Bit-depth info / question
- Thread starter aughii
- Start date
From all that I have read about digital sound in the last 6years or so more bits does not result in better resolution or accuracy or less distortion, merely a lower noise floor (and 16bits is dithered, 24bits need not be AFAIK).
The BBC's FM signals have been distributed over PCM links for many years at, I believe, about 13.5 bits (it is non-linear). That is entirely adequate for FM stereo S/N and I have never heard anyone complain of R3 FM sound quality! There are many who say DAB is not as good.
Dave.
The BBC's FM signals have been distributed over PCM links for many years at, I believe, about 13.5 bits (it is non-linear). That is entirely adequate for FM stereo S/N and I have never heard anyone complain of R3 FM sound quality! There are many who say DAB is not as good.
Dave.
Bobbsy
Boring Old Git
If you're interested in the BBC radio distribution system for FM radio, there's a good article HERE.
It's worth noting that the best S/N ratio the signal was capable of was about 78dB--but this was deemed more than adequate for the distribution network because of the limitations in the analogue FM broadcast chain. It's also worth noting that they weren't starting on a "clean page" in engineering terms--they had to fit within the bandwidth/capabilities of the GPO circuits they had to use due to the then-monopoly for telecoms carriage.
The numbers kind of put things into perspective when we debate the S/N on even the cheapest home gear these days.
...and I'm another who says DAB isn't as good but that's because DAB is a compressed (MP2) signal at relatively low bit rates so the can cram in more channels.
It's worth noting that the best S/N ratio the signal was capable of was about 78dB--but this was deemed more than adequate for the distribution network because of the limitations in the analogue FM broadcast chain. It's also worth noting that they weren't starting on a "clean page" in engineering terms--they had to fit within the bandwidth/capabilities of the GPO circuits they had to use due to the then-monopoly for telecoms carriage.
The numbers kind of put things into perspective when we debate the S/N on even the cheapest home gear these days.
...and I'm another who says DAB isn't as good but that's because DAB is a compressed (MP2) signal at relatively low bit rates so the can cram in more channels.
dgatwood
is out. Leave a message.
Yeah, that's pretty much unavoidable given that, by definition, each bit in a binary number represents a factor of two more than the next smallest bit.
(Yes, you could do some weird nonlinear mapping, I suppose, but....)
snow lizard
Dedicated Slacker
ecc83 said:
More bits results in the dynamics of the audio being quantized more accurately. The decibel gets sliced into smaller chunks.
It's not usually the most limiting factor. A state of the art converter running at 16 bits might still sound better than an El-Cheap-O running at 24.
No, it doesn't, each bit is still ~6dB you just get more of them and thus more "legroom".
Yes. The limiting factor in digital systems is the hardware, not the "theory".
Dave.
snow lizard
Dedicated Slacker
Each bit is 6 dB. This is linear.
The way a binary word works is 2 (the number of values available in a bit, 0 or 1) to the power of how many bits you have. This is exponential, not linear.
2^16 = 65,536
This is how many levels you can quantize at in a 16 bit word. 16 bit audio has a 96 dB range.
2^24 = 16,777,216
24 bit audio has a 144 dB range in theory.
96 divided by 65,536 vs. 144 divided by 16,777,216.
Not the same thing at all, but probably not the most important thing from a practical standpoint either.
The way a binary word works is 2 (the number of values available in a bit, 0 or 1) to the power of how many bits you have. This is exponential, not linear.
2^16 = 65,536
This is how many levels you can quantize at in a 16 bit word. 16 bit audio has a 96 dB range.
2^24 = 16,777,216
24 bit audio has a 144 dB range in theory.
96 divided by 65,536 vs. 144 divided by 16,777,216.
Not the same thing at all, but probably not the most important thing from a practical standpoint either.
Not my area of expertize but I have seen the "resolution" argument debunked several times by experts I have been reading for decades.
A similar argument is made for sampling rate, i.e. more samples per second MUST means finer "resolution"? Again not so, simply an increase in the upper cutoff frequency*
This is analagous to doubling tape speed. 30ips will get you into bat territory but there are no other intrinsic benefits apart from a small noise improvement, especially now that head drive and playback electronics can be made many dBs better than any tape (don't mean they always ARE tho!).
*Mind you many AIs don't bother. The same filters are used for 44.1 and 48kHz!
I shall continue to be guided by the likes of Hugh Ford and Dan Mills.
Dave.
Mo Facta
Farts of Nature
What is the "resolution argument" just so we're on the same page?
Also, I'm not so sure your second argument there is correct. At 44.1 and 48, there are only roughly TWO samples possible to represent a full cycle of a 20kHz wave (by simple math calculations of sampling rate vs wave frequency). At 88.2/96 roughly four. At 192, roughly eight. This, by logic, means there are more samples to represent the high frequencies, not just an increase in bandwidth. Doesn't that equate to a THEORETICAL higher resolution? You can test this by generating a 20kHz wave in an audio editor and then zooming down to the sample level. Do it at multiple sampling rates and see what results.
Whether or not our ears tell us it's a higher quality signal is another thing altogether. That is the realm of opinion. It's very difficult to prove what sounds better to me on paper.
Cheers
Lt. Bob
Spread the Daf!
how about don't come into a forum asking a question and then tell a LONG time member how he should post and what links he should post.
If you don't want to read the wiki link fine ...... but get rude to a longtime member who just tried to help and almost surely knows more about this stuff than you do and you'll get run off quickly.
bouldersoundguy
Well-known member
The theory about sampling frequency is that two samples is enough for the DAC to accurately reconstruct the waveform. More than that isn't necessary for more accuracy at that frequency, it's to extend the frequency response. Actually I think you need just slightly more that two samples, which is why the sampling frequency has to be more than double the highest frequency.
bouldersoundguy
Well-known member
I think that when these systems were designed they needed sufficient space between 20kHz and the Nyquist frequency to accommodate the slope of the analog brick wall filter. Now with oversampling converters the analog filters can be way more gradual and the final filtering is done digitally before downsampling.
Lt. Bob
Spread the Daf!
ah ....... thanks. So they don't really use brick wall filters anymore and do that by oversampling to push the freqs higher where they put the more gradual filters?
Farview
Well-known member
Besides the fact that this is a thread about bit depth and you seem to be talking about sample rate...Also, I'm not so sure your second argument there is correct. At 44.1 and 48, there are only roughly TWO samples possible to represent a full cycle of a 20kHz wave (by simple math calculations of sampling rate vs wave frequency). At 88.2/96 roughly four. At 192, roughly eight. This, by logic, means there are more samples to represent the high frequencies, not just an increase in bandwidth. Doesn't that equate to a THEORETICAL higher resolution? You can test this by generating a 20kHz wave in an audio editor and then zooming down to the sample level. Do it at multiple sampling rates and see what results.
The thing that everyone seems to get tripped up on with sample rate and nyquist is that there is no more resolution to be gotten out of a 20khz sine wave. Once you can express the wave, that's it.
If there were any more to capture with a higher sample rate, it would be at a frequency above 20khz.
The only thing a higher sample rate allows you to do is record higher frequency waveforms. That's it.
snow lizard
Dedicated Slacker
ecc83 said:
It was proven that Harry Nyquist was not wrong several decades ago. I think Dan Lavry went into a bunch of detail about this years ago at PSW. Something about a buffer range to make it easy to build a filter with a decent slope.
This is cycles, not bits but I think I understand where you're coming from.
For me it's important to consider that these aren't the only variables in determining how the system will work and sound. There has to be a point of diminishing returns somewhere.
ecc83 said:
Again, that's something quite a bit different. 30 IPS will usually allow a tape machine to record flat up to 20 kHz whereas 15 IPS might not. It changes depending on which machine you're using and how it's calibrated. The benefit of getting the top octave to be flat has to be weighed against what it sounds like. A lot of people like to record at 15 IPS because tape machines all have their own head bump in the low end. It hypes the bass frequencies similar to proximity effect from microphones and can add a sense of size that works well for a lot of music. When you switch to 30 IPS the head bump moves up an octave and usually flattens out. The difference in high end might be difficult to notice where the difference in low end usually isn't.
By comparison, digital is pretty flat. Changing sample rates doesn't do anything like this.
Found a decent link on the subject.
Also, I'm pretty sure the voltage values used to print dynamics to tape aren't quantized, which comes back to the bit depth discussion.
At the end of the day the money question is still what does it sound like?
Last edited:
Farview
Well-known member
Db is a scale that puts a linear numbering system to an exponential thing. You can't take the Db scale and treat it as if it is exponential too.
In other words, the actual voltage change with a 6db move will depend on the actual voltage you started at. The ratio that a db represents remains the same.
Db represents a ratio, not an integer.
snow lizard
Dedicated Slacker
Farview said:
Almost. dB is a scale that puts a linear numbering system on a logarithmic thing.
Farview said:
This is PCM audio. I think there was a thread a few years back where we figured out that dBv values properly logged out correspond pretty closely but not exactly to the 6 dB per bit thing. Essentially the bit depth slices the volt into equal steps. More bits, smaller slices. The actual voltage change results in a 6 dB increase at approximately twice the voltage from where you started at.
This is kind of like trig, where it can be a confusing subject to understand until you get the math. Then it's easy.
For perspective I should point out that I agree with anyone thinking that this has nothing to do with gain staging. Overcooking your levels to chase the "resolution" is basically sacrificing something critical to preserve something with a much lower priority.
Ethan Winer
Acoustics Expert
It might seem that way, but all that's affected is the noise floor. What the "resolution" arguments miss is the reconstruction filter built into every D/A converter. This smooths the output waveform thus removing the steps. The attached figure is from my Audio Expert book, and it shows how the output waveform is identical to the input, within the frequency and noise limits of the current sample rate and bit depth.
--Ethan
