Windows Vista has a broken rate converter which gets invoked when an
input audio stream at 48kHz sampel rate is requested. I've no idea why
our application can't get exclusive access to the audio input device
and have a unconverted stream direct at 48kHz.
To get around this our down sampling filter for audio input from 48kHz
to 12kHz is disaabled by default on Windows Vista, instead we request
a 12kHz stream and process it directly.
This default behviour can be overriden by specifying the following
settings value:
[Tune]
Audio\DisableInputResampling=false
This settings value defaults to true on Windows Vista and false
everywhere else so normally needn't be present.
git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@3588 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
The filter used for 4 times down-sampling cannot handle sample streams
where the hardware or drivers deliver chunks of data that are not
multiples of 4 frames long. This seems to be prevalent on some Linux
platforms. Also de-interleaving of single channel audio from stereo
streams was no longer supported.
I have changed the input strategy to de-interleave the incoming
sample stream into an intermediate buffer large enough to hold all the
samples required for a single unit of processing (one basic waterfall
interval) and apply the down-sampling filter to the whole intermediate
buffer just prior dispatch to the FFT generator.
This now means that we are now using the ubiquitous 48kHz hardware
sample rate for both input and output of audio across all platforms
and decoding a single channel of a stereo stream is again
supported. The down-sampling to 12kHz is done with a high quality FIR
49-tap low pass filter specifically designed by Joe (K1JT) for
operation in a 4kHz bandwidth.
git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@3585 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
Change source URLs in teh CMake scripts for the kvasd binaries.
git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@3563 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
Audio input can be mono, left of stereo pair or, right of stereo
pair. Audio output can be mono, left of stereo pair, right of stereo
pair or, both of stereo pair (the same output goes to both channels in
both mode). Settings are remembered between sessions.
Stream channel suport is implemented mainly in the new AudioDevice
class which is now the base class of Modulator and Detector.
Audio channels are selected on the configuration screen. Only
supported channel configurations per device can be selected.
Audio output volume (actually attenuation) is now possible from the
GUI. I have added a slider control to the main window; I don't
necessarily propose this as a final release location for the widget as
I understand that changes to the main screen are sensitive. This
location is just a starting suggestion for a trial. The volume
(attenuation) setting is remembered between sessions and is not device
dependent. This addresses all issues of volume setting on *nix
versions since there is no need to use pavucontrol to set audio
levels. The volume (attenuation) action is logarithmic.
Shaped CW keying has been implemented in Modulator although it is
currently disabled as I am not 100% happy wth the implementation. If
you want to try it define the C++ preprocessor macro WSJT_SOFT_KEYING
in your build.
The Modulator instance has been moved to the same thread as the
SoundOutput instance as it should have been since the output callback
already operates in that thread. Cross thread slots are now correctly
called in a thread safe way as a result.
A number of files where in the SVN repository with DOS line endings
which I have removed. SVN users on Windows need set the config for
native line endings so that DOS line endings are automatically
stripped on checkin.
The DevSetup class now holds it's UI o the heap to reduce imapact on
build dependencies.
The application settings are now passed to objects from the main.cpp
file. Management of settings are moved to the responsible classes (top
level windows). This has involved a few settings moving groups so
users will see some settings reverting to default values on the first
run of an update.
Persistance of top level windows geometry and position is now handled
in the recommened manner (constructor for load, closeEvent for store
in modal windows and, hideEvent for store in modeless dialogs).
The MainWindow class now holds its children as members rather than
global variables.
The LogQSO class now hides its implementation and takes responsibility
for its own settings and widows rendering parameters. A new settings
file group is implemented to persist the LogQSO class settings.
The WideGraph class now hides its implementation and manages its own
settings and window rendering parameters.
--This line, and those below, will be ignored--
M Modulator.cpp
M rigclass.cpp
M widegraph.cpp
M signalmeter.cpp
M soundin.cpp
M soundout.cpp
M mainwindow.h
M main.cpp
M meterwidget.h
M devsetup.cpp
M mainwindow.ui
M Detector.cpp
M logqso.h
M rigclass.h
M mainwindow.cpp
M meterwidget.cpp
M soundin.h
M devsetup.ui
M wsjtx.pro
M devsetup.h
M logqso.cpp
M Modulator.hpp
M psk_reporter.cpp
M killbyname.cpp
M Detector.hpp
M signalmeter.h
M widegraph.h
M psk_reporter.h
M soundout.h
M PSKReporter.h
M lib/afc65b.f90
M lib/gran.c
M lib/usleep.c
M lib/afc9.f90
M lib/wrapkarn.c
A AudioDevice.hpp
git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@3542 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
Currently only Qt5 or above is known to work with this code. It may be
possible to backport it to Qt4 if required.
Audio output goes back to a separate thread to try and minimize
stutters in streaming on Windows particularly.
A crash on Linux due to mishandling of stereo audio output has been
fixed and both left and right channels are now correctly synthesised
with identical contents.
Rigs are enumerated directly from hamlib API rather than running a sub
process reading output of rigctl -l. This was initially done to get
rid of some GUI thread blocking in the configuration dialog, but is
generally a better way of doing it anyway.
Some refactoring in MainWindow to accomodate the audio streaming,
modulation and detecting classes.
Exit handling for application refactored to use signals rather than
brute force event loop exit. This was required to get correct thread
shutdown semantics.
The GUI update timer is now stopped during application shutdown which
is necessary to stop crashes when shutting down gracefully with
signals and window close() calls.
There is an outstanding issue with Linux audio streams; the QAudio
Input/Output classes create a new stream name each time a stream is
started. This doesn't play well with PulseAudio utilities such as
pavucontrol to set stream volume as settings are lost every tx
period. I have tried to keep a single stream for all output but there
are problems restarting it that haven't been resolved yet.
The QtCreator project file has been rearranged a little because it
passes all the object files to the linker rather than using an archive
library. Since the GNU linker is single pass; the object files need to
be in a logical order with definitions appearing afer references to
them. This was required to avoid a linking error.
The lib/Makefile.linux has been enhanced to use the fortran compiler
to locate the correct version of the Fortran library to use. This is
necessary on the latest Linux distros because the unversioned symlink
to compiler support libraries is no longer provided. This only an
issue with mixed programming language links where the linker driver
for one language has to link support libraraies for another language.
git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@3532 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
