About Sample rate?

[jayyft]

I fairly new to recording and there's something I heven't quite figured out. It's about the sample rate. I can turn it to a lower setting and it helps with delay when recording and monitoring. Does lowering the sample rate affect only playback while recording or is it also lowering the quality of what is being recorded?

Thanks

[Tarktones]

Changing the sample rate (http://www.sweetwater.com/expert-center/glossary/t--SampleRate) affects how much data you're using to store the audio information. While the quality is drastically affected when you're below the Nyquist Frequency (http://www.sweetwater.com/expert-center/glossary/t--NyquistFrequency) (which for us is about 40kHz), once you are safely above it (like CD's at 44.1kHz) there is very little discernible difference between audio recorded at one sample rate vs. another as far as human hearing is concerned.

So the reason you're getting more delay (latency) when recording at higher sample rates is because your computer is having to process twice the amount of data recording at something like 88.2kHz or 96kHz than it would if you were doing it at 44.1kHz. While in theory it's "lowering the quality" of what's being recorded, 99 out of 100 won't be able to hear a difference even in critical listening environments. As long as you're not setting your sample rate at something like 32kHz, you'll be fine.

It's general consensus that the difference in bit-rate (24 bit vs. 16 bit) is immensely more audible than difference in sample rate. Recording at 24 bit vs. 16 is (IMHO) critical.

[brianbfw]

yes, sample rate does effect the quality of the recording, but it's debatable how much better it actually makes it.

44/1khz is standard CD format. most likely though, the source material was recorded at high sample rate, such as 48Khz or even 96Khz. samples rates can even go as high as 192Khz now.

here's my experience. primarily, i'm a classical guitarist, but i also have a recording studio, where i record and mix many different styles.

bit-depth makes a far more impact then sample rate. i always record at 24-bit. you must dither-down later to 16-bit.

for rock, pop, styles, i've don't not notice any differences between 44.1khz and the 96khz, maybe a little in the upper highs (cymbals, etc).

when i'm doing classical, jazz, or anything with soft passages, or acoustic music, it's always 96Khz. i can defintely tell the difference with the higher sample rate for this type of music.

[Rad]

Well, my take is there's several different things going on here, so let me try to tackle each in turn.
Firstly, how much audible differences you will hear between 44.1 and 96 kHz sample rates mostly depends on quality of the convertors you use. By now, there seems to be a widespread agreement among studio professionals that on high-quality studio converters, such as the AKM 5385 and 5394, differences are almost impossible to tell, as already mentioned by Tarktones. Not so with medium and lower quality convertors, where a switch from 44.1 to 96 may be quite audible, but not because of 96k's inherent superiority, but because mid- and low-quality convertors have harder time getting good sound of low sample rates. The bottom line: Choose the sample rate which on your equipment sounds best, but do not be mislead that it inherently and always sounds better than other sample rates on all other equipment. At least two highly professional studios I know work on 44.1k/24 bit for most of the time.

Now, regarding 16 vs 24 bit. There's a widespread confusion of which some time ago I myself fell a victim, that 24 bits give you more "resolution", the same way as 1024x800 pixel computer screen gives you a better resolution than a 800x600. This is sadly wrong and is nicely deobfuscated in Nika Aldrich's book "Digital Audio for the Audio Engineer". Without getting into tech details, 24 bit is preferable for studio because it gives you bigger headroom above the noise floor so you can freely record 8-9 dB below the zero mark without worrying about unexpected peaks that may clip. I know it sounds strange at first, but it is explained there (and quite well) that every extra bit gives you 6 dB additional dynamic range. This is, regarding the theory. In practice, since every convertor has a distortion floor, 16-bit may sound "worse" especially on medium to low-quality convertors. Distortion usually occurs the most when the last bit "fills up", and with 24 bits, this last bit is sent an additional 48 dB down closer to the noise floor, so it is harder to hear.

[Tarktones]

Now, regarding 16 vs 24 bit. There's a widespread confusion of which some time ago I myself fell a victim, that 24 bits give you more "resolution", the same way as 1024x800 pixel computer screen gives you a better resolution than a 800x600. This is sadly wrong and is nicely deobfuscated in Nika Aldrich's book "Digital Audio for the Audio Engineer". Without getting into tech details, 24 bit is preferable for studio because it gives you bigger headroom above the noise floor so you can freely record 8-9 dB below the zero mark without worrying about unexpected peaks that may clip. I know it sounds strange at first, but it is explained there (and quite well) that every extra bit gives you 6 dB additional dynamic range.

Not to nitpick, I'm not a fan of that analogy. In the computer screen analogy, pixel count is more equivalent to sample rate and bit depth is more like color resolution. The following is not to patronize Rad in any way shape or form because I know he knows all of this probably a lot better than I do, but is just my attempt at better explaining it to the original poster.

Think of it this way: there's an image you want to capture. Sample rate is the equivalent of how many pixels you want to use to capture the image. The bit depth is the equivalent of color resolution (how many different colors each pixel can be). That's exactly what bit depth is in audio too. Each sample is a measurement of a signal's voltage at a given point in time. The bit depth determines how many different values it can represent.

Think of it like graph paper. You have to draw a picture but you can only do it by connecting one intersection of grid lines to the next. The more grid lines you have, the more accurate your picture will be. Your sample rate sets the number of grid lines on your X-axis and the bit-depth determines the number of grid lines on your Y-axis.

For example, an audio file at 2 bits will only have 4 degrees of resolution: 00 (0% bottom), 01 (33.3% of maximum), 10 (66.6% of maximum) and 11 (100% of maximum). As you can imagine, this would not sound very accurate. Nor would a picture on graph paper with only possible drawing points on the Y-axis be very recognizable. However 8 bits gives 256 different points to represent the wave form. So a sheet of paper with 256 grid markers might allow that picture. When you get up to 16 bits (because it's all exponential) you've got a total of 65,536 different steps to represent a signal. You could probably get a pretty accurate picture with all those possible steps. But the human ear can still notice the difference in audio when it has 65,536 steps at 16 bits vs. 16,777,216 steps at 24 bits.

Ugh... I'm a windbag. Somebody stop me.

[Rad]

I have to admit that Tarktones' analogy about color resolution is actually better than mine. I hadn't thought of it this way, but it is a nice illustration.

One can also think of bit depth as of how many decimal places you use to record the voltage of the signal when a sample is taken. The analogy is rough, because in binary system there's "binary places", not decimal ones, but the idea is the same. More bits, just as more decimal places, give you more precise measurements of the voltage every time when a sample is taken.

Now, about the effect of this on the quality of the audio is where I think I may differ from Tarktones (even though we're both fans of Al di Meola).
Suppose we are in the 16-bit world, so our measurement of the signal is not very precise: the signal's true value is, say, 0.441832, but instead we measure it as 0.4418. This is not very accurate, so the remaininig 0.000032 is called quantization error. If the true signal however is 0.4418, it will be measured correctly: as 0.4418, with a quantization error of 0.
Hence, these errors are non-random: the amount of error heavily depends on what particular voltage the signal has. To the human ear, non-random errors sound as distortion:generally speaking, in a harsh way. By contrast, random errors sound as just noise - that's what noise is, in essence! Thus, a major problem facing any convertor whatsoever, is how to avoid the distortion caused by quantization error. Fortunately, the analog equipment that precedes the convertor (preamps, even plain old wires) turns out to generate enough low-level noise for it to be comparable in magnitude to the quantization steps of the convertor. Thus, what would have been a quantization distortion (if the true signal was fed into the convertor), is now just quantization noise, because instead of the true signal coming from the mic, we have fed into the convertor a true singal plus a miniscule amount of analog noise that makes the otherwise non-random quantization errors random, i.e. just plain noise.
So, the difference between a 24-bit convertor and a 16-bit one is that the 16-bit has more quantization noise. It is just another way of characterizing accuracy: a more accurate recording has less noise! This is why people think of extra bits as just giving extra dynamic range above the noise floor, and this is what I was trying to get at by saying that it's unlike the computer monitor analogy.

This being said, while I have some technical bacground to make understanding the math behind this a bit easier, I am by no means an expert on this. There's a forum member called DPD who I believe used to work as a convertor designer, so he would be able to explain this probably a lot better than myself (and correct me if I am wrong.)

[Tarktones]

Thus, what would have been a quantization distortion (if the true signal was fed into the convertor), is now just quantization noise, because instead of the true signal coming from the mic, we have fed into the convertor a true singal plus a miniscule amount of analog noise that makes the otherwise non-random quantization errors random, i.e. just plain noise.

So, the difference between a 24-bit convertor and a 16-bit one is that the 16-bit has more quantization noise. It is just another way of characterizing accuracy: a more accurate recording has less noise! This is why people think of extra bits as just giving extra dynamic range above the noise floor, and this is what I was trying to get at by saying that it's unlike the computer monitor analogy.

I'd never thought about it like that. Hence why I love this board; always something new to learn. Great explanation!

[dpd]

Maximum signal level (24 bit or 16 bit) is set by the converter's reference voltage. These days, IIRC, it's around 3.3 Vpp / ~1 Vrms / ~0 dBV.

Minimum signal level is set by the system's noise floor. Theoretically, at the same sample rate, 24 bit is ~48 dB quieter than 16 bit (6.02 dB/bit). So - as long as the rest of the electronics don't have a noise problem - the 24 bit converter has greater dynamic range, but it comes from being quieter.

Now, the converter designer still has to worry about all kinds of distortion products that are harder to control in the 24 bit converter, since it's self-noise is so much lower than a 16 bit. This requires great execution in the design of the chip layout, the modulator, decimating filters, etc. So, good quality converters, especially at 24 bit, are crucial to improved sound. I'd venture the statement that 24 bit converters may be more sensitive to clocking stability issues, since the noise products are harder to suppress below the noise floor.

In order to maintain peak converter performance it's critical to not clip the converter and that means you must maintain the peak signal level below 0 dB Full Scale (FS). Real random signals (e.g. voice, music, noise) all have an average level (you may have heard the term 'RMS') and a peak level. The ratio of peak level over the RMS level is called the 'crest factor' and can run as high as 18-20 dB over the RMS level for very dynamic material, which translates to 3+ bits in the converter.

So, a good rule of thumb is to set your average level to about -20 dB FS. (look at a lot of early digital recorders and you'll see that the 'calibration' level is set there.) You do that and you'll have gobs of clean headroom while recording AND you won't need a limiter or compressor when tracking. In doing so, you give up 3 bits of dynamic range BELOW the RMS level - which leaves you with only 13 bits of range under the RMS in a 16 bit system but 21 bits in a 24 bit system. In a 24 bit system you can have your cake (superb dynamic range) and eat it too (lots of headroom).

IMO, Pushing the input levels on a 24 bit system is just dumb. You simply don't need to do it. I'll take that extra clarity on the upper end of the dynamic range any day and never hit the noise floor on the bottom end. Perfect.

What is 'resolution'? In the 24 vs 16 bit scenario, it means the voltage level change needed at its input for 1 bit bit change in the converter's output. The 24 bit converter has a step size that is 256 times smaller (the extra 8 bits) than the 16 bit converter. So, it can resolve finer detail in the signals. You can see the effects of this on a modern digital oscilloscope compared to an early one, for example. Modern ones use 12 bits vs 8 bits on the older ones. You simply can see (resolve) far finer details on the systems with higher bit depth.

PS - good to see you back, Rad!

[Rad]

Good to have you around, DPD! (and Tarktones!)

A very interesting post from which I'm continuing to learn more and more about convertor design.
One thing I'm still a bit unclear about though, is the ultimate quality change implied by 24 bits. So far we all agreed that 24 bits has higher dynamic range. If you are saying that this allows one to hear detail in the recording that would otherwise have been masked below the noise floor in 16-bit, I'm with you. But I'm less clear if, apart from and in addition to that, you get any more detail out of 24 bits.
My gut feeling about that is no, because of the following:
When the signal voltage coincides exactly with a quantization step, it's accurately measured, so there's no room for improvement. When it doesn't, it generates a quantization error, which is smaller in 24 bit than 16 bit, so the room for improvement is to go from 16 to 24 bit. Nothing else besides these 2 scenarios can happen, so the extra benefit of having 24 is entirely translated into the wider dynamic range (or at least I've seen this view defended by Nika Aldrich and some of the moderators, like DAS, I think).

This whole thing in translating accuracy of measurement into dynamic range is not entirely intuitive, though, so it would be nice to know if I'm right.

[jayyft]

That was a really nice in depth conversation ya'll had! It was fun watching everyone go all out on my (what I thought) simple question. I want to thank everyone, because I now have a much better understanding of sample rate and so forth. Just so I'm clear, if I turn down the sample rate in my prefrences to Logic 8, the playback will seem to have no delay, but technically it shouldn't sound as good, although audibly it may not be noticable (depends on the ear).

So what I plan on doing is having the sample rate up and lowering it as the musician I'm tracking feels he needs a more undelayed playback, which seems to be the case on really fast lead guitarists.

In my current setup I'm using a Rosetta 200 with a x-fire card with an AU 610and a great river. I must say I am stunned by the set up, I knew for the money it had to be great, but it's different to actually hear the sound playing back. Anyways, I figured since I had so many knowledgeable people writing on this thread, I wanted to ask a different question about gear, and it's about mics.

As you could see I broke the bank on the converter and pre-amps so now I'm saving for a couple more mics for the near future. What I have now is an AT4047, AT2020, and an sm57. I got the 4047 recently after listening to it on the AT website through a pre-amp (the great riverwhich I already had) and loved the character of it. I haven't done enough testing on the 4047 yet, but so far I can tell I'm going to really like it. My question was for a mic to record guitar cabs with 100W to 50W marshall heads playing through them? As of now, I'm going to be trying out the 57 close up and the 4047 further away. What would everyone recommend if I wanted to get a third mic to mix and match with. This mic could not be more than $700. I'm leaning towards a Beyerdynamic M 160. What do ya'll think, any experience with other mics I would not have thought of to try?

[DAS]

Good to have you around, DPD! (and Tarktones!)

So far we all agreed that 24 bits has higher dynamic range. If you are saying that this allows one to hear detail in the recording that would otherwise have been masked below the noise floor in 16-bit, I'm with you. But I'm less clear if, apart from and in addition to that, you get any more detail out of 24 bits.
My gut feeling about that is no, because of the following:
When the signal voltage coincides exactly with a quantization step, it's accurately measured, so there's no room for improvement. When it doesn't, it generates a quantization error, which is smaller in 24 bit than 16 bit, so the room for improvement is to go from 16 to 24 bit. Nothing else besides these 2 scenarios can happen, so the extra benefit of having 24 is entirely translated into the wider dynamic range (or at least I've seen this view defended by Nika Aldrich and some of the moderators, like DAS, I think).

This whole thing in translating accuracy of measurement into dynamic range is not entirely intuitive, though, so it would be nice to know if I'm right.

Rad, you are on the right track. It sounds like you have a good grasp of the material. dpd's comments about the resolution of the voltage are, of course, correct, but it's not contradictory to what you're saying. Guess where those small changes in voltage values show up? Down at low levels in the audio. So when we say more bits give us more resolution in this context that resolution shows up in our ability to better "resolve" low level information. In a 16 bit converter that low level info is lost as the numeric values are simply quantized to the nearest quantization point we have, resulting in the aforementioned quantization distortion. If there is enough noise present to cause those quantization steps to become correlated to the noise instead of the signal then it is merely heard as noise.

It's important to understand that ALL converters have quantization distortion. The question is where is that distortion relative to the noise and the signal. For instance, one might not need any more than 16 bits to record a rock band in a noisy environment because, A) the dynamic range of the band arguably wouldn't require a lot of low level detail, and B) even if A were not true the noise levels present could cause any potential details below that 16th bit to just show up as noise. Strictly speaking one only needs enough dynamic range to capture the dynamic range of the material in question in the environment where it is being recorded.

Now, in a multi-track world where serious manipulation of individual signals occurs after the recording we do tend to prefer 24 bit because we are sure to capture all of the possible dynamic range that we might need to use in subsequent processing, and we don't have to be as careful with recording levels in the process.

Hope this makes sense.

[brianbfw]

i also have a rosetta 200 and i love it!! fantastic sound.

although the sm57 is a favorite of mine on guitar cab's, sometimes a nice large diaphram condenser is good if you want a more tame sound.

i have a Neuman TLM-103 which sounds great on guitar cab's. i've used it with my 100W as well as my 5watt gibson. sounds great on both.

the sm57 gives me that bright van-halen"ish" tone. the neuman give me a nice mellow tone.

[Rad]

Thanks to all for the good discussion.

It just occurred to me that Cubase comes with a plugin called Bit Crusher in which you can simulate bit depths anywhere from 8 to 24 bits. Just set it to whatever you like and listen carefully. That's a very good non-technical way to understand the effect of bit depth on audio. For example, one thing you'll notice right away is that 8 bits is a lot noisier! (and now after the discussion you will know why... quantization noise matters).

Again, a good discussion.

[dpd]

It just occurred to me that Cubase comes with a plugin called Bit Crusher in which you can simulate bit depths anywhere from 8 to 24 bits. Just set it to whatever you like and listen carefully. That's a very good non-technical way to understand the effect of bit depth on audio. For example, one thing you'll notice right away is that 8 bits is a lot noisier! (and now after the discussion you will know why... quantization noise matters).

You can do the same with hearing the effects of various dither algorithms - dither an 8 bit version of your mix with the different types. Pretty interesting experiment.

Agreed - good discussion!