This is only a preliminary version. There is certainly some room for
improvement, but the basic functionality is already in place. That
should be sufficient for daily use, and possibles issues can always be
fixed when discovered.
After the new firmware upgrade, up to three gas mixes are available
instead of only two. This causes the parsing to fail because there is
now an extra 6 byte gasmix block in the header.
Unfortunately, we can't rely on the firmware version to detect whether
this extra gasmix is present or not. In theory the dive computer can
contain dives in both the old and the new format. Because the firmware
version is a property of the device, and not stored inside each dive,
using the firmware version would either cause the old or the new dives
to fail parsing. This appears to be the approach DM4 is using.
According to some sources on the internet, the logbook gets erased
during the firmware update. However, according to the memory dumps I
received, that appears to be incorrect. The old dives are still
present, and it seems it's only DM4 that fails to download them.
So we need an alternative solution. After a detailed analysis of all
the Suunto dives at my disposal, I noticed something interesting. The
first 5 bytes of each dive are almost static. There are only 5
different variations:
0061906216
0061A06216
0062909118
0062C07118
0063C07118
The interpretation of these bytes is currently unknown, but the second
byte might be some kind of data format version. Each model always has
the same value here, and for the D6i it changes after the firmware
update:
0x61: D4, D6, D9, Cobra 2, Cobra3, Vyper 2, Vyper Air
0x62: D4i, D6i (old firmware), D9tx, HelO2
0x63: D6i (new firmware)
This can't be coincidence, so we use this byte to detect the presence
of the extra gasmix.
The HelO2, D4i, D6i and D9tx all use the same data format for the gas
mixes. The only difference is the number of gas mixes and the initial
byte offset. With this knowledge, we can easily use the same code for
all models. An additional advantage is that because the profile
configuration data is stored immediately after the gasmix section, we
can also replace the hardcoded offset with a simple calculation.
This macro was used to compensate for the fact that the 4 bytes at the
start of each dive, containing the previous and next dive pointers, are
stripped. With the SKIP macro the byte offset remained the same as in
the documentation. Nowadays, this compatibility isn't necessary anymore
and it only makes interpreting the raw binary data more difficult.
The max depth is stored in imperial units (feet), and should be
converted into metric units (meter). But for some reason this unit
conversion was omitted.
Apparantly there are also two different type of serial numbers present,
and their interpretation depends on the application. The Windows Dive
Organizer application shows both a serial number (byte offset 0x04) and
a warranty number (byte offset 0x0C). However, the Mac OS X Divers'
Diary application shows the number at byte offset 0x0C as the serial
number. Very confusing. For now, we just stick to the number at byte
offset 0x0C, because that's the number that is shown by the device
itself.
The serial numbers were not decoded at all. The raw bytes where simply
converted into an arbitrary 32 bit integer. Now the serial number
matches the real serial number as shown by the device.
The Cobalt tracks only the NDL time in the samples. There are however
two indications when the dive changes into a decompression dive, but
they don't always occur at the same time and may or may not appear in
the same sample. The first indication is that the NDL time goes to zero
and the other is the first occurrence of the "deco schedule computed"
bit (0x02) in the violation byte of the sample.
As soon as the NDL time goes to zero, an attempt is made to generate a
deco schedule. Depending upon the algorithm used, a schedule may or may
not have any stops, even though the NDL time is zero. If the schedule
does generate deco stops, the deco schedule computed bit is set in the
violation byte of the sample. This bit is set each time a non-zero
schedule is computed. But because this bit is immediately cleared after
the sample has been stored, and only set again at the completion of the
next schedule computation, not every sample will have this bit set.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The Matrix uses the same communication protocol as the Nemo Wide 2,
except that the sending the request packets with just a single write
operation doesn't seem to work. That's very surprising because it
caused no problems for the Nemo Wide 2 or the Icon HD!
The first two bytes of each request packet are probably some kind of
command type. These two bytes are answerred immediately with an ACK
byte (0xAA). Once this ACK byte has been received, the payload of the
command (if any) can be sent, and finally the response packet can be
received.
I suspect that when trying to send the entire command at once, the
device somehow doesn't receive the payload bytes correctly. Maybe it's
still busy processing those first two bytes, which causes the remainder
of the packet to get dropped? That might explain why the version
commands is not affected, because it doesn't have any payload bytes!
The application shouldn't have to deal with the delay between packets.
If the default value isn't good enough, that should be fixed internally
and not on the application side.
The second gas change event (type 0x06) contains both the oxygen and
helium percentages. These are now reported correctly with the new
GASCHANGE2 event.
The D9 family has begin and end of the deco event and we can therefore
convert this to the deco sample. For compatibility with existing software
we keep the events around.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The SAFETYSTOP is conceptually somewhere in between the NDL and the
DECOSTOP, so it makes sense to re-order the constants in the enum to
reflect this order.
These are based on the documentation we have and have been tested and
verified against actual dive data (with the exception of the pO2 and
ascent speed violations).
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
Two issues:
- the OSTC counts its gases 1-based, not 0-based
- dives don't always start with the first gas. Simply create a gas change
event right after the first time sample that informs the application what
the first gas mix is
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
The version.h header file is generated by autoconf and therefore
located in the build directory and not the source directory. If
building out-of-tree, and a version.h header file is accidentally
present in the source tree, the wrong file will be picked up.
By reversing the order of the include directories, the build directory
is searched first, and the correct header file will be used.
For bug reports it's very convenient to know the exact version. For
release builds, the standard version triplet (major.minor.micro) is
more than sufficient, but that's not the case for development builds.
Due to the post-release version increment, development builds already
have a version number that is distinct from previous releases, but
including the git commit SHA1 is even more accurate.
On Windows, the git commit SHA1 is also embedded in the version
resource.
The new vendor event provides a mechanism to deliver auxiliary data,
which is automatically retrieved during the data transfer, but not
accessible through the library interface otherwise. Possible examples
include handshake data and/or device identification data.
This event is mainly intended for diagnostic purposes, in combination
with the memory dumping support. Very few applications will actually
need it for anything else.
If the first attempt fails, that might indicate the device isn't ready
yet to service requests. In that case immediately retrying again isn't
the right solution. Adding a small delay seems to increase the success
rate, so it's a good idea anyway, regardless of the underlying reason.
The application shouldn't have to deal with the maximum number of
retries. If the default value isn't good enough, that should be fixed
internally and not on the application side.
The d9 backend already reads the version info, to autodetect the
protocol variant. When doing the same in the vyper2 backend, we don't
have to read the version info again when downloading the dives.
The hardcoded version strings are now grouped into arrays, according to
their corresponding layout descriptor. The main advantage of using
arrays is that new versions strings can now easily be added, without
having to modify any code.
The Nemo Wide 2 uses the same communication protocol as the Icon HD,
except for two differences:
The Nemo Wide 2 requires a different baudrate and parity setting.
Unfortunately it doesn't seem possible to autodetect the correct
protocol variant at runtime. Attempting to proceed with incorrect
settings will render the device unresponsive and requires a battery
reset to recover.Therefore the model code needs to be provided as an
extra parameter, when opening the connection.
The Nemo Wide 2 also appears to have trouble downloading the entire
memory with a single request. Therefore the single large request is
split into many smaller ones. The offical Mares application uses 256
byte packets, and so do we. The Icon HD keeps using the large packets
because they are significant faster.
The extra model parameter breaks backwards compatibility!
The Petrel is slightly different from the Predator because the device
reorders the internal ringbuffer before sending the data. The most
recent dive is always the first one, and there is no need to search for
it, like we have to do for the Predator.
Apparantly, the windows wingdi.h header file already defines the
ERROR macro. By defining the NOGDI macro before including the
windows.h header file, we can prevent the wingdi.h file from being
included and thus avoid the warning. We don't need that header for
anything anyway.
Because the libusb header file includes the windows.h file
explicitly, it needs the same fix.
The recently introduced CNS support caused a regression for older
firmware versions. If a firmware doesn't support a certain sample
yet, the corresponding sample divisor and size are both equal to
zero (the default value for unused bytes). However, because the
size was always checked, regardless of whether the sample is
actually present or not, this zero size caused the parsing to fail.
To fix the regression, the size is now only checked when the
divisor indicates the sample is actually present.
The example application is updated to support the new setpoint,
ppO2, CNS and decompression samples that have been introduced. The
NDL event is removed again.
Having these as events seems less useful since for many dive computers
there are data with every sample - so it makes much more sense to have
these as part of the sample.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
So far only OSTC and Shearwater Predator are supported. For the OSTC we
support CNS and setpoint changes in the samples (the current hardware
doesn't actually support ppO2 sensors and for the older hw that does I
don't have the correct encoding information).
For the Predator we support only the "average ppO2 during the sample".
The Predator also gives us a CNS value at the end of the dive - I don't
quite know yet how to deliver that back to the consumer. Possibly as CNS
value in the very last sample? That would at least be consistent.
Signed-off-by: Dirk Hohndel <dirk@hohndel.org>
