Schiit Magni Mesh DAC Review

[VientoB]

Amir actually wrote this about the Magni Unity (a listening test)

Yeah, I guess that's the amp reaching it's limit.

 

[rebbiputzmaker]

Amir actually wrote this about the Magni Unity (a listening test)

Well that was pretty accurate. Clipping sounds bad, another amplifier has more power and is cleaner. That’s very informative.

 

[solderdude]

Possible as 118dB peaks would be the limit the Magni - E3 combo can hit.
The DX5II in S.E. mode has the same output power level though.

 

[rebbiputzmaker]

Possible as 118dB peaks would be the limit the Magni - E3 combo can hit.
The also mentioned DX5II in S.E. mode has the exact same output power level though.

That’s actually not good at all. Headroom is important, clean amplifiers are important. More transducer damage is done by amplifier clipping than too much power.

 

[solderdude]

More transducer damage is done by amplifier clipping than too much power.

For speakers there are cases where that may happen but not so with headphones.
They get damaged by too much power or a DC voltage.

 

[VientoB]

That’s actually not good at all. Headroom is important, clean amplifiers are important. More transducer damage is done by amplifier clipping than too much power.

Nah, too much power is too much power.

 

[rebbiputzmaker]

For speakers there are cases where that may happen but not so with headphones.
They get damaged by too much power or a DC voltage.

I do own headphones, and iems but I consider myself an occasional headphone user. Speaker much more. That was what I was thinking about mostly. For me it’s hard to imagine playing headphones that loud and long to damage them, without developing serious hearing loss eventually.

 

[Joe Smith]

I have the Magni 2 and Magni 3 amps, they work fine for “sane” listening levels. I think the Magni amps have always been aimed at a mid-range user, not people who have very power hungry cans or those who want to play at very high “bleeding” volumes.

 

[Csaba Kelemen]

Well, it’s not a disaster and most performance is respectable, but the USB noise interference raises eyebrows. Price is low-ish but performance on the whole isn’t competitive with the best in its price class. Plus, Schiit has shown they have engineering chops to make excellent gear—just not here. Taking all of the above into account, realizing the deficiencies will probably not be audible unless you have a really noisy USB source, I’m voting “Fine.” However, it is an almost “Not Terrible.” Really on the line here.

Technically, it is very necessary to increase the number of taps. AKM and ESS chips only work with a few hundred, Mesh 18k. The 16-bit precision signal reconstruction of CD PCM requires mega taps.
Amirm does not measure the bit accuracy of reconstruction, it really should.

 

[Matias]

Technically, it is very necessary to increase the number of taps. AKM and ESS chips only work with a few hundred, Mesh 18k. The 16-bit precision signal reconstruction of CD PCM requires mega taps.
Amirm does not measure the bit accuracy of reconstruction, it really should.

Actually it seems all these extra taps are not helping much with low level resolution...

Example Topping DX5 II (ESS ES9039Q2M).

SMSL SU-1 (AKM AK4493S).

 

[Csaba Kelemen]

Actually it seems all these extra taps are not helping much with low level resolution...

Example Topping DX5 II (ESS ES9039Q2M).

SMSL SU-1 (AKM AK4493S).

I mean the difference between the ideal complex signal and the reconstructed one. This cannot be measured directly, but it would be telling.

 

[VientoB]

I mean the difference between the ideal complex signal and the reconstructed one. This cannot be measured directly, but it would be telling.

Why would it be telling?

 

[Csaba Kelemen]

Why would it be telling?

Because this is the most accurate characterization of signal fidelity, that is, the maximum of signal deviation.
It would show the justification for increasing the tap number.

 

[VientoB]

Because this is the most accurate characterization of signal fidelity, that is, the maximum of signal deviation.
It would show the justification for increasing the tap number.

Would it be audible? If it doesn't show up in the usual measurements why would it be audible?

 

[Csaba Kelemen]

Would it be audible? If it doesn't show up in the usual measurements why would it be audible?

Better spatial perception, more natural sound.
But that's not the point. The fundamental problem in principle is that the DAC error is not characterized by a general and expected value in technical/scientific life, that is, the maximum deviation from the ideal. Which could be expressed very clearly here by the actual bit resolution. This can be 4-5 bits for ESS/AKG chips with CD PCM input, Mesh maybe 8-10 and Dcs and some Chord systems reach 16, the value that the Compact Disc marketing promised back in 1983.
The SINAD value is misleading because it does not show the phase error.

 

[amirm]

Better spatial perception, more natural sound.

Those aspects are in recording and somewhat modified by the room/speaker. A DAC has nothing to do with them, much less their filtering.

 

[amirm]

The SINAD value is misleading because it does not show the phase error.

What phase error?

As to filtering, I show that with extreme resolution:

Noise floor which represents the type of thing you are talking about, is only average here. Its cut-off slope is also average like any IC filter rather than a high-tap FIR as used by likes of Chord.

 

[Csaba Kelemen]

Those aspects are in recording and somewhat modified by the room/speaker. A DAC has nothing to do with them, much less their filtering.

As I know, the task of filtering is to reconstruct the signal shape, by interpolation. If this is not good enough, the spatial sense and naturalness deteriorate. The theory requires Whittaker-Shannon interpolation and a sufficiently good approximation requires a lot of computation.
Excellent SINAD values can also be achieved with less computing power, but it is misleading to conclude from this that the signal reconstruction is accurate.
At least, this is the conclusion I came to while studying PCM.
Can you please point out where I am wrong?

 

[kyuu]

As I know, the task of filtering is to reconstruct the signal shape, by interpolation. If this is not good enough, the spatial sense and naturalness deteriorate. The theory requires Whittaker-Shannon interpolation and a sufficiently good approximation requires a lot of computation.
Excellent SINAD values can also be achieved with less computing power, but it is misleading to conclude from this that the signal reconstruction is accurate.
At least, this is the conclusion I came to while studying PCM.
Can you please point out where I am wrong?

Apparently you never came across Shannon-Nyquist while you were studying. I suggest watching this video:

 

[amirm]

As I know, the task of filtering is to reconstruct the signal shape, by interpolation.

That's not what it does. It is a simple low pass filter to get rid of high frequency artifacts. There is no intelligence to shape anything. Most of the time, the job is not interpolation but sampling down.