YA Feature request, compensation automatization

[jschwender]

Dear John,
The "Add harmonic distortion" is a nice idea in the generator. It can be used to reduce distortion on a measurement signal, in case for a distortion measurement on an amplifier, for example. This is not simple to do as lot of settings have to be tuned in combination. There is also a technical limitation to the result:

As an example lets look to the 3rd harmonic (h2) here, which is uncompensated at -95.3 dB The compensation now should do a subtraction of exact the same value from the harmonic value. That means a subtraction of a integer value from the harmonic. If you step through Levels you get the following values for h3 residuals:
Level result
98dB -106.9
97dB -110.4
96dB -117.6
95dB -123.7 <-- minimum
94dB -111.1
93dB -105.7
As you can see the difference between the steps get larger the closer they are to the perfect compensation. In worst case the compensation Level is offset by 1/2 dB step, which results in a residual of -120 dB. If you are more lucky,and the hx has a level value of -xxx.00 dB, then the compensation with integer steps works perfectly and the hx actually disappears in the noise. This example is simplified, as the same applies also to the Phase quantization. In real life the harmonics on an amp output for example vary with both the output level and the load, so the whole compensation process is only stable for a single operating point. That means if you really want to use it: lot of manually tweaking…
Two suggestions to improve:
1. add one digit to the Level and the Phase input fields – that should be simple to implement, and would already help.
2. insert a button, which grabs default values from the RTA window and puts them into the corresponding "add harmonics" fields, with Phase-180°. That would make this specific application very very simple.
Dear John, i just hope you find these suggestions helpful enough to consider putting it in your (assumed) huge pile of other to-do's.
greetings

 

[John Mulcahy]

I'll put those on the list.

 

[jschwender]

Dear John, thank you, higly appreciated!

 

[lorrain]

All this sounds very interesting but I have difficulty to understand your example.
Could you give me more explanation.
I understand that you are trying to determine the distortion to place in the signal generator to have any more distortion at the output of your amplifier under test.
What is the benefit to do this?
Thanks.


.

 

[jschwender]

The purpose is to reduce the distortion, not increase. In plain theory you can create a distortion free signal on an sound card output, even though the sound card has distortion (= adds harmonic signal components). This can be archieved by subtracting the harmonic component from the generator signal. The limitation is, that this mostly works for a distinct signal level, load and frequency. It can be used for distortion measurement.

 

[lorrain]

Thank you very much for your answear.
It is clear.

 

[trobbins]

I don't recall getting similar large step level jumps in resultant %HD when tweaking a simple loopback on an EMU0404 over 2 years ago, but can support the request as a 'nice to have' enhancement to the nulling capability. The nulled HD levels I measured were just observable out of the noise floor when using 64k FFT length and 32 averages (this was before coherent averaging option was available from May 2020, so I may get keen to do this again), but were down at the 0.000014% 2HD level.

As the nulling process comprises HD contributions from the DAC and ADC paths of the soundcard (or soundcard + additional circuitry or device) in a loopback configuration, it can be worthwhile stepping the DAC level by say +/- 5-10dB and seeing if a resultant nulled level is higher/lower than the previous measurement, as a way to assessing whether the DAC or ADC contribute similar or significantly disproportionate HD contributions. Getting down in to the 0.0000x% HD levels needs some confidence that the DAC HD levels are actually in that range if the aim is to have an alternative (and cheap) signal generator to say using a Victor oscillator.

 

[jschwender]

In my example i could compensate second harmonic to the same (almost perfect) level (-137dBV) but for the third harmonic the minimum was reached at -121dBV, no matter what i tried to tune. If that compensation works down to -137dBV or not is just a matter of luck, depending on if the harmonic vector fits to the level/angle quantization or not. I would be surprised if you can archive better results with just any harmonics and any device you test. If you test that with more than just one harmonic you should see that effect with high probability. Therefore with finer quantization on the compensation control, the compensation will work for more than just the "lucky" harmonics. In my opinion this is what makes the implementation of harmonics compensation useful in the first place. Otherwise it works only partially, as said. If you calculate the maximum level jump that can occur: it is 25 dB for integer steps of the compensation amount, in case the phase matches perfectly.

 

[trobbins]

I just re-did the nulling for my EMU0404 and this time used coherent averaging and many averages, with rectangular window and 64k FFT at 1kHz nominal. 2HD% was trimmed to about 0.000015% and all the higher harmonics were trimmable to well under 0.00001%. There was a bit of jitter in the HD levels, and I did go up/down a tone magnitude mV step to find a better trim outcome. The coherent averaging certainly helped to better see each harmonic for trimming, but the higher harmonics still ended up in the noise floor and nearly at the 0.000001% minimum resolution of the distortion measurement box. But yes in my situation, a finer trim resolution on the added 2nd HD may allow it too to be pushed below 0.00001% (ie. lower than -140dB below fundamental).

If the aim is to be able to generate a very low HD sine generator output from a soundcard then there is still the disconnect of confirming what HD is generated from the DAC side versus the ADC side, and then having a method to trim the DAC side and to confirm it is appropriately trimmed.

 

[jschwender]

Interesting result. This is a valuable hint: I did not try to adjust tone magnitude. I just tried with a fixed given magnitude, but when i think about it: tuning the magnitude must have an influence on the subtraction result. Nevertheless, finer compensation steps would make it not necessary to both tune the tone magnitude in conjucntion with each HD magnitude and phase. My calculation suggests, that one digit more in magnitude is helpful, and for the phase it is likely not even necessary to add a digit, as the function for the residual is pretty flat.