what do bits and Mhz mean?

Discussion in 'Archived Threads 2001-2004' started by BurkeP, Jul 17, 2001.

  1. BurkeP

    BurkeP Extra

    Joined:
    Mar 8, 2001
    Messages:
    16
    Likes Received:
    0
    Trophy Points:
    0
    I've been noticing that some DVD players have 27 mhz, 10-bit Video D/A converters, while others have 54 mhz, 12-bit converters. For audio, many have 96mhz, 24-bit, while other go higher. I assume higher is better, but what technically are these numbers telling me, and in what ways and how noticeably do they impact the picture and the sound?
     
  2. Kieran Coghlan

    Kieran Coghlan Second Unit

    Joined:
    Oct 26, 1998
    Messages:
    262
    Likes Received:
    0
    Trophy Points:
    0
    Well, you ask a very big question, and honestly I'm surprised more people don't ask it. I'll do my best to keep this to a brief summary, rather than the novel that it could become.
    To keep this simple, I'm going to refer to sound as a simple sine wave, even though it's more complex than that. If you're not familiar with what a sine wave looks like, I'll post a graph of one later... let me know.
    Digital sampling of analog sound (and video, but it's a similar set of concepts so I'll stick to sound) is what you're asking about. Sound travels through any medium as a wave. In air, the wave consists of regions of compressed air and "rarefied" air. The compressed regions are usually referred to as the peaks of the wave, and the rarefied regions are called the troughs (pronounced "troffs"), or valleys. The wave propogates, or travels at the speed of sound. (this velocity varies depending on the medium, and the ambient temperature and pressure as well, but that's not important.) The frequency of the sound wave, also known as the pitch, can be thought of as the number of complete waves (one peak and one trough) that pass by a single point in space PER SECOND. The units of frequency are called Hertz, named after a famous physicist, and are abrieviated "Hz" (always with a capital 'H' by the way, and the word "Hertz" when used as a unit, is both singular and plural... no need to add an "s" or take away the "z"... one Hertz, two Hertz same word). Hz can be used to measure the frequency of any cyclic phenomena, not just sound waves. So, sound pitch (how "high" or "low" the sound seems to your ear) is measured in Hz. Human hearing ranges from about 20 Hz, to about 20,000 Hz (also abrieviated 20 kHz for kiloHertz or 1,000 Hz). Now, sound waves (actually any sort of wave) have two other key characteristics: Amplitude, and Wavelength. Wavelength is defined as the distance between any two consecutive peaks (or troughs) on a wave. Wavelegth and frequency are actually very similar. In fact, the wavelength is simply the product of the speed of sound, times the reciprocal of the frequency. Amplitude, is a measure of the amount of energy that the wave carries with it. This can be visualized as the "height" or "size" of a 2-d sine wave, measured from the bottom of one trough, to the top of the adjacent peak. You can also think of Amplitude of a sound wave as the volume (how loud is it) of the sound.
    So, now that we've defined some basic terms, on to your question, which pertains to digital sampling of an analog audio waveform. Here's how digital sampling works in a nutshell: An analog sound signal is converted to an analog electrical signal by something like a microphone. This analog electrical signal is sent to the sampler, which is basically a specialized computer. The sampler reads this analog signal, and takes a sample of it. A sample consists of a record of just one characteristic about the electrical wave: the amplitude when the sample was taken. This one sample represents the value, or amplitude of the sound wave, at one specific instant in time. The sampler then takes another sample, immediately following the first sample. This process is repeated constantly, and these data are recorded in a digital form, which can be done because all that they consist of is just numbers, specifically the amplitude at various points along the wave.
    Now, in order for a computer to recreate an analog waveform from these data, it must also know when in time each data point was recorded relative to the orignal waveform. This is a known entity, though, because the sampler records samples at a steady rate, one after another, so all it has to do is place the sample back in order at that same rate. This rate is called the "Sampling Frequency," and is also measured in Hz, more often though kHz or MHz are used. So now, the computer can reassemble the samples in order, properly spaced, and then interpolate a new analog waveform from these data, similarly to how your mind would interpolate a "picture" from the lines of a completed "connect the dots" drawing.
    That is basically how digital sampling works. However, of course, there's a lot more to it, which you might have guessed, as I haven't really answered your question.
    Now to address though, the question of quality: How close to the original waveform is the new re-created waveform? Well, if you think about it, this depends on two things: How many samples per wave does the sampler take, and, how accurately does the sampler measure the value of the amplitude at each sample? First I'll address the number of samples taken:
    Obviously, the more samples per waveform that can be taken, the better, right? Like a connect the dots drawing, if you had an infinite number of dots to connect to make the drawing, the curves and details would be very precise, instead of a bunch of straight lines that vaguely looked like the picture. (forget about how long it would take to connect the dots... [​IMG] ) Another way to think of it is comparing a circle to a polygon: A pentagon looks more like a circle than a square, and an octagon looks more like a circle than a pentagon, while a 1,000 sided polygon (Megalogon?) would look almost just like a circle, depending on how closely you looked at it. So, how many samples is "enough" to determine what the original sound wave REALLY looked like? A famous physicist named Nyquist came up with a theorem, which states (more or less) that, using some pretty tricky mathematics (Nyquist's words, not mine... [​IMG] ), one only needs to know the location of two points on a wave form, to completely determine its frequency and amplitude. So, when creating that digital audio format we all know and love, CD-Audio, the designing engineers and scientists, wanted to use the minimum number of samples necessary to accurately recreate the original audio waveform. The reason why they were constrained in the number of samples they could use, is because the more samples you take, the bigger the digital file becomes, and they only had 650MB (more of less) of data storage to work with, namely the capacity of a Compact Disc. So, to determine the minimum sampling frequency reaquired, one must consider the HIGHEST frequency (pitch) sound that will be sampled. Well, since human hearing (on average) only goes out to 20kHz, there was not much reason to worry about any frequencies beyond 20,000 Hz. So, at 20 kHz, with 2 samples per wavelength (a'la Nyquist) we get a sampling frequency of 40,000 Hz. It turns out that for CD audio standard, a sampling frequency of 44,100 Hz was adopted. I'm not sure why they added an extra 4100 samples per second, but they did.
    Now, what about the other issue I brought up, namely the accuracy with which the sampler measures and records the amplitude of the electrical signal? This is where your question of 'bits' comes in to play. The amplitude of an electrical analog wave signal is really a measure of the voltage at that specific time. The sampler reads the voltage, and records that value as a binary number. I won't get into the math of converting decimal numbers to binary, so I'll just stick with decimal examples for now, and you keep in mind that the data is really recorded in binary. If a sampler only had 2 digits of accuracy when recording a voltage value, then you could never know the difference between .504 volts and .495 volts, as they would both be recorded as .50 volts. So, as you can see, the more digits that a sampler can use to record a voltage value, the more precise each sample will be. Well, this acuaracy is measured by the number of bits that the sampler can use to record the value of any one number. An 8-bit sampler can only use 8 places to record a number, each place can be a 1 or a 0. So a sampler could record a range 00000000 to 11111111 and anything in between (i.e. 10101010 or 10010010, etc.) But it could only use 8 places, or bits, to record the binary number. So obviously, the more bits that the sampler can use to read the voltage of the electrical signal of the audio waveform, the better, right? RIGHT! By the way, the number of bits that a sampler uses to record an analog signal, is also referred to as the "resolution" of that digital recording. And, it turns out that CD-Audio uses a standard of 16-bit sampling. So, PCM/CD-A etc, is also known as 16 bit, 44.1 kHz digital audio.
    With the advent of more powerful computers (and therefore samplers) and storage media with much larger storage capacities (i.e. DVD discs) the digital audio cogniscienti have decided it's time to re-vamp the digital audio standard, and improve the quality of CD-Audio. This is why you see new formats with rates of 48 kHz, 96 kHz, and higher, and resolutions of 20bit, 24bit, and higher. DVD-Audio has a standard (IIRC) of 24bit/96kHz for 5.1 channel audio, and 48bit/192kHz (I think) for 2-channel audio.
    SACD is another animal entirely, and I don't understand it well enough yet to explain how its rates and resolutions relate to what I've just explained.
    I hope this has been helpful, and not TOO long. [​IMG]
    ------------------
    -Kieran
    My HT Page
    [Edited last by Kieran Coghlan on July 17, 2001 at 03:07 PM]
    [Edited last by Kieran Coghlan on July 17, 2001 at 03:11 PM]
     
  3. StephenK

    StephenK Stunt Coordinator

    Joined:
    Jun 1, 1999
    Messages:
    226
    Likes Received:
    0
    Trophy Points:
    0
    NM
    [Edited last by StephenK on July 17, 2001 at 03:51 PM]
     
  4. Bob McElfresh

    Bob McElfresh Producer

    Joined:
    May 22, 1999
    Messages:
    5,182
    Likes Received:
    0
    Trophy Points:
    0
    Wow! Quite a post Kieran.
    Let me give a simpler explination:
    Burke: you are mixing 2 different things:
    - The processing power in a DVD player
    - The number of bits used to record a CD.
    DVD PLAYERS: When "Secrets of Home Theater and High Fidelity" did a comparison of DVD player, they noticed that the players that had the faster internal CPU did a better job with some video problems shown by slower models.
    (I think this solved some artifacts - better smoothing, or some such).
    So for DVD players, the faster units tend to give a bit better picture.
    BITS ON A CD: Think of this - you want to see how good the insulation is in your house. You install a thermometer and record the temp inside and outside every 3 hours and then draw a graph.
    So in a day, you can show a graph with 6 temperature readings.
    This is 6 samples per day.
    But you would get a better graph if you could take the temperature every 5 minutes.
    This would be 288 samples per day.
    This graph would be much more detailed. And you could tell a lot more about your house.
    The standard CD takes a sample of music 16 thousand times per second.
    There are some new CD's where they took a sample 96 thousand times per second. (Six times more). And some DVD players and CD players will read this disk.
    Since it takes more samples, it is believed that the music sounds better.
    Does this help?
     
  5. Kieran Coghlan

    Kieran Coghlan Second Unit

    Joined:
    Oct 26, 1998
    Messages:
    262
    Likes Received:
    0
    Trophy Points:
    0
    Bob, thanks.
    However, it's not 16 thousand times per second for CD-A, it's 44.1 thousand times per second, and your example didn't explain the accuracy (bits) issue. Using your example, the bits would be representative of how many significant digits you could measure the temperature to.
    ------------------
    -Kieran
    My HT Page
     
  6. BurkeP

    BurkeP Extra

    Joined:
    Mar 8, 2001
    Messages:
    16
    Likes Received:
    0
    Trophy Points:
    0
    Bob and Kieran,
    Thanks for the responses - I think I'm starting to get it. But one question - the sampling rates you describe would seem to apply to the recording of music - once it's in software form (on DVD or CD), the sampling rate is fixed, so there's a limit to how much digital information the converter can convert to analog, right? Does that mean that these converters have the ability to deal with media sampled at a higher rate, even though most CD's and DVD's don't yet include digital recordings at those rates?
    The bits issues seems clear - thanks.
     
  7. RicP

    RicP Screenwriter

    Joined:
    Feb 29, 2000
    Messages:
    1,126
    Likes Received:
    0
    Trophy Points:
    0
     
  8. John Kotches

    John Kotches Cinematographer

    Joined:
    Mar 14, 2000
    Messages:
    2,635
    Likes Received:
    0
    Trophy Points:
    0
    Ric,
    Mostly excellent post, but I noticed you've glossed over DSDs shortcoming: noise.
    Above 8 KHz the noise starts a drastic increase.
    Also, your point WRT frequency vs. bits available I don't find to be ringing completely true to my ears.
    As you lower frequency, PCM also has more samples.
    With DSD at zero crossing you have constant switching between full + and full -, and it is possible to have switching distortion when you cycle at that frequency.
    Have you spent much time listening to 24bit/192KHz stereo yet? I have a couple of discs in the house and I've been able to give it a serious listen.
    Until you've experienced 24bit/192KHz stereo, you don't know what PCM is really capable of.
    BTW, I'm not SACD bashing I find it to be excellent reproduction, but so is 24bit/192KHz.
    Either way they are miles ahead of Redbook CD.
    Ed Meitners a good egg, and I have a lot of respect for his opinion as well as Bob Ludwigs.
    Regards,
    ------------------
    John Kotches
    Contributing Writer
    Secrets of Home Theater and High Fidelity
     
  9. RicP

    RicP Screenwriter

    Joined:
    Feb 29, 2000
    Messages:
    1,126
    Likes Received:
    0
    Trophy Points:
    0
    John,
    Good point regarding the noise but thats essentialy what the Pro 8-bit DSD was designed to overcome. The advanced noise shaping algorithms take advantage of the Pro DSD stream to effectively reshape the waveform to place the noise well out of the audible spectrum. Does this cause artifacts? Hard to say conclusively, but it does sound great.
    WRT 24/192 I've heard some, not much but I wasn't that much more impressed than I was with 24/96 (which sounds pretty good).
    I mean if you believe in Nyquist and the PCM noise shaping technology then 192Khz is WAY overkill for PCM wave capture and should "in theory" not capture any more information than 96Khz sampling does. [​IMG]
    ------------------
    Ric Perrott
    My Theater ;My DVD's
    "RicP's posts are SUPERBLY ERUDITE!" - David Manning, Ridgefield Press
     
  10. Bob Segno

    Bob Segno Agent

    Joined:
    Nov 27, 2000
    Messages:
    46
    Likes Received:
    0
    Trophy Points:
    0
    VERY interesting reading from everyone.However, I think I'm really baffeled by all of this very deep explination. [​IMG]
    ------------------
     
  11. Scott L

    Scott L Producer

    Joined:
    Feb 29, 2000
    Messages:
    4,457
    Likes Received:
    1
    Trophy Points:
    0
    This is why High-res audio goes beyond 20 kHz and increases into the 192 kHz and even 2.8 mHz range. It seems that the frequencies that supposedly were beyond our sense of hearing are the frequencies that make recorded audio all the more realistic.
     
  12. Wayne_T

    Wayne_T Stunt Coordinator

    Joined:
    Jun 15, 2001
    Messages:
    236
    Likes Received:
    0
    Trophy Points:
    0
    Kieran, that was a great post. Although I majored in Physics (many moons ago) I still found it helpful. You must be, or must have been at some point, a teacher.
    I have recently added a DVD-A (Panasonic RP91) player to my system, and have a couple of questions to further explore the concepts you guys have explained so well.
    1. The Panasonic "remasters" the 16/44.1 PCM signal and creates a 88.2 KHz signal which it sends to my receiver. I don't know if this is 16/88.2 or 24/88.2, as the receiver only displays the sampling frequency. My understanding is that the unit somehow interpolates the harmonics above 20KHz and adds them in. In any case, I find the result to be quite pleasing. Can you elaborate on what is happening here?
    2. What is the ability of the rest of my system to reproduce frequencies above 20KHz and on up to close to 100KHz? I have listened several times to my one and only (so far) DVD-A recording and it does sound terrific, but that could just be because of the added resolution. I'm reasonably confident that my amps can go up to 100KHz but I doubt very much that most speakers can, including my Paradigm Reference. More and more of us are adding DVD-A and SACD capabilities to our systems, but can our other components keep up with the high frequencies? And how would we know short of measuring with instruments?
    ------------------
    [Edited last by Wayne_T on July 18, 2001 at 08:54 PM]
     
  13. Bob McElfresh

    Bob McElfresh Producer

    Joined:
    May 22, 1999
    Messages:
    5,182
    Likes Received:
    0
    Trophy Points:
    0
    Kieran: Darn! good catch. Cd's are sampled at 44 thousand times per second, not 16.
    And yes, I did ignore a major issue.
    Burke & Bob: Let me go back to the home-temperature example:
    So now you are sampling every 5 minutes. But your thermometer is one of those strips that change color. And they only show temperature changes every 5 degrees. So your data is something like: 60,60,65,65,70,70,70...
    So it takes a big jump in temperature before your reading changes. And your graph is very ... blocky.
    So you run to Radio Shack and buy a digital thermometer that displays temperature changes as small as 0.1 degree.
    So now your graph is a lot smoother because you can record much smaller changes.
    So when you sample music for a CD, how small of a change can you detect/record?
    This is where the number of bits (size of the number) comes in.
    An audio CD is recorded with 16 bits. This means that a single sample can be a number from 0 - 65,536. So a CD can record very small changes in the sound.
    Is this enough? Yes. Can we do better? Yes.
    The newer audio CD's use more bits in a sample (20 or 24 bits) which means they can record much smaller changes in sound.
    Conclusion: you can get a better recording (sound or house-temperature) if you increase 2 things:
    - The number of times you take a reading per day/hour/second (This is the KiloHertz numbers you read)
    - The maximum number you can record (so you can detect smaller changes). (This is the 16 bit, 20 bit, 24 bit numbers you hear about)
    Is this clear?
    BurkeP: You asked an interesting question:
     
  14. AlbertA

    AlbertA Stunt Coordinator

    Joined:
    Nov 3, 1999
    Messages:
    107
    Likes Received:
    0
    Trophy Points:
    0
    Scott:
    Actually, brick wall filters are a mere
    mathematical idealization. In practice, filters
    have a transition region to reach the stopband.
    Commonly, the corner frequency of the filters
    used for CD players are at 20 KHz, effectively moving the
    transition region out of the audible range.
    Why they chose exactly 44.1KHz? I don't know. They could have as well chosen 48 KHz.
    Hi-res audio doesn't go as high as 2.8Mhz or 192Khz.
    SACD and DVD-Audio have similar response-- about 100 Khz.
    Personally I don't understand why in DVD Audio they
    simply chose to use PCM again, intead of ADPCM or simply DPCM, maybe copyright issues with SACD?
    These techniques are nothing new, so I wonder.
    RicP: although I do believe in Nyquist,
    think about voice. The common assumption is that voice signals have a bandwidth of 3.5 Khz. So telephone companies sample them at 8Khz. But we all know how much better our voices would sound if they were sampled at a higher rate.
    However I am aware that I am not considering that repeated
    tests have shown that the human ear cannot hear above 20Khz on the average.
     
  15. Saurav

    Saurav Cinematographer

    Joined:
    Feb 15, 2001
    Messages:
    2,174
    Likes Received:
    0
    Trophy Points:
    0
     
  16. Scott L

    Scott L Producer

    Joined:
    Feb 29, 2000
    Messages:
    4,457
    Likes Received:
    1
    Trophy Points:
    0
     
  17. BurkeP

    BurkeP Extra

    Joined:
    Mar 8, 2001
    Messages:
    16
    Likes Received:
    0
    Trophy Points:
    0
    I know my questions aren't up to the sophisticated bent this conversation is taking, but you guys are a great source of info. Aren't we mixing two kinds of frequency here? Isn't 44.1khz the sampling frequency for PCM - in other words, the CD sends the receiver 44,100 16-bit numbers per second? Whereas the 20hz-20khz is the frequency of the sound - the number of crests that hit your ear per second? If that's the case, why would the sampling rate be filtered in the 20k range, since that's the rate information is being read at rather than a measure of the frequency of the sound that's being transmitted? I hope this question is clear to you guys.
     
  18. BurkeP

    BurkeP Extra

    Joined:
    Mar 8, 2001
    Messages:
    16
    Likes Received:
    0
    Trophy Points:
    0
    Sorry for the multiple post, but if my assumption above is right (that sampling frequency has little to do with the frequency of the sound being reproduced) then you could certainly have a format that samples at a 2.8mhz rate, but only reproduces sound from 20hz-20khz, right? It would just be a more detailed reproduction. I may be totally wrong, but it seems like the frequencies of two different things are being confused in the thread.
    [Edited last by BurkeP on July 19, 2001 at 01:46 PM]
     
  19. Kieran Coghlan

    Kieran Coghlan Second Unit

    Joined:
    Oct 26, 1998
    Messages:
    262
    Likes Received:
    0
    Trophy Points:
    0
    Just because a new format has sampling rates up to 192kHz or so, doesn't mean that the goal is to reproduce sound in the 95kHz range (1/2 of the sampling frequency). Nyquist's theory works, but there's some that claim it doesn't work very well. Recreating harmonics beyond human auditory range is one thing, but reproducing them (i.e. through a speaker) is another thing altogether. As several have already said, most speakers, even very high end ones, only have a range of maybe a half octave above 20kHz. It's not capturing those "higher harmonics" that gives better sound quality, IMO. It's the fact that the stuff we CAN hear, and that normal speakers CAN reproduce, e.g. sounds up to 20kHz or so, is better (more accurately) rendered with more than two samples per waveform. Strong believers in Nyquist's theorem will claim extra samples are just not necessary. I tend to think it can help, but I'm not knowledgable enough about Nyquist's work to really discuss the matter too deeply.
    BurkeP: You are correct, that using a 24bit/96kHz DAC (Digital to Analog Converter) for cd audio (i.e. 16/44.1) would not really give you anything more in the way of sound quality, except that a new 24/96 DAC is possibly of better quality than a 16/44.1 DAC.
    Scott L:Although I didn't know about the filter at 22.05 Hz, I'm not surprised by it. Also, although I don't KNOW, I'd agree with Albert that it's probably not a "brick wall" as hardly anything in electronics is, especially not filters. There almost HAS to be a slope of some sort, and I doubt the slope is infinite! [​IMG]
    Wayne_T:
     
  20. Kieran Coghlan

    Kieran Coghlan Second Unit

    Joined:
    Oct 26, 1998
    Messages:
    262
    Likes Received:
    0
    Trophy Points:
    0
    BurkeP: Yes, you could have a 2.8MHz (remember capital "H"... [​IMG] and a small "m" means "milli," 1/1000, not "Mega," 10^6.) sampling rate, and only sample frequencies up to 20kHz... and yes it would give you many more samples per waveform (140 samples on a 20kHz signal, to be precise). However, it would mostly be a waste of space on the storage medium, and a waste of processing power. I don't believe that people are confusing or switching sampling rate with sonic frequency. Yes, both use the same units (Hz) but they are different things. If our posts haven't clearly differntiated between the two, sorry... [​IMG] I for one will try to be more clear.
    Scott L didn't say that the sampling rate was filtered at 22.05kHz, he said that the source was filtered at 22.05kHz. (This is half the sampling rate.) This means that the sampler is not getting any sound over 22.05kHz to sample. But it wouldn't matter anyway, since anything above 22.05kHz SOUND frequency, would have less than two samples per waveform (due to the 44.1kHz SAMPLING frequency) and could not be accurately re-constructed anyway. This actually, I presume, would be the reason for having a filter at 22.05 in the first place: to prevent the computer from trying to reconstruct waveforms for which it has fewer than two samples per wave. Funny things can result when a computer tries to do this without some very advanced algorhythms.
    ------------------
    -Kieran
    My HT Page
     

Share This Page