Distortion Surface: old measurement, new approach

My example is deliberately over-smooth.

 

Ok see, now I want a program that lets me map the individual harmonic surfaces, overlays them all on a glorious 3d plot, then make it interactive and let me put in coefficients for each harmonic. I want multi-visualizations to really draw out potential patterns.

 

Near 3000 Hz, I think you mean

Interpolation. The input data are sparse - see @rfernand 's table, there are 4 5 data columns: 50, 100, 500, 1000 & 5000 Hz - so different contouring routines used by different programs might be expected to produce different results (and some might have options for contouring algorithm or degree of smoothing). Here's yet another example (yes that's a log scale along the bottom, but I couldn't figure out how to make this program number it correctly):



The low-distortion area in @atomicbob 's 3F graph is between columns of data points. Obviously to be comparable, graphs for different amps produced from these sorts of data distributions should use the same contouring parameters.

 

It's just the depth of the colormap. Bob is using a more coarse map (with ~20 levels), while the others are probably using smoother maps (256?).

 

Yes...

I get the issue with interpolation/color mapping, etc. But colormap on AtomicBob's matched the 3d plot/stealth fighter plot. The other colormaps do not appear to match. If the data input was enough to create the stealth fighter shape in 3d, then I think it would be enough data to generate a similar looking profile.

I guess that's my point. If the graphs use the same data set, then the hope would be that they generate similar graphs... or yet we come up with yet another way for other sites to fudge/tailor to results to their "benefactors".

 

No, it's more than that. The 'hole' centred around -5 dBu & 3kHz in Bob's 3F graph is down in the high -90s dB distortion because a trough has been interpolated into that area from the high -90s data points at the top of the 5k column. But in the @rfernand & @earnmyturns graphs, and mine, the same areas are ~10dB less -ve.

 

Many bad decisions on where or whether to drill expensive oil wells have been made on interesting contouring of sparse data by promoters keen to sell their prospects. I'm happy to be retired

 

Combine with Gedlee Distortion?

https://www.superbestaudiofriends.o...-distortion-non-coherent-distortion-ncd.6391/

 

Contours, they make the "hole" more evident:

 

Huh huh huh… hole… heh heh heh

 

I am delighted to see so many exploring the possibilities this new approach has to offer.

Taking the plunge and updating my PSIPlot from v8 to v10.5 I find the addition of macros, templates, projects, etc. All of which facilitate automation and less work from input to outputs desired.

Some thoughts on why the data capture was so time consuming as a manual process:

First step was to run a sweep at a particular frequency. Then select each trace individually and turn off the trace marks which are distracting to the analysis at hand. And again select each trace individually for export.


Upon selection of a trace press the button to display a list of values for the trace. Then select export to CSV. Note the columns this method will export.


Now manually enter a descriptive filename for the CSV data to reside in. Repeat for each trace on the sweep.
Then sweep for each frequency. Many manual operations.

dScope automation has multiple methods. I use a hybrid of measurement setups used in an auto-sequence and the built-in VBScript language which can take care of housekeeping at the end of a measurement sweep:

With this combination approach I have been able to export only the data I wish, into columns desired and reduce many manual steps. Now the data capture is about 30 minutes instead of several hours with great reduction in potential errors from operator manual filename entries. There still is a need for a shell script to concatenate the data output and then copy results to a directory associated with the component being measured.

Next is import into my plotting program using a template followed by a macro to generate the desired outputs.

Last step is into a picture manipulation editor for watermarking and any desired annotation.

 

(This might be better in the distortion thread, but...): For history buffs, the source for this that's usually quoted is

DEL Shorter (1950) The influence of high-order products on non-linear distortion: Electronic Engineering 22, p. 152-153.

I haven't been able to find this online, but a preliminary report available here which also gives the formula is

DEL Shorter (1949) An investigation into non-linear distortion - first interim report: BBC Research Department Report 1949/30, 7pp.

Some great quotes in this one, e.g.: "... cases arise in which the non-linear distortion observed subjectively is not proportional to that expected from the customary measurements."; "It has long been recognized that high-order harmonics are more offensive than low ..."; and the one in my sig. There's also reference to a 1937 Radio Manufacturers Association publication noting the same effects and proposing a (different) weighting factor.

Whatever sort of correction might be found most appropriate, whether it be a BBC-like weighting or something like @atomicbob 's crap factor, it's interesting that the observed-vs-measured (THD) disconnect has been talked about for so long yet apparently remains unknown to certain "Audio Science" experts.

 

Just a thought, - maps for harmonic distortions may exhibit correlation between shapes and "second order" sound characteristics ("airy", "smooth", "open" etc)

 

Great update! When you have the intended exported data format, please post it here, and I'll try to package and share my Python script to read it and generate graphs like the ones I shared above.

 

Here are the current formats with about 90% of the process automated. Still more manual steps in this latest development than I would like. Have been splitting my time between this and creating All-In-One measurement suite for iFi Diablo which has some unique challenges combing DAC and Amp into a single measurement set with multiple 0 dBFS output levels depending on mode.

The contour map Legend will range all the way to Red if I obtain data above -48 dBu or create a fake data set to generate the Legend and then make a template from that. On the todo list.

Frequency data sets are acquired using a 1-2-5 sequence, which was always the plan. However given the entirely manual process at the beginning I chose to leave out 200 and 2000 Hz.
Data set frequencies: 50, 100, 200, 500, 1000, 2000, 5000 Hz.

iFi Diablo at 300R in ECO mode, balanced output.

 

DSHA-3F 300R load:

 

For those wishing to play with visualizations I've attached the DSHA-3F sweep data sets.

 

Limited to 16 colors this time:



And only the relevant region:


Really appreciate the culture of sharing data. Offer stands to set up a (private) GitHub repo for data and code sharing in the spirit of reproducibility.