Review and Measurements of RME ADI-2 DAC

[dc655321]

I don't understand how Benchmark can do what they purport to do without closed-loop control on the outputs.
Apologies for the ignorance, but do most amp output stages operate open-loop?

 

[SIY]

Very few. Closed loop including the output stage is nearly universal.

 

[DonH56]

I don't understand how Benchmark can do what they purport to do without closed-loop control on the outputs.
Apologies for the ignorance, but do most amp output stages operate open-loop?

Yes, virtually all of them. The output stage is not all that linear without feedback, and achieving high performance (including very low output impedance) is impractical without feedback.

The anti-feedback bias these days is, to use a technical term, nuts. NOT saying that is what you meant! It continually amazes how marketing works to make old stuff better until it's not and must be new and improved...

 

[JohnPM]

That variation in distortion measurement at the output is usually due to back emf from the load. The lower the amp's output impedance the less it should appear, but it is from the load, not the source.

 

[xxie]

When we change the load to headphones, the distortion profile wildly changes unfortunately. So there is some secret sauce in Benchmark headphone amps that the RME does not have.

Benchmark referenced this paper: http://www.ti.com/lit/an/slyt630/slyt630.pdf, it propose a R/C network at amplifier frontend, figure 7 in the paper seems like mach Benchmark's amplifier behavior.

 

[Ron Kuper]

Seems like a dream DAC.

I wonder:
1. If it can be stackable for doing multichannel (2 units or more) using the USB ASIO driver, the ADI-2 Pro manual hints little about this capability ("multi-interface", if I'm not confusing it) and the ADI-2 DAC hint even less. Could then be just about the best Multichannel DAC/opAMP at about $800 less than the closest contender (exaSound e38) if only 5.1ch is needed (or $1800 less for 2.2ch), with much more features and probably a better headphone AMP(s, as you get more than one unit at this price).

2. How would the $599 iFi Micro iDSD BL be compared in both DAC and headphone AMP section. Benefits are a native DSD to headphones (unlike conversion to PCM as in the RME) and MQA support.
I would guess that channel balance would be less performant due to an analog volume control, compared with the DSP based one in the RME.

RME answered the first one in their forum -
https://www.forum.rme-audio.de/viewtopic.php?id=26975

No native stacking for the ADI-2 DAC, but possible with AES/SPDIF multichannel interface like the $257 USB MiniDSP U-DIO8.

Also this (from an email) -
Q: With AES (external clock) we are now trusting the external AES device's clock. In case it is of lesser quality compared to the internal ADI-2 one, would the SQ theoretically be lower than with the ADI-2 connected directly with USB or when configured to use internal clock?

A: No, due to Steadyclck, http://www.rme-audio.de/en/support/techinfo/steadyclock.php


Would be interesting to measure such a setup for multichannel.

 

[Sythrix]

The "diversion" is microscopic. Ohm's Law.

If the measuring instrument has an input of 1 megohm, which wouldn't be unusual, then the power going its way would be less than 0.01% even with the highest impedance 'phones, so nothing to worry about for the non-expert!

Notes taken. I still don't understand a lot but I'm sure that will change if I spend enough time here.

Fun and learning, right?

 

[Dro]

That variation in distortion measurement at the output is usually due to back emf from the load. The lower the amp's output impedance the less it should appear, but it is from the load, not the source.

What are the implications on what the headphones actually output? Does higher distortion coming from the load make the load even more distorted?

Edit: What I am trying to get at: Does it feed back? Ie headphone load feeds distortion into the electrical signal, will this then go back to affect the acoustic signal? Or is it just some signal we can measure, without any changes as to what actually comes out of the mechanical driver?

 

[RayDunzl]

What are the implications on what the headphones actually output? Does higher distortion coming from the load make the load even more distorted?

Hmm...

Well, just out of curiosity, based loosely on the above question...

Here is the right earpiece of my HD650 outputting a signal, as it "listens" to the stereo across the room playing the news on TV.

Speakers -> air -> HD650 -> UMC202HD (with input preamp turned on) -> PC

 

[mindbomb]

The confusing thing to me is how benchmark says it affects amplifiers even though they have low output impedance, when it seems output impedance is the only thing affecting this. Is the idea that some amps have higher output impedance at certain frequencies and the specifications are misleading? Also, is the reason that the grado headphones do particularly poorly due to high resistance on the cable?

 

[Dro]

The confusing thing to me is how benchmark says it affects amplifiers even though they have low output impedance, when it seems output impedance is the only thing affecting this. Is the idea that some amps have higher output impedance at certain frequencies and the specifications are misleading? Also, is the reason that the grado headphones do particularly poorly due to high resistance on the cable?

Impedance is a function of frequency. Doesn't mean it has to depend on frequency, but it can. If you have a capacitor in your output, you will get a higher impedance at low frequency, resulting in rolled off bass.

I think Grados are mainly distorted because of their design. It is their "house sound". Boosted mid bass, rolled of lower bass, boosted upper midrange, rolled off upper treble and tons of distortion.

 

[dc655321]

Benchmark referenced this paper: http://www.ti.com/lit/an/slyt630/slyt630.pdf, it propose a R/C network at amplifier frontend, figure 7 in the paper seems like mach Benchmark's amplifier behavior.

That paper is a gold mine. Thanks for pointing it out!

 

[amirm]

Benchmark referenced this paper: http://www.ti.com/lit/an/slyt630/slyt630.pdf, it propose a R/C network at amplifier frontend, figure 7 in the paper seems like mach Benchmark's amplifier behavior.

That is it! I had seen that link but somehow did not click on it. It fully explains the topic and solution. Here is the final THD graph matching the results we are seeing:

Thanks a bunch for posting it.

Interesting that both Benchmark and TI give this solution away but it doesn't seem to be implemented elsewhere.

 

[dc655321]

Interesting that both Benchmark and TI give this solution away but it doesn't seem to be implemented elsewhere.

I was going to ask how widespread the approach in this paper may be.
Pessimistically, are most headphone amp makers getting this wrong?

 

[Sythrix]

I was going to ask how widespread the approach in this paper may be.
Pessimistically, are most headphone amp makers getting this wrong?

Seconded. The levels we’re seeing here, even at this early stage of discovery, should have everyone asking this question. Still want to see more samples of various kinds, but I’d be lying if this whole thing isn’t making me look twice at all the amps I have.

Would also like to know if there are any unforeseen drawbacks to this method as well.

 

[mindbomb]

That is it! I had seen that link but somehow did not click on it. It fully explains the topic and solution. Here is the final THD graph matching the results we are seeing:

But in that case, the traditional amplifier had a 47.5 ohm output impedance. It doesn't explain how low output impedance amps do poorly at this test.

 

[Dro]

Seconded. The levels we’re seeing here, even at this early stage of discovery, should have everyone asking this question. Still want to see more samples of various kinds, but I’d be lying if this whole thing isn’t making me look twice at all the amps I have.

Could be that it has been considered insignificant by the manufacturers who know about this. Getting -60 dB THD from your amp does not add much to the -20 dB you get from your Grado already. But this is just me guessing.

 

[Blumlein 88]

I'd seen this too somewhere. I just assumed these expensive headphone amps would do this. Another case where an honest measurement site has lots to offer. You don't even have to know why something misbehaves to show it does. Test results are paying off.

Also looks to me like in some designs it would be possible to add The Rx/Cx compensation to an existing amp. I really think this topic of the TI paper (which incidentally quotes a John Siau paper as a source) is worthwhile. Looks like maybe Benchmark figured it out and cued in TI.

I reckon this is the Siau paper.
https://cdn.shopify.com/s/files/1/0321/7609/files/Headphone_Amplifier_Performance_-_Part_2.pdf?1361

 

[DonH56]

Adding an input zero/snubber circuit to prevent HF signals beyond the feedback's ability to compensate from going through the amplifier has been around "forever". I used the idea decades ago for a new differential amplifier circuit topology (hey, even hairy-knuckled engineers invent something now and then). I do not remember which ones, but some amps targeting ESLs included the same circuit to help compensate their HF capacitive load.

 

[nodoubt]

Mytek are ok, but then their designer publicly states that the dac’s performance can be improved with a linear power supply!
For me that rules them out straightaway.
Keith

you said that right...for 2 grand or more, a dac should come in at full fighting trim, primed and ready to go.
i understand something that costs a hundred bucks or so saying this..