Fix Vectorscope

[KinMaynard]

Vectorscope not functioning as intended. Run test data through external vectorscope & verify current performance.

[KinMaynard]

Okay so I now have some control data to use to better understand what is wrong with the vectorscope. I've created images comparing the expected vectorscope output to the current output. Here are the findings:

Mono
All the mono test data is displaying mostly correctly as straight lines down the middle. The silence is displaying as a line of dots down the middle presumably because of some dithering in the file (not actual silence) which is to be expected. The only thing that is perhaps different than expected is that the bottom and top of the lines are not at the actual bottom and top of the axis like they are in the control data. For some reason the limits are shrunk on both ends.

Stereo
Silence: No dots on the bottom line of the axis like in the control. Otherwise similar enough.

White Noise: Unique enough to warrant an image. It looks like there are limits in the current plot that create a square edge. There seem to be limits just up until the edge of the curved polar axis. The data doesn't seem to fully populate the plot like in the control. There seems to be a density of dots however like in the control and it is possible that if the control were cumulative like the current plot it would look fully dense like the current output.

Chirp: Looks like a straight line down the center (must be a mono file). Has the same difference of a limit not reaching the edge of the polar plot in the current plot.

Sin: It looks like the current plot is both a mirror of the control & it extents up at the top creating the top edge of what looks like a square (similar to the top edge limit in the noise plot).

Lopez: Similar issue with the current plot having constraints inside the full extent of the polar plot. Creates a square limit similar to the sin & white noise plots. Looks dense & has left and right dots like the control though.

So overall there seems to be some kind of issue with an axis limit creating a square limit for the data instead of letting it fully populate the plot. There also seems to be a possible mirroring behavior happening. These both may be due to how I reflected half the data across the x axis so all of it would appear on a 180 degree polar plot instead of a 360 which may have been a mistake. Let's try undoing that and see what it looks like.

[KinMaynard]

Okay looking around the code and running some tests reveals a little.

Relevant code:

# take absolute value of the array to flip all data into 180 degrees of polar plot
absarray = np.absolute(array)

# converting cartesian coordinates to polar
r = np.sqrt(np.sum(np.square(array), axis=1))
theta = np.arctan2(absarray[:,0], array[:,1])

So when I run this instead just plotting the array instead of taking the absolute value:

theta = np.arctan2(array[:,0], array[:,1])

and disable this to show the whole polar plot instead of a wedge.

ax.set_thetamax(180)

This is the output:

This suggests to me that the way I am calculating the conversion to polar from Cartesian creates this box that is visible here.

[KinMaynard]

Okay so looking at the conversion it looks correct. However I've neglected to mention that I am converting the audio to mid/side before plotting mid on Y and side on X. So I'm going to try disabling conversion because it is probably creating these constraints.

This did not fix the issue. The mid side does make sense. I'm going to enable axis labels to see where this square limit behavior terminates at.

So this reveals that one of the issues is simply that the outer gridline is above 0 dBFS. The mono plots terminating earlier than the control was not true, the plot just has an axis above 0dBFS. So the first thing to take care of is to lower the outer grid to 1.

[KinMaynard]

I've lowered the axis limit to 1.0. This makes most plots appear similar to the control in terms of terminating at the same place. However it does look like the control plots limits are calculated based on the size of the data so you get the dithering in the silent files displayed as well as the 0dBFS signals too. By setting a hard limit the dithering is no longer visible in the white noise files and a lot of the files are not 0dBFS so they terminate -3 db or otherwise. It could be an option to calculate the axis limit based on the size of the input.

I'd like to lower the dot size and see what the white noise stereo plot looks like. Here is the plot with an s value of .001 and no absolute value of the array:

• So let's work on setting the r limit based on the data. I'll want to find the max value of the array and set the r limit to that.

[KinMaynard]

So this is the stereo sin wave in the current build. It looks like it extends above the control and also like it is on the opposite side.

[KinMaynard]

So it almost looks like I haven't actually converted my Cartesian coordinates into polar. Like its plotting a Cartesian plot on a polar plot just tilted 45 degrees. I know I originally imagined the mid and side values as coordinates on a Cartesian plane with side amplitudes being the X coordinate and mid amplitudes being the Y coordinate. I tried to convert those to polar but since the r values aren't reaching the polar edge and seem bounded by a Cartesian plane I think I need to reevaluate how I'm converting my r values.

[KinMaynard]

Okay getting lost here, I want to do a sanity check on my conversion from Cartesian coordinates to polar. I'm going to find a reproduce a simple polar plot from Cartesian coordinates with my current conversion and see what the results are. This should show me if it's my conversion process that is causing the issues or if it is elsewhere.

[KinMaynard]

Wonderful news! By running the sanity check I've found that the numpy.arctan2 function outputs radians not degrees. So by adding a numpy function for converting radians to degrees like so:

theta = np.degrees(np.arctan2(array[:,0], array[:,1]))

# instead of 
theta = np.arctan2(absarray[:,0], array[:,1])

We get these outputs as a result:
Mono
Silence

White Noise

Chirp

Sin

Sweep

Stereo
Silence

White Noise

Chirp

Sin

Full song

These tell an interesting story. The first thing to remark is that all the mono plots seem tilted to the left by ~25 degrees. The stereo white noise plot is looking good. There no longer seems to be a square shaped limit on the data. However in the full lopez song output we see that it looks like noise (which might actually be accurate to how a cumulative plot of it would be in the control without the dots fading out). The stereo sin plot also seems to rotate fully around the axis before being settled on a ~ 38 degree tilt to the right.

So this leaves us with a few issues:

• Tilting plots (~25 degrees left of center, one ~38 degree off 0 degrees)
• stereo sin wave plot high amp values spin around

[KinMaynard]

I think another sanity check is in order with this new numpy conversion. Let's see if anything must be done (I have a hunch that even though arctan2 care about quadrants that np.degrees doesn't?).

[KinMaynard]

So a lot has changed. From comparing lissajous and polar vectorscopes there seems to be something simpler going on. I think that the mid side encoding is unnecessary and I think that swapping the signs of the bottom half of the lissajous vectorscope transforms it into the top half then something needs to happen to give it a circular upper bound (multiply by pi?).

• So I want to take in a stereo or a mono file as input (if mono double the mono file into a 2 channel array * .5).
• Split the file into left and right plotting them as x and y on a cartesian plane
• rotate the plane by 45 degrees counterclockwise
• take all negative values in the left channel and put them in the right channel as positive values
• take all negative values in the right channel and put them into the left as positive values
• multiply all values by pi gives theta
• plotting the values gives r
• try plot

[KinMaynard]

Okay I'm going to build a lissajous vectorscope first before fixing the dot per sample polar vectorscope since they seem to be closer in function that I originally thought. I've built a small prototype and it needs some adjustments:

• Tilt plot 45 degrees
• tilt data 45 degrees
• disable axis ticks
• add labels for left and right
• add labels for negative left and right
• label color same as other plots
• spines same color as other plots
• title same formatting as other plots
• change from scatter to plot (faster)

[KinMaynard]

Okay so in trying to rotate the plot, axes or spines & data 45 degrees I've come across lots of roadblocks. It seems that one way to do this I've found involved using a floating axes. I tried using this but it broke my ability to change the axes parameters (color etc). I finally found out that this is due to floating axes being a subclass of the 'axisartist toolkit' which has it's own API that works differently than a typical axes. So now I need to redo the original checklist with this floating axis api and see if there are any more roadblocks. If this doesn't work for some reason, the next best thing seems to be to only rotate the data and not the spines.

• Tilt plot 45 degrees
• tilt data 45 degrees
• disable axis ticks
• add labels for left and right
• add labels for negative left and right
• label color same as other plots
• spines same color as other plots
• title same formatting as other plots
• change from scatter to plot (faster)
• coordinate plane spines

Also I need the test data on my other computer as well.

[KinMaynard]

So now I want to make the lissajous vectorscope part of the polar vectorscope function and have a radio button that switches between them.

• Add radio button for plot type
• add lissajous to vectorscope
• test alone & in visualizer