Amplifier distortion testing using a modestly priced audio interface

[phofman]

However, I never got the 2i2 to work with the 1010LT output. FlexASIO kept failing. The ADCiso was fine, however

ASIO was not designed for merging two independently-running soundcards. The ASIO bridges which allow different capture and playback cards will always fail at some moment.

REW runs its generator and analyzer independently. In ASIO connector the two sides must run synchronously, due to the ASIO limitation. But in windows the correct solution is using REW's WASAPI exclusive connector - java -> EXCL: devices. Each direction will run independently.

Two (speculating on this), when measuring an unbalanced amplifier, an unbalanced audio interface may have an advantage over a balanced one due to the common ground of output and input.

That introduces a ground loop. Any current running through the amplifier ground will split and run through the soundcard ground too, creating an error voltage. Differential input is a major advantage - just connect hot and cold analyzer lines to the amp output - see the last picture of post https://audiosciencereview.com/foru...-measuring-interface-for-the-soundcard.10445/ . Eventually you can multiply the reported voltage value by two, to get the actual value.

 

[DavidR]

ASIO was not designed for merging two independently-running soundcards. The ASIO bridges which allow different capture and playback cards will always fail at some moment. REW runs its generator and analyzer independently. In ASIO connector the two sides must run synchronously, due to the ASIO limitation. But in windows the correct solution is using REW's WASAPI exclusive connector - java -> EXCL: devices. Each direction will run independently.

FlexASIO worked well with the ADCiso and the 1010LT. I used the WDM-KS mode with the WASAPI and Exclusive options set. The Scarlett 2i2 has been more problematic. It did work well with the ADCiso (after some FlexASIO issues) and with the 1010LT when it fed the balanced input only. The problem was when the amp was inserted into the test, since it's an unbalanced input amp. The output to the amp fluctuated without a probe connected. All devices are plugged into the same power socket.

That introduces a ground loop. Any current running through the amplifier ground will split and run through the soundcard ground too, creating an error voltage. Differential input is a major advantage - just connect hot and cold analyzer lines to the amp output - see the last picture of post https://audiosciencereview.com/foru...-measuring-interface-for-the-soundcard.10445/ . Eventually you can multiply the reported voltage value by two, to get the actual value.

Yes, I connected only the positive of the probe when using the unbalanced I/O in the 1010LT. What I meant was that the amp and test rig are both plugged into the same power circuit. That's what I meant by common ground. There is no ground or neutral connection directly from the amp via the probe.

Years ago when trying to measure my dipole system it was erratic. The DSP PC and speaker amp (KM-X1) were plugged into a socket across the room from the test rig. After a bit I realized the ground was the problem. I ran an extension cord from that socket to the test rig. Problem solved.

 

[DavidR]

If you really do more than occasional amplifier testing, probably the best option for those without an AP rig is the current Quant Asylum device. One can work around quite a bit using an audio interface by various means. The Quant Asylum makes it all pretty easy and it is capable of getting some good results. It still may not be good enough to probe the full limits of the very state of the art, but it gets close enough that once something is running up against its limits it is good enough not to worry about at all.

QA403 Audio Analyzer

14-Nov-2025: The QA403 is back in stock Affordable and Powerful Audio Analysis The QA403 is our 4th generation audio analyzer. Since 2012 we've been providing audio test and measurement to companies around the world. And with each iteration, the capability and programmability of the...

quantasylum.com

QuantAsylum. Innovative tools for audio engineers.

Since 2010, QuantAsylum has been developing unique products for engineers working in unique spaces.

quantasylum.com

I don't plan to measure a lot of amps, but certainly all I have now and preamps, etc. Were I to do that I would probably invest in the QA403. I wasn't aware of the audio interface issues for this when I bought the 2i2 as it was suggested for speaker distortion tests and is probably adequate for that. Later I wanted to measure my old amps for distortion, so I started with what I had on hand for an amp test rig. This became more of an academic exercise to learn details of distortion tests. At this point my costs would have made it better to start with the QA403. The purchase of the ADCiso was partly about planning to convert my LPs to digital, although that is probably over-kill.

 

[DavidR]

Testing has shown one issue with the KM-X1 under test. I have one in constant use and another one to test. I bought a third one, the currently tested one, as a spare given how old these are and how cheap they were on ebay. The one tested now required all relays to be replaced (had all been bypassed) and was way out of spec for the bias on all channels that I corrected. After initial tests of all channels I opened it and checked bias again. Small corrections made, but the tests of each channel remain mostly unchanged.

This is the overlay of all six channels. Note that 10-turn pot output was fed to the 1010LT balanced input.

Two channels (designated 3 and 6) are much worse than the other four. HD2 was as much as 10dB worse than the best. All six channels are supposed to be the same (I think) at 100W, though in stereo mode the power rating is higher at 130w. It's curious that the two poorer channels are the surround channels. Even so, at 5W all channels are meeting or exceeding the 60W rating in the documentation for distortion. Now that I have a high power resistor I'll run tests at 60W for each channel. Kenwood doesn't distinguish the distortion rating as to whether or not it is with two channels powered or just one.

One other observation is that the distortion when first powered on was a bit worse than when hot. I could watch the numbers improve real time in REW as it heated up.

As an aside it has two transformers. When in stereo mode it's a dual mono setup.

This brings up a point I made much earlier. A high quality old PCI card such as the 1010LT is surprisingly good for testing unbalanced amps, although the self noise of the card is high. But for investigating HD components it's rather good, especially with software such as REW now. So for an "inexpensive audio interface" for use in a test rig, the only devices in this test loop are the 1010LT, a 10-turn pot, a voltage divider and a probe. How much cheaper could it be? Maybe a bit more if one needed to buy a good card, either used PCI or a newer PCI-E.

Edit: Add the cost of power resistors, but that's needed for any test rig.

 

[phofman]

FlexASIO worked well with the ADCiso and the 1010LT. I used the WDM-KS mode with the WASAPI and Exclusive options set.

That's pure coincidence that crystals in ADCiso and 1010LT have close frequencies and it takes a (long) while for them to run apart so much that buffers before the ASIO duplex switchBuffer call under/overflow. Since a single call serves both directions simultaneously, it's always only a matter of shorter or longer time.

Yes, I connected only the positive of the probe when using the unbalanced I/O in the 1010LT. What I meant was that the amp and test rig are both plugged into the same power circuit. That's what I meant by common ground. There is no ground or neutral connection directly from the amp via the probe.

That means you are measuring a signal between amp live output terminal and the input ground terminal. However there can be quite a lot going on between the input GND and output GND terminals inside the amp which will be part of your measurements. However this will not be seen by the output device which would be connected between output live and GND terminals. However, if your analyzer reads the difference between the output terminals using its differential input, it will see exactly what the output device would see.

 

[DavidR]

That's pure coincidence that crystals in ADCiso and 1010LT have close frequencies and it takes a (long) while for them to run apart so much that buffers before the ASIO duplex switchBuffer call under/overflow. Since a single call serves both directions simultaneously, it's always only a matter of shorter or longer time.

I am not well versed in any of those details. REW is able to and reports the clock delta. That may be what allows it to make measurements when two clocks are involved.

That means you are measuring a signal between amp live output terminal and the input ground terminal. However there can be quite a lot going on between the input GND and output GND terminals inside the amp which will be part of your measurements. However this will not be seen by the output device which would be connected between output live and GND terminals. However, if your analyzer reads the difference between the output terminals using its differential input, it will see exactly what the output device would see.

I think I need to correct myself. When using the unbalanced input for feedback to the 1010LT I only connected the hot. Typical way to measure speakers with a feedback probe into an unbalanced card. When feeding the balanced input to the 1010LT I made a voltage divider (due to input limit) and connected one side of the probe to the middle of the divider and the other probe side to the amp output neutral. I'll need another divider when I test the amp at high power. With three input devices to test and the 1010LT having both balanced and unbalanced inputs I occasionally mis-state what I did.

Edit: What I've determined is that for testing of unbalanced devices, the 1010LT using the balanced input is by a small margin the best option given my current hardware. Even the ADCiso is a bit lower, but I attribute that to the clock issue. The 1010LT-only loop has one clock and REW can be set for the Rectangular window. REW even provides a suggestion to use the Rectangular window when the clock delta is zero.

 

[phofman]

I am not well versed in any of those details. REW is able to and reports the clock delta. That may be what allows it to make measurements when two clocks are involved.

I was talking about ASIO interface which joins the two streams to one callback. REW has to honor that single callback when using ASIO. However, when using WASAPI, its generator and analyzer can run independently.

When using the unbalanced input for feedback to the 1010LT I only connected the hot.

OK, what is the signal return path for the 1010LT input if you connect only the hot wire from the amp output? What path does the 1010LT input current take to return back to the amp output? I am afraid it will go through some rather noisy ground routes.

However, if using the 1010LT differential input, the return path would be via the cold wire of the differential input directly to the amp output GND terminal - just what a connected speaker would see.

It may not be crucial for measuring amp outputs because speakers are not extremely sensitive, but e.g. when measuring low-signal chains like a preamp or a DAC, the added noise could be significant.

Also, an amp can have bridged output where both output terminals are "hot". Connecting a single-ended measurement can then cause quite some issues, see e.g. https://www.avnirvana.com/threads/c...ke-small-thiele-parameters-measurement.14861/

 

[Blumlein 88]

I don't plan to measure a lot of amps, but certainly all I have now and preamps, etc. Were I to do that I would probably invest in the QA403. I wasn't aware of the audio interface issues for this when I bought the 2i2 as it was suggested for speaker distortion tests and is probably adequate for that. Later I wanted to measure my old amps for distortion, so I started with what I had on hand for an amp test rig. This became more of an academic exercise to learn details of distortion tests. At this point my costs would have made it better to start with the QA403. The purchase of the ADCiso was partly about planning to convert my LPs to digital, although that is probably over-kill.

I've more or less done the same thing. You get sucked into working around the audio interface using it for other purposes. So a QA403 could be a better choice for what ends up being about the same money. Even if amp testing is not on the menu, you might spend money on a better dedicated ADC of higher quality like the E1DA Cosmos.

 

[DavidR]

OK, what is the signal return path for the 1010LT input if you connect only the hot wire from the amp output? What path does the 1010LT input current take to return back to the amp output? I am afraid it will go through some rather noisy ground routes.

The KM-X1 and test rig are both plugged into the same outlet, so the negative of each is referenced to the same potential point. Connecting only the positive of the probe is standard practice for the speaker measurement systems I've used. I first used and still use the LAUD/Fiji sound card I purchased in 1998. I've also used SoundEasy and other software for measurements through the Fiji and a Delta 410. In both cases only the positive of a probe is connected for SPL and impedance measurements. Impedance uses the delta of two probes across a reference resistance, but still no neutral connection.

I can also say that when I first connected the probe to the amp recently I absent mindedly connected positive and neutral. The REW measurement was garbage. The moment I disconnected the neutral the measurement was as expected and posted here. There was evidently a ground loop problem.

Also, for reference I'm attaching pictures from the LAUD manual that explicitly says to not connect the neutral of the probe(s). Certainly for the balanced measurements both ends of the probe must be connected.

 

[DavidR]

Still testing and learning. I finished with the KM-X1 and went back to the V3 Mono/Monitor1/Scarlett 2i2 combo since I now have a high power low inductance resistor. Later I'll test the V3 Mono/2i2/ADCiso setup again. One detail became apparent as I tested the KM-X1 having to do with the feedback voltage. I had originally thought that the optimum voltage would be that which resulted from the examining the REW S-THD vs Level of the audio interface (1010LT) and confirming that with the RTA, then using the 10-turn pot to set the input as I posted earlier. The problem with this is that there are two variables, output (generator) and input. The optimum points (sweet spot) of output and input likely don't coincide, whether 1010LT or 2i2.

I ran the REW S-THD vs Level with the V3 Mono/Monitor1/2i2 loop again with the new power resistor. The Monitor1 was set to maximum essentially bypassing it other than the previously noted slight voltage drop. It became obvious that this also wasn't providing what is needed to find the optimum input for a sweet spot, although that was a good starting point. The results of this don't provide an obvious optimum input point because there are now three variables in the S-THD, interface output, amp and interface input.

The distortion levels of this loop vary due to the often extreme range of distortion products that occur with even small level changes in any one of the three devices. The APx system has its auto range function for the input to provide the best SNR range. No doubt its electronics are optimized for that. The ADCiso is somewhat like that with its input level switch settings. Most audio interfaces are not. The best point (sweet spot) can't be precisely determined due to the issue of having two or three variables involved in a loopback test, with or without amp.

What to do? My current testing is for 5W into 8 ohms. The goal was to find the optimum input voltage for all future distortion testing using the amp to ensure that its influence will be minimal. What I decided was to make a set of manual S-THD vs Level tests with the output to the amp fixed for 5W. This eliminated both the 2i2 output and the amp as variable s since they remained unchanged. OF course later tests I'll make with different amp power output will have the 2i2 output as a variable, but for any fixed power level the 2i2 output is not a variable.

The input to the 2i2 was controlled manually by the Monitor1. This was tedious, but enlightening. I chose the input range that included the upper area to the point at which the HD2/HD3 started a continuous rise. This was -7.0 to -4.9 dBFS with a 0.1dB step as close as I could do with the Monitor1. The REW level display makes setting that easy. The RTA HD components do often change dramatically with 0.1dBFS input level change into the 2i2.

This is an overlay of the best and worst RTA results for the 0.1dBFS step tests.

Out of curiosity I ran an RTA 0.05dB above and below what I thought was the best RTA above, -6.90dBFS. I was surprised that -6.85dBFS was even better.

At this point I thought I was finished. Not so. Out of more curiosity I ran an RTA with a 0.01dB step change in a narrow range including the above "best". This takes a very light touch on the Monitor1 given the audio taper close to the top of the range, but it can be done. They were all done well after the amp had reached a stable temperature. Rather than try to cull the graphs for display I collated the distortion panels that shows how dramatically the distortion numbers change with just a 0.01dB step. A 0.1dB step may look good, but there's no guarantee that any of those step points are optimum. It may be that this is more of an issue with the Scarlett 2i2, but the 1010LT was similar. I'd like to know the sensitivity of an APx system to this since it operates within a range for input.

I chose the -6.87dBFS to be the input level for all future testing.

This is why I will use the Monitor1 (balanced) or 10-turn pot (unbalanced) to set the input to the same, optimum feedback voltage for distortion measurements. This will eliminate the input variability. Output power variability will remain since it must change to supply the required voltage to the amp...unless I add an output potentiometer. Then the output of the audio interface can be set to a high, unchanging level and the pot can then be set manually for the desired voltage. The same scheme I'm using for the input. This still only provides a more consistent result for my comparisons. Comparing my results to those by others isn't valid, but it would still be less problematic. My current experience makes me think that the only valid comparisons by different systems would be those that use the same equipment, such as the same APx model. Having Amir testing all the devices with the same system is a real benefit.

However, on that last comment it reminds me that I don't place a lot of significance in the single device quality of SINAD. I would not make a choice based on that, I would want to compare the detailed measurements he provides. SINAD is like shooting at a target with a shotgun rather than a rifle. In my opinion, of course.

 

[DavidR]

My plan did not work out as I had hoped. I have two 8-ohm low inductance resistors now. In order to compare this test scheme to Amir's results I needed to test using 4 ohms rather the 8 ohms of sand cast resistance. The distortion levels were significantly higher at the input voltage to the 2i2 I described above. Ultimately I ran a stepped THD in REW to view the distortion. I had to re-select a 2i2 input based on that. The previous one for the 8-ohm low impedance inductor did provide the best result.

Part of the problem is, I think, as I mentioned above, the variability of the 2i2 output. The amp magnifies the difference. The 2i2 input for 5W went from 6.32V for 8 ohms to 4.47V for 4 ohms. Therefore the 2i2 output was changed and its distortion components at that level were different. I can't say exactly how much change there is in V3 Mono distortion given the 2i2 variability with level both output and input. I suspect the 2i2 output change is higher. I don't think that the 2i2 (or any) loopback can be used reliably given that it includes input and output distortion at any level.

But in testing I found something that I thought could not happen. The Monitor1, a passive attenuator, has an impact on distortion in one case, at least according to REW RTA results. I made a series of tests of it, changing only the Monitor1 level for specific settings with constant V3 Mono feedback voltage, then those same settings, but altering the 2i2 output to the V3 until the 2i2 input voltage was the same for each Monitor1 level. Afterwards I made measurements for linearity at different settings.

Linearity at various levels looked good. The one issue is with the 100 (max) setting to 90. I noted early on that the level change is practically zero. Level doesn't start to drop until 90 or below. However, the distortion components changed, primarily HD2. A lot. I ran S-THD for a direct connection to the 2i2, then with the Monitor1 connected, but with the dial at 100 (max). It isn't a complete bypass, given the small level drop, but HD2 increased a lot. Lower dial settings shifted the curve higher, but that's expected given the lower level, so SNR increased. I can only guess that it has to do with the symmetry of the level of the positive and negative of the attenuator in the Monitor1. I can't see what else it could be given that it's a purely passive unit.

Here are the S-THD results for HD2. Note that the 2i2 direct result (black) without the Monitor1 in the loop remains below 105dBr above the level where all settings diverge. What is most curious is that HD2 with the Monitor1 at 100 (maximum setting) starts a nearly constant rise as level increases. This is counter to what I would have expected. Given the small attenuation of the Monitor1 at full setting I would expected only a slight shift to the right since this is the graph of the input relative to the output. The lower Monitor1 settings do shift right as expected, although they vary somewhat in shape with some noise difference. I just can't find an explanation for this behavior.

HD2:

HD3 is interesting in how different it is from HD2. Until the Monitor1 setting is low (very low input voltage to the 2i2) the odd harmonics remain almost identical. My observations in my testing is that even harmonics have a much larger variability with signal level, both output and input. If I set change the 2i2 output (higher power) and adjust the Monitor1 so that the input to the 2i2 is same for any output power, the even harmonics change much more than the odd.

HD3:

I have another test plan to make. After some online research I found a site that demonstrates how odd and even harmonics in a balanced signal are affected by positive/negative symmetry. The only mode I can test (that I can think of for now) is vertical P/N symmetry. Other signal deviations that don't follow a perfect sine wave I can't produce, but since the Monitor1 is only passive, I can test and compare to it. I now have two linear unbalanced potentiometers. My plan is to make my own balanced passive attenuator using these two combined. I'll be able to manually set the symmetry of P/N, from nearly perfect to various amounts of asymmetry to examine the linearity change (if any) and distortion introduced. I wanted a more precise ability to set feedback level compared to that of the Monitor1 that is very sensitive to finger pressure for precise settings. If/when I get some results, I'll post that. If it works out it will be a cheaper, more precise passive attenuator for testing than most passive units such as the Monitor1 given that with those there is no way to account for any imprecision between P/N. It will be more time consuming to make changes given the need to set both P/N precisely to the same point, but since it's only for testing that won't be much bother.

The other reason for this manual balanced pot is to use it on the 2i2 output. I want to test it with constant output to the pot and adjust it for the amp input. This will let me examine the 2i2 output to see if there is an optimal output voltage for lowest output distortion to the amp. All of my testing makes it apparent that the limitation for amp testing using the 2i2 is its output distortion that changes with level. I'll also use the ADCiso at some point for input in place of the 2i2 input.

This is all overkill for my initial intention of testing with an inexpensive audio interface, but it's been very instructive to me vis-a-vis distortion in and testing of components.

 

[DavidR]

One final graph of the V3 Mono using the Monitor1/2i2 combo at what I determined to be the optimal 2i2 input level for the HD components as a group.

Red is the 256k FFT, classical average. Green is using a 512k FFT. Black is the 512k FFT with Coherent averaging. All three have essentially the same HD levels.

Amir made a valid point in my early tests that used an 8 ohm resistance load. Now that my tests use a 4 ohm low inductance resistance, here's the graph again by Amir for comparison.

There are a few distinct differences. Amir's tests have a lower noise floor and somewhat better HD3 whereas mine show better HD2. But considering the cost differential my results are fairly competitive. This with the poor 2i2 output distortion. I hope to improve this when I construct my hand-made balanced passive preamp to use on the 2i2 output. Of course my testing lacks many of the additional tests available to the APx systems. Those that can be done using REW have been somewhat disappointing due to the 2i2 limitations, SINAD a prime example.

 

[DavidR]

While continuing testing, I wanted to confirm some previous results such as the graph above. I've had occasional problems getting repeatable results, often when changing cabling and probes. The first big problem was a balanced cable. All of mine are fairly recent purchases, random ones over time. I avoided what looked cheap. I purchased a short one to eliminate a 1m cable coiled to connect. I thought the Monitor1 input socket might be failing, touching that cable made the RTA results change, but the problem was the short cable. The tip and ring on the RTS end both would rotate. Both connections failed. The TRS tip of the other cable of the pair also would rotate. I lost a lot of time trying to find the reason for the repeatability issue. I connected a high quality 1m cable again, problem solved. Or so I thought.

I also have found another issue with distortion measurement, the probe cabling. I'm using the Scarlett 2i2, Monitor1 and V3 Mono as one configuration. The other one is 2i2 (generator), V3 Mono and ADCiso with a directly connected probe. After eliminating the rotating-ring cable issue the REW results were still nowhere near the previous low HD levels. I have two probes, both direct, one with a male TRS for the feedback, the other with a female one. Due to cabling I purchased, I need both. The RTA results were good now with good cables, just nowhere as good as previous results. One of two probes has the ground wire with an alligator clip, but the V3 Mono has floating outputs (I think), so it is not connected. Starting with 2i2 probe loopback and REW running the RTA continuously I tested for other noise sources, the V3 cooling fan (produces small spikes if too close), the V3 power supply, the monitor above it all. Nothing fixed it. Then I touched the loose ground wire. The HD2 made a significant change, for the better. So I tested various placements of just the ground wire and the whole cable, though the probe wires are short. Every positional move of the probe cable changed the RTA results, almost exclusively the even-order HD, primarily HD2. Eventually I simply put the alligator clip onto the base of the ground wire itself and lo-and-behold the distortion measurements returned roughly to the previous very low results. Still, moving the cable around changed that. I found the best cable position to be as shown in the pictures below. I also coiled the probe wiring in the unlikely event that might help.

I'm dumbfounded about this. I switched to the probe without a ground wire soldered onto the probe end for the 2i2. The results were never better than minimal. Reconnecting the probe with the loose ground wire was required to get the excellent results, although with the ADCiso you can see that I found a good probe wire postion needed to best results. I can see no reason for any of this whatsoever, but it's absolutely repeatable. The even-order HD components are extremely sensitive to his, odd-order very little. Maybe the feedback probe should be different somehow. They are both direct connections, no voltage divider of any kind. I would appreciate any feedback that might explain this issue.

 

[DavidR]

I replaced the taper resistors in the Behringer with hand-matched 1%. These measure much better than the original ones. No before/after comparison graphs to show, but the distortion results did improve a bit, though not dramatically. But that should improve any measurements. Continuing to experiment for optimal use for measurements I found a number of interesting details. From here on all measurements with the 2i2 include the new taper resistors in the Monitor1.

I ran new 2i2 loopback measurements. I settled on -21dBFS as optimal. It's possible to find different input voltages that result in better specific HD components, but that always also results in changes in other component results, so it's a balancing act. I found this to be true with both the 2i2 and the Delta 1010LT and I suspect it's likely the case for almost any audio interface. The current "best" for the 2i2/Monitor1 combo is below. Note the noise level that impacts the SINAD. Noise dominates.

This is the previous "best". Note that the numbers are values are for the coherent measurement. That primarily improved the odd-order even-order harmonic results.

The next one is a comparison of two feedback voltage levels. This was controlled using the Monitor1 to adjust it. I experimented with a number of Monitor1 settings. These were two of the best. Two things to note are the odd-order components and the noise. There is a lot of variation in all HD components with feedback level. With lower feedback voltage the noise worsens in a repeatable way. Always. The individual HD components may change dramatically one way or the other, but the noise change is specific to level. This is another reason why noise is not high on my list of importance, thus neither is SINAD. That is a good, rough idea of how an interface will perform, but I doubt that it has much impact on audibility. The same goes for THD, in my opinion. My focus is primarily HD2 and HD3.

My goal for amp/preamp testing is to find the optimal I/O levels for distortion testing, not noise. But when an interface is used with varying input levels, for example recording a performance, there's no way to avoid changes in the noise impact for SNR. As an example of noise change, the next graph shows the 2i2 loopback S-THD Noise for various input settings of the Monitor1. Each S-THD was run from REW output of -60dBFS to the level that provided 0dBFS at the input (Mon50 the exception). X-axis is the 2i2 input graph.

It's obvious (unless I'm interpreting the REW graph wrong) that for any given 2i2 output, the input noise dBr is affected by the Monitor1 setting, meaning that the input noise as measured changes. As a loopback, it's the sum of 2i2 output and input noise. The HD components change as well, but those changes are totally unpredictable.

I have a series of tests of the V3 Mono/Monitor1/2i2 to post later that will demonstrate in more detail the problem with measurements made with different input voltages.

 

[DavidR]

This is the best result I can get with the V3 Mono using the Scarlett 2i2 loopback with the Monitor1 in the feedback of the loop. This is the balanced loop. Later I'll show tests using the V3 unbalanced input. I'll have more to say about "best" later as well. There's an issue with using an average of the RTA that has been constant in all tests I've made.

This a comparison of the Scarlett 2i2 RTA over the V3 Mono RTA for the same Monitor1 Dial setting and the same 2i2 input voltage. The 2i2 input is always set for 12dB gain.

The 2i2 was measured for loopback with the REW generator and the Monitor1 set for what I considered the best 2i2 input for the loopback. In this case, -15.63dBFS. The Monitor1 was then left untouched.

Next the V3 Mono was driven for 5W into 4 ohms using the 2i2. The Monitor1 was adjusted so that the V3 feedback probe into the 2i2 input matched the 2i2 loopback test input level as closely as possible.

This meant that the differences on the two tests were the generator output level (for both tests) and the V3 gain for its test. Given the limited ability to isolate to a single variable this is the best test I could devise to start.

The results are show in the next graph. The 2i2 RTA (red) overlays the V3 Mono RTA.

Right away it's obvious that the V3 Mono adds a lot of higher harmonic distortion. It also adds some HD2-HD8, but not a lot. It's hard to determine how much influence the 2i2 output has on the V3 measurement. It's made more problematic due to having to set the REW generator for different output to achieve the same 2i2 input. Two variables make it unclear. The 2i2 loopback requires higher 2i2 output to match the V3 loop feedback with the Monitor1 fixed since the V3 has gain.

The next step will be to make the balanced attenuator using the two 10-turn pots. This will allow setting the 2i2 output to the same level to eliminate that as a variable. The result will be constant 2i2 output, identical 2i2 input and all level adjustments made using passive attenuators.

On that note I'll be posting the unbalanced V3 Mono tests made using the unbalanced, single pot I had on-hand with feedback going to a balanced input. Maybe tomorrow.

 

[DavidR]

In studying the variations in distortion with levels of audio interface output and input I started to notice some trends. One was the wide variations of all distortion components. Another was the limitations with level, both the low end and, to my surprise, the high end. I expected better distortion quality from higher output, especially since so many manufacturers publish distortion for high output with amps in particular touting THD for rated output. So far I'm seeing more variation in the interfaces (PC cards and USB) with I/O (I exclude the ADCiso) than amps. Old PC cards seem much more linear in distortion vs. level. That seems to be related to intended use of USB interfaces where the output may often seldom be used in its upper range or is there for short term dynamic response. But the distortion increase at the upper end is significant nevertheless.

Another issue is noise. The SNR in my measurements in neither the "sweet spot" nor those I find "optimal" come close to the level in documents. It seems to me that this is similar to THD numbers. Makes for good advertising or, in the case of SINAD, a roughly useful reference for comparison, but I've been using the Delta 410 and 1010LT for many years. My measurements show the 1010LT best noise to be at the upper extreme, an area that almost never has content when I play music. My system uses the Ultimate Equalizer (UE) from Bodzio Software (SoundEasy author) for a dipole system. I set the playback setting to +3dB for the 1010LT now (due to CD digital levels) that still has some headroom. This provides near maximum DAC output. An 8-channel preamp controls the signal to the amp. Except for a few extreme recordings that clip until I drop the UE level to prevent digital clipping, I have never noticed any noise issues. I've never even had a protection cap on the tweeter. The 1010LT seems to have around -80dB as its best noise level. In the area of common signal level I think it's around -70dB. But the HD components are admirably low. This explains why I tend to dismiss SINAD as a useful figure of merit. I focus on HD2 and HD3 given that most higher components are even lower.

I spent a lot of time experimenting with signal levels, input and output, to characterize my hardware and study distortion. The graphs below help to explain how I arrived at my positions on it. The distortion variations are actually very disappointing to me. Very often one HD component is decreasing, but one or more other ones are increasing. I essentially found not a single region in the sound cards nor the USB interface where there was a smooth, constant area of distortion components. They are all more roller coaster in nature. This makes it very difficult to find a true "sweet spot". Granted that this has been, so far, related to it being the sum of output and input distortions, but when I set a constant output level or input level, the undulations remained.

The graphs below demonstrate the issue using the Scarlett 2i2 Gen 4. They include THD, HD2 and HD3 for different levels. Note that there are multiple curves on each graph. Each curve is the result of a different setting of the Monitor1, from maximum (almost pure passthrough) to 0dBFS or to the limit of the 2i2 output.

The first graphs show measured distortion vs. generator output level.

The next graphs show measured distortion vs. input level in the same order.

It took me a while to realize that in REW, switching the graphs from generator to input shifted the graphing based on the dBFS of generator or input. These are zoomed in to see detail.

I hope to see improvement when I build the balanced passive attenuator using the 10-turn pots for output control. On that note, I'll be posting measurements of the V3 Mono using a single 10-turn pot for output control soon.

 

[DavidR]

I've yet to make the balanced passive attenuator, but I do have the single 10-turn attenuator set up. I ran tests using that between the unbalanced output of the 1010LT to the unbalanced input of the V3 Mono. This does not allow use of the REW S-THD option to find the "sweet spot" for 1010LT output. The Monitor1 was connected between V3 Mono output probe and the 1010LT balanced input. The goal was to find the best measurement settings for V3 Mono @5W into 4 ohms.

The Monitor1 was set to the point previous determined to be the optimum for 1010LT balanced input and remained fixed throughout the test. The effort to find the best output to the V3 required a "manual stepped-THD". That is, the 10-turn pot was set to maximum while the REW output was set so that the V3 output was 5W (4.47V) to start. This provided the lowest possible REW output for 5W.

The "stepped THD" was accomplished by manually increasing the REW output by 0.1dBFS, then adjusting the 10-turn pot to reduce the V3 output back to 4.47V, making note of significant setting results along the way. As it turned out, each 0.1dBF change increased the V3 output by exactly 0.05V. This continued until the REW output was higher than about -10dBFS. Above this point the 1010LT output distortion continually increased.

I made a few observations during this. The upper range of the 1010LT output was similar to the 2i2 in that above a certain point the distortion made it unusable for measuring another device. Not surprising I suppose. Another one is that since the best area of low self-noise is in the unusable area, again in the upper output range. Due to this the SINAD/Noise numbers at the settings that were noted as best for distortion results when measuring the V3 are certainly not impressive. In fact, they are disappointing. There's no way to obtain the lowest distortion measurement results of the amp without it being where the audio interface self-noise isn't a factor. At least in my experience with both the 1010LT and the 2i2. I'll know more about the 2i2 when I make the 10-turn pot attenuator for its output control, but I expect that to be no different. Part of this is, of course, due to the relatively high self-noise of these two even near or at full output.

I ran a long series of V3 measurements at various settings of REW and potentiometers. These are the most representative. These all used the 1010LT unbalanced output to balanced input. The 10-turn pot was on the output with the Monitor1 on the feedback probe (direct, no voltage divider) to the input.

First up had the 10-turn pot at 9.7 (near maximum) that required to REW set at the lowest value that still provided 5W into 4 ohms. The left graph is the classical from the right overlaid on the coherent measurement. I'll have more to say on this later.

Next is the result with the REW output at -22.72dBFS (bad edit in the graph comments) with input at -11.51dBFS vs -19.25dBFS above. The 10-turn pot was again set so that the V3 output 5W (9.0, not 90 as typed). The most obvious differences are that the noise floor is little changed and most distortion components are similar with the exception that H2 in the coherent response is significantly worse. This also will be addressed later.

Next is a comparison with the same REW input, but with REW 0.0dBFS output vs -15.0dBFS output. The Monitor1 was untouched for the V3 feedback, only REW and the 10-turn pot were changed. There are significant differences. The most obvious is that the noise floor and noise are much lower with 0.0dBFS output, not surprising The HHD levels are similar. However, what is notable is that all even-order harmonics are worse, especially HD2. All graphics up to this point have been for a 32 average.

This brings me to the point I alluded to earlier. The first four graphs were what I call "cherry picked". What that means is that I've been aware from early on since working with REW that for an average, in these cases the 32 average default, that the harmonic component values can and usually do change rather dramatically if the 32 average is allowed to continue. The change is most often biggest in HD2, though all even-order will change a lot. The latter has more to do with which "sweet spot" is chosen. I focused on HD2 when capturing a graph update. HD2 very often had 10dB or more swings over time. HHD changed several dBs as well. Noise generally remained stable.

To follow up on that I started making longer averages. The default in REW for "Forever" with "Stop at" the initial default of 100. Later extended that, at one point letting it run overnight to more than 5000. Eventually I found that 100 was fairly close to longer averages, though not always. But what was always the case was that the longer it ran, the lower the numbers. That is, the HD components generally were worse. At some point they stabilize, but I would say that the the default of 100 is adequate. Examples are in the graphs below.

The right shows the classical measurement for 32 (red) and 1000 (black) average results. The left is another input level using classical and coherent measurements with classical 32 (red) and coherent 1000 (black). Both of these input levels were points I had made note of during testing that looked promising. Note how close they are for most HD components.

I've posted the above graphs because I have questions related to two aspects. One is about the typical 32 average often used. In my testing it does not appear to be adequate. This holds for both the 1010LT and the 2i2. I noticed this in measuring the V3 Mono and the Kenwood KM-X1 as well as for audio interfaces in loopback. An average of 100 or more seems to be more accurate. I ran a number of long term measurements, both classical and coherent. As can be seen above, the coherent numbers, even for different input levels (good ones, that is) tend to converge to nearly the same numbers. Cherry picking a continuing 32 point average provides widely varying results, but 32 seems to be a common average reported here, though even fewer at times, but 100 would be more accurate.

This made me curious about the APx results. The company site AP More about Averaging provides some details. It has the option for number of averages as does REW. My interpretation is the Synchronous is analogous to Coherent in REW and Power analogous to Classical (non-coherent that is) in REW.

Synchronous averaging​

Synchronous averaging operates on a time domain acquisition. Synchronous averaging is useful to examine coherent time domain waveforms by reducing the average level (but not the variance) of noise and other non-coherent signals. Since the frequency domain results in a measurement are derived from the time domain acquisition, synchronous averaging affects spectrum results as well.

Power averaging​

Power averaging (spectrum averaging) operates only on a frequency domain result. The power spectra for all averaged acquisitions are summed and then divided by the number of acquisitions. Power averaging helps reveal coherent frequency components by reducing the variance (but not the average level) of noise and other non-coherent signals.

My other question is directed to Amir if he's reading this. What type of average is used in the APx and which averaging? I have not been able to find that info here.

 

[phofman]

@DavidR: IMO the decision to use coherent or standard/magnitude averaging in REW is important. I do not know exactly what the AP synchronous averaging does, how it does averaging in time domain. But REW does basically this:

Standard/magnitude-based averaging - for each FFT pass the magnitude (i.e. the length) of each bin vector value is averaged.

Coherent averaging - AFAIK in vector analyzers this averaging is called vector averaging - Based on the fundamental phase in the given FFT, measured vectors of all bins are rotated to correspond to e.g. zero phase of the fundamental (basically time-shifting to time corresponding to start of the period of the fundamental frequency). Then all bins are vector-averaged with the previous FFT run, i.e. not only their magnitude/length, but also direction/phase. Since fundamental-related components (i.e. the fundamental and harmonics) are coherent with the fundamental phase, after time-shifting to the "zero time" of the fundamental they will have the same phase at each FFT run, and the averaging will basically average only their magnitude. All the other bins with components not coherent with the fundamental will have a phase different (basically random) every processed run, and the vector averaging will suppress them much more. This applies not only to random noise, but also to other spectrum components not synchronous with the fundamental - e.g. switched-supply artefacts, artefacts from digital circuits processing, etc.

IME (from working on https://www.diyaudio.com/community/...on-compensation-for-measurement-setup.328871/ ) when measuring small distortions, the phases of tiny harmonics vary due to noise too, and coherent averaging reduces them to some extent too, yielding (IMO artificially) better THD numbers than standard averaging.

I do not know if what the AP synchronous averaging does is equivalent to REW coherent averaging, it may be.

 

[Holmz]

…
IME (from working on https://www.diyaudio.com/community/...on-compensation-for-measurement-setup.328871/ ) when measuring small distortions, the phases of tiny harmonics vary due to noise too, and coherent averaging reduces them to some extent too, yielding (IMO artificially) better THD numbers than standard averaging.
…

Non-coherent gives the typical measurement noise of 1/SQRT(N), whereas coherent is 1/N.
Pick what N is, but in your example it would be # of FFT “frames”.

 

[DavidR]

I'm trying to tie this with the layman's description of the AP and what Amir uses. In complex math when you rotate, the magnitude remains fixed, phase rotates. Real/reactive components both change accordingly. I'm trying to correlate your two descriptions as they relate to complex math. It's easier if reference for "length" is the magnitude or the real component. But I may be over-thinking this. I may even have my descriptions in reverse for standard/coherent. I may re-read this later after my head clears and update it if needed. Please point out anything that may be incorrect (or inverted by averaging type).

@DavidR: IMO the decision to use coherent or standard/magnitude averaging in REW is important. I do not know exactly what the AP synchronous averaging does, how it does averaging in time domain. But REW does basically this:

Standard/magnitude-based averaging - for each FFT pass the magnitude (i.e. the length) of each bin vector value is averaged.

This sounds like the magnitude component being averaged that does not change with rotation. There would is no need to rotate prior to magnitude component averaging. Ignores phase altogether. Then this is the REW standard (classical) response.

Coherent averaging - AFAIK in vector analyzers this averaging is called vector averaging - Based on the fundamental phase in the given FFT, measured vectors of all bins are rotated to correspond to e.g. zero phase of the fundamental (basically time-shifting to time corresponding to start of the period of the fundamental frequency). Then all bins are vector-averaged with the previous FFT run, i.e. not only their magnitude/length, but also direction/phase. Since fundamental-related components (i.e. the fundamental and harmonics) are coherent with the fundamental phase, after time-shifting to the "zero time" of the fundamental they will have the same phase at each FFT run, and the averaging will basically average only their magnitude.

It would seem to me that this is the real/reactive components being averaged after rotation since magnitude does not change with rotation, real does. Thus real would change, but never be equal in value to magnitude unless the phase of the harmonic being rotated is equal to the original phase of the fundamental such that the result is phase equal to zero, likely (almost) never the case. Thus the resultant average would tend to be more negative in dB both for the real value and the vector average of a larger real value. This would explain some of the distinctly more negative HD dB components in coherent averaging.

Edit: Rotation prior to averaging would be a key point.
Edit 2: Added underlined text.

All the other bins with components not coherent with the fundamental will have a phase different (basically random) every processed run, and the vector averaging will suppress them much more. This applies not only to random noise, but also to other spectrum components not synchronous with the fundamental - e.g. switched-supply artefacts, artefacts from digital circuits processing, etc.

This has been mentioned to me in other posts and is the one area that is clear and makes perfect sense.

IME (from working on https://www.diyaudio.com/community/...on-compensation-for-measurement-setup.328871/ ) when measuring small distortions, the phases of tiny harmonics vary due to noise too, and coherent averaging reduces them to some extent too, yielding (IMO artificially) better THD numbers than standard averaging.

Same as previous comment.

I do not know if what the AP synchronous averaging does is equivalent to REW coherent averaging, it may be.

This is what I'm hoping to determine. That's why I'd like for Amir to chime in. That is, which AP averaging does he use and how many samples are averaged.

BTW, thanks for the detailed descriptions.