Amplifier distortion testing using a modestly priced audio interface

[DavidR]

The ADCiso discussion about cascaded inputs prompted me to experiment with the REW Virtual Balanced Input option. It was dismissed in comments for typical acoustic measurements, but I thought that a split input to two channels might somehow improve distortion measurements. I made a number of tests with the 1010LT/V3Mono setup I had in place and connected a balance splitter that I have to provide the same input to the two balanced input channels of the 1010LT. Output was 1010LT unbalanced to the V3 unbalanced input with the probe return to Monitor1/1010LT balanced input. The results are interesting to say the least.

REW reported the noise level improved, but only 2.1-2.2dBFS. No matter the input level. Every time. I had thought it would be 3.0dBFS. There were distinct improvements in some individual harmonic distortion components, most noticeable in HD2.

REW reported the channel difference to be 0.02dBFS higher on 1010LT that approximately 1.0mV. I'll add some graphs soon.

These are the classic RTA 32 average a single channel (left) beside the two channel (right) response.

For easier comparison this is the overlay of 2 channels (black) on 1 channel(red).

I also ran 100 and 1000 average tests. This is an overlay of 2 channel (black) on single channel (red) for a classic 1000 average.

I also ran 100 and 1000 average tests. This is an overlay of 2 channel (black) on single channel (red) for a coherent 1000 average.

My thoughts are that it is advantageous to have a probe return to two channels through a splitter. Given that there is a measured difference by REW between channels it may be that being able to adjust one independently such that the two channels match in level could provide even better distortion results. I'm speculating, I'll have to find a way to do determine that experimentally.

I forgot to mention above that I used the REW Virtual Balanced option in Sum mode. REW sums the two channels, then divides by 0.5 according to the docs.

I plan to do the same testing using the Scarlett 2i2 Gen 4.

 

[DavidR]

Each time I thought I'd exhausted all means of improving results another one (or two) is presented. I had to go back to doing loopback testing for the 1010LT. This is for unbalanced output to balanced input. I've yet to move to the 2i2. The REW author pointed out that individual channels can be calibrated. The distortion results improved with more consistency from run to run after calibration. It also improved the results when using the Virtual Balanced option in REW. At this point I'm convinced that channel calibration makes use of Virtual Balanced inputs (in Sum mode) a valid option. I plan to always use it at this point. Noise level is improved 2.2-2.6 dBFS in every test of made. HD2 improved as well.

Again I thought I'd exhausted all options. Then in reading through older threads there was a discussion on added distortion. The REW author showed the option for added distortion. This opened a another avenue for improvement that for my case was significant, primarily in one aspect. Again, noise level. But the reason is quite different. Up until now I've had to set the REW output level at -15dBFS or lower to get the best distortion results. It was always a balancing act of the distortion components. Some would go up, some down with an output level change, input as well, but I have settled on that though I may now experiment more with it. Use of added distortion allows me set the REW output to near maximum, -0.60dBFS. Above that the distortion is significantly worsened. This resulted in significant improvement in the measured noise level as would be expected.

Once again all options were exhausted. Or so I thought. I found that the Delta Control Panel is not totally bypassed. The Master does not change output level, but the Output tab channel sliders do. The 1010LT is like the 2i2 in that regard, the latter having digital output control through its control panel and being best with some gain. I had done all testing with the gain (attenuation) at default (0db). Experimenting with it I settled on ~ -1.8dB (the slider resolution is rough). This allowed me to set the REW output at maximum, -0.01dBFS (0.0 is unusable). This allows use of the lowest noise level, though there wasn't much change in the HD components. For reference, much of my earlier testing before these changes needed to be where noise was in the -70dBFS range. It's better than -93 now. I will be testing the 2i2 for this later.

One other comment before I post graphs. I had been using 256kFFT up to now. Testing showed that the distortion results changed very little with a 32kFFT, so I've settled on that for all future testing (much faster). It also is what Amir uses, although he uses only a four average where I prefer to use 100 because results of multiple runs almost always converge whereas even a 32 average often has a lot of fluctuations of HD component levels.

These graphs are for the 1010LT loopback, unbalanced output to balanced input, a PCI card designed in 1998 (I think), though mine are stamped 2001. Bought off ebay, no cap upgrades.

First this shows the distortion for -1.0dBFS that is very poor for HD2 and HD3, typical of the upper range. The first five graphs have a 100 average.

This is the result of setting the added harmonics using the REW built-in control "Set for RTA". This does compensate for DAC and ADC, but the 1010LT spec indicates that the THD for output and input are identical <0.002% at 0dBFS. Seems safe to assume that both contribute roughly equally, at least if the levels are the same. In my case, the input dBFS is much lower than the output.

One thing to keep in mind is that this must be reset for any changes in the output. I would say that it should be done if there is a change in the temperature of the audio interface. Warmup is important and in my opinion it takes more than a few minutes.

Next shows the results for -3.0dBFS. Note that significant change in noise with this relatively small change. Distortion components are both up and down.

This also shows the result of being able to set the output maximum, -0.01dBFS, by having the Delta control panel setting with some attenuation. There isn't a real improvement of the HD component levels, but there are some differences, primarily in the noise level that is not specified for the 1010LT. Another is that the HHD value is much worse that can be attributed to the additional output where the HD components above HD9 are not compensated by added distortion. This also is for the combined DAC/ADC, so I'm guessing that each side is around -95dB at 0dBFS. Not bad for a card about 25 years old.

For comparison this is the classic average 100. With added harmonics the HD components are buried in the noise floor. This is why most measurements were made using coherent averaging to push the noise floor down.

Last are graphs made with settings similar to what Amir uses, 32kFFT with a four average. I assume he's using classic as well, not coherent averaging.

Classic averaging.

Coherent averaging.

My opinion (subject to change) at this point (for REW) is:

• Calibrate all used channels
• Use a splitter to input the same signal to two input channels
• Use the REW option in preferences for Virtual Balanced Input set for Sum to improve the noise measurement
• A 32kFFT is adequate
• If the HD components are near or in the noise floor, use coherent averaging to push the noise floor lower
• Use a 100 (or more if desired) average for most accurate results
• Use REW added distortions and if the card control panel allows it, adjust the gain/attenuation to allow for close to 0dBFS output
• Always reset the added distortion values after any changes in the output

 

[DavidR]

I made the balanced attenuator from two 10-turn pots and made a lot of tests using the 2i2 in loopback. But first, I want to point out that my earlier problems with repeatability were likely due to poor quality cables and adaptors. This surprised me being balanced, but the problem was physical. I had a pair of 0.5m TRS/XLS cables and a pair of TRS/XLS adaptors rather than buy more dedicated cables. These didn't look like poor quality, but were inexpensive. In the end the problems were with the TRS plugs. The tips eventually rotated causing poor connection internally or even breaking a wire off. The tip of one XLR-to-TRS adaptor actually detached inside the 2i2 socket. Fortunately I was able to extract it by opening the 2i2. The socket used is not a totally sealed part.

Only name brand parts for me going forward. At least my other long cables were all good. Next I'll be posting some results of the use of the balanced attenuator.

 

[DavidR]

First up are results of the 2i2 without the attenuator in the loop. This was to set a baseline. If the desire is to examine the HD components only, this is fairly satisfactory, although the HD component values aren't quite as low as will be seen later. This one is also only for a selected "sweet spot", though this can't be called definitive. For this I used a balanced splitter cable and with/without added harmonic distortion. I ran many tests of various 2i2 loop settings, but this is probably the best representative. Using the REW added harmonic option dramatically lowers the measured distortion into the noise floor, but this is combined DAC/ADC results. More on that later.

The key takeaway here is that the noise is dominant for almost any reasonable loopback level for the basic 2i2. It won't be adequate for testing other components for SINAD, but is probably adequate for many/most amps other than those of highest quality (i.e. best HD and noise).

 

[DavidR]

I made a balanced attenuator from two 10-turn pots. I was disappointed when I finally started because it was past a return date to Amazon. I ordered two of a single item listed and expected to get two identical pots. However, despite both being stamped as Bourns it was obvious that they were of two different manufactures. The total resistances were a bit different and the minimum resistance of the output as well. But since this was more proof-of-concept it seemed just as well to have mis-matched pots. Even for matched, their never identical so the difference became moot. The fact that it works makes it even better to have that mismatch in the proof.

I have a spool of Mogami Star-Quad cable, so I used that to wire the internals of the balanced attenuator. I'd use a different box next time, it's not easy trying to work with that in short lengths due to stiffness. My soldering also was not ideal (especially my first attempt had a mistake). In the end it works well. Initially I left it open and tapped the pots with the scope probes internal. Later I closed it and used a balanced splitter for easy access with no change in distortion measurements.

This is the test rig setup.

I learned that the Rigol can be connected to a network so along with the laptop on my local network I used TightVNC Viewer and worked from my main PC much of the time.

 

[DavidR]

One surprise was that using a splitter into the 2i2 that worked so well for the unbalanced tests made essentially no difference. In fact, it was detrimental to the HD results, so from then on I dispensed with that. I can only think that it has to do with the balanced nature.

What prompted me to investigate this scheme was when I read some info about distortion and what causes what. Specifically, the detail that even-order harmonic distortion is affected by the symmetry of AC voltage. That is, if the positive and negative portions of the cycle are not perfect mirror-image (so-to-speak), then it introduces even-order distortion. Other factors affect odd-order, but I confirmed in my testing that the primary impact is on HD2, though HD3 was affected to a degree while others are affected even less. When making changes HD3 settled rather quickly while HD2 was extremely sensitive to any imbalance in P/N symmetry.

I initially intended only to measure voltages of P and N, but quickly realized that would be more difficult and slow so I connected the Rigol. This made it a lot easier to get the two pot settings close to whatever voltage output was needed since they both must be changed and ultimately need to be nearly identical in output. What I found was that I could get close to the desired voltage using the Levels display in REW, set the P-side to half the desired voltage via DVM and then the N-side could be tweaked. I subsequently confirmed final voltage with a DVM, but even more interesting is that once set that way the N-side setting was not necessarily ideal. I found that by watching the REW RTA HD2 display in continuous single-tone distortion mode I could adjust the N-side for minimum HD2. When HD2 is minimum the desired voltage is correct. It's extremely sensitive when close to minimum so small changes in the pot change HD2 a lot. Subsequent DVM measurement always confirmed the exact full voltage. That is, the P/N ratio was nearly perfect, mirror image if you will. I tried using the Rigol display, but it doesn't have the visual resolution plus the DVM option in it is based on percentages in the display, so a DVM is required. I'll put resulting REW graphs in subsequent posts.

 

[DavidR]

REW RTA graphs using the balanced dual 10-turn pot attenuator. There appears to be (at least) two benefits to using such an attenuator. First is setting the P/N ratio such that harmonic distortion is at the lowest point possible in a loopback. I also hope that it carries over to amp testing that I will run on the V3 Mono. Second is being able to set the voltage output on the source to a much higher level that is in the better range for noise of the source component. That will become obvious in the graphs to follow.

I've distilled the list of graphs down to these few (I made a LOT of tests). This one seemed like the likely best one judging by a stepped THD. The 2i2 output dial was set at maximum (as were all tests made using the 2i2). The input gain was set to 0dB as one test. HD components are very good, but noise dominates so SINAD is useless. This run set the REW generator to provide 2i2 output of 4.0V the connected to the attenuator that was then set to provide 0.422V for loopback. That setting is what is required for the V3 Mono @5W into 4 ohms to be tested later. The note about using the oscilloscope was wrong, it required a DVM.

I thought that the next one might be the best overall. The 2i2 gain was set to-12dB that I had found to be the best compromise for low ADC noise. I thought this would be best because the 2i2 raw loopback is best at around -20dBFS that provided about -15dBFS input in raw loopback testing. HD components were better as was HHD, but noise remained the limitation.

I wanted to see just how high the 2i2 output could be used. This one was similar to the above, but the generator was set to -0.01dBFS 4.94V. Close to 0dBFS would trip the REW distortion shutoff. The attenuator was again set to provide 0.422V. This one provided the best HHD content. The note is wrong, it's only the added distortion graph. I began noting the attenuator 10TPot settings (range from 0.00-10.00). This one provided the best HHD, but the noise still was the limitation, although higher any testing I recall having made with the Monitor1 in the loopback.

Next I attempted to find the settings that provided the maximum SINAD possible in this scheme. For this one I assumed, from examining the stepped THD, that around -6.0dBFS would be optimal. It was definitely an improvement. At this point the HD components were becoming much worse, but largely corrected using the REW added harmonic distortion (HD) option. The non-HD is shown for the numbers with the overlay of non-HD vs added HD beside it. The noise, thus SINAD, is definitely better.

Finally we have the best result. I was really surprised at how high the 2i2 could be used. In raw form, it can't. But using coherent averaging along with added harmonic distortion the results are excellent. Here REW was again set to -0.01dBFS 4.94V, 2i2 gain again 12dB, but with the 10TPot attenuator manually set to provide near maximum input to the 2i2 ADC, that being -0.13dBFS. You can't get much higher in both output and input for the 2i2, yet with added HD the noise is now 105dB. The attenuator was approaching its limit at the low end. Small changes were having a big impact.

Here is the RTA with no added harmonic distortion. The HD components were getting seriously poor.

With added harmonic distortion overlaid.

HHD is definitely worse, but that's to be expected because the REW added HD is only applied to HD 2-9. This looks to be good enough to measure an amp such as the V3 Mono for distortion an is approaching the ability to measure SINAD. Amir's V3 Mono measurement showed about 100dB for SINAD. This is for RTA of course. This can't compete with the myriad other tests the the AP runs, but for basic distortion and SINAD other than the best amps, I found this to be a surprising result. This is using a 2i2 Gen 4.

I plan to go back to trying the 2i2/ADCiso loop to see how that compares. But I think that having the same clock for both DAC and ADC may be a distinct advantage. After that I'll test the V3 Mono gain using this setup.

Edit: I ran a test using of the 2i2/ADCiso, had thought that may have done something wrong the first time I tested that combination. Even using the 10V (highest impedance) setting the 2i2 can't output its maximum voltage, so they can't be used together. I had hoped to be able to set the added harmonics for the 2i2 DAC side using it with the superb ADCiso that would allow me to use added harmonics that were almost exclusive of the ADC side, but the 2i2 is not adequate for that. The max output was down approximately 0.5V due to the low ADC impedance.