Twisted Pear Buffalo DAC

[Torpedo]

I have noticed on needled drops people have sent me the 24/96 usually sounds better since it is recorded directly from the soundcard. Whereas going to 16/44 requires some downconverting algorithm. There are surprising differences in SQ between different algorithms.

Not just that. Try recording a vinyl or straight with a microphone into the computer or a decent pro recorder, using different sample rates and bit depths, and reproduce them in a good system from the own recording device to avoid conversion down to 16/44.1

IME more important than sampling rate is bit depth. Just 20/44.1 is a nice improvement over standard RB, and to my ears, the difference from RB to 24/44.1 is way more interesting and noticeable than going to 16/96.

The problem I see on those upsampling and upconverting DACs is that they cannot invent information that's lacking, so they may be good to use filters off the audible range, but not sure that's really an interesting addition to the sound obtained using "normal" DACs

[grawk]

Not with analog. An analog volume control reduces the analog signal and the digital bits would remain the same. A digital volume control does cause a bit loss.

I didn't mean "bits" as in 0 or 1, I meant bits as in, little pieces. If you can't hear it, it's not there. Analog or digital.

[grawk]

Not just that. Try recording a vinyl or straight with a microphone into the computer or a decent pro recorder, using different sample rates and bit depths, and reproduce them in a good system from the own recording device to avoid conversion down to 16/44.1

IME more important than sampling rate is bit depth. Just 20/44.1 is a nice improvement over standard RB, and to my ears, the difference from RB to 24/44.1 is way more interesting and noticeable than going to 16/96.

The problem I see on those upsampling and upconverting DACs is that they cannot invent information that's lacking, so they may be good to use filters off the audible range, but not sure that's really an interesting addition to the sound obtained using "normal" DACs

That's a lot more complicated than the bits available. At record time, more bits is helpful because it lets you keep the analog portion of the recording process in the sweet spot while leaving enough headroom that you don't run out of bits. Take the resulting file, and normalize it, and then convert it to 16 bits and you'll get exactly the same sound. 24 bits is a HUGE benefit for the recordist, it's just not necessary post-mastering.

[Dreadhead]

I have noticed on needled drops people have sent me the 24/96 usually sounds better since it is recorded directly from the soundcard. Whereas going to 16/44 requires some downconverting algorithm. There are surprising differences in SQ between different algorithms.

Or because some DACs do weird things to make the two sound different. The chord DAC64 is a good example of this (at least to my friends ears since he sold it afterward).

Truncation from 24 to 16 should be like this (just an example):

24bit: .9230232332102301239123091233120931209

16bit: .923023233210230123912309

Sample rate conversion on the other hand can be a bit of a problem and there are different theories on how to do this that have their own problems. In the end each is an interpolation.

There have been tests done where people have taken a 24/96 stream and created a DBT box that either left it alone or resampled and trunctated it (correctly) and no one could tell the difference. Actually if I remember correctly someone could tell but he admitted it was because the "click" when the stream engaged was slightly different not anything to do with the music.

[Dusty Chalk]

I didn't mean "bits" as in 0 or 1, I meant bits as in, little pieces. If you can't hear it, it's not there. Analog or digital.

But if the gain of your power amp is extremely high, such that you have to turn the "pre" way down, the analog won't lose any information, whereas the digital will.

The problem I see on those upsampling and upconverting DACs is that they cannot invent information that's lacking, so they may be good to use filters off the audible range, but not sure that's really an interesting addition to the sound obtained using "normal" DACs

Yes and no -- a lot of people are digitizing their own analog these days, so there's your source. Also, companies like Reference Recordings and 2L and Chesky and ECM haven't totally given up on high-res digital -- cf. "HRx". Build it, and they will come, I say optimistically.

[grawk]

I still disagree. If you turn it down to where you would have lost information in digital, you'll have lost information in analog, because you won't be able to hear it.

[Dreadhead]

I still disagree. If you turn it down to where you would have lost information in digital, you'll have lost information in analog, because you won't be able to hear it.

Yup. When stuff is below the noise floor it's below the noise floor and that's it.

[Torpedo]

That's a lot more complicated than the bits available. At record time, more bits is helpful because it lets you keep the analog portion of the recording process in the sweet spot while leaving enough headroom that you don't run out of bits. Take the resulting file, and normalize it, and then convert it to 16 bits and you'll get exactly the same sound. 24 bits is a HUGE benefit for the recordist, it's just not necessary post-mastering.

I didn't mean otherwise. Just saying that when you are using as your source a standard RB shinny disc, any bit depth or higher sampling rate used during the recording stage, won't come back no matter how much upsampling and upconverting you use in the process of conversion to analog.

OTOH if we had the chance to reproduce 24/88.1 files which were recorded originally that way directly, sound would be much better than RB, even if the DAC used isn't top notch. IMHO post processing of a 16/44.1 file can't do miracles, just "sound shaping".

[Dusty Chalk]

Truncation from 24 to 16 should be like this (just an example)...There have been tests done where people have taken a 24/96 stream and created a DBT box that either left it alone or resampled and trunctated it (correctly) ...

I'm curious as to your usage of the word "should" here -- you do know that in audio, when you reduce your bit depth, you need to dither it down, not just truncate it. You've heard of quantization error, yes?

24bit: .923023233210230123912309999999999999

16bit: .923023233210230123912309

That would be truncation, but I dare say that would not be what you would want. And it's not just simply rounding, since you're dealing with frequency components.

I mean, there are whole fields of study as to the correct dither algorithm to use when going from a higher bit depth to a lower one -- Sony was promoting the snot out of their "SBM" technology before they came up with DSD.

[grawk]

I disagree 100%. Especially when you start with 88.2/24. You could rather easily produce 44/16 that is indistinguishable in post. Sample rate conversion algorythms have a lot more trouble with 96->44 than 88->44, because with 88/44 they just drop half the samples. You lose the freqs over 22khz, but you can't hear those anyway. And the bits from 24->16 are just headroom unless you're listening over 96db.

[Dusty Chalk]

I still disagree. If you turn it down to where you would have lost information in digital, you'll have lost information in analog, because you won't be able to hear it.

You're not understanding my example -- I specifically set up a situation where you amplified it back up into the hearing range. In the analog case, you won't have lost information; in the digital case, you will have.

[grawk]

Dusty, the dithering happens with the least significant bit. Yes, if you just truncate, you can get problems, but that's why you normalize first, then use noise shaping. At -96db, you're not going to hear the noise shaping, or anything else for that matter.

[grawk]

Ok, yes, if you turn the volume down digitally, then turn it back up in the analog realm, you can lose information. If you're doing that, tho, you're a dope.

[Torpedo]

Just try it yourself man, there's a clear difference. Make your own recording from a pure analogue source and use different sample rates and bit depths, then listen with no downconversion.

[grawk]

I've done it myself a lot. I used to do a lot of location sound recording. Sample rate conversion can be heard, bit depth changes can't, if done with any care at all.

[Torpedo]

And you didn't notice any difference from having recorded the very same signal at 16/44 than at 24/88? Surprising, that doesn't match my experience at all.

[grawk]

I didn't do 24/88, I did 24/44, because I'm not a bat. I do it sometimes for the theoretical advantage, but in the field resources are scarce, so I didn't capture data I didn't need. Bit depth I needed, higher frequencies than 22k I didn't.

It's theoretically possible in dac designs with no brickwall filter at 22khz (most have them) you might hear a difference, but I doubt it. But you're really not going to hear anything 96db below the loudest point recorded unless you're playing that loudest point recorded at 120dB.

[Dusty Chalk]

You lose the freqs over 22khz, but you can't hear those anyway.

I think we're going to have to agree to disagree. Not only do I disagree with your conclusions, you're beginning to introduce new opinions with which I disagree, and I'm going to start calling you names, too.

(I've long been a proponent of the opinion that just because you can't hear individual frequencies above 20kHz, that doesn't mean it doesn't change the nature of what you can hear audibly.)

Also, I don't think you understand my "resolution" vs. "dynamic range" argument at all.

[grawk]

Dusty, I agree with you in theory, but most dacs filter that off anyway.

And I apologize for calling you a dope. That's just shorthand for "doing something that is seriously non-optimal, whether done in the digital or analog realm.

[Dusty Chalk]

I'm not worried about it.

[Torpedo]

I didn't do 24/88, I did 24/44, because I'm not a bat. I do it sometimes for the theoretical advantage, but in the field resources are scarce, so I didn't capture data I didn't need. Bit depth I needed, higher frequencies than 22k I didn't.

It's theoretically possible in dac designs with no brickwall filter at 22khz (most have them) you might hear a difference, but I doubt it. But you're really not going to hear anything 96db below the loudest point recorded unless you're playing that loudest point recorded at 120dB.

Theoretically you're absolutely right and there's no rational advantage in using a higher sampling rate. However you'd have to agree that having twice the points to reconstruct the analog signal should make it kind of "smoother" at any frequency in the audible range. That which is just an intuitive observation and which may be wrong, IME makes an audible difference, and I'm not a bat either. Maybe the maths used to reconstruct a 10KHz wave from 4 sampled points aren't as perfected as they are to reconstruct that same wave using 8 samples. In any case I insist that the bigger difference I noticed for making the digital recording closer to the vinyl used, was by increasing the bit depth more than the sampling rate, despite it was noticeable, IMO wasn't as evident.

[Dreadhead]

I'm curious as to your usage of the word "should" here -- you do know that in audio, when you reduce your bit depth, you need to dither it down, not just truncate it. You've heard of quantization error, yes?

24bit: .923023233210230123912309999999999999

16bit: .923023233210230123912309

That would be truncation, but I dare say that would not be what you would want. And it's not just simply rounding, since you're dealing with frequency components.

I mean, there are whole fields of study as to the correct dither algorithm to use when going from a higher bit depth to a lower one -- Sony was promoting the snot out of their "SBM" technology before they came up with DSD.

True enough I didn't discuss noise shaping to prevent the harmoinc rounding but as Grawk pointed out it's on the inaudible LSB.

[Dreadhead]

Theoretically you're absolutely right and there's no rational advantage in using a higher sampling rate. However you'd have to agree that having twice the points to reconstruct the analog signal should make it kind of "smoother" at any frequency in the audible range. That which is just an intuitive observation and which may be wrong, IME makes an audible difference, and I'm not a bat either. Maybe the maths used to reconstruct a 10KHz wave from 4 sampled points aren't as perfected as they are to reconstruct that same wave using 8 samples. In any case I insist that the bigger difference I noticed for making the digital recording closer to the vinyl used, was by increasing the bit depth more than the sampling rate, despite it was noticeable, IMO wasn't as evident.

Well then people were doing the bit depth conversion wrong (maybe the normalization?) because the noise floor on the LP is well above anything in that the 24to16 bit truncation should see.

[grawk]

Theoretically you're absolutely right and there's no rational advantage in using a higher sampling rate. However you'd have to agree that having twice the points to reconstruct the analog signal should make it kind of "smoother" at any frequency in the audible range. That which is just an intuitive observation and which may be wrong, IME makes an audible difference, and I'm not a bat either. Maybe the maths used to reconstruct a 10KHz wave from 4 sampled points aren't as perfected as they are to reconstruct that same wave using 8 samples. In any case I insist that the bigger difference I noticed for making the digital recording closer to the vinyl used, was by increasing the bit depth more than the sampling rate, despite it was noticeable, IMO wasn't as evident.

I'll still argue the difference was at record time, and in the sample rate conversion done. The final sample rate and bit depth isn't going to be the significant factor. The way frequencies are recreated is based on sine waves already, so even tho the digital representation of it is smoother, the final output will be the same. Where it's possible you'll get differences is from harmonic distortion introduced by frequencies greater than 22khz but it would take a hell of a signal path to recreate those correctly anyway.

I'd guess that the variability you hear is in the DAC design, and the optimization for specific sample rates. Chances are it's better at some frequencies than others, and that's the improvement you're hearing.

Edited January 13, 2009 by grawk

[Dreadhead]

Dusty, I agree with you in theory, but most dacs filter that off anyway.

And I apologize for calling you a dope. That's just shorthand for "doing something that is seriously non-optimal, whether done in the digital or analog realm.

We are a low pass filter (with weird response and varying effectiveness). A low pass filter does not care what it's blocking out it just blocks it out. A lot of dacs have a brick wall filter on them because you don't want high frequency noise output into op-amps that have problems with them.

Sample rate down-sampling has to be done exceedingly carefully but most just use a higher order interpolation and then a brick wall to prevent the noise leakage and of course noise shaping or dither.