Digital Audio and Amplifier Noise Floor Comparison - Is 16bit/44.1kHz All We Need ???

[tmtomh]

56 minutes ago, Paul R said:

 

That's okay -  but you are now invited to provide the correct information.  I would expect the extra 48db of lower noise floor means the limiting noise floor would be the analog devices - i.e. the mic and pre-amplifier. Also, modern converters are only using about 20-22db of the 24bits. 

 

 

 

 

 

Well, the noise floor of 16-bit is -96dB (putting aside that -120dB is achievable in practice with the use of noise-shaping dither). If -96dB were 20% above the noise floor of most equipment, then most equipment would have a noise floor of -115 to -120dB (the former being 96 x 1.2 and the latter being 96 divided by 0.8).

 

Based on measurements of a lot of equipment at sites like audioscience review, very few DACs and virtually no amplifiers are capable of -115dB and pretty much none are capable of -120. And the vast majority of amps fall short of even -96dB, while for DACs -96dB is middle of the pack.

[Paul R]

3 hours ago, tmtomh said:

 

Well, the noise floor of 16-bit is -96dB (putting aside that -120dB is achievable in practice with the use of noise-shaping dither). If -96dB were 20% above the noise floor of most equipment, then most equipment would have a noise floor of -115 to -120dB (the former being 96 x 1.2 and the latter being 96 divided by 0.8).

 

Based on measurements of a lot of equipment at sites like audioscience review, very few DACs and virtually no amplifiers are capable of -115dB and pretty much none are capable of -120. And the vast majority of amps fall short of even -96dB, while for DACs -96dB is middle of the pack.

The Benchmark AHB2 Power Amp is rated at  132db. That's about 20 to 30 db quieter than a lot of very expensive reference amps. 

An iFi iDSD Pro is 119db. That's pretty good, but there are other DACs with better specs out there. 

 

 

 

 

 

 

 

[tmtomh]

6 hours ago, Paul R said:

The Benchmark AHB2 Power Amp is rated at  132db. That's about 20 to 30 db quieter than a lot of very expensive reference amps. 

An iFi iDSD Pro is 119db. That's pretty good, but there are other DACs with better specs out there. 

 

 

 

 

 

 

 

 

The Benchmark amp is an excellent piece of kit, but I would not necessarily believe -132dB:

 

https://www.audiosciencereview.com/forum/index.php?threads/review-and-measurements-of-benchmark-ahb2-amp.7628/

 

 

[Shadders]

Hi,

The specification is A-weighted, so please correct if wrong, removes noise due to the filter. The ASR review probably does not use A-weighting.

Regards,

Shadders.

[mansr]

2 hours ago, tmtomh said:

The Benchmark amp is an excellent piece of kit, but I would not necessarily believe -132dB:

There's probably some creativity involved in arriving at that figure.

[pkane2001]

1 minute ago, mansr said:

There's probably some creativity involved in arriving at that figure.

 

For unweighted 20-20kHz SNR the spec says 130dB. Here's what Amir measured:

 

 

[mansr]

10 minutes ago, pkane2001 said:

For unweighted 20-20kHz SNR the spec says 130dB. Here's what Amir measured:

 

THD+N according to Amir is 113 dB. Still very impressive.

[pkane2001]

4 minutes ago, mansr said:

THD+N according to Amir is 113 dB. Still very impressive.

 

THD+N is better than any of my DACs

 

[Paul R]

3 hours ago, tmtomh said:

 

The Benchmark amp is an excellent piece of kit, but I would not necessarily believe -132dB:

 

https://www.audiosciencereview.com/forum/index.php?threads/review-and-measurements-of-benchmark-ahb2-amp.7628/

 

 

 

[The Computer Audiophile]

1 minute ago, Paul R said:

 

So @tmtomh was right 😁

[Paul R]

4 minutes ago, The Computer Audiophile said:

So @tmtomh was right 😁

130+ vs 133 - yep. 

 

The 24 bit noise floor is 148db,  so instead of 20db under, it is 18db under. 

[lucretius]

The Benchmark ABH2 is remarkable.  Now, if you could put it, along with the HPA4 and DAC3 in jewellery casework ... Oh but to dream.

[psjug]

6 minutes ago, Paul R said:

130+ vs 133 - yep. 

 

The 24 bit noise floor is 148db,  so instead of 20db under, it is 18db under. 

 

Best measuring amplifier you can find is the same as "most equipment"?

[Paul R]

4 minutes ago, psjug said:

 

Best measuring amplifier you can find is the same as "most equipment"?

 

You are welcome to pick anything you want to measure by.  All I said was that 24bit is below the noise floor of most equipment, and 16bit is above same. Exactly how much is dependent upon equipment chosen, but in general, that holds true. 

 

I am unclear exactly what your objection is. Here are a couple other amps. 

 

The now $379 Outlaw Audio M2200 amp is 112db A weighted, and still falls within that criteria. 

The $1700  Peachtree Nova 150 I use a lot is 105db A weighted. 

 

[psjug]

24 minutes ago, Paul R said:

 

You are welcome to pick anything you want to measure by.  All I said was that 24bit is below the noise floor of most equipment, and 16bit is above same. Exactly how much is dependent upon equipment chosen, but in general, that holds true. 

 

I am unclear exactly what your objection is. Here are a couple other amps. 

 

The now $379 Outlaw Audio M2200 amp is 112db A weighted, and still falls within that criteria. 

The $1700  Peachtree Nova 150 I use a lot is 105db A weighted. 

 

I just thought it was interesting that you used spec from the Benchmark amplifier which was about where you said ordinary equipment should measure.

 

As for your own amplifier, the A weighting is likely to be giving a big improvement in the spec since it's a class D amplifier.  Not to be down on your amp, just to point out what is realistic.  If you don't care about the A-weighting of the amp spec, then you shouldn't care about noise shaping with 16 bit to get much better SNR.

 

Edit: In other words, I suggest that if you are using A-weighted spec to try to match equipment to 16-bit audio, then you should use the effective SNR for dithered 16-bit, so something like 115 to 120 dB.

[Paul R]

31 minutes ago, psjug said:

I just thought it was interesting that you used spec from the Benchmark amplifier which was about where you said ordinary equipment should measure.

 

As for your own amplifier, the A weighting is likely to be giving a big improvement in the spec since it's a class D amplifier.  Not to be down on your amp, just to point out what is realistic.  If you don't care about the A-weighting of the amp spec, then you shouldn't care about noise shaping with 16 bit to get much better SNR.

 

Edit: In other words, I suggest that if you are using A-weighted spec to try to match equipment to 16-bit audio, then you should use the effective SNR for dithered 16-bit, so something like 115 to 120 dB.

 

A weighted specs are often all that is easily available, but that is a good point. What would you suggest are figures you would be happier with? I generally prefer A weighted specs myself.

 

However,  it does not follow that noise shaping to boost the 16bit noise floor is the same thing, or even related. I think you are conflating two different subjects there.

 

That boost to 120db in 16bit PCM by “noise shaping” has other costs. Depending upon the exact scheme, you get degradations in coherency and an unstable noise floor. Techniques commonly used to restore that, such as windowing, can boost the amplitude of some frequencies and have other effects. 

 

In short, you are hearing a significantly modified sound, whereas those issues do not exist if you simply move to 24bit. I find it rather silly to defend 16 bit recording, once you take that into account. 

 

[Edit - perhaps you could be a little more clear on the technical point you are trying to make? ]

[psjug]

19 minutes ago, Paul R said:

 

A weighted specs are often all that is easily available, but that is a good point. What would you suggest are figures you would be happier with? I generally prefer A weighted specs myself.

 

However,  it does not follow that noise shaping to boost the 16bit noise floor is the same thing, or even related. I think you are conflating two different subjects there.

 

 

I agree that my numbers might not make sense.  I meant more to make the general point.

 

I don't worry about it so much myself.  If I could afford the benchmark I would buy it, just because it is a great amp not because I think I need 130dB.  And if I owned that amp I wouldn't then go out and seek 24 bit audio; I'd still be happy with 16-bit.

[Shadders]

22 minutes ago, psjug said:

 

I agree that my numbers might not make sense.  I meant more to make the general point.

 

I don't worry about it so much myself.  If I could afford the benchmark I would buy it, just because it is a great amp not because I think I need 130dB.  And if I owned that amp I wouldn't then go out and seek 24 bit audio; I'd still be happy with 16-bit.

Hi,

If you buy Bob Cordell's new book, there is an example design in chapter 4, with all required ancillary circuits, which has 125watts in 8ohms, and 1kHz THD is 0.0008%, and 20kHz THD is 0.004%, close to the Benchmark. Although the S/N is 100dB A-weighted.

Regards,

Shadders.

[tmtomh]

2 hours ago, Paul R said:

 

 

There are multiple ways to measure a noise floor - the traditional SNR as noted above, and the THD-N/SINAD as I'd noted previously.

 

Moreover, the effective noise floor of CD is -120dB with noise-shaping dither.

 

I'm really not trying to win an argument for the sake of winning an argument. My point simply is that one has to go to some lengths to put together an equipment chain that can top the effective noise floor of a properly produced 16-bit CD/digital file, and therefore it's not accurate that 16-bit's noise floor is generally above that of most equipment, even most decent equipment.

 

I am hopeful that in the future, the specs of the very best Hypex-based amps like the Benchmark, along with the specs of the very best DACs, will become more common, especially in more affordable equipment. But for now, the majority of amps and DACs don't have noise floors that reliably or significantly exceed that of properly dithered 16-bit sources. (And that's putting aside the fact that it's almost impossible to be in a practical listening space where a -96dB signal can be heard when the volume level is set so that peaks are not at ear-damaging levels.)

[Paul R]

9 minutes ago, tmtomh said:

 

There are multiple ways to measure a noise floor - the traditional SNR as noted above, and the THD-N/SINAD as I'd noted previously.

 

Moreover, the effective noise floor of CD is -120dB with noise-shaping dither.

 

I'm really not trying to win an argument for the sake of winning an argument. My point simply is that one has to go to some lengths to put together an equipment chain that can top the effective noise floor of a properly produced 16-bit CD/digital file, and therefore it's not accurate that 16-bit's noise floor is generally above that of most equipment, even most decent equipment.

 

I am hopeful that in the future, the specs of the very best Hypex-based amps like the Benchmark, along with the specs of the very best DACs, will become more common, especially in more affordable equipment. But for now, the majority of amps and DACs don't have noise floors that reliably or significantly exceed that of properly dithered 16-bit sources. (And that's putting aside the fact that it's almost impossible to be in a practical listening space where a -96dB signal can be heard when the volume level is set so that peaks are not at ear-damaging levels.)

 

Where we disagree is that 16 bit noise floors are below the noise floor of most equipment. I think you are simply wrong there, 16bit is well above the noise floor for most equipment these days.  And of course 16bit noise shaping adds some peculiar distortions not present at all in 24bit.

 

[Rt66indierock]

9 minutes ago, Paul R said:

 

Where we disagree is that 16 bit noise floors are below the noise floor of most equipment. I think you are simply wrong there, 16bit is well above the noise floor for most equipment these days.  And of course 16bit noise shaping adds some peculiar distortions not present at all in 24bit.

 

 

Find me some amps with a S/N ratio of 96 dB at one watt of output. The quick and dirty method to estimate this is your numbers minus 20 dB. I don't think you will find many.

[esldude]

A-wtd is usually a 4-6 db improvement over no weighting.  So it isn't like unweighted results are suddenly terrible.  More common is fudging on real values or some trickery in gain staging that gives a number you won't see in normal use.  As Rt66indierock said, your noise floor is much closer to the signal at 1 watt levels.  Of course with common speakers you only need mid-80 db SNR to be so low it will never be heard at 1 watt.  Plus your SNR with dithered 16 bit is worse by a half bit to bit.  So rather pointless to engage in petty arguments without putting all this in context. 

 

I've found it interesting no one ever points out that with shaped dither and 16 bit audio there is a real benefit to high sample rates.  Here is where I took silence, and saved it in 16 bit with shaped dither in both 44/16 and 192/16.  I then filtered out everything above 20 khz in the 192 file.  Notice the shaped 44 has a noise level of -84 db while the filtered 192 is -120 db.  This is what would happen if your speakers (and your ears) don't respond above 20 khz.  Of course no one does 192 at 16 bit do they.  Getting back on topic, something like this could have been done with MQA without requiring licensing, messing with undecoded fidelity or causing other problems.  

 

[Paul R]

1 hour ago, Rt66indierock said:

 

Find me some amps with a S/N ratio of 96 dB at one watt of output. The quick and dirty method to estimate this is your numbers minus 20 dB. I don't think you will find many.

 

I believe both the Benchmark and maybe the Outlaw monoblocks meet that criteria.

[Rt66indierock]

12 minutes ago, esldude said:

A-wtd is usually a 4-6 db improvement over no weighting.  So it isn't like unweighted results are suddenly terrible.  More common is fudging on real values or some trickery in gain staging that gives a number you won't see in normal use.  As Rt66indierock said, your noise floor is much closer to the signal at 1 watt levels.  Of course with common speakers you only need mid-80 db SNR to be so low it will never be heard at 1 watt.  Plus your SNR with dithered 16 bit is worse by a half bit to bit.  So rather pointless to engage in petty arguments without putting all this in context. 

 

I've found it interesting no one ever points out that with shaped dither and 16 bit audio there is a real benefit to high sample rates.  Here is where I took silence, and saved it in 16 bit with shaped dither in both 44/16 and 192/16.  I then filtered out everything above 20 khz in the 192 file.  Notice the shaped 44 has a noise level of -84 db while the filtered 192 is -120 db.  This is what would happen if your speakers (and your ears) don't respond above 20 khz.  Of course no one does 192 at 16 bit do they.  Getting back on topic, something like this could have been done with MQA without requiring licensing, messing with undecoded fidelity or causing other problems.  

 

 

In my defense, I'm going to keep using my Klipsch Heresy's at home so I need something quiet. John Siau of Benchmark recently wrote and made some graphs about how quiet your stuff has to be to hear certain bit depths. 

[Paul R]

27 minutes ago, esldude said:

A-wtd is usually a 4-6 db improvement over no weighting.  So it isn't like unweighted results are suddenly terrible.  More common is fudging on real values or some trickery in gain staging that gives a number you won't see in normal use.  As Rt66indierock said, your noise floor is much closer to the signal at 1 watt levels.  Of course with common speakers you only need mid-80 db SNR to be so low it will never be heard at 1 watt.  Plus your SNR with dithered 16 bit is worse by a half bit to bit.  So rather pointless to engage in petty arguments without putting all this in context. 

 

I've found it interesting no one ever points out that with shaped dither and 16 bit audio there is a real benefit to high sample rates.  Here is where I took silence, and saved it in 16 bit with shaped dither in both 44/16 and 192/16.  I then filtered out everything above 20 khz in the 192 file.  Notice the shaped 44 has a noise level of -84 db while the filtered 192 is -120 db.  This is what would happen if your speakers (and your ears) don't respond above 20 khz.  Of course no one does 192 at 16 bit do they.  Getting back on topic, something like this could have been done with MQA without requiring licensing, messing with undecoded fidelity or causing other problems.  

 

 

Well said. Did you get the typical “bathtub” output? 

 

I just find find it hard to argue that 16bit, even with noise shaping is preferable to 24 bit. Equipment specs are only going to keep getting better and better.