===============================================================================
IT++ version 4.0.1                                         Released: 2007-12-15
-------------------------------------------------------------------------------

This is the first bug-fix release of the current stable itpp-4-0 branch. As
usual, the maintenance release of the IT++ library is backward compatible
with previous versions from the same branch (4.0.x), because only bug-fixes
and modifications that do not break the interface have been included.

Here is a brief summary of the most important changes included in this
release:


Modifications:
--------------

* Fortran compiler is no longer required for detection of some BLAS/LAPACK
  implementations. For instance, the BLAS and LAPACK libraries provided by
  Intel MKL or Apple's Accelerate Framework can be used without Fortran
  compiler installed.

* Flags and libraries returned by itpp-config and "pkg-config itpp" commands
  are adjusted to the linker requirements of a particular platform and
  library type. With this change, the use of "--static" switch should be no
  longer necessary. This should also fix linking problems to a shared
  library on Mac OS X.

* Major improvement in calculation performance of SVD and QR decomposition
  routines, especially for larger matrices.

* "operator*(Vec &v, Mat &m)" documented as deprecated, because its
  behaviour is equivalent to "outer_product(v, m.get_row(0))".

* "-Werror" flag is not used any more by default when compiling the
  debugging library.

* Documentation improvements of svd() and diag() functions

* Latest MKL and ACML libraries should be automatically detected on both 32-
  and 64-bit systems.

* The MS Visual C++ project files now use environment variables (INCLUDE,
  LIB and PATH) to find the location of header and library files of MKL and
  ACML. Besides, multi-threaded DLL libraries are built by default.

* "min" and "max" macros are now unconditionally undefined in config.h and
  config_msvc.h. IT++ uses std::min() and std::max() functions.

* Autoconf 2.60 is now required when building from SVN.

* Improved configure checks for isfinite(), isinf() and isnan() functions.
  This should improve portability of the library on other UNIX and BSD
  systems.

* The configuration summary is logged in the config.log file.

* Added "itpp_debug.pc" pkg-config's file for linking with the debugging
  library.


Bug-fixes:
----------

* Fixed compilation errors when using GCC 4.3 (bug-report [1848375])

* Fixed LDPC tutorial examples (bug-report [1827041])

* Added a workaround for a compilation error of a few explicit
  instantiations in converters.cpp when using GCC 3.3.x with
  "-finline-functions" flag enabled (bug-report [1818395])

* The distributed files "itpp.spec" and "itpp-config.1" are no longer
  removed when invoking a "make distclean" command (bug-report [1818391]).

* Fixed wrong calculation of bessely() caused by incompatible
  implementations of gamma() and lgamma() functions. This problem was only
  visible when using an Intel C++ compiler.

* Prevent warnings about string to char* conversions when using GCC >= 4.2


===============================================================================
IT++ version 4.0.0                                         Released: 2007-10-14
-------------------------------------------------------------------------------

IT++ 4.0.0 is the new stable release based on itpp-3-99 development branch.
It supersedes our previous 3.10.x stable versions. After over one and a half
year of intensive development it includes a lot of new features and
improvements with respect to IT++ 3.10.12 release (the previous stable
version).

A brief summary of the most important additions and changes is listed below.
Should you need more detailed history of changes that are new in this
release, please have a look at the NEWS-3.99 and ChangeLog files.


New features:
-------------

### Base module: ###

* New functions for initializing vectors and matrices from strings
  implemented. Main features of the improved parser are as follows:
  - Decimal, octal and hexadecimal notation support according to C/C++
    conventions, i.e. octal numbers starts with `0' (e.g. `0133', `077'),
    decimal numbers starts with `1-9' or are single `0' digits, hexadecimal
    numbers starts with `0x' or `0X' prefix and have at least one `0-9',
    `a-f' or `A-F' digit after (e.g. 0xFF, 0x0, 0x070).
  - Mixing of various representations in one string
  - Signed numbers supported, i.e. with leading `+', `-' signs, e.g.
    `-1000', `+0133', `-0x1'
  - Vector values separated with spaces ` ', tabs `\t' or comma `,'
  - Matlab's "a:b" or "a:b:c" notation for increasing and decreasing
    values, e.g. "0:2:10", "-9:-18", "4:-1:0", "0x0:0x8:0xFF"
  - Matrix rows separated with a single semicolon `;'
  - `NaN' (not a number) and `Inf' (infinity) values properly interpreted
  - Detection of syntax errors during parsing

* New features of the communication channels:
  - Implemented missing "Gauss I" and "Gauss II" Doppler spectra required by
    a few COST207 channel models
  - Channel_Specification and TDL_Channel classes extended with support of
    LOS (Rice) fading for any tap. Removed Doppler spectrum of type `Rice',
    since its generation method was the same as of the `Jakes' one. To
    obtain the `Rice' spectrum, a relative power, and optionally relative
    Doppler, needs to be defined for certain taps.
  - Implemented exponential channel profile generation function

* New implementation of the Normal_RNG and Complex_Normal_RNG classes, which
  uses so-called Ziggurat algorithm. The new generators are about 4-5 times
  faster than the previous ones using Box-Mueller method (feature request
  [1707969] and [1711727]).

* New Sort class implementation included with four different algorithms:
  Introsort, Heap sort, Quick sort and Insertion sort. The implementation of
  the new sorting functions is much faster than our previous QS()
  implementation due to the use of low level pointer arithmetic (feature
  request [1746876]).

* Added repmat() functions, which expand an input matrix or vector to a
  larger one by concatenating multiple copies of the input one.

* New methods in Vec and Mat classes: elem_mult_inplace(), elem_mult_out(),
  elem_div_out(), elem_mult_sum(), elem_div_sum() and sumsum()

* New methods: set_rows() and set_cols() implemented in the Mat class

* Added missing left(), right() and mid() methods to the Array class

### Communications module: ###

* Added new classes for Low-Density Parity Check (LDPC) codes. So-called
  Block LDPC (B-LDPC) codes, or Quasi-Cyclic LDPC codes are supported as
  well.

* Added support for MacKay's "alist" representation of sparse GF(2)
  matrices

* Added systematic codewords support in BCH and RS encoder/decoder classes
  (feature request [1418250])

* Added a new function waterfilling() to solve water-filling problems

### Statistics module: ###

* Created a new set of Mixture of Gaussians (MOG) classes (Statistics
  module)

* Added templated Histogram class provided (feature request [1451288])

### Other: ###

* MS Visual C++ .NET project file for linking IT++ with ACML added. Also
  added MSVC++ project files for building test programs.


Modifications:
--------------

### Base module: ###

* Major changes in IT++ file format related classes:
  - the new file format (version 3) is incompatible with version 2, which
    was used in IT++ 3.10.x series; for backward compatibility the previous
    format can be handled using the it_file_old and it_ifile_old classes,
    which are considered as deprecated and will be removed from the library
    in future
  - data is always saved using a little endian byte ordering; files files
    written on machines using big or little endian ordering should be
    identical; endianity variable removed from the data_header structure
  - all size variables are stored using fixed-width 64-bit unsigned integer
    type, and therefore are independent of the architecture word size (32-
    or 64-bit)
  - short and int data types are explicitly casted to fixed-with types
    int16_t, uint16_t, int32_t and uint32_t, since C/C++ standards do not
    assume fixed widths of these types
  - cleaner and more efficient implementation of the write_data_header()
    function
  - added missing implementation of a pack() function, which compacts an
    opened it_file by removing free space between data blocks
  - added optional description filed which can be saved with each variable
  - added missing operators and methods for reading and writing svec and
    smat data types
  - test program improved
  - itload.m and itsave.m M-files updated to be compatible with the latest
    IT++ file format

* Improved and reorganised existing error handling macros:
  - using standard NDEBUG flag for controlling it_assert_debug() macro,
    which replaced a three-level ASSERT_LEVEL definition
  - for backward compatibility it_assert0() and it_assert1() now behaves the
    same as it_assert_debug()
  - added new it_info() and it_info_debug() macros for printing information
    messages; it_info_debug() is also dependent on the NDEBUG compile-time
    definition
  - added it_error_msg_style() function for changing the format of
    it_assert(), it_error() and it_warning() messages; if style is set to
    Minimum, file name and line number of the error/assert/warning macro is
    not printed
  - error functions extended to handle stream operators like `<<' or
    `std::endl' in concatenation with strings (feature request [1569867])

* Improved the consistency of the alloc(), free() and set_size() methods in
  the Array, Mat and Vec classes

* General performance improvements of various operators in Vec and Mat
  classes by using a copy_vector() function where possible. The
  copy_vector() function use BLAS 1 routines for double and complex<double>
  arguments.

* The outer_product() function now uses an extra argument "hermitian", which
  controls the calculation for complex vectors. By default "hermitian" is
  set to false, which results in (x * y^T) calculation. If "hermitian" is
  true, (x * y^H) is calculated. This solution is backward compatible with
  3.10.x interface of outer_product(). Final fix for bug [1727923].

### Communications module: ###

* Major redesign of 1D and 2D modulator classes. From now, there is only one
  templated base class Modulator, which can handle 1D (real) and 2D
  (complex) constellations. Other modulators inherits from this base class.
  BPSK and PAM modulators are split into: BPSK_c/PAM_c classes which have
  complex-valued interfaces, and BPSK/PAM that have real-valued interfaces.
  Besides, demodulate_soft_bits() functions take an additional parameter
  "method" to switch between Log-MAP or approximate demodulation. The input
  signal for the Modulator_2D has now a slightly different meaning. Also an
  improper definition of the QPSK constellation is fixed (it was a 4-QAM
  modulator previously).

* Important updates in the communication channels:
  - Fading generators inheritance model redesigned. Now there are three
    types of fading: Independent, Static and Correlated. The Correlated
    fading can be generated using Rice (MEDS), FIR or IFFT methods. The
    default fading type is set to `Independent'.
  - The delay profile discretization function is improved and moved to the
    TDL_Channel class. An additional parameter `sampling_time' is required
    when setting the channel profile using the Channel_Specification object.
    It is now possible to discretize the channel with too low sampling
    frequency, but only if the taps being merged into a new single tap use
    the same Doppler spectrum type.
  - Cleaned up the interfaces of various fading generator classes

* Improved interfaces of Modulator_ND classes, so they are now more similar
  to 1D and 2D modulators. BTW, implemented Zero-Forcing (ZF) Log-MAP soft
  demodulator.

### Signal processing module: ###

* Memory allocation for double and complex<double> based vectors and
  matrices uses 16-byte alignment. This results in much better performance
  (up to 30% improvement) of FFT/IFFT and DCT/IDCT operations when using
  FFTW external library (feature request [1418707]).

### Other: ###

* Removed dependency on the external CBLAS library by providing direct
  interfaces to Fortran BLAS routines (feature request [1746876]).

* New "Signal Processing", "Statistics" and "Numerical Optimization" modules
  created from a subset of classes and functions from the "Base" module

* Restructuring of the source files in the "Base" module. Some of the
  functions grouped in "Linear Algebra" and "Mathematical and Miscellaneous
  Functions" sub-modules.

* Modularization added to the configuration scripts. It is now possible to
  select a limited set of modules, which will be built and installed as the
  IT++ library.

* Improvements in itpp-config and pkg-config's itpp.pc scripts. The changes
  are intended to follow the distinction between linking to a static and
  shared version of the library, which is used in pkg-config's itpp.pc file.
  By default "itpp-config --libs" returns "-I${prefix}/lib -litpp", whereas
  when "--static" is added in front of "--libs", the returned string is
  extended with a set of external library names required for proper static
  linking. Moreover, an additional "--debug" switch has been added, which
  changes the output of "itpp-config --cflags" and "itpp-config --libs" by
  providing debugging flags and debugging library name. For backward
  compatibility "--cflags-opt", "--cflags-debug", "--libs-opt" and
  "--libs-debug" still exist, but they are not documented and their use is
  deprecated.
