=25),
chirp_common.PowerLevel("Low", watts=10)]
UV5001_POWER_LEVELS = [chirp_common.PowerLevel("High", watts=50),
chirp_common.PowerLevel("Low", watts=10)]

# this must be defined globaly
POWER_LEVELS = None

# valid chars on the LCD, Note that " " (space) is stored as "\xFF"
VALID_CHARS = chirp_common.CHARSET_ALPHANUMERIC + \
"`{|}!\"#$%&'()*+,-./:;<=>?@[]^_"


##### ID strings #####################################################

# BTECH UV2501 pre-production units
UV2501pp_id = "\x01\x03\x00\x01\x07\x09\x04\x00"
UV2501pp_id += "\x00\x05\x02\x00\x57\x48\x4B\x4A"
UV2501pp_id += "\x31\x36\x38\x4D\x49\x4E\x4D\x32"
UV2501pp_id += "\x43\x32\x39\x34\x55\x38\x38\x30"
UV2501pp_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV2501pp_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV2501pp_id += "\x55"
# fingerprint for the saved images (pre-production units)
UV2501pp_fid = "M2C294"


# BTECH UV2501 pre-production units
UV2501pp2_id = "\x01\x03\x06\x01\x07\x04\x04\x00"
UV2501pp2_id += "\x00\x04\x08\x00\x57\x48\x4B\x4A"
UV2501pp2_id += "\x31\x36\x38\x4D\x49\x4E\x4D\x32"
UV2501pp2_id += "\x39\x32\x30\x34\x55\x38\x38\x30"
UV2501pp2_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV2501pp2_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV2501pp2_id += "\x55"
# fingerprint for the saved images (pre-production units)
UV2501pp2_fid = "M29204"


# B-TECH UV-2501 first generation (1G)
UV2501_id = "\x01\x03\x00\x01\x07\x09\x04\x00"
UV2501_id += "\x00\x05\x02\x00\x57\x48\x4B\x4A"
UV2501_id += "\x31\x36\x38\x4D\x49\x4E\x4D\x32"
UV2501_id += "\x39\x32\x30\x34\x55\x38\x38\x30"
UV2501_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV2501_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV2501_id += "\x55"
# fingerprint for the saved images
UV2501_fid = "M29204"


# B-TECH UV-2501 second generation (2G)
UV2501G2_id = "\x01\x03\x00\x01\x07\x09\x04\x00"
UV2501G2_id += "\x00\x05\x02\x00\x57\x48\x4B\x4A"
UV2501G2_id += "\x31\x36\x38\x4D\x49\x4E\x42\x54"
UV2501G2_id += "\x47\x32\x31\x34\x55\x38\x38\x30"
UV2501G2_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV2501G2_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV2501G2_id += "\x55"
# fingerprint for the saved images
UV2501G2_fid = "BTG214"


# NOTE:
# About the ID2 for the 2501+220, that is a representative amount of data
# that must no pass the 16 bytes count, if right padding on the log
# please remove it from here, alse remove the header

# B-TECH UV-2501+220 pre-production units
UV2501_220pp_id = "\x01\x03\x00\x01\x07\x09\x00\x00"
UV2501_220pp_id += "\x00\x00\x4D\x49\x4E\x31\x32\x35"
UV2501_220pp_id += "\x02\x01\x00\x02\x03\x00\x00\x00"
UV2501_220pp_id += "\x00\x00\x4D\x33\x43\x32\x38\x31"
UV2501_220pp_id += "\x04\x00\x00\x05\x02\x00\x00\x00"
UV2501_220pp_id += "\x00\x00\x00\x00\x00\x00\x00\x00"
UV2501_220pp_id += "\x55"
# fingerprint for the saved images (pre-production units)
UV2501_220pp_fid = "M3C281"
# extra block read for the 2501+220 pre-production units
UV2501_220pp_id2 = " 280528"


# B-TECH UV-2501+220
UV2501_220_id = "\x01\x03\x00\x01\x07\x09\x00\x00"
UV2501_220_id += "\x00\x00\x4D\x49\x4E\x31\x32\x35"
UV2501_220_id += "\x02\x01\x00\x02\x03\x00\x00\x00"
UV2501_220_id += "\x00\x00\x4D\x33\x47\x32\x30\x31"
UV2501_220_id += "\x04\x00\x00\x05\x02\x00\x00\x00"
UV2501_220_id += "\x00\x00\x00\x00\x00\x00\x00\x00"
UV2501_220_id += "\x55"
# fingerprint for the saved images
UV2501_220_fid = "M3G201"
# extra block read for the 2501+220
UV2501_220_id2 = UV2501_220pp_id2


# B-TECH UV-5001 pre-production units
UV5001pp_id = "\x01\x03\x06\x01\x07\x04\x04\x00"
UV5001pp_id += "\x00\x04\x08\x00\x57\x48\x4B\x4A"
UV5001pp_id += "\x55\x56\x2D\x31\x36\x38\x56\x31"
UV5001pp_id += "\x39\x32\x30\x34\x55\x38\x38\x30"
UV5001pp_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV5001pp_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV5001pp_id += "\x55"
# fingerprint for the saved images
UV5001pp_fid = "V19204"


# B-TECH UV-5001 alpha units
UV5001alpha_id = "\x01\x03\x00\x01\x07\x09\x04\x00"
UV5001alpha_id += "\x00\x05\x02\x00\x57\x48\x4B\x4A"
UV5001alpha_id += "\x55\x56\x31\x36\x38\x38\x56\x32"
UV5001alpha_id += "\x38\x32\x30\x34\x55\x38\x38\x30"
UV5001alpha_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV5001alpha_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV5001alpha_id += "\x55"
# fingerprint for the saved images
UV5001alpha_fid = "V19204"


# B-TECH UV-5001 first generation (1G)
UV5001_id = "\x01\x03\x00\x01\x07\x09\x04\x00"
UV5001_id += "\x00\x05\x02\x00\x57\x48\x4B\x4A"
UV5001_id += "\x55\x56\x2D\x31\x36\x38\x56\x31"
UV5001_id += "\x39\x32\x30\x34\x55\x38\x38\x30"
UV5001_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV5001_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV5001_id += "\x55"
# fingerprint for the saved images
UV5001_fid = "V19204"


# B-TECH UV-5001 second generation (2G)
UV5001G2_id = "\x01\x03\x06\x01\x07\x04\x04\x00"
UV5001G2_id += "\x00\x04\x08\x00\x57\x48\x4B\x4A"
UV5001G2_id += "\x55\x56\x35\x30\x30\x31\x42\x54"
UV5001G2_id += "\x47\x32\x31\x34\x55\x38\x38\x30"
UV5001G2_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV5001G2_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
UV5001G2_id += "\x55"
# fingerprint for the saved images
UV5001G2_fid = "V2G204"


# WACCOM Mini-8900
MINI8900_id = "\x01\x03\x06\x01\x07\x04\x04\x00"
MINI8900_id += "\x00\x04\x08\x00\x57\x48\x4B\x4A"
MINI8900_id += "\x48\x54\x59\x32\x38\x38\x4D\x32"
MINI8900_id += "\x38\x38\x35\x34\x55\x38\x38\x30"
MINI8900_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
MINI8900_id += "\x30\x30\x30\x30\x30\x30\x30\x30"
MINI8900_id += "\x55"
# fingerprint for the saved images
MINI8900_fid = "M28854"


#### MAGICS
# for the Waccom Mini-8900
MSTRING_MINI8900 = "\x55\xA5\xB5\x45\x55\x45\x4d\x02"
# for the B-TECH UV-2501+220 (including pre production ones)
MSTRING_220 = "\x55\x20\x15\x12\x12\x01\x4d\x02"
# magic string for all other models
MSTRING = "\x55\x20\x15\x09\x20\x45\x4d\x02"


def _rawrecv(radio, amount):
"""Raw read from the radio device, new approach, this time a byte at
a time as the original driver, the receive data has to be atomic"""
data = ""

try:
tdiff = 0
start = time.time()
maxtime = amount * 0.009

while len(data) < amount and tdiff < maxtime:
d = radio.pipe.read(1)
if len(d) == 1:
data += d

# Delta time
tdiff = time.time() - start

# DEBUG
if debug is True:
LOG.debug("time diff %.04f maxtime %.04f, data: %d" %
(tdiff, maxtime, len(data)))

# DEBUG
if debug is True:
LOG.debug("<== (%d) bytes:\n\n%s" %
(len(data), util.hexprint(data)))

if len(data) < amount:
LOG.error("Short reading %d bytes from the %d requested." %
(len(data), amount))

except:
raise errors.RadioError("Error reading data from radio")

return data


def _rawsend(radio, data):
"""Raw send to the radio device"""
try:
for byte in data:
radio.pipe.write(byte)
time.sleep(0.003)

# DEBUG
if debug is True:
LOG.debug("==> (%d) bytes:\n\n%s" %
(len(data), util.hexprint(data)))
except:
raise errors.RadioError("Error sending data to radio")


def _make_frame(cmd, addr, length, data=""):
"""Pack the info in the headder format"""
frame = "\x06" + struct.pack(">BHB", ord(cmd), addr, length)
# add the data if set
if len(data) != 0:
frame += data

return frame


def _send(radio, frame, pause=0):
"""Generic send data to the radio"""
_rawsend(radio, frame)

# make a *optional* pause, to allow to build for an answer
if pause != 0:
time.sleep(pause)


def _recv(radio, addr):
"""Get data from the radio """
# 1 byte ACK +
# 4 bytes header +
# data of length of data (as I see always 0x40 = 64 bytes)

# catching ack
ack = _rawrecv(radio, 1)

# checking for a response
if len(ack) != 1:
msg = "No response in the read of the block #0x%04x" % addr
LOG.error(msg)
raise errors.RadioError(msg)

# valid data
if ack != ACK_CMD:
msg = "Bad ack received from radio in block 0x%04x" % addr
LOG.error(msg)
LOG.debug("Bad ACK was 0x%02x" % ord(ack))
raise errors.RadioError(msg)

# Get the header + basic sanitize
hdr = _rawrecv(radio, 4)
if len(hdr) != 4:
msg = "Short header for block: 0x%04x" % addr
LOG.error(msg)
raise errors.RadioError(msg)

# receive and validate the header
c, a, l = struct.unpack(">BHB", hdr)
if a != addr or l != BLOCK_SIZE or c != ord("X"):
msg = "Invalid answer for block 0x%04x:" % addr
LOG.error(msg)
LOG.debug("CMD: %s ADDR: %04x SIZE: %02x" % (c, a, l))
raise errors.RadioError(msg)

# Get the data
data = _rawrecv(radio, l)

# basic validation
if len(data) != l:
msg = "Short block of data in block #0x%04x" % addr
LOG.error(msg)
raise errors.RadioError(msg)

return data


def _do_magic(radio, status):
"""Try to put the radio in program mode and get the ident string
it will make multiple tries"""

# how many tries
tries = 5

# prep the data to show in the UI
status.cur = 0
status.msg = "Identifying the radio..."
status.max = len(radio._magic) * tries
radio.status_fn(status)
mc = 0

try:
# do the magic
for magic in radio._magic:
# we try a few times
for a in range(0, tries):
# Update the UI
status.cur = (mc * tries) + a
radio.status_fn(status)

# cleaning the serial buffer, try wrapped
try:
radio.pipe.flushInput()
except:
msg = "Error with a serial rx buffer flush at _do_magic"
LOG.error(msg)
raise errors.RadioError(msg)

# send the magic a byte at a time
for byte in magic:
ack = _rawrecv(radio, 1)
_send(radio, byte)

# A explicit time delay, with a longer one for the UV-5001
if "5001" in radio.MODEL:
time.sleep(0.5)
else:
time.sleep(0.1)

# Now you get a x06 of ACK if all goes well
ack = _rawrecv(radio, 1)

if ack == "\x06":
# DEBUG
LOG.info("Magic ACK received")
status.msg = "Positive Ident!"
status.cur = status.max
radio.status_fn(status)

return True

# increment the count of magics to send, this is for the UI status
mc += 1

# wait between tries for different MAGICs to allow the radio to
# timeout, this is an experimental fature
time.sleep(3)

except errors.RadioError:
raise
except Exception, e:
msg = "Unknown error sending Magic to radio:\n%s" % e
raise errors.RadioError(msg)

return False


def _do_ident(radio, status):
"""Put the radio in PROGRAM mode & identify it"""
# set the serial discipline
radio.pipe.setBaudrate(9600)
radio.pipe.setParity("N")
radio.pipe.setTimeout(0.005)
# cleaning the serial buffer, try wrapped
try:
radio.pipe.flushInput()
except:
msg = "Error with a serial rx buffer flush at _do_ident"
LOG.error(msg)
raise errors.RadioError(msg)

# do the magic trick
if _do_magic(radio, status) is False:
msg = "Radio did not respond to magic string, check your cable."
LOG.error(msg)
raise errors.RadioError(msg)

# Ok, get the ident string
ident = _rawrecv(radio, 49)

# basic check for the ident
if len(ident) != 49:
msg = "Radio send a sort ident block, you need to increase maxtime."
LOG.error(msg)
raise errors.RadioError(msg)

# check if ident is OK
if not ident in radio.IDENT:
# bad ident
msg = "Incorrect model ID, got this:\n\n"
msg += util.hexprint(ident)
LOG.debug(msg)
raise errors.RadioError("Radio identification failed.")

# DEBUG
LOG.info("Positive ident, this is a %s" % radio.MODEL)

# Ok, we have a radio in the other end, we need a pause here
time.sleep(0.01)

# the 2501+220 has one more check:
# reading the block 0x3DF0 to see if it's a code inside
if "+220" in radio.MODEL:
# DEBUG
LOG.debug("This is a BTECH UV-2501+220, requesting the extra ID")
# send the read request
_send(radio, _make_frame("S", 0x3DF0, 16), 0.04)
id2 = _rawrecv(radio, 20)
# WARNING !!!!!!
# Different versions send as response with a different amount of data
# it seems that it's padded with \xff, \x20 and some times with \x00
# we just care about the first 16, our magic string is in there
if len(id2) < 16:
msg = "The extra UV-2501+220 ID is short, aborting."
# DEBUG
LOG.error(msg)
raise errors.RadioError(msg)

# ok, check for it, any of the correct If must be in the received data
itis = False
for eid in radio._id2:
if eid in id2:
# DEBUG
LOG.info("Confirmed, this is a BTECH UV-2501+220")
# set the flag and exit
itis = True
break

# It is a UV-2501+220?
if itis is False:
msg = "The extra UV-2501+220 ID is wrong, aborting."
# DEBUG
LOG.error(msg)
LOG.debug("Full extra ID on the 2501+220 is: \n%s" %
util.hexprint(id2))
raise errors.RadioError(msg)

return True


def _download(radio):
"""Get the memory map"""

# UI progress
status = chirp_common.Status()

# put radio in program mode and identify it
_do_ident(radio, status)

# the first dummy packet for all model but the 2501+220
if not "+220" in radio.MODEL:
# In the logs we have found that the first block is discarded
# this is the \x05 in ack one, so we will simulate it here
_send(radio, _make_frame("S", 0, BLOCK_SIZE), 0.1)
discard = _rawrecv(radio, BLOCK_SIZE)

if debug is True:
LOG.info("Dummy first block read done, got this:\n\n")
LOG.debug(util.hexprint(discard))

# reset the progress bar in the UI
status.max = MEM_SIZE / BLOCK_SIZE
status.msg = "Cloning from radio..."
status.cur = 0
radio.status_fn(status)

data = ""
for addr in range(0, MEM_SIZE, BLOCK_SIZE):
# flush input, as per the original driver behavior, try wrapped
try:
radio.pipe.flushInput()
except:
msg = "Error with a serial rx buffer flush at _download"
LOG.error(msg)
raise errors.RadioError(msg)

# sending the read request
_send(radio, _make_frame("S", addr, BLOCK_SIZE), 0.1)

# read
d = _recv(radio, addr)

# aggregate the data
data += d

# UI Update
status.cur = addr / BLOCK_SIZE
status.msg = "Cloning from radio..."
radio.status_fn(status)

return data


def _upload(radio):
"""Upload procedure"""

# The UPLOAD mem is restricted to lower than 0x3100,
# so we will overide that here localy
MEM_SIZE = 0x3100

# UI progress
status = chirp_common.Status()

# put radio in program mode and identify it
_do_ident(radio, status)

# get the data to upload to radio
data = radio.get_mmap()

# Reset the UI progress
status.max = MEM_SIZE / TX_BLOCK_SIZE
status.cur = 0
status.msg = "Cloning to radio..."
radio.status_fn(status)

# the fun start here
for addr in range(0, MEM_SIZE, TX_BLOCK_SIZE):
# flush input, as per the original driver behavior, try wrapped
try:
radio.pipe.flushInput()
except:
msg = "Error with a serial rx buffer flush at _upload"
LOG.error(msg)
raise errors.RadioError(msg)

# sending the data
d = data[addr:addr + TX_BLOCK_SIZE]
_send(radio, _make_frame("X", addr, TX_BLOCK_SIZE, d), 0.015)

# receiving the response
ack = _rawrecv(radio, 1)

# basic check
if len(ack) != 1:
msg = "No response in the write of block #0x%04x" % addr
LOG.error(msg)
raise errors.RadioError(msg)

if not ack in "\x06\x05":
msg = "Bad ack writing block 0x%04x:" % addr
LOG.info(msg)
raise errors.RadioError(msg)

# UI Update
status.cur = addr / TX_BLOCK_SIZE
status.msg = "Cloning to radio..."
radio.status_fn(status)


def model_match(cls, data):
"""Match the opened/downloaded image to the correct version"""
rid = data[0x3f70:0x3f76]

if rid in cls._fileid:
return True

return False


def _decode_ranges(low, high):
"""Unpack the data in the ranges zones in the memap and return
a tuple with the integer corresponding to the Mhz it means"""
ilow = int(low[0]) * 100 \
+ int(low[1]) * 10 \
+ int(low[2])
ihigh = int(high[0]) * 100 \
+ int(high[1]) * 10 \
+ int(high[2])
ilow *= 1000000
ihigh *= 1000000

return (ilow, ihigh)


class btech(chirp_common.CloneModeRadio, chirp_common.ExperimentalRadio):
"""BTECH's UV-5001 and alike radios"""
VENDOR = "BTECH"
MODEL = ""
IDENT = ""
_vhf_range = (130000000, 179000000)
_220_range = (220000000, 240000000)
_uhf_range = (400000000, 520000000)
_upper = 199
_magic = None
_fileid = None

@classmethod
def get_prompts(cls):
rp = chirp_common.RadioPrompts()
rp.experimental = \
('This driver is experimental and for personal use only.\n'
'\n'
'Please keep a copy of you memories with the original software '
'if you treasure them, this is the first release and may contain'
' bugs.\n'
'\n'
'You will miss the setting tab, we are working on it. Your '
'success/failure story is appreciated, visit the Chirp\'s '
'website and drop us a comment or just say THANKS if it works '
'for you.\n'
)
rp.pre_download = _(dedent("""\
Follow this instructions to download your info:

1 - Turn off your radio
2 - Connect your interface cable
3 - Turn on your radio
4 - Do the download of your radio data

"""))
rp.pre_upload = _(dedent("""\
Follow this instructions to upload your info:

1 - Turn off your radio
2 - Connect your interface cable
3 - Turn on your radio
4 - Do the upload of your radio data

"""))
return rp

def get_features(self):
"""Get the radio's features"""

# we will use the following var as global
global POWER_LEVELS

rf = chirp_common.RadioFeatures()
rf.has_settings = False
rf.has_bank = False
rf.has_tuning_step = False
rf.can_odd_split = True
rf.has_name = True
rf.has_offset = True
rf.has_mode = True
rf.has_dtcs = True
rf.has_rx_dtcs = True
rf.has_dtcs_polarity = True
rf.has_ctone = True
rf.has_cross = True
rf.valid_modes = MODES
rf.valid_characters = VALID_CHARS
rf.valid_name_length = NAME_LENGTH
rf.valid_duplexes = ["", "-", "+", "split", "off"]
rf.valid_tmodes = ['', 'Tone', 'TSQL', 'DTCS', 'Cross']
rf.valid_cross_modes = [
"Tone->Tone",
"DTCS->",
"->DTCS",
"Tone->DTCS",
"DTCS->Tone",
"->Tone",
"DTCS->DTCS"]
rf.valid_skips = SKIP_VALUES
rf.valid_dtcs_codes = DTCS
rf.memory_bounds = (0, self._upper)

# power levels
if self.MODEL == "UV-5001":
POWER_LEVELS = UV5001_POWER_LEVELS # Higher power (50W)
else:
POWER_LEVELS = NORMAL_POWER_LEVELS # Lower power (25W)

rf.valid_power_levels = POWER_LEVELS

# bands
rf.valid_bands = [self._vhf_range, self._uhf_range]

# 2501+220
if self.MODEL == "UV-2501+220":
rf.valid_bands.append(self._220_range)

return rf

def sync_in(self):
"""Download from radio"""
data = _download(self)
self._mmap = memmap.MemoryMap(data)
self.process_mmap()

def sync_out(self):
"""Upload to radio"""
try:
_upload(self)
except errors.RadioError:
raise
except Exception, e:
raise errors.RadioError("Error: %s" % e)

def set_options(self):
"""This is to read the options from the image and set it in the
environment, for now just the limits of the freqs in the VHF/UHF
ranges"""

# setting the correct ranges for each radio type
if self.MODEL == "UV-2501+220":
# the model 2501+220 has a segment in 220
# and a different position in the memmap
ranges = self._memobj.ranges220
else:
ranges = self._memobj.ranges

# the normal dual bands
vhf = _decode_ranges(ranges.vhf_low, ranges.vhf_high)
uhf = _decode_ranges(ranges.uhf_low, ranges.uhf_high)

# DEBUG
LOG.info("Radio ranges: VHF %d to %d" % vhf)
LOG.info("Radio ranges: UHF %d to %d" % uhf)

# 220Mhz case
if self.MODEL == "UV-2501+220":
vhf2 = _decode_ranges(ranges.vhf2_low, ranges.vhf2_high)
LOG.info("Radio ranges: VHF(220) %d to %d" % vhf2)
self._220_range = vhf2

# set the class with the real data
self._vhf_range = vhf
self._uhf_range = uhf

def process_mmap(self):
"""Process the mem map into the mem object"""

# Get it
self._memobj = bitwise.parse(MEM_FORMAT, self._mmap)

# load specific parameters from the radio image
self.set_options()

def get_raw_memory(self, number):
return repr(self._memobj.memory[number])

def _decode_tone(self, val):
"""Parse the tone data to decode from mem, it returns:
Mode (''|DTCS|Tone), Value (None|###), Polarity (None,N,R)"""
pol = None

if val in [0, 65535]:
return '', None, None
elif val > 0x0258:
a = val / 10.0
return 'Tone', a, pol
else:
if val > 0x69:
index = val - 0x6A
pol = "R"
else:
index = val - 1
pol = "N"

tone = DTCS[index]
return 'DTCS', tone, pol

def _encode_tone(self, memval, mode, val, pol):
"""Parse the tone data to encode from UI to mem"""
if mode == '' or mode is None:
memval.set_raw("\x00\x00")
elif mode == 'Tone':
memval.set_value(val * 10)
elif mode == 'DTCS':
# detect the index in the DTCS list
try:
index = DTCS.index(val)
if pol == "N":
index += 1
else:
index += 0x6A
memval.set_value(index)
except:
msg = "Digital Tone '%d' is not supported" % value
LOG.error(msg)
raise errors.RadioError(msg)
else:
msg = "Internal error: invalid mode '%s'" % mode
LOG.error(msg)
raise errors.InvalidDataError(msg)

def get_memory(self, number):
"""Get the mem representation from the radio image"""
_mem = self._memobj.memory[number]
_names = self._memobj.names[number]

# Create a high-level memory object to return to the UI
mem = chirp_common.Memory()

# Memory number
mem.number = number

if _mem.get_raw()[0] == "\xFF":
mem.empty = True
return mem

# Freq and offset
mem.freq = int(_mem.rxfreq) * 10
# tx freq can be blank
if _mem.get_raw()[4] == "\xFF":
# TX freq not set
mem.offset = 0
mem.duplex = "off"
else:
# TX feq set
offset = (int(_mem.txfreq) * 10) - mem.freq
if offset != 0:
if offset > 70000000: # 70 Mhz
mem.duplex = "split"
mem.offset = int(_mem.txfreq) * 10
elif offset < 0:
mem.offset = abs(offset)
mem.duplex = "-"
elif offset > 0:
mem.offset = offset
mem.duplex = "+"
else:
mem.offset = 0

# name TAG of the channel
mem.name = str(_names.name).rstrip("\xFF").replace("\xFF", " ")

# power
mem.power = POWER_LEVELS[int(_mem.power)]

# wide/narrow
mem.mode = MODES[int(_mem.wide)]

# skip
mem.skip = SKIP_VALUES[_mem.add]

# tone data
rxtone = txtone = None
txtone = self._decode_tone(_mem.txtone)
rxtone = self._decode_tone(_mem.rxtone)
chirp_common.split_tone_decode(mem, txtone, rxtone)

# Extra
mem.extra = RadioSettingGroup("extra", "Extra")

spmute = RadioSetting("spmute", "Speaker mute",
RadioSettingValueBoolean(bool(_mem.spmute)))
mem.extra.append(spmute)

scramble = RadioSetting("scramble", "Scramble",
RadioSettingValueBoolean(bool(_mem.scramble)))
mem.extra.append(scramble)

bcl = RadioSetting("bcl", "Busy channel lockout",
RadioSettingValueBoolean(bool(_mem.bcl)))
mem.extra.append(bcl)

pttid = RadioSetting("pttid", "PTT ID",
RadioSettingValueList(PTTID_LIST,
PTTID_LIST[_mem.pttid]))
mem.extra.append(pttid)

pttidcode = RadioSetting("scode", "PTT ID signal code",
RadioSettingValueList(
PTTIDCODE_LIST,
PTTIDCODE_LIST[_mem.scode]))
mem.extra.append(pttidcode)

optsig = RadioSetting("optsig", "Optional signaling",
RadioSettingValueList(
OPTSIG_LIST,
OPTSIG_LIST[_mem.optsig]))
mem.extra.append(optsig)

return mem

def set_memory(self, mem):
"""Set the memory data in the eeprom img from the UI"""
# get the eprom representation of this channel
_mem = self._memobj.memory[mem.number]
_names = self._memobj.names[mem.number]

# if empty memmory
if mem.empty:
# the channel itself
_mem.set_raw("\xFF" * 16)
# the name tag
_names.set_raw("\xFF" * 16)
return

# frequency
_mem.rxfreq = mem.freq / 10

# duplex
if mem.duplex == "+":
_mem.txfreq = (mem.freq + mem.offset) / 10
elif mem.duplex == "-":
_mem.txfreq = (mem.freq - mem.offset) / 10
elif mem.duplex == "off":
for i in _mem.txfreq:
i.set_raw("\xFF")
elif mem.duplex == "split":
_mem.txfreq = mem.offset / 10
else:
_mem.txfreq = mem.freq / 10

# tone data
((txmode, txtone, txpol), (rxmode, rxtone, rxpol)) = \
chirp_common.split_tone_encode(mem)
self._encode_tone(_mem.txtone, txmode, txtone, txpol)
self._encode_tone(_mem.rxtone, rxmode, rxtone, rxpol)

# name TAG of the channel
if len(mem.name) < NAME_LENGTH:
# we must pad to NAME_LENGTH chars, " " = "\xFF"
mem.name = str(mem.name).ljust(NAME_LENGTH, " ")
_names.name = str(mem.name).replace(" ", "\xFF")

# power, # default power level is high
_mem.power = 0 if mem.power is None else POWER_LEVELS.index(mem.power)

# wide/marrow
_mem.wide = MODES.index(mem.mode)

# scan add property
_mem.add = SKIP_VALUES.index(mem.skip)

# reseting unknowns, this have to be set by hand
_mem.unknown1 = 0
_mem.unknown2 = 0
_mem.unknown3 = 0
_mem.unknown4 = 0
_mem.unknown5 = 0
_mem.unknown6 = 0

# extra settings
if len(mem.extra) > 0:
# there are setting, parse
for setting in mem.extra:
setattr(_mem, setting.get_name(), setting.value)
else:
# there is no extra settings, load defaults
_mem.spmute = 0
_mem.optsig = 0
_mem.scramble = 0
_mem.bcl = 0
_mem.pttid = 0
_mem.scode = 0

return mem

@classmethod
def match_model(cls, filedata, filename):
match_size = False
match_model = False

# testing the file data size
if len(filedata) == MEM_SIZE:
match_size = True

# testing the firmware model fingerprint
match_model = model_match(cls, filedata)

if match_size and match_model:
return True
else:
return False


# Note:
# the order in the lists in the _magic, IDENT and _fileid is important
# we put the most common units first, the policy is as follows:

# - First lastest (newer) units, as they will be the most common
# - Second the former latest version, and recursively...
# - At the end the pre-production unitst (pp) as this will be unique

@directory.register
class UV2501(btech):
"""Baofeng Tech UV2501"""
MODEL = "UV-2501"
_magic = [MSTRING, ]
IDENT = [UV2501G2_id, UV2501_id, UV2501pp_id, UV2501pp2_id]
_fileid = [UV2501G2_fid, UV2501_fid, UV2501pp_fid]


@directory.register
class UV2501_220(btech):
"""Baofeng Tech UV2501+220"""
MODEL = "UV-2501+220"
_magic = [MSTRING_220, ]
IDENT = [UV2501_220_id, UV2501_220pp_id]
_fileid = [UV2501_220_fid, UV2501_220pp_fid]
_id2 = [UV2501_220_id2, UV2501_220pp_id2]


@directory.register
class UV5001(btech):
"""Baofeng Tech UV5001"""
MODEL = "UV-5001"
_magic = [MSTRING, ]
IDENT = [UV5001G2_id, UV5001_id, UV5001pp_id]
_fileid = [UV5001G2_fid, UV5001_fid, UV5001pp_fid]


@directory.register
class MINI8900(btech):
"""WACCOM MINI-8900"""
VENDOR = "WACCOM"
MODEL = "MINI-8900"
_magic = [MSTRING_MINI8900, ]
IDENT = [MINI8900_id, UV5001alpha_id]
_fileid = [MINI8900_fid, ]
(3-3/9)