Schiit (MultiBit) Bifrost

Thanks! I appreciate the corrections!

Schiit has said that they want to deliver the same samples to the DAC chip as they are delivered from the playback device. Normal sinc interpolation will not only change the samples (this is not a bad thing, but contrary to Schiit's goal, for better or for worse) but make it difficult to do a closed-form analysis to figure what it will actually deliver, on a sample-by-sample basis. Or flipping it around, being able to do a closed-form analysis is the only way they can guarantee that they'll deliver the original samples. I'm guessing that's why they do what they do.

Why would original samples deliver more space or soundstage or whatever? Now I'm really speculating in the weeds, but if one considers that the low-level information that tells us about an acoustic space lives in the lower bits of a sample (because it's usually softer than the music itself), then it *may* not be such a large leap to see how an interpolated sample could alter those cues since it will be lowest bits that change the most in an calculation. The comments above about precision are perhaps relevant to this? Anyway, just guesses, so please add salt!

 

I don't have Matlab. I'll have to adapt this code to Mathematica.

 

Even doing 4x up sampling, the original samples are conserved. Looks like better results are achieved if a second pass through the up-sample/FIR filter as opposed to doing a direct 4x up sampling tho. Additionally the frequency content is the pretty much the same in the first 1/M part of the spectrum where M is the amount up sampled, since the result is expanding the frequency spectrum to contain higher frequencies.



Perhaps an explanation cannot be achieved without divulging the secret sauce of the filter, but currently I'm not seeing how the Parks-McClellan filter design process is at fault. It simply tries to find the optimal Chebychev filter for a given pass band with ripple tolerance, and stop band with ripple tolerance.

 

Wait, is there a filter design out there called Porky-McDonalds? That would be the most awesome name ever.

 

Yeah, sure, and the next guy would name his something about burritos. Right. You're living in a fantasy land, join the real world.

 

Thinking about this more, the original samples ought to lie on the reconstructed curve because they're part of the curve that we're reconstructing. What Schiit claims is bit-for-bit identical. So I wonder what the difference is between the reconstructed samples from the P-M method vs. the original samples input into them, because there will be limits to the precision of the computation. In theory, it ought to be around or below the noise floor of the DAC, otherwise the differences would produce errors that rise above the noise floor. But it might be interesting to see what the difference is, and perhaps the FFT of the difference signal as well.

 

But then it's not a linear filter if you force the "non-interpolated" samples to the input ones. This could create low level distortions that you won't see on a Matlab plot but will still be at audible levels.

Can you run an anlaysis on your data to see if it's actually the case?

 

I'll be curious to see where this lands on Marv's DAC chart du awesome.

 

Hopefully now that all the DACs are done we can finally get the long fabled megacomboburrito filter white paper we've all been wanting.

Hopefully Mr Moffat will be able to fill in some of the blanks without giving away too much of the proprietary bits.

 

Just waiting for schiit to release the "mike" with 128000 taps

 

Has anyone been able to compare the Bifrost Multibit with the Gungnir Multibit using SE output? Particularly, I am interested how they compare with a EC BW as amp. Since BW is single ended, I was considering a DAC-19 in my future instead of a Gungnir Multibit, but now the Bifrost Multibit offers a fresh and equally cost efficient alternative

 

FWIW, shipped my Bifrost back for the MB upgrade today. When it returns I'll make it available to him to measure & compare to his Yggdrasil and Gungnir Multibit.

 

"Wish I had something like a DAC19 around to compare it again to, but the PCM1704 does have a very distinctive sound for better or worse. I'm pretty sure this DAC would sound better if it were put in a bigger chassis with lights, meters, and lead weights."

I've owned an AGD DAC19-DSP for several years now and I'm surprised at all the great reviews that the 10th Anniversary Edition is receiving.

It was my first DAC and while not awful it eventually took a DIY balanced power supply, vibration isolation, an OR5 using the Short Block, a decent digital coax cable and a coax attenuator to really make it sing.

When I first received it I was using a cheap DVD player for a transport and it didn’t sound much better than the analog outs of that DVD player.

Kingwa’s new USB/digital implementation must be much better than what came in my DAC19-DSP.

I’ve also read that the Master 7 and the new DAC19 use a different digital filter so perhaps that explains some of the difference.

My experience with sources makes me cringe whenever I hear recommendations for spending more money on phones and/or speakers before purchasing a better DAC and/or source.

Even the Mid-Fi HD-650s can show you things that you don’t want to see with your upstream equipment and for me once you see them they're hard to ignore.

Looking to buy a Bifrost Multibit for work as the Denon DVD-910/Magni 2 combo I'm using now can sound a little grating after an hour or two.

 

I am super curious how this compares with the Gungnir DS version. Wondering if it's worth paying the extra $200 over the Bifrost Multibit. I know the Gungnir has features like balanced output and VCXO reclocking, however is that enough to sound better then a MB Bifrost?

Edit:

Anyone look at the new Bifrost product photos?

http://schiit.com/products/bifrost

Why is there a straw coming out of the transformer?

 

That's the music straw. All the music comes through there. It's why it sounds so good.

 

Ha!

After studying it a bit I think it's attached to a grounding plane. Which probably means the transformer has some internal shielding now.