iFi nano intermodulation distortion

[mansr]

I've noticed that the iFi nano has quite a bit of intermodulation distortion at high frequencies. Playing two tones at 60 kHz and 61 kHz gives this output:

 

 

These distortion levels are easily audible and far higher than with any other DAC I have on hand. Of course real music doesn't have that much high-frequency content so it's not a big deal, but I still find it curious.

[Jud]

I've noticed that the iFi nano has quite a bit of intermodulation distortion at high frequencies. Playing two tones at 60 kHz and 61 kHz gives this output:

 

[ATTACH=CONFIG]24837[/ATTACH]

 

These distortion levels are easily audible and far higher than with any other DAC I have on hand. Of course real music doesn't have that much high-frequency content so it's not a big deal, but I still find it curious.

 

What happens if you run the same input through software upsampling with, e.g., HQPlayer?

[mansr]

What happens if you run the same input through software upsampling with, e.g., HQPlayer?

 

The graph above is what I got operating the DAC at 192 kHz. Feeding it PCM at 384 kHz, bypassing the internal upsampling filters, makes things worse:

 

 

Note the smattering of spikes 1 kHz apart around 20 kHz.

 

With software conversion to DSD256 we get this:

 

 

Here the 1 kHz intermodulation product and its harmonics are somewhat reduced (still audible), and the high-frequency artefacts are gone. However, the noise floor is annoyingly high (easily audible), probably caused by some interaction between the 60 kHz signal and DSD noise. A 1 kHz tone doesn't have this problem.

[jhwalker]

The graph above is what I got operating the DAC at 192 kHz. Feeding it PCM at 384 kHz, bypassing the internal upsampling filters, makes things worse:

 

[ATTACH=CONFIG]24845[/ATTACH]

 

Note the smattering of spikes 1 kHz apart around 20 kHz.

 

With software conversion to DSD256 we get this:

 

[ATTACH=CONFIG]24846[/ATTACH]

 

Here the 1 kHz intermodulation product and its harmonics are somewhat reduced (still audible), and the high-frequency artefacts are gone. However, the noise floor is annoyingly high (easily audible), probably caused by some interaction between the 60 kHz signal and DSD noise. A 1 kHz tone doesn't have this problem.

 

 

I'm wondering if the Micro has the same issue - I can hear clearly the difference between the two (the Nano sounds a lot more "harsh" and digital vs. the Micro).

 

Also, are you using the latest (5.1) firmware on the Nano? I've heard (and can testify to) a clear benefit to the newer firmware.

[mansr]

I'm wondering if the Micro has the same issue - I can hear clearly the difference between the two (the Nano sounds a lot more "harsh" and digital vs. the Micro).

 

I don't have a Micro, but you can easily test this yourself.

 

Also, are you using the latest (5.1) firmware on the Nano? I've heard (and can testify to) a clear benefit to the newer firmware.

 

Yes, I am.

[jhwalker]

I don't have a Micro, but you can easily test this yourself.

 

"Easy" is a relative term What hardware / software are you using to digitize and measure the output?

 

I don't have anything now, but might consider buying something so I could do some of my own analysis, as you suggest.

[john57]

I doubt that I have any music recorded that has any musical signals anyway near 60kHz. I do not see the point of this exercise to inject a 60kHz signal.

[sdolezalek]

I doubt that I have any music recorded that has any musical signals anyway near 60kHz. I do not see the point of this exercise to inject a 60kHz signal.

 

It is because although we can't hear 60kHz sounds in cables or in hi-rez recordings, we can in DACS...

[mansr]

I don't have a Micro, but you can easily test this yourself.

 

"Easy" is a relative term What hardware / software are you using to digitize and measure the output?

 

I don't have anything now, but might consider buying something so I could do some of my own analysis, as you suggest.

 

You just need to play a test signal like the one I used. If there is intermodulation of this magnitude you'll hear a 1 kHz tone very clearly.

 

I used sox to synthesise the signal. If you want I can provide a file.

 

For the recording I used a Steinberg UR242. It's not stellar, but not bad either for the price. I did the graphs with Octave, a free Matlab clone.

[mansr]

I doubt that I have any music recorded that has any musical signals anyway near 60kHz. I do not see the point of this exercise to inject a 60kHz signal.

 

The effect is present around 30 kHz too, though not as severe.

 

I did this test because I noticed a problem, not to go looking for one.

[AMR/iFi audio]

Hi,

 

If one looks at other tests for HD, one will see that the iDSD nano has a very tube like distortion spectrum and not the lowest distortion one might find, but instead distortion that is low compared to much actual tube gear and of a spectrum that is generally masked by the human hearing.

 

iFi nano iDSD

 

Harmonic distortion has equivalent counterparts in IMD and that is what is measured, the 2nd harmonic of the mix of 60kHz & 61kHz is 1kHz and one can also see the 3rd, 4th and 5th harmonic, however anything higher order is absent.

 

In order to produce audible IMD products with music one would need very high levels of ultrasonic content, which if it can be found at all, is exceedingly rare.

 

If you still find an issue, just open a Support Ticket, send through the screenshots and data and we'll get the techies to have a look.

 

Cheers.

[mansr]

Hi,

 

If one looks at other tests for HD, one will see that the iDSD nano has a very tube like distortion spectrum and not the lowest distortion one might find, but instead distortion that is low compared to much actual tube gear and of a spectrum that is generally masked by the human hearing.

 

iFi nano iDSD

 

Harmonic distortion has equivalent counterparts in IMD and that is what is measured, the 2nd harmonic of the mix of 60kHz & 61kHz is 1kHz and one can also see the 3rd, 4th and 5th harmonic, however anything higher order is absent.

 

In order to produce audible IMD products with music one would need very high levels of ultrasonic content, which if it can be found at all, is exceedingly rare.

 

If you still find an issue, just open a Support Ticket, send through the screenshots and data and we'll get the techies to have a look.

 

Cheers.

 

The IMD is audible even with a 15+16 kHz test signal, but with music there is probably enough low-frequency content to mask it. I recorded the nano playing some music with high-frequency content, and aside from a slight roll-off upwards of 17 kHz (which could equally be caused by my recording equipment), the capture tracks the original very closely.

 

I only posted this because I thought someone else might also find it interesting, not as a complaint.

[Jud]

The graph above is what I got operating the DAC at 192 kHz. Feeding it PCM at 384 kHz, bypassing the internal upsampling filters, makes things worse:

 

[ATTACH=CONFIG]24845[/ATTACH]

 

Note the smattering of spikes 1 kHz apart around 20 kHz.

 

With software conversion to DSD256 we get this:

 

[ATTACH=CONFIG]24846[/ATTACH]

 

Here the 1 kHz intermodulation product and its harmonics are somewhat reduced (still audible), and the high-frequency artefacts are gone. However, the noise floor is annoyingly high (easily audible), probably caused by some interaction between the 60 kHz signal and DSD noise. A 1 kHz tone doesn't have this problem.

 

What did you use to obtain the sample rates/formats?

[mansr]

What did you use to obtain the sample rates/formats?

 

I used sox.

[sdolezalek]

The IMD is audible even with a 15+16 kHz test signal, but with music there is probably enough low-frequency content to mask it. I recorded the nano playing some music with high-frequency content, and aside from a slight roll-off upwards of 17 kHz (which could equally be caused by my recording equipment), the capture tracks the original very closely.

 

I only posted this because I thought someone else might also find it interesting, not as a complaint.

 

Notwithstanding my earlier teasing comment, I wonder how much of the audio equipment we surround ourselves with is susceptible to similar "out-of-band" noise that does have "in-band" IM or harmonic distortion consequences. Further, as Jud has frequently cited, dealing with these same distortions are likely to be the real benefit of hi-res files given the filtering choices they permit. But, given that most manufacturers only publish their 20Hz-20kHz specs, absent testing like yours, we are unlikely to know the degree to which significant out-of-band issues might have more audible consequences.

 

Further, I assume that in feeding equipment for testing purposes, manufacturers make sure they are only using material that doesn't contain the out-of-band spikes that you triggered with the iDSD Nano, so you wouldn't see the in-band IM distortion show up in those tests. On the other hand, it does support those who have argued here that using ultra-wide band hi-res files could actually cause equipment problems otherwise solvable with a 20kHz brickwall filter...

[jhwalker]

Thanks for posting this - very interesting.

 

I tested with my own Nano and Micro (60kHz + 62kHz tone, both set to ~75% max volume - yes, I know they have different maximum output, but these actually sounded pretty much the same volume suprisingly). Results:

 

1) Nano - 2kHz intermodulation tone audible with minimum phase filter - inaudible with standard filter

2) Micro - 2kHz intermodulation tone audible with bit perfect filter - less prominent, but still audible with minimum phase filter - inaudible with standard filter

 

I also tested with my LH Labs Pulse - inaudible in every mode.

[Jud]

I used sox.

 

Would you be able to try an HQPlayer filter (e.g., poly-sinc or poly-sinc-2s) and modulator (ASDM7 or DSD7) to see what the results are?

[mansr]

Would you be able to try an HQPlayer filter (e.g., poly-sinc or poly-sinc-2s) and modulator (ASDM7 or DSD7) to see what the results are?

 

The exact result varies with the filter settings, but it's always a clearly audible tone along with some hiss just like with sox.

[mansr]

Thanks for posting this - very interesting.

 

I tested with my own Nano and Micro (60kHz + 62kHz tone, both set to ~75% max volume - yes, I know they have different maximum output, but these actually sounded pretty much the same volume suprisingly). Results:

 

1) Nano - 2kHz intermodulation tone audible with minimum phase filter - inaudible with standard filter

2) Micro - 2kHz intermodulation tone audible with bit perfect filter - less prominent, but still audible with minimum phase filter - inaudible with standard filter

 

This is odd and also not what I'm getting on the Nano. At sample rates high enough to represent a 60 kHz signal, there is barely any difference between the filter settings, and I can neither hear nor measure any appreciable difference in the output.

 

Note that the filter switch labels on the Nano are misleading. The "minimum phase" setting actually selects a linear phase filter with slow roll-off (details are in the DAC chip datasheet) which reduces the ringing.

[AMR/iFi audio]

The IMD is audible even with a 15+16 kHz test signal, but with music there is probably enough low-frequency content to mask it. I recorded the nano playing some music with high-frequency content, and aside from a slight roll-off upwards of 17 kHz (which could equally be caused by my recording equipment), the capture tracks the original very closely.

 

I only posted this because I thought someone else might also find it interesting, not as a complaint.

 

Hi,

 

Thanks and much appreciated!

 

The linked Video provides an excellent (visual) illustration of the levels of ultrasonic content in actual music (and other) recordings, as well as IMD test signals and shows that IMD Tests with very high frequencies and high levels are not any indication of what the system will face in practice:

 

[mansr]

Hi,

 

Thanks and much appreciated!

 

The linked Video provides an excellent (visual) illustration of the levels of ultrasonic content in actual music (and other) recordings, as well as IMD test signals and shows that IMD Tests with very high frequencies and high levels are not any indication of what the system will face in practice:

 

 

Yes, real music content usually falls off quickly above 15 kHz if not sooner. This distortion could still be relevant if you plan on using the DAC as a signal generator for testing other equipment.

 

The Nano is a fine little DAC for normal listening, and I wouldn't suggest anyone avoid it based on this.

[Miska]

This is one common reason why the limited interpolation filters to 352.8/384k rates in DAC chips causes issues, even if the source data frequency content doesn't go high. Such create image frequencies around multiples of the digital filter output rate (352.8/384) and due to the frequency mirror property the intermodulation difference tones fall into audio band. Even if DAC itself doesn't posses this IMD problem in it's analog output stage, insufficient analog filtering leaks those to the following amplifiers where intermodulation distortion may get worse...

 

Due to the mirror property, lowest and thus strongest audio frequencies are also closest to each other and thus the difference tone is also lowest frequency. The frequency difference is naturally 2x the frequency of source tone.

 

Analog filter stages inside DAC need to have high bandwidth and low distortion at very wide frequency band due to this.

 

Another note is that whenever you use ADC to convert output of a DAC, you need to take into account quality of the ADC's anti-alias filter. So that the images or modulator noise don't end up being aliased into pass-band, in case the DAC's analog reconstruction filters don't do very good job (common problem especially with PCM input to DACs due to above limitation on the digital filter side). So study the ADC first with known clean full-scale frequency sweep input that reaches well beyond output sampling rate of the ADC!

[YashN]

I'm wondering if the Micro has the same issue - I can hear clearly the difference between the two (the Nano sounds a lot more "harsh" and digital vs. the Micro).

 

John, do you still have both? I wonder if you have tried or can try the Nano, but with the GND disconnected, running on the internal battery - the glare goes away if I do this.

 

To AMR/iFi: is there anything we need to be aware of long-term if disconnecting the GND of the USB cable with the Nano or there's nothing to be worried about?

[AMR/iFi audio]

To AMR/iFi: is there anything we need to be aware of long-term if disconnecting the GND of the USB cable with the Nano or there's nothing to be worried about?

 

If USB ground or Vbus are disconnected on the iDSD nano, it will simply not work.

 

 

If it does work with the ground disconnected, it means there is an extra ground/earth in the system which is unwanted, unpredictable and may cause malfunction or damage if things go off the rail. So it is not recommended.

 

 

To disconnect USB ground (and Vbus) to prevent ground/earth loops in such a case, using the new iDefender is the safe and reliable way, or if you have na iUSB 3.0 micro or iUSB Power from iFi you can engage the isoEarth/isoGround switch which does the same as the iDefender.

[YashN]

If USB ground or Vbus are disconnected on the iDSD nano, it will simply not work.

 

Actually, only VBus disconnection causes the disconnection. If the internal battery is charged and only GND is disconnected, it works. Of course, only works while the battery has charge.

 

If it does work with the ground disconnected, it means there is an extra ground/earth in the system which is unwanted, unpredictable and may cause malfunction or damage if things go off the rail. So it is not recommended.

 

Wouldn't it be functioning because it is internally powered by the battery with an internal GND reference?

 

Thanks a lot for your response, I appreciate it.