Virtual balanced questions to John

[dlr]

I've been using REW for audio interface and amplifier distortion measurements and following threads at AudioScienceReview. One ADC specifically for hardware measurements, the E1DA ADCiso that I purchased, has an option to combine stereo inputs into mono that when used with a splitter to both inputs provides an additional 3db improvement in its inherent noise impact. That triggered my test of using a splitter into my PC for testing and using the REW Virtual Balanced option in Sum mode. This produced a 2.1dB noise improvement with better even-order harmonic component results. I suppose 1.2dB may be due to the two channels that are not perfectly matched.

Two, when using the Diff mode that is used for pseudo-balanced use, the noise is then calculated against the difference which makes the distortion components very high in comparison. I don't see a benefit for distortion testing in Diff mode, but if that diff magnitude could then be graphed as the fundamental in some way it would be useful for measuring channel imbalance. It may not be worth the effort, that I understand. The HD components are displayed with the fundamental being used for normalization of the HD components, but having for the Diff mode it would be another piece of data for analysis.

My results can be found here.

Edit: Question 2 was spur-of-the-moment. It's not valid for the RTA, but would be for the S-THD or a normal measurement.

 

[dlr]

I ran standard measurements to test the Virtual Balanced mode with the splitter to two channels as above. Overlays are single channel, two channel in Sum mode and two channels in Diff mode. The REW Levels display showed about 0.02dBFS difference with a 1kHz tone that was about 1mV.

I'm not sure how to interpret the results other than the two channels start to vary below about 2kHz.

 

[dlr]

This is comparison of single-channel vs dual-channel (Virtual Balanced Sum mode) multitone results. The noise floor shows improvement from about 3kHz on down. This brings me back to my question 1 above, can the dual-channel setup be considered valid? It's accepted for hardware done this way (e.g. E1DA ADCiso), so why not when done in software?

 

[phofman]

I played with the virtual symmetrical mode not in REW, but directly in linux audio chain https://www.diyaudio.com/community/...om-regular-soundcard-in-linux-results.326747/ (the principle is identical). The results were good, but it required calibrating the channels for same gain prior to measuring. I am not sure that is done in REW. Even a minimal gain difference (which is almost always the case for a 2ch HW without software adjustment) ruins the difference effect.

 

[dlr]

I played with the virtual symmetrical mode not in REW, but directly in linux audio chain https://www.diyaudio.com/community/...om-regular-soundcard-in-linux-results.326747/ (the principle is identical). The results were good, but it required calibrating the channels for same gain prior to measuring. I am not sure that is done in REW. Even a minimal gain difference (which is almost always the case for a 2ch HW without software adjustment) ruins the difference effect.

Thanks for that info. I had thought that might be an issue. The two channels are close, but that's probably not good enough. My idea for testing the Scarlett 2i2 is to insert a 10-turn pot on the 2i2 input as I did with the 1010LT for more precise settings, but I think I'll also use them on the 1010LT input (since it's still set up) to adjust the input on one balanced input channel to try to precisely match the two channels to test the impact of the voltage delta between channels. Another issue will likely be that the delta may change with voltage level, so it's probably not a solid scheme. Hardware such as the ADCiso can provide the precision necessary I suppose. Still, I'm doing this more as an academic exercise at this point, though I had hoped for better.

Edit: I started to read your diyaudio thread. I'm using the 1010LT balanced input now, but I may put the 10-turn pot on an unbalanced input to match another unbalanced one to see what happens. It's not ideal, but this is to test the scheme to see the results of hardware matching vs. your test in software. There will still be software (REW) involved, but that will be for inverting one input channel. REW level display will provide the channel data.

 

[phofman]

IMO much easier (and technically viable) is to precisely fine-tune the gain in the digital domain. It's very unlikely a HW would surpass the precision the digital gain offers. IMO ADCiso has no advantage in precise gain alignment - it has input amplifiers just like any other ADC.

10-turn pots - the inexpensive chinese ones introduce minor distortion, the quality ones are quite expensive. I use them in my project for digital distortion compensation https://www.diyaudio.com/community/...ion-for-measurement-setup.328871/post-6187405

 

[John Mulcahy]

The results were good, but it required calibrating the channels for same gain prior to measuring. I am not sure that is done in REW.

REW uses the input full scale voltage calibration data (see Calibrating the input level) to match the channels.

 

[dlr]

REW uses the input full scale voltage calibration data (see Calibrating the input level) to match the channels.

Thanks for pointing this out. I calibrated two channels and ran quite a few tests. In doing that I found that the control panel for the Delta 1010LT ASIO that I've been using is active when using REW. I had thought it to be bypassed. The gain can be changed per channel as well as inverting the signal of each. I calibrated the channels again after setting the gain and input levels for the latest tests.

Long story short, using the virtual balanced feature gave distortion results almost identical to single channel results. This surprised me. Prior to calibration there was significant improvement, but this was at different gain settings, so part of that problem was, I think, more gain of a lower input signal.

After the calibration I ran the Scope feature in REW. The Ch1-Ch2 shows the difference after calibration. One issue may be that my DVM may not have enough resolution for a good calibration. Another may just be internal audio card noise. The Scope Ch1-Ch2 result is shown below. It shows a fair amount of noise in the diff. No way to know how much is imprecise calibration and/or input channel differences.

I plan to test the Scarlett 2i2 for this as well, but I'm becoming more skeptical of being able to use modestly priced audio interfaces for amp testing. I hope that the 2i2 results are more consistent. At least most settings of it are in the digital domain with channel gain more precisely controlled.

In any case, thanks for the calibration link.

Note that theCh1/Ch2 scales are 1V/div whereas the Ch1-Ch2 scale is 1mV/div.

 

[John Mulcahy]

Looks like the channels have a fractional sample timing offset.

 

[phofman]

Maybe it's like this:

sin(x) - sin(x-0.001) = 5.×10^-7 sin(x) + 0.001 cos(x)

IIUC that 1mV subtracted signal is cos(x), i.e. phase shifted by pi/2.

The channels are slightly phase shifted against each other at that frequency - to be expected for real-world HW, they cannot have identical complex transfer function.

 

[phofman]

Long story short, using the virtual balanced feature gave distortion results almost identical to single channel results.

IIUC the subtraction can suppress only common-mode artefacts in both channels which is not harmonic distortion.

 

[dlr]

Looks like the channels have a fractional sample timing offset.

It would have to be extremely small. Both channels overlay perfectly for as much as I can tell with decreasing scale. I may be that it's not apparent in the channel overlay, but shows up as the noise in the sum. Of course any timing of sampling is built into the ASIO provided by Delta. I have no experience whatsoever with how the ASIO does the sampling.

I just ran another test using the Scope in REW. Channel 2 was inverted using the Delta control panel. The scope was set for Ch1+Ch2. I also decreased the scale for both channels. As far as I could tell there was perfect overlay in the first graph above with perfect symmetry here, but minute differences may not be observable. Despite that, here's the result. Note that the horizontal scale is a bit more spread here.

For ease of comparison, here's the diff version with scales as above.

 

[phofman]

It's always perfect cosine to the fundamental sine, IMO just a different phase shift between the channels as that simple equation yields.

 

[dlr]

It's always perfect cosine to the fundamental sine, IMO just a different phase shift between the channels as that simple equation yields.

I had noticed the phase shift, thought it must have been how REW processed the two channels, but it did seem odd to be 90 degrees. Interesting to see your equation for that.

My earlier comments relating to calibration has to do with the noise present in the sum/diff responses. That may be primarily due to poor calibration due to the DVM precision not being good enough. It only goes to two decimal places. Or it could be some drift in the input channels as well. I doubt I'll be able to improve on it in any case.

Prior to calibrating the channels the virtual balanced mode had made improvements in all aspects, close to a 3db improvement in noise and much better HD components which was evidently due to lack of calibration. After calibration there is essentially no improvement in noise nor HD which is disappointing since when done by hardware in the ADCiso that the developer has built into it there is improvement. I hoped for similar results in REW, but that limits of calibration and, I suppose, non-linearities between channels, the single channel mode is the only valid method otherwise. phofman, I understand your point about CMRR, but the ADCiso does have improved THD in its two-channel mono mode. I have the balanced splitter on hand because of that, I have the ADCiso Grade A.

As an aside after I realized that the Delta control panel was still in the loop I experimented with different gain settings and found a much better "sweet spot", very low gain with high signal input via the amp feedback probe into the balanced pot (Monitor1).

One more observation that has an influence on the results reported here I believe. I really need to make the single channel balanced attenuator using two 10-turn pots. Setting the Monitor1 that has the typical 4-pot attenuator (stereo) is problematic. Setting precise levels means just slight pressure one way or the other when close. Since each balanced channel is actually two potentiometers for P/N changes, I noticed that pressure one way provides slightly different changes than pressure the other way. Any change in the ratio of P to N (that should be constant) means a change in distortion results. As has been pointed out at ASR, a balanced passive attenuator should be high quality. The Monitor1 doesn't provide that for optimal distortion testing. Digital control would be ideal of course, but that adds expense.

 

[phofman]

I understand your point about CMRR, but the ADCiso does have improved THD in its two-channel mono mode.

But that's not related to some differential mode, but simply averaging the vector distortions of the two channels (each a bit different) improves the resultant distortion, statistically.