Major Bug Fixes for Release 12

This document describes major bug fixes for the following:

• MATLAB 6.0
• Communications Blockset 2.0
• Optimization 2.1
• Signal Processing Toolbox
• Communications Toolbox 2.0
• Wavelet Toolbox
• Data Acquisition Toolbox
• Symbolic Toolbox 2.1.2 Bug Fixes

Stereo sound: recording both channels

Recording a stereo .wav file now works correctly. Previously, the recorded sound was distorted (missing a channel) on playback.

Bitmaps stretched evenly when printing

Stretched bitmaps are now stretched evenly when printing. Previously, one column of the stretched image would occasionally print at double width.

Resize figure window using original units during printing

The printing code now calls each figure's ResizeFcn (if the figure has one) with the figure units unchanged. Previously, ResizeFcn was called after setting the figure units to paper units.

Colormaps correct when printing two or more images

The color postscript driver now preserves the colors of multiple images (or axes) printed on same page. Previously, some incorrect colors may have appeared in the output.

Print to metafile at higher than screen resolution

The -r (resolution) flag now allows you to specify any resolution when printing to a metafile. Previously, you could only print to metafiles at your screen resolution.

IFFT2 uses correct scale factor for N-dimensional inputs

IFFT2 now applies the correct scale factor in the case where the input has 3 or more dimensions and the user supplies the optional mrows and mcols input arguments. Previously, IFFT2's output was off by a constant scale factor in this case.

GRIDDATA more robust

GRIDDATA now uses DELAUNAYN to compute its triangles instead of DELAUNAY. This is more robust in many cases. Previously, GRIDDATA used to return NaNs when used on certain cases.

DELAUNAY reverts to its MATLAB 5.2 behavior

A MATLAB 5.2 bug in DELAUNAY was fixed for MATLAB 5.3 but this caused different results in some cases. In some cases, the warning "Degenerate case: collinear points. Possible incorrect triangulation." was issued. A modification to this bug fix returns the MATLAB 5.2 answer in MATLAB 6.0.

INTERP1(X,Y,XI,'cubic') no longer calls 'spline' method on irregularly spaced data

INTERP1(X,Y,XI,'cubic') now calls 'pchip' and returns cubic interpolation even when the data is not equally spaced. Previously, INTERP1 used to call 'spline' when the data was not equally spaced. The old behavior is still accessible through INTERP1(X,Y,XI,'v5cubic').

DBLQUAD no longer accepts a trace argument

DBLQUAD used to accept a trace argument, ostensibly in order to pass it on to quad. This was simply a placeholder and had no effect on the function, so it was removed.

ODE solvers produce output at all time points specified in TSPAN.

SOLVER(ODEFUN,TSPAN,Y0) called with a vector TSPAN now evaluates the solution at all time points specified in TSPAN. Previously, due to rounding errors, the solution was not always evaluated at the last point of TSPAN.

Figure property KeyPressFcn

The figure property KeyPressFcn now generates an event for all keys pressed. Previously, modifier keys (such as Control and Esc) did not generate an event to execute the KeyPressFcn callback.

Some blocks do not support Real-Time Workshop code generation

• Eye and scatter diagrams
• Continuous-time voltage-controlled oscillator
• Passband modulators and demodulators
• CPM modulators and demodulators
• Multipath Rayleigh Fading Channel
• Rician Fading Channel

Some blocks are not compatible with the Simulink Accelerator

The digital modulators produce continuous-time outputs and process datain continuous time

The M-FSK demodulators use incorrect correlation

The complex Rayleigh fading channels have incorrect characteristics

Probabilities of ones in output from the Binary Vector Noise Generatorare incorrect

Gaussian Noise Generator uses wrong sample time

PN Sequence Generator repeats its output

Eye and scatter plot blocks produce an error if you close the figure windowand run the simulation again

Scrambler and Descrambler give incorrect results for non-base-2 calculations

Scrambler and Descrambler ignore all elements other than the first in acalculation base vector

Scrambler and Descrambler negate the results when the initial states containnegative numbers

The AWGN Channel produces all-zero outputs when given continuous-time inputsin both of the Signal to noise ratio (Es/No and SNR) modes

Segmentation violation when blocks are given invalid or empty parameters

QUADPROG and LINPROG: empty f vector assumes f=zeros(n,1).

In calls such as
    QUADPROG(H,f,A,b)
or
    LINPROG(f,A,b)
if f = [], f is treated as if f = zeros(n,1). Previously, f = [] would sometimes cause poor solutions or incorrect messages about lack of convergence.

QUADPROG iterates no more than 'MaxIter' iterations.

QUADPROG will iterate no more than the value of 'MaxIter'. Previously, QUADPROG would iterate up to the value of 'MaxIter' + 1 iterations.

LINPROG: option 'Display' set to 'off' suppresses all messages.

Setting 'Display' to 'off' in the options passed to LINPROG now suppresses all messages. Previously, messages about exit conditions were sometimes printed.

FSOLVE sets the EXITFLAG output argument correctly when 'Display' is 'off' .

When 'Display' is set to 'off', the EXITFLAG output argument from FSOLVE is set correctly. Previously, the EXITFLAG would sometimes have the incorrect value.

LSQLIN: memory usage reduced in medium-scale algorithm.

LSQLIN with 'LargeScale' set to 'off' uses less memory by using an "economy-sized" QR factorization when possible. Previously, if the first argument was a large m-by-n matrix where m>>n, an out of memory message was generated. It is still possible to have a large enough matrix that the out of memory message is generated, but "large enough" is bigger than it was before.

FMINUNC, LSQCURVEFIT, and LSQNONLIN line search algorithms no longer use " \ ".

Line search algorithms for FMINUNC, LSQNONLIN, and LSQCURVEFIT avoid divide by zero messages. Previously, the algorithms used backslash (" \ ") in such a way that extraneous divide by zero warnings were sometimes printed to the screen.

 

xcorr computes cross-correlation correctly

We now compute the cross-correlation as an estimate of Rxy(k) = E{x(n+k)*conj(y(n)} = E{x(n)*conj(y(n-k)}. The
speed of the algorithm has also been vastly improved, especially for long vectors.


lpc now returns the prediction error variance correctly


The error variance is computed based on the correctly scaled auto-correlation sequence (the biased estimate).


fircls correctly handles frequency steps bigger than 0.01

The fircls function no longer produces the warning "Empty matrix multiplication with unequal inner dimensions" when the frequency step is bigger than 0.01.


Made remez more robust

remez no longer generates the warning "Subscript indices must be integer values".


Improved help for hilbert

Help for hilbert now states that when the input is complex, only the real part is used. Also, hilbert generates a warning if the input is complex.


Fixed axis labeling in freqz

freqz now labels the x-axis correctly in all cases.


AR spectral estimation methods calculate the one-sided PSD

The power spectral estimation (PSD) functions pburg, pcov, pmcov, and pyulear now calculate the one-sided PSD.


pmtm now computes one-sided and two-sided PSD

The power spectral estimation (PSD) function pmtm now accepts an optional string to specify one-sided or two-sided PSD.


Improved handling of input string flag to ss2sos

ss2sos now correctly allows the direction flag string to be specified as either the 5th or 6th input argument.


tf2latc correctly normalizes transfer function

tf2latc now correctly normalizes the numerator and denominator when the leading coefficient of the denominator is not equal to one.

VITERBI is slow and does not decode correctly

DDEMOD and DDEMODCE do not produce correct symbol error rates

DMOD and DMODCE generate incorrect waveform for MSK and FSK

GFADD, GFSUB, GFDIV, GFMUL, GFCONV and GFDECONV return incorrect answers

• Correct answers for prime and extension Galois fields, including prime fields, GF(p), where p>2.
• Correct handling of -Inf and negative values for extension Galois fields.
• Enhanced help descriptions to better distinguish the purposes of the functions.
• Improved input parameter checking.

GFMINPOL returns incorrect answers when first input is -Inf or when p>2. The function also sometimes crashes

GFPLUS and RSENCODE returns incorrect answers for negative inputs

GFLINEQ returns incorrect answers in prime Galois fields of order greaterthan 2

GFPRIMDF produces "out of memory" messages for degrees higher than 24

DECODE using the cyclic decoder option does not decode (23,12) Golay codecorrectly

GFPRIMFD finds incorrect primitive polynomial

GFTUPLE returns incorrect answers when m=1

GFPRIMCK, GFTRUNC, GFADD and GFFILTER causes segmentation violations

GFPRIMCK returns incorrect answers if p > 2 or inputs are large.

DECODE incorrectly decodes block codes

RCOSFLT does not correctly filter and upsample the input signal

EYESCAT is difficult to use and plots I and Q components together

ADEMOD  ignores the phase offset parameter under the 'pm' option andhas no sensitivity parameter under the 'fm' option

ADEMOD 'pm' option - sensitivity parameter is required and causes a dcoffset

RANDINT hangs when the range is large

RANDBIT output is not random

Histograms for a level 1 analysis

The histogram tool now correctly shows the histogram plots for a level 1 analysis in both the 1-D and 2-D tools.

Number of colors used in the Continous Wavelet 1-D tool

The maximum number of colors available is no longer set, but determined based on platform.

Number of axes available to view

The View Axes palette (formerly labeled, Zoom Axes) is now updated with the correct number of pushbuttons after preforming a 2-D analysis.

Correct wavelet family in Cascade Algorithm demo

The demo now uses the correct family to approximate the wavelet function.

New Channel Properties

1. Two read-only properties, NativeOffset and NativeScaling, have been added to analog input and analog output channels. These properties can be used to convert data in native format to doubles with the formula:
   slope = ai.Channel(1).NativeScaling;
   intercept = ai.Channel(1).NativeOffset;
   data = slope*nativeData(:,1) + intercept;
   

Daqhwinfo

1. The ConversionExtraScaling and ConversionOffset fields have been removed from daqhwinfo('adaptor') where adpator is winsound, nidaq or hpe1432. The Channel properties NativeOffset and NativeScaling should be used to convert data in native format to doubles.
2. If an older version of the comcat.dll is on your computer, DAQHWINFO with no input arguments will not work. However, DAQHWINFO with input arguments will work and the rest of the toolbox will work. For additional information, please refer to: http://www.mathworks.com/support/solutions/v5/23852.shtml

Properties

1. When the SamplesPerTrigger property is set to a small number and the TriggerRepeat property is set to a large number, and the analog input object is started, your machine may lock up. This will generally occur when SampleRate/SamplesPerTrigger is greater than 1000 and the TriggerRepeat property is greater than 1000. If the analog input object's properties follow the characteristics described above, you will receive the following warning message:
   Warning: Engine Block Size is small. Check SamplesPerTrigger.
   Starting may lock up the machine.
   
   As a workaround, the SamplesPerTrigger property should be set to the number of samples you would like to acquire and TriggerRepeat should be set to 0.

Data Logging

1. The UserData information is not returned from a .daq file.

Putdata

1. The BitsPerSample, InputRange, SensorRange and UnitsRange properties should not be modified after calling PUTDATA.
2. A channel cannot be added after calling PUTDATA.

Adaptors - Nidaq

1. When running at a sampling rate of 5000 or higher and with a TransferMode value of Interrupt, there may be a considerable decline in system performance.
2. The TriggerChannel for National Instruments 61110 and 61111 boards is always the first channel in the channel array.
3. It is recommended that the SampleRate property is the last property set with SETVERIFY before starting since the SampleRate value is dependent upon the number of channels and the ChannelSkew property value.
4. When using the 1200 series hardware, channels must be added in reverse order. If invalid channels are given, the data acquisition engine will create the number of requested channels with valid hardware ids. The hardware ids used can be determined from the object's display or from the channel's HwChannel property value.
5. Only one digital I/O object should be pointing to the same hardware. To perform separate tasks on the hardware lines, all the necessary lines should be added to the same object and separate line groups should be created to perform the separate tasks.
6. All channels contained within an analog input object must have all unipolar InputRange property values or all bipolar InputRange property values. A channel array cannot contain channels with a unipolar InputRange and channels with a bipolar InputRange.
7. When using mux boards with National Instruments, the channels must be added in a specific order. For additional information, please refer to the online help for addmuxchannel.
8. The PCI-1200 does not work properly with sample rates below 30-40 Hz if the TransferMode is set to SingleDMA. If the TransferMode value is Interrupt, the PCI-1200 works properly.

Adaptors - Hpe1432

1. For the HP E1432A and HP E1433A boards with the Arbitrary Source Output option, if an analog input object is created and started, the source will also be started. Due to this behavior of the hardware, a TriggerType of Manual should not be used for the HP E1432A and HP E1433A hardware with the Arbitrary Source Output option.
2. For analog output objects, it is recommended that the SampleRate or Span property is the last property set with SETVERIFY before starting since the SampleRate and Span values are dependent upon the number of channels contained in the analog output object.
3. The first time an analog input or analog output object is connected to the HP hardware, a list of available HP hardware is determined and all the hardware is initialized. Due to an HP limitation, this list is not updated during a MATLAB session. Therefore, if you install a new board or remove an existing board while MATLAB is running, MATLAB will not see the new configuration and the old configuration will still be used. To have MATLAB see the new configuration, you must restart MATLAB.
4. Due to an HP limitation, the minimum SampleRate on the HP E1433A is 20Hz and the minimum Span on the HP E1433A is 7.8125.
5. The first channel in the TriggerChannel property will be used to trigger the object.
6. The HP E1434A channels 1 and 2 and channels 3 and 4 share a 56000 DSP. Therefore, certain operational aspects will be coupled between the channels in each pair. For example, both channels in each pair will have the same RampRate.
7. Channels must be added in increasing order and a channel array cannot contain repeated channels.
8. All hardware is initialized the first time an HP object is created. When all HP objects are deleted from the data acquisition engine, all the hardware is closed.
9. It is recommended that if an object is created which spans multiple boards, that the object is created with the logical addresses listed in the same order in which they were found. The DAQHWINFO command with no input arguments will list the InstalledBoardIds in the order in which the logical addresses were found.

Improved handling of control-c

Pressing cntrl-c to stop execution during a symbolic computation no longer causes a segmentation violation.

SYM is more robust

SYM is now more robust at parsing its inputs. In earlier releases, there were cases where SYM would create invalid matrices that would produce error messages when these matrices were used.

EQ produces a logical array

EQ, the overloaded function used when checking for equality with ==, now returns a logical array. In earlier releases a non-logical array was returned.

SYMVAR produces a column-shaped output

SYMVAR now always produces a column cell array. In earlier releases, a row was sometimes produced.

Major Bug Fixes From Recent Releases

This document describes major bug fixes for the following:

• MATLAB 5.3.1

Invisible figures cause focus changes to be slow (UNIX only)

Creating invisible figure windows caused Matlab to thrash the X server during focus changes. This his been fixed.

Invalid movies caused a segmentation violation

Previously, calling the movie command with a movie array in which one or more elements were unitialized caused a segmentation violation. This now causes an error.

getframe sometimes fails to capture uicontrols (UNIX only)

The getframe function now always captures uicontrols.

getframe fails with truecolor using less than 24 bits (UNIX only)

The getframe function now supports truecolor XWindows visuals using fewer than 24 bits.

Memory leaked when converting arrays from MATLAB to Java

Extensive use of the MATLAB/Java interface could result in exhaustion of system memory. This memory leak has been plugged.

Using CAT with cell or struct arrays may result in a crash

Certain usages of CAT with cell or struct arrays or empty arrays can result in a segmentation violation. This regression from MATLAB 5.2 has been corrected.

MATLAB may crash while reading a corrupt MAT-file

It is possible to encounter a segmentation violation when reading a corrupted MAT-file due to misplaced error check. This problem has been corrected.

Incorrect results when doing complex division in array initialization

A constant expression involving division by an imaginary number (e.g. 3/i) produced random results when used in forming an array.

Displaying structures with long field names can cause an assertion

Displaying structures with long field names can cause an "Attempt to free previously freed memory" assertion due to a flow-of-control error. This problem has been corrected.

Colon indexing with sparse cell arrys can cause an assertion

An expression of the form b{1} = sparse(eye(4)); b{1}(1,1) would cause an "Attempt to free previously freed memory" assertion. This problem has been fixed.

Error occurs when using GUIDE Control Panel

Opening the GUIDE Control Panel on Microsoft Windows with no open Figure windows can cause an error under certain circumstances. This problem has been corrected.

Old P-files may cause crash when displaying cell arrays or structures

Cell array or structure expressions that display their results (i.e., are not terminated with a semicolon) may result in a crash when run in MATLAB 5.3 if they were saved in a P-file with MATLAB 5.2 or earlier.

MATLAB may crash on HP systems with HP's Y2K patch (HP700 only)

A symbol name conflict between one of our third party libraries and one of HP's system libraries can cause MATLAB to crash when creating new files. This issue has been resolved via an updated library.

MATLAB may fail to start when running on IRIX 6.5 (SGI/SGI64 only)

IRIX 6.5 no longer ships with certain math libraries that MATLAB depends on. The missing libraries are now provided as part of the MATLAB distribution in $matlab/sys/os/sgi[64].