44.1 or 48kHz?

[JSchmo_Bass]

Recently I did a read-up on Digital-Analog converters (DAC/ADC).

This a deep technical rabbit hole to fall down for sure, but also quite a burgeoning industry is being made selling high quality DACs to audiophiles to listen to high res streams of music. And studios whom we may send stems also have various systems in place. So it raises questions for the home producer...

Sigma Delta DACs- today’s most common- it seems do conversion between PCM sampling and DSD inside the chip. There is also oversampling, and filtering at various stages. Apparently as a result with these type of ADC/DAC, some sample rate frequencies may be “native” than others to a chip in that the generate less error/noise than others depending on the chip design with respect to the DSD part. Forgive me if I muck up a the engineering with the above.

While there seems to be much debate and hyperbole over these differences in the audiophile world, one thing that does seems to be generally agreed upon is that one ought to avoid format conversions as much as possible throughout the entire chain of signal from musician to ears.

That seems wise.

But it made me wonder: is one sample rate more “native” to the DP’s converters than another? Do anyone know if our DP machines perform or sound different or better at 48/24 vs 44.1/24? Or is it equally facile inside the box?

My natural inclination was to just work at 48/24, the video standard. because I assumed simply that a “little more must be a little better” and disk storage is cheap, and I had no intention of pressing CD’s.

I guess there are equally relevant questions about whether a DAW does processing with equal fidelity at different frequencies? it’s all just “math” but math includes errors and rounding. And whether the “distribution system” (which for me consists of BandCamp and about 5 followers, at least for now!) converts WAV files better at one setting vs another...? More rabbit holes to explore!

 

[thornev]

When recording, 48 is always better than 44.1. It's science. Capturing 48 thousand samples per second is better than 44.1. And it's better to have that quality all through the mixing process so that all your fudging of the sound is done with that quality. Keep it at 48 until the very end when you have to convert your master for CD which requires 44.1.

I'm not sure what you mean by "native". The hardware can process 48 or not. Maybe you're asking if our machines are upsampling? I hope not because that's crap. Filling in the holes with a guess is not quality to me. But then I've never done a scientific analysis about upsampling.

 

[JSchmo_Bass]

I thought the same as you.

My (very rudimenary) knowledge would have suggested that indeed if the ADC is operating purely in PCM sampling mode throughout, than I am hard pressed to see how this would in not fact be true...more samples per second = better ( and noise can be shifted further up spectrum).

But what I read (note high potential for error and misunderstanding!) is that most modern ADC and DAC chips first sample as DSD (a different style of sampling, MUCH faster but only a single bit) and then convert to PCM in order to have editable files.

In that conversion, I read, there can be other errors introduced, and those errors may vary depending on the various sample rates that are in play between the signal and the chip’s inherent process.

I doubt this is all such a big issue - all those sources also say that the rest of the circuits (and power supply and clock) matter at least as much as the actual DAC’s.

I was just more curious than anything. I love to nerd out on this... and afterall, the DP give us the choice and the documents say nothing about whether it matters.

(it does matter whether you choose 24 bit vs 16 bit for sure)

 

[Arjan P]

When recording, 48 is always better than 44.1. It's science. Capturing 48 thousand samples per second is better than 44.1.

That is true - if you are recording for dogs. For humans it makes no difference and I would say; if your end product will be at 44.1k, then start with 44.1k so no conversion at all needs to be done. When talking sample rate, you immediately have to think Nyquist frequency. In simple terms: if your goal is to have an expression range of 0 - 20kHz, your sample frequency needs to be at least twice that number. Both 44.1 and 48 kHz are fit for that purpose and I always use 44.1 because I hardly ever produce for video.

I did experiment with 88.2k and 96k since my equipment handles it, but I only heard a difference in a very quiet acoustic recording (it was guitar at the time). In conclusion I found the hassle of switching between sample rates not worth it (the DM mixer needs to restart after a SR change - with more consequences).

 

[thornev]

P - Good point. But some of us are dogs.

 

[-mjk-]

I work at 24b/48k. I mixdown at 24b/48k. What matters is how it sounds.

If you really want to go down a rabbit hole, study the effects of jitter on digital audio. That will keep you up at night!

 

[dctdct]

Been there, done that decades ago in my HiFi/HighEnd years.
My lesson was (after spending too much time and money in the jitter rabbit hole, and many more):
a) The higher the sampling rate, the higher the bad jitter effects against the high frequencies of the user signal
b) Less AD DA and format conversions reduce the chance to introduce jitter and other unwanted artifacts
c) What you see on a measuring instrument is there, true; but this doesn't mean you will also hear all of it at the end of your chain

For my purpouse, I really don't care anymore about a few % difference on this or that value of the chain. Compared to the digital technology in the eighties I am very comfortable with the devices we use today.

To quote @-mjk- :
What matters is how it sounds.

To quote @Arjan P :
...In conclusion I found the hassle of switching between sample rates not worth it

 

[Phil Tipping]

I did a simple measurement comparing 44 vs 48 https://www.tascamforums.com/posts/38005, and although the result is as expected, I hadn't realised it also impacts its use as a straight mixer, whether or not you're recording.

 

[JSchmo_Bass]

Hmm. so essentially the collective guidance is...

- Avoid unnecessary rate conversions but don’t obsess about them either, usually not audible.

Similarly any inherent 44.1 vs 48 difference while measurable is also marginal (mainly very high frequencies)

Generally it seems it circles back to choose a rate that fits your usual destination and otherwise don’t worry about it.

To anyone’s knowledge, there no inherent “preferred” sample rate in the DP’s chips set (if one has no other reason to select)?

 

[-mjk-]

I did a simple measurement comparing 44 vs 48 https://www.tascamforums.com/posts/38005, and although the result is as expected, I hadn't realised it also impacts its use as a straight mixer, whether or not you're recording.

But it does make sense given that the converters are immediately after the inputs.

 

[-mjk-]

@dctdct said:

For my purpouse, I really don't care anymore about a few % difference on this or that value of the chain. Compared to the digital technology in the eighties I am very comfortable with the devices we use today.

I couldn't agree more.

 

[Mark Richards]

Not much to add to this discussion that hasn't already been said. IMO the bit rate is much more important, since it affects headroom directly.

Also, as can be seen in this chart there isn't a heck of a lot going on above about 15kHz.

 

[Arjan P]

I did a simple measurement comparing 44 vs 48 https://www.tascamforums.com/posts/38005, and although the result is as expected, I hadn't realised it also impacts its use as a straight mixer, whether or not you're recording.

Phil, indeed your test shows as expected - and according to the Nyquist theorem. But what remains in the dark if we're talking about digital mixing, is the internal resolution of the machine. With multiple audio streams being mixed, the machine will have to calculate at a higher resolution - and this type of information is usually hard to find (since the DP allows 24-bit streams, my guess would be; at least 32-bit resolution). Ofcourse, the sample frequency is less relevant here, since it only tells us how many calculations need to take place per second (44,100 or 48,000).

Current DAWs for instance use 64-bit floating point calculation, which allows for great accuracy and also allows internal signals to go over 0 dB (that's due to the fp). BTW, this was also true with the previous 32-bit fp. It should be interesting to find out what these machines internally do.

 

[Arjan P]

Similarly any inherent 44.1 vs 48 difference while measurable is also marginal (mainly very high frequencies)

Much more important IMO - and that's why I stress it again - is the choice for 24 bit over 16 bit. Yes, it's 50% more data but it gives you so much more dynamic range (and the luxury of staying way clear from digital overs).

 

[skier]

Man, you guys really did go down this rabbit hole and it's a deep one. The problem is that the assertions made are all correct, but the requirements of what we do can make one better than another depending on the circumstances.

For example, starting with sample rate: converting from one sample rate to another is called "gear boxing", and when we go from one rate to another, we lose a tiny bit of fidelity and gain some noise. If we drop from a higher multiple, such as 88.2kHz to 44.1kHz, or 96kHz to 48kHz, the damage/noise is much less. Recording at 44.1kHz and staying there is better; or recording at 48kHz and staying there is equally good; so the best thing is to make no change in sample rate from recording to final product. That said, sample rate conversion software is better than ever and the conversions these days are usually transparent to most of us. I used to record at 44.1kHz, but changed to recording at 48kHz years ago when I was preparing music for video. As for the argument that humans don't hear higher than 20kHz, that has been my experience - I can't hear that high, and no one has yet passed a double-blind test that I've conducted. Yet, that doesn't mean that no one can hear higher, just that I haven't yet met them. Plus, 88.2kHz or 96kHz use up twice the disk space of 44.1 or 48kHz sampling. And I won't even talk about 176.4kHz or 192kHz sample rates.

As far as bit depth, a higher setting is an advantage for the reasons that @Arjan P mentioned above: it gives you much more dynamic range and you don't need to be so close to 0dB. When you mix the tracks, it can exceed odB for some parts of a cycle if you start too close to 0dB and digital distortion is really ugly. Many, if not most plug-ins actually process their effect at 32 bits to be able to use that headroom in the processing and stay below 0dB and above the noise, but their output goes back to the bit depth you selected whether 24 or 16 bits. 24bit will use up 50% more disk space, but most of us feel it's worth it for the extra dynamic range.

Now, all the foregoing notwithstanding, because conversion software is so much better these days as I mentioned above, disk space is so cheap, the fact most people these days listen to music in lower fidelity formats, such as MP3, and on lower fidelity speakers than studio monitors, unless you can hear the lower fidelity, it won't really matter as @mjk stated. With everything we do in recording and reproducing music, how it sounds is truly all the matters because no one but us will know, and no one will care what sample rate or bit depth we used, or whether we added dither or anything else - they'll only care about whether or not the song moves them. With that as my credo, I've found it liberating to spend less time on these concerns than I used to, especially because once I've changed the EQ (added frequency distortion), used distortion on my guitar (added harmonic distortion), used some phasing or flanging on an instrument (added phase distortion), used delay and reverb on some of the tracks (added amplitude, frequency, and time distortion), and used compression of many tracks, some sub mixes, and the final stereo mix to glue it all together better (added multiple forms of amplitude distortion all over the place), then why should I be worried about what sample rate I used because it's effect and sonics are so much lower in level that they're swamped below audibility by all the forms of distortion I've added - just something to think about.

If it sounds good, it is good.

 

[Sam Trenholme]

The Tascam DP-32SD and similar recorders use the AK5381 analog to digital converter. This is a variable rate A/D converter which can run from any frequency between 4kHz and 48kHz . Looking at the DP block diagram on page 39 of the DP service manual, it looks like the AK5381 is running in “slave” mode (which makes sense, since all four stereo A/D converters must run in parallel), so the A/D converter runs at a slightly slower rate when sampling at 44.1k instead of 48k. Looking at the data sheet, it appears that the pre-decimated input is a 2.8224mHz (44.1k) or 3.072mHz (48k) 2-bit signal. [1]

Point being, the A/D converters do not have a “native” sampling frequency; they are being externally clocked by the DP-32SD so will run equally well at 44.1kHz or 48kHz. [2]

[1] The converter could do “standard” DSD64 if the chip was changed internally to add a different decimator, but would not be able to do DSD128/256/512.

[2] The chip runs at 48kHz sample rate when run in “master” mode, which doesn’t appear to be (Edit: is not; the block diagram near the DAC shows that the A/D and D/A chips are given either a 22.5792MHz or 24.576MHz clock depending on whether we are in 44.1k or 48k mode. That would explain the click I get when I go to and from the tuner; the tuner may run only in 48k mode) the case with the DP-32SD.

 

[Noto]

Hi Sam,

Does the service manual state what type/brand of mic preamps are in the DP32sd?

 

[Sam Trenholme]

Looking at that block diagram, it looks like the preamps don’t use a chip (it’s simply marked “Audio in circuit”, without mention of a particular op amp being used). Most likely, it uses discrete surface mount components to amplify the signal as needed. (Edit: Not in the service manual, but the schematic shows that the op amp [i.e. preamp] is a BA4580RF-E2, which is, scanning over its datasheet and looking at its 25 cent price, an inexpensive clean transistor based design. There are no tubes or transformers in the A/D signal path, which is expected for this price point.)

As an aside, I speculate the reason why Behringer is making a bunch of 1970s analog synth designs is because those, by and large, can be made with discrete components on a surface mounted board without needing too many chips, so they are more practical to manufacture during the current COVID-19 chip shortage. So that’s why we got the 2600 (1971 design, discrete components) instead of any new polyphonic synths (lots of chips for the voices, assigning the voices, and patch storage).

 

[-mjk-]

Whatever the reason, I'm just happy there's a affordable ARP 2600 clone out there!

 

[JSchmo_Bass]

Hey Just wanted to thank you guys. Net concusion I’m going to continue to use 24/48.

My main distribution will be to upload my 24/48 WAV files to platforms like BandCamp... (one does wonder at the conversion and decimations that happen on these platforms..but that’s a moving target and probably irrelevant...and for those who really care, the downloads are free).

Sam do you also have the Service Manuals for the Model 12, 16 and 24? I am curious how the ADC compare. Has Tascam “upgraded” or stayed steady.

My ears have come to really appreciate the Model 12’s sound. I may just be my tastes but I really like it’s “sound”.

I have a mind to open them just to take a peek and a snap a picture (no touch for me, I know my limits)....