chipmuenk / pyfda Goto Github PK

• Selection range of zoom tools is not visible with dark theme (black on black)
• The stimulus in the plot_impz.py is not visible
• Grid lines of all QTableView > QHeader widgets are dark instead of white
• Separator lines of matplotlibs NavigationToolbar are white instead of dark
• Icon are hardly visible with dark theme

Selecting different sets of icons depending on the theme could be achieved by:

• using QFileSelector (qt5 only, unfortunately)
• working with QResource by specifying different prefixes, e.g.

  <qresource prefix="/images/icons/dark">
  <file alias="quantize.svg">quantize.svg</file>
  </qresource>
  
  

This would require compiling to e.g. qrc_resources_dark.py and qrc_resources_light.py and make a conditional import.

The following style sheet could be used or modified:
https://github.com/ColinDuquesnoy/QDarkStyleSheet

Currently, all error and warning messages generated within the filter design routines are printed to the console (where you'll probably miss them).

A pop-up window or a text field in the GUI should display them instead (or additionally).

When trying to reduce the horizontal size of the input widget area, it is either shown or hidden completely when using the splitter.

Probably individual scroll areas are needed for each input widget.

First of this tool looks great! thank you so much for your effort in putting this together.

I'm trying to use this to plot / design low and high pass filters though noticed that in changing the Fc value the filters are not changing. The filters always seem to be calculated using the default Fc value for the filter type. Is there a known work around for this issue? The filter order does seem to be working properly.

This would be implemented in plot_widgets/plot_impz.py.

Differentiator with minimum order selected fails - "unknown design method DIFFmin": minimum order should not be selectable

Hilbert: Target specifications shows empty subwidget

Wanted to install the new version. Used the described command "python setup.py -install" form the mainpage. If you look in the setup help using "python setup.py --help", the install command is "python setup.py install". Without the "-" which works perfectly fine :)

When a filter is designed manually by saving P/Z or b,a, the filter design method changes to "manual" but the display of the widgets is not updated.
This would require triggering FilterSpecs.update_UI().

Shrink the status window to a smaller size and place it under the plot widget tab.

For manual filter order design algorithms, target frequency and amplitude specs should be displayed although they are not used for the design process.

They are needed though for displaying the spec limits. Currently, this doesn't work.

The root cause for this lies in the module filter_specs.py where the amp. and freq. parameters for minimum order filter design are misused as target specs. This has to be fixed by adding 'targ' parameters in the filter specifications of all filter modules and using them for the spec limits.

Horizontal and vertical stretches are not removed when not needed, creating layout faults

... hopefully, this is easy once the crash issue has been fixed.

pyFDA should detect at startup whether pyQt5 is available and set a corresponding global flag for doing something like

if PYQT5: from PyQt5 import QtGui from PyQt5.QtCore import pyqtSignal else: from PyQt4 import QtGui from PyQt4.QtCore import pyqtSignal

For details see http://pyqt.sourceforge.net/Docs/PyQt5/pyqt4_differences.html

• only new-style signal-slot connections have been used so far, so this shouldn't be a problem
• "disconnect() takes no arguments and disconnects all connections to the QObject instance" - this could be problematic
• "PyQt4’s QtGui module has been split into PyQt5’s QtGui, QtPrintSupport and QtWidgets modules" - this needs to be checked

The base directory that is defined in pyfda_rc throws a warning in pyfdax.py when it doesn't exist. This might confuse users, change it to level INFO.

This also gives a message during pip installation:

Detected hard-coded path in text file bin/pyfdax
Detected hard-coded path in text file bin/pyfdax_no_term

In contrast to FIR filter design, IIR filters only can be designed with identical amplitude specs for both pass- resp. stopbands.

The UI does not allow to enter e.b. A_PB2 or A_SB2, hence, prior settings are used for displaying the specs in H(f) and the info tab.

Solution: Copy A_PB -> A_PB2 resp. A_SB -> A_SB2 in the filter design routines.

For some filter designs, dynamic subwidgets with text entries, combo boxes etc. are created (e.g. grid_density for equiripple designs).

Changing these subwidgets does not trigger the signal sigSpecsChanged.

Reminder: Check whether signals are properly disconnected when reloading filters !

Hello. Thank you for this useful tool. I installed it on my laptop with Windows 10, which is set to 150% scaling, 144 dpi. Unfortunately, everything was so tiny, to the point of being unreadable.

A possible reason is that you specify font sizes in absolute pixels (12px) which probably does not allow scaling on high dpi displays. I've changed font sizes to points, by replacing in pyfda_rc.py, 12px with 10pt (three occurrences). This way, the font scales well and the interface looks good both on standard display and on the high dpi one.

The second problem was that the left pane was way too narrow, most of the list box items were replaced with ellipsis. For my own purposes, I changed the pane width (pyfdax.py, line 97) to setMaximumWidth(420) and now it looks good on my display. Maybe this value should be set automatically based on screen width. Maybe this line is not even necessary.
Edit: I've just checked this on my laptop: removing this line gives a nice looking layout.

An another thing: tabs labels were not fitting in the tabs. You specify the minimum width of tab as 10ex (pdyfa_rc.py, line 165). ex is the height unit, width should be specified with em because it makes sense. Changing this to min-width: 5em; gave the result that looks very nice on my screen.

Please consider making these changes in your code.

Certain choices caused these errors:

[  ERROR] [pyfda.pyfda_lib:1147] could not broadcast input array from shape (0) into shape (2)
[  ERROR] [pyfda.filter_factory:216] Err. Code 18:

Error calling method 'BPman' of class 'Butter':
The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
[WARNING] [pyfda.input_widgets.filter_specs:286] start_design_filt:
Attribute not found.
Unknown design method.

But it is hard to tell what's wrong from messages.

Screen looks like this:

... as the title says, forgotten to rename it at some point.

When creating a design that fulfills the specs at the corners of pass and stop band but violates them somewhere inbetween, this is not flagged in the Info tab (input_widgets/input_info.py).

The reason is that so far only the band corners are checked ...

When adding / deleting poles or zeros, the gain factor k remains unchanged. This changes the overall filter gain e.g. in the passband.
File: input_widgets/input_pz.py
Method: normalizeCoeffs()

The QClipboard widget is an elegant way for importing (e.g. coefficients, P/Z, ...) and exporting (e.g. screenshots) data - this would enhance interfaces and functionality.

See e.g. http://python.6.x6.nabble.com/Paste-entire-column-into-QTableView-from-Excel-td5016995.html

Clipboard functionality has been started as qcopy_to_clipboard() and qcopy_from_clipboard() in the library pyfda_qt.lib: The following steps are required / are working already:

• qcopy_to_clipboard() with all cells as float
• qcopy_to_clipboard() with selected cells in current format: second column (empty) is always copied as empty string. When only some cells are selected, number, order and content are not correct.
• qcopy_from_clipboard() as float or in selected format
• combine with csv export and import
• import / export in other formats than float doesn't work yet

Particularly, if somebody isn't already an expert, they will be wondering what Fxx and Axx are.

Some entry fields are not yet filtered through safe_eval() and thus potentially unsafe and / or throw errors in the console. They should also use event_filter for displaying a limiting number of digits.

• Update simple_eval() to new version
• Update all QLineEdit fields to use safe_eval() with backup parameter
• Add a paramter to safe_eval() to check / convert e.g. for integer type
• Implement QEvent filters to allow editing the values with full resolution

Integer QLineEdits:

• select_filter.py -> set_filter_order (): filter order
• equiripple.py: grid_density
• ma.py: M, stages
• plot_impz.py: number of data points

Float QLineEdits:

• plot3D: various fields
• Windowed FIR: various fields
• filter_coeffs.py -> set_coeffs_zero: scale (float instead of int), W, WI, WF

Add another error handler to display debug,info, error messages to user on the main display. In addition to logging errors to error log and terminal window, they will also appear on the main GUI display.

In some cases it is not possible to design a filter although the displayed filter specifications seem to be o.k. It seems that the corner frequencies are passed in the wrong order to the actual filter design routine.

This bug occurs from time to time when changing the response type (e.g. from bandpass to bandstop).

Most likely, the bug is in input_widgets/input_freq_specs.py, input_specs.py or input_targ_specs.py.

... e.g. from 0 ... f_S/2 to 0 ... f_S and back. The graph is displayed correctly but the axis stays the same.

• Scaling parameter is not used yet
• Illegal values in CSD return zero
• int_places calculation for CSD doesn't always make sense
• Frmt2float for fractional hex yields wrong results, probably due to incorrect use of MSB
• WI < 0 yields wrong scaling for decimal

The QTableWidget for viewing and editing pole/zeros is still missing.

The active plot tab is redrawn when the size of the canvas is changed via an event filter in plotTabWidgets that calls the redraw routines of all plot widgets after a delay of 500 ms.

However, the hidden plot tabs are not affected by this, requiring a manual redraw (the brush icon).

This could be solved by only redrawing the active tab window upon resize and redrawing the other tabs when they become active.

I've noticed an issue in the calculation of group delay and magnitude response when designing a highly constrained filter: Lowpass, IIR (N=10), Chebychev 2, Fc = 0.01, A_sb = 60 . This issue appears to be caused by the use of the transfer function polynomials in the calculation and I have been able to correct the group delay by changing the calculation to use zero-pole values instead of the polynomials (see code below). I did some testing and it looks like the magnitude response appears to be corrected if the filter is designed in the analogue domain however this affects subsequent frequency and group delay calculations and the coefficients need to be transformed to the digital domain.

'''
add to pyfda/plot_widgets/plot_tau_g.py (line 82) 
remove/comment out currently defined w and tau_g variables
'''
        zz = fb.fil[0]['zpk'][0]
        pp = fb.fil[0]['zpk'][1]

        G = np.zeros(rc.params['N_FFT'])
        w = np.arange(0,1, 1/rc.params['N_FFT']) * np.pi
        for i in range(zz.shape[0]):
            rz = np.abs(zz[i])
            az = np.angle(zz[i])
            G += (-rz*np.cos(w-az) + rz**2) / (rz**2 - 2*rz*np.cos(w-az) + 1)
        for i in range(pp.shape[0]):
            rp = np.abs(pp[i])
            ap = np.angle(pp[i])
            G -= (-rp*np.cos(w-ap) + rp**2) / (rp**2 - 2*rp*np.cos(w-ap) + 1)
        tau_g = G

How to reproduce the issue:

• run python -m pyfda.input_widgets.input_filter
  (Lowpass, FIR, Equiripple should be selected)
• select 'Minimum Order'
• select 'Bandpass' ... and:

379: self.fo = foList[0] # use first list entry from filterTree

TypeError: 'dict_keys' object does not support indexing

Here, a minimum order bandpass is not contained in the filter dict. Consequently, it should not be selectable at all. But it is and it produces the error above.

Each time the design method, response type or filter order (min/man) is changed, the specs for frequency, amplitude, filter order and target amp/frequency are updated, firing a local sigSpecsChanged signal in each subwidget.

These signals are connected to input_widgets.input_specs.sigSpecsChanged which is connected
to input_widgets.input_tab_widgets.sigSpecsChanged.

Finally, in pyfdax.py, the signal triggers pltTabWidgets.update_view, updating the subplots four times causing unneccessary delay.

ToDo: make creation of sigSpecsChanged in the subwidgets more intelligent (was the update caused by changing individual specs or by reloading all?)

Add new IIR non-causal elliptical filter. This creates an elliptic filter with the square root of the Amplitude specs, then later the code will square the filter coefficients to create a filter with zero phase that could be applied to signal data that has been stored to a file. Applying these filter coefficients is not done in time order, so the input signal data would need to have been stored.
Update displays to show correct Frequency response, phase response, poles/zeros, group delay and impulse response for the new elliptic filter.

Especially for complex P/Z and coefficients, the input fields become very wide. For low-resolution displays it would be nice to reduce the number of displayed widgets.

However, internally and when editing a field, the full resolution needs to be maintained.
This could be achieved by keeping a shadow copy.

• Display coefficients with the number of digits from pyfda_rc.params['FMT_ba'] and fixpoint formats
• Display P/Z with selectable number of digits from pyfda_rc.params['FMT_pz'] and formats

RuntimeError: the PyQt4.QtCore and PyQt5.QtCore modules both wrap the QObject class

This is a regression that had been fixed with PR #11 and was re-introduced by me while adding pyQt5 compatibility and hunting down another but.

As far as I understand the problem, matplotlib inherits from QObject which is ambiguous when both libraries are present on the system.

Solution: Specifiy the backend for matplotlib when it is imported for the first time:

    import matplotlib
    matplotlib.use("Qt5Agg")

Currently, it installs conf files into the python package directory:

	/usr/local/lib/python2.7/site-packages/pyfda/pyfda_log.conf
	/usr/local/lib/python2.7/site-packages/pyfda/pyfda_log_debug.conf

Then the app fails:

IOError: [Errno 13] Permission denied: '/usr/local/lib/python2.7/site-packages/pyfda/pyfda.log'

because this is a wrong location for the log file.

When creating a complex valued filter (e.g. by deleting a single pole), the zpk2sos algorithm throws an error message (probably also true for tf2sos).

This condition should be detected in pyfda_lib.fil_convert, and in this case fil[0]['sos'] should be set to 'None'.

File: pyfda/hdl_generation/hdl_specs.py

The file name / path entered in the file dialog in the methods createHDL() and simFixPoint() is not passed to the actual myhdl routines, hence all resulting files are created in the default directory.

When closing the application by closing the window or by clicking the "Quit" button, the python kernel crashes (Windows 7, 64 bit). I don't know whether the same happens under other OS's.

File: pyfda/hdl_generation/hdl_specs.py

Since integrating the SOS (second-order sections) from scipy.signal, the designed filter can be retrieved in SOS format from the central dict via

fb.fil[0]['sos']

Using this information to simulate and design fixpoint filters using myHDL would be a major step towards an universal fixpoint filter synthesis tool.

When the coefficient table is displayed for the first time, the columns are not labelled b and a but 1 and 2.

It would be very helpful to create a design report for a generated filter (PDF) containing relevant plots and tables.

A very promising approach is described in

https://towardsdatascience.com/creating-pdf-reports-with-python-pdfkit-and-jinja2-templates-64a89158fa2d

where HTML to PDF conversion is achieved with pdfkit, HTML generation is controlled with jinja2 (needs to be installed).

As an alterative, PDF could be generated directly using fpdf:
https://stackoverflow.com/questions/47195075/insert-base64-image-to-pdf-using-pyfpdf

When choosing e.g. npz file format, *.npz should be added automatically when the extension is missing - under Windows this works but not under Linux.

In some cases, default corner frequencies e.g. for pass- and stop band are identical. This means, something goes wrong between reading frequency specs from the global filter dict and displaying / storing them back to the filter dict. Could be related to issue #6.

Forgot to relabel the tablewidget header in input_pz._clear_table()

Display P/Z in various formats using itemDelegates similar to filter_coeffs.py

• Cartesian
• Polar
• SOS

Sort the P/Z according to:

• Frequency
• Magnitude
• Real part
• Imaginary part

More options in the editor

• Option for mirroring P/Z (w/ and without copying) at the UC or the x-axis
• Option for limiting P/Z to a selectable magnitude while maintaining the angle

Interactive Editing
When a dedicated signal-slot connection between the editor and the plot window has been established, it should be easy to:

• Select P/Z in the editor and highlight it in the plot window

Providing interactive options as explained in https://stackoverflow.com/questions/29277080/efficient-matplotlib-redrawing allows to

• Select P/Z in the plot window and highlight it in the editor
• Add / Delete P/Z in the plot window
• Drag & Drop P / Z in the plot window

The method PlotTabWidgets.update_view is called twice when filter / filter order are changed, triggered by the signal sigFiltChanged.

The first time is triggered by select_filter.py -> load_filter_order() (filter / filter order have been edited) and the second time by filter_specs.py -> update_UI.

Debug information should be presented more structured and detailed (input_widgets/input_info.py):

• One window for all debug info (filter tree, filter dict), selectable via combo box
• Select debug level from UI
• Display logging info
• Display versions of Python interpreter and imported modules
• Export debug info to text files

Shrink name of new elliptic zero phase filter in pullDown menu.
Improve name of file_io pullDown menu for custom text file output