R2026a

New Features, Bug Fixes, Compatibility Considerations

 Deep Learning: Downsample sequence and image data using wavelet pooling layers

This release introduces the waveletPooling1dLayer and waveletPooling2dLayer objects. The layers enable you to downsample features within a deep learning network using wavelet decomposition and reconstruction. The layers support adaptive and nonadaptive wavelet pooling. In adaptive wavelet pooling, scaling factors on the detail and approximation coefficients are modified through gradient descent or learning. You can specify layer parameters such as:

• Which wavelet to use in the pooling operation

• Analysis (decomposition) level

• Synthesis (reconstruction) level

• Fixed or learnable lowpass (scaling) and highpass (wavelet) subband gains.

• Which detail coefficients to include in the pooling

waveletPooling1dLayer and waveletPooling2dLayer support real-valued inputs in the "CBT" and "SSCB" format, respectively.

 New Orthogonal Wavelet Family: Complex-valued least asymmetric Daubechies wavelets

Wavelet Toolbox™ now supports complex-valued least asymmetric Daubechies wavelets for critically sampled 1-D and 2-D discrete wavelet analysis. The new wavelets, with family short name "csym", are also known as complex symlets.

You can use these wavelets to obtain a minimally redundant discrete wavelet transform of complex-valued data, instead of having to treat the real and imaginary parts of the data separately. Complex symlets are sensitive to diagonal features in image data and have the highest number of vanishing moments for a given support width. The associated scaling filters have near-linear phase response and are symmetric for odd numbers of vanishing moments and are least asymmetric for even numbers of vanishing moments.

This release also introduces two functions related to complex symlets.

• csymwavf — You can use the family short name with csymwavf to obtain the associated scaling filter. You can then use orthfilt to obtain the four filters associated with the wavelet.

• daubfactors — You can also use daubfactors to obtain the scaling filter. daubfactors unifies the functionality of dbaux and symaux, as well, enabling you to obtain the scaling filters of the three Daubechies families, "csym", "db", and "sym" through a single interface.

These functions support complex symlets:

• 1-D Discrete Wavelet Analysis — dwt, idwt, wavedec, waverec, dwpt, idwpt, appcoef, detcoef, and wrcoef

• 2-D Discrete Wavelet Analysis — dwt2, idwt2, wavedec2, waverec2, appcoef2, detcoef2, and wrcoef2

• Wavelet Filters — wfilters, orthfilt, and isorthwfb

• Wavelet Management — wavemngr and waveinfo

Create Plot Live Editor Task: Visualize continuous wavelet transform function output

You can now use the Create Plot Live Editor task to interactively visualize the output of the cwt function. You can select different chart types and set optional parameters. The task also automatically generates code that becomes part of your live script.

 Continuous Wavelet Transform: Specify zero padding as an extension mode

The Wavelet Time-Frequency Analyzer app, the cwt and wsst functions, and the cwtfilterbank object now support zero padding as a boundary-extension mode.

 Compatibility Considerations

Use the new Boundary name-value argument to specify the extension mode when using cwt and wsst. For more information, see cwt and wsst Functions: ExtendSignal name-value argument will be removed.

Wavelet Scattering: Downsample filters while preserving halfband frequency support

You can now use the new FilterDownsampling property of waveletScattering to downsample the wavelet and scaling filters in a wavelet scattering network to halfband support without introducing aliasing. The analytic wavelets remain analytic and bandlimited. The object function filterbank now returns, on a base-2 logarithmic scale, the downsampling factors in the filter bank metadata.

The default value of FilterDownsampling maintains compatibility with all previous releases of waveletScattering. MATLAB^® code written using releases prior to R2026a will continue to work as you intended. Code updates are not necessary.

cwt and modwtmra Functions: Plot in target axes and apps

The cwt and modwtmra functions now support a new Parent name-value argument that allows you to specify a target container to plot the scalogram and multiresolution analysis (MRA) decomposition. The cwt function supports Axes, UIAxes, and Panel as targets. modwtmra supports Panel objects. To learn how to target a parent object for a plot, see Plot Spectral Representations of Signal in App Designer.

Deep Learning: Code generation for continuous wavelet transform

The dlcwt and dlicwt functions and cwtLayer and icwtLayer objects all support C/C++ and GPU code generation.

You must have MATLAB Coder™ to generate C/C++ code. To generate GPU code, you must have GPU Coder™.

GPU Computing: Accelerate tffilt on your GPU

The tffilt function now supports gpuArray objects.

You must have Parallel Computing Toolbox™ to use gpuArray objects with supported functions. For more information, see Run MATLAB Functions on a GPU (Parallel Computing Toolbox). To find which GPUs are supported, see GPU Computing Requirements (Parallel Computing Toolbox).

 Functionality being removed or changed

cwt and wsst Functions: ExtendSignal name-value argument will be removed

Still runs

The cwt and wsst name-value argument ExtendSignal will be removed in a future release. To specify the boundary-extension mode, use the new name-value argument Boundary instead. The Boundary name-value argument supports zero padding as an extension mode and makes cwt and wsst more consistent with other wavelet transform APIs, such as dwpt and dldwt. You cannot specify Boundary and ExtendSignal simultaneously.

Functionality	Result	Use Instead
ExtendSignal=0	Still runs	Boundary="periodic"
ExtendSignal=1	Still runs	Boundary="reflection"

dbaux and symaux functions will be removed

Still runs

The dbaux and symaux functions will be removed in a future release. Use daubfactors instead. This function unifies the functionality of dbaux and symaux, as well, enabling you to obtain the scaling filters of the three Daubechies families, "csym", "db", and "sym", through a single interface.