Release Note Report Major Bug Fixes This document describes major bug fixes for the following: Communications Blockset Communications Toolbox Database ToolboxDSP BlocksetEmbedded Target for TI C6000 DSPFilter Design ToolboxFuzzy Logic ToolboxImage Processing ToolboxMapping ToolboxMATLABModel-Based Calibration ToolboxOptimization ToolboxSignal Processing ToolboxSimulinkSimulink AcceleratorSpline ToolboxStateflowStateflow CoderStatistics ToolboxWavelet ToolboxxPC Target Communications Blockset Click on a problem area listed below to read how it has been fixed. Fading channels List of demos for which code can now be generated using Real Time Workshop Passband FSK and CPM modulators and demodulators are more accurate Reed-Solomon blocks now encode and decode correctly Fading channels The accuracy of Doppler spread of the Rayleigh and Rician Fading Channel Blocks has been improved. The blocks now give better results for high sampling rates and small Doppler frequencies, as specified in communication standards such as WCDMA. List of demos for which code can now be generated using Real Time Workshop The following demos now generate code using Real Time Workshop: dmt_sim dvbt_sim tstgraycod phasenoise_sim dmt_alt_sim tstconvcod Passband FSK and CPM modulators and demodulators are more accurate The passband FSK and CPM modulators have been modified to generate more accurate waveforms by performing FIR interpolation when upsampling. The FIR filters significantly reduce the levels of any spectral copies revealed by upsampling. They also introduce delay into the modulators. Reed-Solomon blocks now encode and decode correctly The Reed-Solomon blocks now encode and decode signals correctly. Communications Toolbox Click on a problem area listed below to read how it has been fixed. Reed-Solomon decoder corrects up to t errors Reed-Solomon encoder and decoder use more conventional format for data Reed-Solomon decoder corrects up to t errors The new function rsdec accurately decodes Reed-Solomon codes containing up to t errors in each codeword. This new function replaces the earlier functions rsdeco and rsdecode. Reed-Solomon encoder and decoder use more conventional format for data The new functions rsenc and rsdec operate on the new Galois data type, which represents symbols using a decimal format. The formats used in the earlier functions, rsenco, rsencode, rsdeco, and rsdecode, are no longer supported. See "Updating Existing Reed-Solomon M-Code" in the Release Notes to learn how to convert your existing code data to the format used in the new coding functions. Also, the new functions enable you to choose whether parity bits appear at the beginning or end of each codeword. Database Toolbox Click on a problem area listed below to read how it has been fixed. Improved speed of insert function Oracle tables load faster in Visual Query Builder Returning numeric arrays from fetch is faster Stored procedure now returns data as expected Improved speed of insert function The insert function should perform approximately five (or more) times faster than in the previous version. Oracle tables load faster in Visual Query Builder In the Visual Query Builder, when you select an Oracle data source, the tables load more quickly than in the previous version of the Database Toolbox. Returning numeric arrays from fetch is faster When the data return format is set to numeric in setdbprefs or in preferences, running fetch returns data more quickly than in the previous version. Stored procedure now returns data as expected In Version 2.2, running exec with a stored procedure did not return results for some configurations. The problem has been fixed for this version. DSP Blockset Cumulative Sum block behaves properly for frame-based inputs The Cumulative Sum block now behaves properly when given frame-based inputs and set to sum along columns. In this situation, the block now computes the running sum of each column of a frame-based input, where the running sum is independent of the running sums of previous inputs. To get the previous behavior for frame-based inputs when set to sum along columns, set the block to sum along channels. Embedded Target for TI C6000 DSP Sum, Gain, Mult blocks fast replacement TLC is now Big-endian compatible Sum, Gain, Mult blocks fast replacement TLC is now Big-endian compatible. They now return correct numeric results. Filter Design Toolbox Click on a problem area listed below to read how it has been fixed. firlpnorm help text was incorrect QFILT's sos method no longer embeds filter gain in each section QFILT's sos method no longer normalizes the filter after converting it to SOS firlpnorm help text was incorrect For Release 13, the online help text for filter design function firlpnorm is now correct. QFILT's sos method no longer embeds filter gain in each section The QFILT sos method no longer embeds the filter gain in each section. It now puts the filter gain into the ScaleValue property of QFILT. QFILT's sos method no longer normalizes the filter after converting it to SOS The QFILT sos method no longer normalizes the filter coefficients after converting it to second-order sections, therefore care should be taken to avoid overflows by adjusting the coefficient format of the filter. Fuzzy Logic Toolbox Click on a problem area listed below to read how it has been fixed. addvar and the command line API Simulink blocks for TRIMF and TRAPMF now correctly handle vertical slopes addvar and the command line API In Release 12, addvar systematically initialized the new variable with 3 membership functions, thus breaking the command line API, where it is common first to add the variable and then define the membership functions. Release 13 adds logic to allow GUIs to create 3 default membership functions without interfering with command line calls. Simulink blocks for TRIMF and TRAPMF now correctly handle vertical slopes The cases a-b or b=c in trimmf (a-b and c=d in trapmf) caused Divide by Zero and improper results in Simulink. These blocks have been reimplemented using the IF block to always produce answers consistent with the M file implementation Image Processing Toolbox Click on a problem area listed below to read how it has been fixed. cpselect would ask to save points when it wasn't necessary Stray points could appear in cpselect in point-prediction mode dicomread and private data dicomread and signed data from DICOM files fspecial 'laplacian' and 'unsharp' options gray2ind returns the correct output imadjust no longer errors if passed degenerate inputs imerode handles binary erosion correctly imhist on IRIX with input containing NaNs imnoise with the Poisson option pixval could give a confusing error message for figures with normalized units Figures created by imshow Improved performance of dicomread and dicominfo legend box is resizable in cpselect Modified the implementation of the Canny algorithm in the edge function. RLE decoder for DICOM files cpselect would ask to save points when it wasn't necessary When you closed the contol-point selection tool (cpselect), it would ask if you wanted to save control points, even if you had already done so. This has been fixed. Stray points could appear in cpselect in point-prediction mode When using the control-point selection tool (cpselect) in point-prediction mode, sometimes a stray control point would appear that could not be deleted. This has been fixed. dicomread and private data The dicomread function no longer fails when reading from a file with private data using "explicit VR" syntax. dicomread and signed data from DICOM files The dicomread function used no longer fails when reading signed data from pixels which were not "normal" lengths. fspecial 'laplacian' and 'unsharp' options The fspecial function now accepts a parameter value of 0 for the 'laplacian' and 'unsharp' options. gray2ind returns the correct output If the input image to gray2ind was logical, then the output was also logical. The gray2ind function now returns the correct output in such situations. imadjust no longer errors if passed degenerate inputs A call to imadjust with identical low and high limits, such as imadjust(A,[b b],[]), caused imadjust to error out. The problem has been fixed by providing default limits when the input limits are degenerate. imerode handles binary erosion correctly Prior to Release 13, when performing binary erosion with a decomposed structuring element, such as strel(ones(5,5)), the first 32 rows of the output image could be wrong. This has been fixed. imhist on IRIX with input containing NaNs The function imhist no longer causes a segmentation violation on IRIX with input containing NaNs. imnoise with the Poisson option The function imnoise now returns a correctly- sized output instead of a column vector when using the Poisson option. pixval could give a confusing error message for figures with normalized units Turning on pixval for a figure in normalized units could produce a confusing error message. The pixval function now issues a more informative error message if the figure is not in pixel units. Figures created by imshow Prior to Release 13, if invalid input was passed to imshow, it would sometimes error out after creating an invisible figure object. Subsequent calls to imshow would appear to not do anything, because the images were being sent to the invisible figure. The problem has been fixed by adding additional input checking to imshow. Improved performance of dicomread and dicominfo The dicomread and dicominfo functions parse files quicker than in the past, especially for files with a lot of metadata. legend box is resizable in cpselect The legend box for the control-point selection tool (cpselect) was unresizable, which sometimes caused the title bar to be unreadable. This has been fixed. Modified the implementation of the Canny algorithm in the edge function. In the Canny algorithm in the edge function, Gaussian smoothing should be run perpendicular to the direction of differentiation. This has been implemented. RLE decoder for DICOM files The RLE decoder used by dicomread used to only work for 8-bit grayscale images (and only for certain ones at that). The decoder now correctly handles all bit depths and image types. Mapping Toolbox Click on a problem area listed below to read how it has been fixed. Computed distances no longer have spurious imaginary part Coordinates in map legends for regular matrix maps Fixed incorrect UTM coordinates in the southern hemisphere Reading ETOPO5 data files at full resolution Computed distances no longer have spurious imaginary part The round-off problem that caused the return value of the distance function to include a small, nonzero imaginary part for certain inputs has been fixed. Coordinates in map legends for regular matrix maps Version 1.3 fixes setpostn to handle longitude limits that straddle the prime meridian. Fixed incorrect UTM coordinates in the southern hemisphere A typographical error in defaultm in the false northing for Universal Transverse Mercator (UTM) has been corrected. That error had caused incorrect UTM/geographic conversions for points in the southern hemisphere. Reading ETOPO5 data files at full resolution Padding is added as needed in etopo5 when reading different numbers of elements from the northern and southern hemisphere files. MATLAB Click on a problem area listed below to read how it has been fixed. 'cubic' method in griddata no longer produces segmentation fault diff function has improved handling of negative order eigs(A,B,k,sigma) for sparse B now converges filter function has better support for multidimensional data log10(10^i) now returns i, exactly lsqr now converges for rectangular systems sqrtm now returns correct results for singular Hermitian matrix Array Editor supports more than 10,000 elements Import Wizard opens large files more quickly Improved accuracy for denormal pivot handling Improved Workspace browser performance with many variables Modal dialog behavior improvements Serial port fixes The lookfor function now automatically displays results Workspace browser allows renaming variables 'cubic' method in griddata no longer produces segmentation fault Previously, the 'cubic' method of the griddata function would occasionally encounter a segmentation fault. This no longer happens. diff function has improved handling of negative order Previously, MATLAB encountered a segmentation fault when computing diff(sparse([1;1]),-1). Now, an error message is displayed if the order is not a real positive integer. eigs(A,B,k,sigma) for sparse B now converges Previously, eigs(A,B,k,sigma) for sparse B and non-scalar sigma (sigma not equal to 'SM' either) applied a reordering to B before factoring it for use in repeated solves. Unfortunately, the reordering was not correctly applied in the subsequent solves and as a result eigs did not converge. The reordering is now being applied correctly and eigs now converges in this case. filter function has better support for multidimensional data For multidimensional inputs, filter sometimes produced incorrect error messages or the wrong results. This no longer happens. log10(10^i) now returns i, exactly Previously, log10 was computed directly as log(x) / log(10), where log is the natural logarithm. The error in this was particularly visible in computations of the form log10(10^i), for example: >> log10(1e6) ans = 5.999999999999999e+00 Now, log10 does some additional scaling to compute log10(10^i) to the nearest bit. lsqr now converges for rectangular systems Previously, lsqr only converged for square, consistent systems A*x=b and returned a solution x where norm(b-A*x) was small. Now, lsqr also converges for rectangular systems and returns a solution x where norm(b-A*x) is not necessarily small, but is minimized, and in fact norm(A'*(b-A*x)) is small. sqrtm now returns correct results for singular Hermitian matrix A matrix square root always exists for Hermitian matrices. Previously, sqrtm(X) sometimes returned a matrix of all NaNs for matrices X that were both Hermitian and singular. Now sqrtm(X) returns the correct results for these matrices. sqrtm is now also faster for Hermitian matrices. Array Editor supports more than 10,000 elements The Array Editor can now show arrays with more than 10,000 elements. It does not support arrays with more than 65,536 (2^16) elements. Import Wizard opens large files more quickly In previous releases, the Import Wizard was sometimes slow when opening large files, but the problem has been fixed in this version. Improved accuracy for denormal pivot handling LU factorizations for full matrices now use a modified version of the LAPACK LU factorization routines and return more accurate results when encountering denormal pivots (i.e., smaller than realmin). In previous versions, results may have included NaNs; for example: A = [2^(-1050) 2^(-1050);0 0] [L,U] = lu(A) A = 1.0e-316 * 0.8289 0.8289 0 0 L = 1 0 NaN 1 U = 1.0e-316 * 0.8289 0.8289 0 NaN In the current version, results are more accurate. For example, using A as above, [L,U] = lu(A) L = 1 0 0 1 U = 1.0e-316 * 0.8289 0.8289 0 0 This improved accuracy carries with it a performance loss for any operations that use lu or lu-based \ (backslash) for factorization of full matrices. Improved Workspace browser performance with many variables There is no longer a performance problem with more than 1000 variables open in the Workspace browser. Modal dialog behavior improvements Modal dialogs on Windows platforms have been improved in terms of common appearance and stability. In Release 12, MATLAB would sometimes lock up due to modal dialogs. In Release 13, this is fixed. Serial port fixes Serial port objects now work with the baud rates 128,000 and 256,000 along with all other standard baud rates listed in the BaudRate property help. MATLAB can now find serial ports if the MATLAB root directory contains an underscore. A memory leak no longer occurs when opening and closing the serial port connection. On Windows, serial port objects now recognize ports greater than COM4. On Solaris, there is no restriction on the number of times that the serial port can be opened. On Linux, there is a significant speed increase when writing data to the serial port. The lookfor function now automatically displays results When you run lookfor, results will now automatically display in the Command Window. You no longer need to press Ctrl+C to display results. While lookfor is running, the status bar shows Busy. Workspace browser allows renaming variables You can now rename variables from within the Workspace Browser. Right-click a variable name, select Rename from the context menu, and edit the name. Model-Based Calibration Toolbox Click on a problem area listed below to read how it has been fixed. 3-D and 4-D design views Breakpoint filling using "ShareCurvThenAve" algorithm Converting variables to aliases Cross Section tool now displays correct model error Differences in the history of 1-D tables Fitting and pruning RBF models Importing data from Excel into a Dataset Local sweep notes kept during data augmentation Optimal designs no longer allow incorrect number of stratified levels Outlier sweeps appearing in the global outlier list Previewing large Latin Hypercube designs now faster Stratified Latin Hypercube designs Viewing imported data in Datasets You can now delete designs without crashing the Model Browser 3-D and 4-D design views The 3-D and 4-D design views will now correctly display all of the points in a design. Previously, for some factor ranges points at the edges of the graph, limits were not displayed. Breakpoint filling using "ShareCurvThenAve" algorithm Breakpoint filling using "ShareCurvThenAve" now works correctly in all cases and is significantly faster than in the previous version. Converting variables to aliases You can now convert all Variable Dictionary items into an alias of another Variable Dictionary item. Previously, this would cause an error in some circumstances where the items were being used elsewhere in the Cage project. Cross Section tool now displays correct model error The Cross Section view of the Model Selection window now displays the correct model error while the input factor lines are being dragged. Previously, the error shown while dragging the line was different from that shown when the mouse button was released. Differences in the history of 1-D tables 1-D tables in Cage now correctly show the difference between two versions in the history. Fitting and pruning RBF models Re-fitting an RBF model after pruning now uses the correct value for log10 (GCV). Importing data from Excel into a Dataset You can now import large amounts of data from Excel into a Dataset. Previously, attempting to do this caused an error to occur. Local sweep notes kept during data augmentation When you add new data to a testplan that already has models, any sweep notes which have been made are now saved when the models are re-fitted. Optimal designs no longer allow incorrect number of stratified levels Optimal designs with stratified candidate sets will now correctly fail if there are not enough stratified levels to support the model. Previously, small numerical errors allowed the optimisation to proceed, but poor conditioning meant that the optimal value oscillated up and down. Outlier sweeps appearing in the global outlier list Previously, if a local sweep was removed from fitting at the local level, the global response features only showed the removed sweep in their list of removed points if one or more outliers was selected and removed at the global level. All removed points are now shown at the global response feature level all of the time. Previewing large Latin Hypercube designs now faster You can now create large Latin Hypercube designs from the Classical Design Browser without MATLAB appearing to hang. Previously a full set of search iterations was erroneously done before the expected waitbar appeared and the search commenced. Stratified Latin Hypercube designs Stratified Latin Hypercube designs now choose their stratified levels so they split the stratified dimension into exactly equal sections. Previously, the levels were "snapped" to the nearest non-stratified level, which depends on the total number of points in the design. This meant that some levels were shifted down slightly and others up. Viewing imported data in Datasets You can now view Datasets that consist entirely of imported data. Previously, viewing these datasets caused Cage to crash. You can now delete designs without crashing the Model Browser You can now delete designs from anywhere in the Design Editor tree without causing errors. Previously, if your best design was at the end of the design tree and you deleted a design, you received an error when you closed the Design Editor. Optimization Toolbox Ill-posed or ill-conditioned problems in fmincon The function fmincon would in some small number of cases hang when the problem was ill-posed or ill-conditioned. In these cases, fmincon no longer hangs. Signal Processing Toolbox Click on a problem area listed below to read how it has been fixed. aryule order limit increased cremez example fixed Frequency inputs for cheb1ord, cheb2ord, ellipord Pole/Zero plot root multiplicity Scaling of magnitude response in magnitude and phase plot SPTool Filter Viewer Print Preview aryule order limit increased The model order in the aryule function is no longer limited to half the input signal length. The order now can be up to the signal length. cremez example fixed The second example on the cremez reference page has been updated and now works correctly. Frequency inputs for cheb1ord, cheb2ord, ellipord For the digital bandstop case, the frequency values returned from the cheb1ord, cheb2ord, and ellipord functions are no longer changed from the input frequency values. Pole/Zero plot root multiplicity The pole/zero plot in FDATool and FVTool now can display multiple roots at a single plot point. Scaling of magnitude response in magnitude and phase plot The scaling of the magnitude response portion of the Magnitude and Phase Analysis plot in both FDATool and FVTool was incorrect for some filters. It now is scaled correctly. SPTool Filter Viewer Print Preview Print Preview, when no markers are displayed in the SPTool Filter Viewer, now works correctly. Simulink Click on a problem area listed below to read how it has been fixed. Block parameter defaults now stored in mdl file Breaking disabled library links from the command line Bus Selector adapts to bus structure changes Changing Configurable Subsystem selection crashed MATLAB Creating Simulink data object classes in the MATLAB work directory Dot Product block now works in loops Drag-selecting blocks froze UNIX Grouping blocks caused divide-by-zero error Improved detection of evenly spaced breakpoints in PreLook-Up Index Search and Look-Up Table n-D blocks Keyboard panning capabilities added to Simulink diagram editor MATLAB and Simulink lockup after ESC during drag operation Multirate models caused Simulink to hang PreLook-Up Index Search, Look-Up Table n-D (Flat interpolation) blocks now return top of table Propagating library changes created recursive links Pulse generator computed sample time Scope Block will not crash when "Save To Workspace" variable name is longer than 31 characters SegV when adding/deleting Simulink blocks with Model Browser open on Unix Selecting Tools->Coverage Settings when all Simulink licenses are checked out Self-modifying masked library blocks had missing connections Shift+drag disconnected a block inside a locked reference block Simulink debugger did not recognize CurrentBlock parameter Block parameter defaults now stored in mdl file The mdl file now has a new section, "BlockParameterDefaults." The default values for the parameters used by the block types present in the model are written into this section. Then, an individual block instance's parameter values are written into the mdl file only if they differ from the default. This technique reduces the size of the mdl file, as well as reducing the file load time. A block diagram parameter, SaveDefaultBlockParams, exists to control whether or not the parameter defaults are saved to the file. By default it is set to true. Breaking disabled library links from the command line Breaking disabled library links from the command-line now works. Bus Selector adapts to bus structure changes In previous releases, adding signals to a bus or otherwise changing its structure required that you update any Bus Selector blocks that select signals from the bus. In the current release, you need to update the Bus Selector only if it selects signals that no longer exist or no longer have unique names. Changing Configurable Subsystem selection crashed MATLAB The Configurable Subsystem has been made faster and more robust when changing block choices. This is especially noticeable for large subsystems. Creating Simulink data object classes in the MATLAB work directory Previous releases of Simulink prevented you from creating Simulink data object classes in the MATLAB directory or any of its subdirectories. The current release allows you to create classes in the work subdirectory of the MATLAB root directory. Dot Product block now works in loops In Release 12.1, a Dot Product block in a loop cannot determine its port dimension and hence causes a simulation error. Now the block assigns a default dimension setting to the port, thereby avoiding the simulation error. Drag-selecting blocks froze UNIX Releasing the left mouse button over the Simulink menu bar when drag- selecting multiple blocks caused UNIX to lock up. This has been fixed. Grouping blocks caused divide-by-zero error Grouping blocks in Simulink into a subsystem caused a divide-by-zero error if the blocks were too close to the top or left edge of the model. This has been fixed. Improved detection of evenly spaced breakpoints in PreLook-Up Index Search and Look-Up Table n-D blocks The option for using a fast index search algorithm that relies on evenly spaced breakpoints for look-up tables was enhanced to better detect evenly spaced breakpoints within a tolerance of value*eps. Note that breakpoint expressions such as [-5:0.1:5] may still need to be written as 0.1*[-50:50] to generate a dataset that is evenly spaced within the required tolerance. The linear and binary search options will of course still work on any strictly increasing breakpoint data set (not just evenly spaced ones). Keyboard panning capabilities added to Simulink diagram editor You can now scroll Simulink diagrams around using the keyboard by pressing the Ctrl key in conjunction with any of the arrow keys. This is sometimes more convenient than manipulating the scroll bars with the mouse. MATLAB and Simulink lockup after ESC during drag operation In previous releases, pressing the Esc key while dragging a block or performing other types of drag operations caused MATLAB and Simulink to become unresponsive. In the current release, pressing the Esc key will abort the drag operation. Multirate models caused Simulink to hang For multirate models that are simulated to very large times, Simulink would eventually miss a sample hit for one of the rates. This would cause Simulink to hang. This has now been fixed and Simulink no longer hangs. PreLook-Up Index Search, Look-Up Table n-D (Flat interpolation) blocks now return top of table The behavior of the PreLook-Up Index Search block (index only mode) and the Look-Up Table n-D block (in flat interpolation mode) were changed to return the last index in the table when the input value is at or above the last breakpoint value as this is more useful than the previous behavior. For the Look-Up Table n-D block, this is done on an individual dimension basis. In previous versions (Simulink 4.0 and 4.1), the index search would deliberately return the next to last index and the n-D flat look-up would deliberately return the next to last table value even when the input value was at or above the last breakpoint. Propagating library changes created recursive links If changes are made to a linked block, propagating such changes to the library via the Link options menu now does not result in recursive links. Pulse generator computed sample time The pulse generator block would somtimes report that the computed sample time is not an integer multiple fo the fixed step size when, in fact, it was. This has been fixed. Scope Block will not crash when "Save To Workspace" variable name is longer than 31 characters Scope blocks with "Save To Workspace" variable names longer than the maximum MATLAB variable length (currently 31 characters) used to crash Simulink in unpredictable ways during simulation. Such variable names will no longer crash Simulink, though they will generate an error. SegV when adding/deleting Simulink blocks with Model Browser open on Unix On Unix platforms, a segmentation violation would occur if you added or deleted blocks in a Simulink model while the Model Browser was open. This has been fixed. Selecting Tools->Coverage Settings when all Simulink licenses are checked out The "Undefined variable names" error is no longer displayed when selecting Menu->Tools->Coverage Settings without an available Simulink license. Self-modifying masked library blocks had missing connections Library subsystems containing self-modifying masked blocks no longer have to be marked as MaskSelfModifiable for the linked blocks to be connected properly. Shift+drag disconnected a block inside a locked reference block In previous releases, holding down the Shift key and dragging a block inside a reference block (i.e., a library link) disconnects the block's signal lines and brings up a dialog box asking if you would like to disable the link. In the current release, Simulink does not disconnect the lines. Simulink debugger did not recognize CurrentBlock parameter In the previous release, you could not use get_param or set_param to set or get the value of the CurrentBlock parameter in the command-line debugger until you selected a block. The Simulink command-line debugger now allows you to set and get the value of this parameter without having to first select a block with the mouse. Simulink Accelerator Click on a problem area listed below to read how it has been fixed. Non-inlined S-functions that accept non-double parameters will now work with the accelerator Simulink Accelerator has correct minor step guard for enable, if, case Simulink Accelerator now functions without an RTW license Simulink Accelerator supports gcc compiler on UNIX Non-inlined S-functions that accept non-double parameters will now work with the accelerator In previous releases, if a non-inlined S-function used non-double parameter then the model would error out while trying to generate accelerator code. This restriction has been relaxed. Simulink Accelerator has correct minor step guard for enable, if, case Under certain circumstances with the Simulink Accelerator, the generated code of models containing enable, if and case subsystems could be activiated during minor time steps, causing incorrect outputs. This has been fixed. Simulink Accelerator now functions without an RTW license The Simulink Accelerator no longer requires a Real-Time Workshop license to function properly. Simulink Accelerator supports gcc compiler on UNIX The Simulink Accelerator now supports use of the gcc compiler on UNIX systems. Spline Toolbox Click on a problem area listed below to read how it has been fixed. pprfn failed when no breaks need be inserted Data edits in splinetool now tracked properly Errors in spcol for various unusual inputs corrected pprfn failed when no breaks need be inserted pprfn now properly returns the function unchanged in case all the additional breaks are already breaks for the function. Data edits in splinetool now tracked properly Each spline fit now remembers the data to which it was fit. If, at some point, you make a previously constructed spline fit the current one, then also the data currently showing get changed to the data to which that spline was fit. Errors in spcol for various unusual inputs corrected spcol now provides a correct sparse version of its output even for knot sequences with unusual end knot multiplicities, and when some sites lie to the right of the last knot while some site appears more than once. None of these cases is likely to occur in ordinary spline work since, in all of them, the resulting matrix is not invertible. Stateflow Click on a problem area listed below to read how it has been fixed. Code generation for RTW target failed for array data initialized from workspace Generated code for RTW target did not include custom code Hierarchical printing for closed charts caused transition and printout errors Infinite flow graph loops in Stateflow did not stop during simulation Mismatched dimensions in assignments not flagged by parser Model file names in Stateflow were limited to 25 characters Pointer and address operator use in generated code caused compile errors Search and Replace errored out with LaTex strings in annotations Smart transitions in resized grouped state were graphically messy Transition actions were handled incorrectly for a confluence of transitions Transition source for event broadcast was highlighted for the rest of simulation Code generation for RTW target failed for array data initialized from workspace Code generation for an RTW target no longer fails when the chart includes output array data initialized from the workspace. (Occurred in Versions 3.0 through 4.1.1.) Generated code for RTW target did not include custom code Code generated for an RTW target now can includes any custom code entered by a user. (Occurred in Versions 4.1 through 4.1.1.) Hierarchical printing for closed charts caused transition and printout errors Hierachical Printing invoked from a Simulink window when the charts were closed no longer causes transition label positions to change and no longer causes incorrect printouts. (Occurred in Version 4.0 through 4.1) Infinite flow graph loops in Stateflow did not stop during simulation In models containing an infinite flow graph loop, you can now stop simulation by using Ctrl-C or the Stop Simulation button in the diagram editor and debug window. Mismatched dimensions in assignments not flagged by parser The Stateflow parser did not flag mismatched dimensions or array lengths in vectorized assignments such as x = y, where x and y are vector data with different dimensions and/or lengths. This no longer causes crashes during runtime. (Occurred in Versions 2.0 through 4.1.) Model file names in Stateflow were limited to 25 characters Models containing Stateflow charts must have file names limited to 25 characters or less. When you try to save a model with more characters, an error results. If you try to load a model with more than 25 characters in its file name, a warning results telling you to reduce the number of characters in the file name to 25 or less. Pointer and address operator use in generated code caused compile errors Certain uses of the pointer and address operators ('->' and '&', respectively) in state and transition actions used to cause uncompilable generated code. (Occurred in Versions 3.0 through 4.1.1). This has been fixed. Search and Replace errored out with LaTex strings in annotations The Stateflow Search and Replace Tool no longer errors out in the presence of LaTeX format strings in Stateflow annotations. (Occurred in Versions 4.0.2 through 4.1.) Smart transitions in resized grouped state were graphically messy When a grouped state containing "smart" transitions was resized, the transitions in the newly resized state are now displayed properly. (Occurred in Version 4.1) Transition actions were handled incorrectly for a confluence of transitions Transition actions are now handled correctly for a confluence of transitions into a single destination such as the junction shown below: Transition source for event broadcast was highlighted for the rest of simulation During simulation an early return from an event broadcast from a transition no longer leaves that transition highlighted for the rest of the simulation. (Occurred in Versions 3.0 through 4.1.) Stateflow Coder Click on a problem area listed below to read how it has been fixed. Charts containing unstructured flow graphs caused compilation errors during Real-Time Workshop builds Missing return statements in some graphical functions Optimization of simple "if" conditional statements caused incorrect code Optimization of state exit functions caused incorrect code to be generated Passing chart input data to graphical functions caused code generation failure for RTW targets Using min/max macros in Stateflow charts caused compilation failures during RTW build Charts containing unstructured flow graphs caused compilation errors during Real-Time Workshop builds Charts containing unstructured flow graphs no longer cause compilation errors during Real-Time Workshop builds. Missing return statements in some graphical functions All graphical functions with a return value and unusual flow graph topology now include a final return statement in the generated code. Optimization of simple "if" conditional statements caused incorrect code Optimization of simple "if" conditional statements no longer causes incorrect code to be generated. (Occurred in Version 4.2) Optimization of state exit functions caused incorrect code to be generated Optimization of state exit functions no olonger causes incorrect code to be generated. (Occurred in Version 4.2) Passing chart input data to graphical functions caused code generation failure for RTW targets Passing chart input data to graphical functions no longer causes code generation failures for Real-Time Workshop targets. Using min/max macros in Stateflow charts caused compilation failures during RTW build Using min/max macros in Stateflow charts caused compilation failures during RTW builds of ert/grt embedded targets. This has been fixed. Statistics Toolbox Click on a problem area listed below to read how it has been fixed. cluster function usable with larger problems dendrogram graph corrected for plotting a subset of nodes pdist more efficient for large data sets ranksum uses a continuity correction Corrected calculation of covratio statistic by regstats function Linear and quadratic discriminant analysis added to classify Negative binomial (nbin*) functions no longer require an integer R value cluster function usable with larger problems The cluster function is no longer recursive, so it no longer encounters a recursion limit when it is applied to a large data set. Consequently, cluster now succeeds with larger data sets than in the past. dendrogram graph corrected for plotting a subset of nodes In a plot of a subset of the nodes in a cluster tree, the dendrogram function now adjusts the Y axis limits to scale the tree properly, and it shows the correct distance values. Before Version 4.0 (Release 13), it was subtracting an offset to position the minimum distance value near zero, and therefore it was not possible to read the correct distance values from the Y axis of the resulting plot. pdist more efficient for large data sets The pdist function now makes more efficient use of memory, so it can be used on larger problems than in previous releases. ranksum uses a continuity correction The ranksum function uses an exact calculation for small sample sizes, and a normal approximation for larger sample sizes. Starting in Version 4.0 (Release 13), the normal approximation now includes a continuity correction. Corrected calculation of covratio statistic by regstats function In Version 3.0 and earlier of the Statistics Toolbox, the covratio statistic was calculated incorrectly. The calculation was fixed in Version 3.1. This statistic measures the ratio of the generalized variance of the coefficient estimates without a given row, to the generalized variance using all the data. Linear and quadratic discriminant analysis added to classify The algorithm that was previously implemented in classify used the Mahalanobis distance between sample points and training groups, with stratified estimates of covariance. The new implementation adds the standard algorithms for linear (default) and quadratic discriminant analysis. Use the type 'mahalanobis' in the new version to duplicate the behavior of the previous version. Negative binomial (nbin*) functions no longer require an integer R value Consistent with a more general interpretation of the negative binomial, these functions now accept any positive value, including nonintegers, for the size parameter R: nbincdf, nbininv, nbinpdf, nbinrnd, nbinstats. Wavelet Toolbox Click on a problem area listed below to read how it has been fixed. cmorwavf function inputs Different results for std2 function in the Wavelet and Image Processing Toolboxes Loading wptree object from a MAT-file Loading complex data into Continuous Wavlet 1-D Using wpdec function with a user-defined function cmorwavf function inputs The two last input arguments are no longer reversed in the cmorwavf function. Different results for std2 function in the Wavelet and Image Processing Toolboxes You no longer get slightly different results from the std (compute standard deviation of matrix elements) function in the Image Processing Toolbox and the Wavelet Toolbox. Loading wptree object from a MAT-file wptree has been updated so that a wptree object can be loaded from a MAT-file. Loading complex data into Continuous Wavlet 1-D An error message stating that the data must be real is displayed when you try to load complex data into the Continuous Wavlet 1-D. Using wpdec function with a user-defined function Using wpdec with a user-defined function no longer produces errors that were caused by wpdec's use of wentropy. xPC Target Polling mode with embedded option Previously, when you created a Stand-alone application, polling mode did not start. Now, it does.