NEWS-3.99

===============================================================================
IT++ version 3.99.3.1                                      Released: 2007-08-10
-------------------------------------------------------------------------------

This minor bug-fix release fixes a serious regression, which appeared in
just released 3.99.3 development version. Here is a short description of the
fixed problem:

* Wrong declaration of zdotu_() BLAS function interface caused segmentation
  faults when IT++ was compiled with "-fomit-frame-pointer" flag. To fix
  this problem a Fortran wrapper function zdotusub_() has been added to
  IT++. This wrapper function comes from NetLib's CBLAS package.


===============================================================================
IT++ version 3.99.3                                        Released: 2007-08-10
-------------------------------------------------------------------------------

This is the last development release of IT++ 3.99.x series. The upcoming
4.0.0 stable release will be based on this version after a certain
stabilisation period. Therefore we encourage everyone to switch to this
version and report any problems and bugs on the IT++ Help forum. The
interface of this release should be compatible with the planned 4.0.x stable
releases. Below you can find a short summary of the recent additions,
modifications and bug-fixes included in this release.


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

* 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.

* Added low level functions scal_vector() and axpy_vector(), which use BLAS
  methods for double and complex<double> vector arguments. The scal_vector()
  function is now used by "*=" and "/=" operators.

* Added support for reading and writing bool variables in it_file class

* Added a few MSVC++ project files for test programs. The project files are
  configured to use 32-bit versions of MKL 9.1.025 and ACML 3.6.0 PGI
  libraries installed into their default locations on Windows platforms.


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

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

* 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.

* BLAS-aided specialisation of the Vec::outer_product() method for vec and
  cvec arguments added.

* More consistent test programs for Mat and Vec classes. Now all operators
  and methods are tested.


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

* Fixed a bug in Mat::set_size() method. When resizing a smaller matrix into
  a bigger one with copying the content, the unused entries were not
  initialised to zeros properly.


===============================================================================
IT++ version 3.99.2                                        Released: 2007-06-20
-------------------------------------------------------------------------------

This is the next development release of the IT++ library. As usual, below is
a list of the most important features, modifications and bug-fixes, since
version 3.99.1 was released. As this is a development version, some
interfaces have changed. Therefore, you might need to update your programs
using this library.


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

* Added B-LDPC Codes implementation, which consists of a BLDPC_Parity and
  BLDPC_Generator classes. The former one can be used for constructing the
  parity-check matrix from so-called base-matrix. Base matrix can be saved
  and loaded from a text file. The BLDPC_Generator class implements an
  efficient encoding algorithm using a preprocessed version of the
  parity-check matrix.

* New implementation of the Normal_RNG sample() function, which uses the
  so-called Ziggurat algorithm. The new generator is about five times faster
  than the previous one using the Box-Mueller method. This fixes feature
  request [1707969].

* New implementation of the Complex_Normal_RNG class, which now uses the
  improved Normal_RNG for generating the real and imaginary parts of the
  complex random sample. The new generator is about four times faster than
  the previous implementation. This closes feature request [1711727].

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

* Implemented set_size() method in the GF2mat class


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

* LDPC_Code::decode() methods changed to return only systematic bits, to be
  compatible with the encode() methods. New methods decode_soft_out() added,
  which return soft LLR values for all bits in a codeword.

* Removed unnecessary integrity_check() execution from the
  LDPC_Code::load_code() function. setup_decoder() method replaced with
  three separate methods, which are more flexible to use. Tutorial example
  updated to reflect recent changes.

* 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

* 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].

* Improvements in MKL detection macros. The latest MKL release 9.1.018 has
  different naming of a lapack shared library (libmkl_lapack32.so and
  libmkl_lapack64.so -> libmkl_lapack.so).

* Removed "-fno-exceptions" flag from CXXFLAGS and CXXFLAGS_DEBUG


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

* Fixed bug [1739500] in hadamard() function. The initial recursion for
  creating Hadamard matrices has to start with a 1x1 matrix H = [1], not
  with a 2x2 matrix. This fix provides correct solution for imat H =
  hadamard(1).

* Fixed bug [1739267] in binom() and binom_i(). The binomial coefficient was
  calculated wrongly for n = k. Instead of 1 the functions result was n+1.

* The functions for reading ppm and pgm images have to use binary mode when
  opening files. This fixes bug [1730610].

* Fixed bug [1724746], i.e. a problem with modified input data in
  ifft_real() function. Additional FFTW_PRESERVE_INPUT flag used during FFTW
  plan creation solves this problem.

* Fixed linking errors when trying to use Hadamard and Walsh-Hadamard
  transform functions. This fixes bug report [1719146].

* Fixed bugs in LDPC_Code::load_code() and LDPC_Code::save_code() functions.
  It could happen that LDPC generator object might be left uninitialised
  when loading the coded from a binary file.

* Fixed the actual number of iterations of the LDPC decoding algorithm. It
  was using one more then expected.

* Fixed bug [1703731] by implementing the missing double2mxArray() and
  double_complex2mxArray() functions in itmex.h

* Included missing <cstdlib> and <limits> header files where necessary. This
  fixes problems with building using GCC 4.3 snapshot.

* Fixed a minor bug in the set() method, when parsing an "a:b" or "a:b:c"
  format. Due to precision problems, the last value from some specific ranges
  was not included in the result, e.g. "0:0.2:3" resulted in a vector with
  2.8 at the end (instead of 3).


===============================================================================
IT++ version 3.99.1                                        Released: 2007-03-22
-------------------------------------------------------------------------------

This is the next development release of the IT++ 3.99.x series. Below is a
brief summary of the most important new features, modifications and bug
fixes, as compared to the previous release 3.99.0. Please note that some
interfaces have changed, so you might need to update your programs linking
to this version.


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

* Random number generator extended with random53_01_lclosed() function,
  which generates [0,1) values in 53-bit resolution.

* Added missing	read/write operators for char variables in it_file and
  it_ifile classes

* MS Visual C++ .NET project file for linking IT++ with ACML added

* Added man page for itpp-config script

* Added test file for a binary it_file format. It is supposed to detect
  wrong endianness detection on various platforms.


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

* LDPC_Parity_Matrix and LDPC_Generator_Matrix classes split into a set of
  specialized classes, which inherits from the base classes LDPC_Parity and
  LDPC_Generator. The new inheritance model provides a flexible interface
  for creating new generation methods and parity check matrix
  representations by end users.

* 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

* Portability improved by replacing "unsigned short int" with 8-bit
  "unsigned char" type for storing data bits in GF2mat

* Improved the speed and portability of external libraries detection macros

* Replaced long with int in many sources, since long and int have the same
  sizes on most IT++ supported architectures. If there is a need for 64-bit
  integers, due to portability reasons (u)int64_t type should be used
  instead of long.


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

* Fixed bug [1661419] by adding default constructors to the QAM, PSK, PAM_c
  and PAM modulator classes. Thanks to Jordy Potman for reporting this bug.

* Fixed a bug in BLAS detection, when an explicit library is passed to the
  script configure using "--with-blas=" option.

* Fixed improper declarations of ls_solve_od() and ls_solve_ud() methods

* Protected inheritance of the Convolutional_Code class replaced with a
  public one. This fixes the problem of assigning the
  Punctured_Convolutional_Code class object to a pointer of Channel_Code
  class.

* Minor documentation fixes in many places


===============================================================================
IT++ version 3.99.0                                        Released: 2007-01-21
-------------------------------------------------------------------------------

This is the first release of the IT++ 3.99.x development series, which is
based on sources from the main SVN branch (trunk). After some period of
stabilization, a new stable release will be created from this branch (IT++
4.0.0). Since this is a development branch, some interfaces might be changed
in future releases. Moreover, some interfaces have changed with respect to
the stable 3.10.x series.

Below is a brief summary of the most important new features, modifications
and bug fixes, as compared to the recent stable 3.10.x releases. All minor
modifications and bug-fixes from 3.10.x series are included in this release
as well.


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

* 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 `;'
  - 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

* Created a new set of Mixture of Gaussians (MOG) classes written by Conrad
  Sanderson (Statistics module)

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

* Added new classes for Low-Density Parity Check (LDPC) codes

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

* Systematic codewords support in BCH and RS encoder/decoder classes
  contributed by Steve Peters (feature request [1418250])

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

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


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

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

* 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.

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

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

* Error functions it_assert(), it_warning() and it_error() extended to
  handle stream operators like `<<' or `std::endl' in concatenation with
  strings, e.g. `it_error("Wrong value x = " << oct << x)' (feature request
  [1569867] by George Jongren)

* Changed the method of detecting the endianness of a system -
  "itpp/base/machdep.h" is no longer used.


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

* Fixed a bug that resulted in different output samples of the IFFT fading
  generator on SPARC Solaris 9 using GCC 3.4.5, than the ones obtained in
  other environments. The problem was caused by a different order of
  calculating the random arguments of the concat() function. This is either
  a bug of GCC for Solaris or its dependence on different endianness, or it
  is not fully specified behavior of C/C++ standards.

* Several other bug-fixes and improvements. For a full list of them, see the
  ChangeLog and NEWS-3.10 files.