Digtal vs Analog; Why Isn't Digital Better? (long)

[sunnysal]

yes that does make sense, thanks. I suppose that the opamps in the output of a CD player have as much effect on what we hear, slew rate, etc. tony

[sfogg]

DrWho,

I just got a chance to try this so I wanted to get back to this.....

"The problem is that the process of removing the 30kHz tone (implementing an EQ) is also going to remove the amplitude changes incured upon the 18kHz tone. The whole 12kHz subharmonic is simply a product of the phase relationship between the two waves. Introducing an EQ is basically adding a third phase relationship that is going to cancel out the effects of the 30kHz tone (so now you're back to the original phase of the 18kHz tone, thus no 12kHz subharmonic)."

That isn't what occurs. The beat frequency is caused by the constructive interference of the two waves in the air. As such the beat frequency is basically another sound that is created from the two waves interacting in the air. If you record it, you get the beat frequency.... even if the recording can not capture the two waves that are creating the beat frequency.

To test this I just made a recording with a sampling rate of 22.05kHz. As you well know that means the highest frequency we could record would be just about 11kHz.

My source was a 20kHz sine wave and a 19kHz sine wave using a function generator on my laptop which can do multiple sines and having that play through its built in speakers. Playing either tone by itself was inaudible, when I played them both together I hear a 1kHz tone from the beat frequency.

When I record those two sine waves on another computer (again sampling at 22.5kHz) neither wave is recorded.... but the 1kHz beat frequency is.

Shawn

[DrWho]

----------------

On 5/10/2005 10:40:10 PM sfogg wrote:

DrWho,

I just got a chance to try this so I wanted to get back to this.....

"The problem is that the process of removing the 30kHz tone (implementing an EQ) is also going to remove the amplitude changes incured upon the 18kHz tone. The whole 12kHz subharmonic is simply a product of the phase relationship between the two waves. Introducing an EQ is basically adding a third phase relationship that is going to cancel out the effects of the 30kHz tone (so now you're back to the original phase of the 18kHz tone, thus no 12kHz subharmonic)."

That isn't what occurs. The beat frequency is caused by the constructive interference of the two waves in the air. As such the beat frequency is basically another sound that is created from the two waves interacting in the air. If you record it, you get the beat frequency.... even if the recording can not capture the two waves that are creating the beat frequency.

To test this I just made a recording with a sampling rate of 22.05kHz. As you well know that means the highest frequency we could record would be just about 11kHz.

My source was a 20kHz sine wave and a 19kHz sine wave using a function generator on my laptop which can do multiple sines and having that play through its built in speakers. Playing either tone by itself was inaudible, when I played them both together I hear a 1kHz tone from the beat frequency.

When I record those two sine waves on another computer (again sampling at 22.5kHz) neither wave is recorded.... but the 1kHz beat frequency is.

Shawn

----------------

Interesting, I'll have to try it myself too then. I must confess that I was simply regurgitating what my prof told me and I never bothered to check it myself because it seemed to make sense (not like it'd be the first time he was wrong either). Do you get the same result by throwing in a low pass filter to take out the upper frequency (instead of just lowering the sampling rate)? One other question that comes to mind is how well are your computer speakers reproducing the 19 and 20 kHz tones? In other words, could the 1kHz tone actually be a distortion in the speaker?

Technically then, what other reason might there be for the flute to sound less grainy when recording at higher sampling rates? The difference is night and day so I'm pretty sure it's not psychological. I suppose a blind AB test might be in order too (something I could totally upload for all y'all to hear for yourself too).

I can't imagine that it would be jitter because jitter is a product of the input signal screwing up the time code that the DAC will be reading. And like you mentioned earlier, all the circuits nowadays are upsampling anyway, yet the same peice of equipment sounds better running the higher sample rate.

[cwm]

"This is fun... isn't it? Kinda like a first person shooter...

Shawn's got the better reflexes..."

He certainly does and I admire the man's patience when he'll put this information down succinctly and some of the posters will come back with comments that confirm that they still "dont get it"!! I think a couple of these guys need a good further tutoring on how digital works------but then again I guess thats what Sfogg is trying to do!!

RJP

[DizRotus]

----------------

On <?xml:namespace prefix = st1 ns = "urn:schemas-microsoft-com:office:smarttags" />

5/11/2005

1:56:42 AM DrWho wrote: <?xml:namespace prefix = o ns = "urn:schemas-microsoft-com:office:office" />

"The difference is night and day so I'm pretty sure it's not psychological. I suppose a blind AB test might be in order too (something I could totally upload for all y'all to hear for yourself too)."

----------------

I would like to hear the comparison.

[Mallette]

This is, indeed, a lot of fun and most instructional. To those who've commented on fearing a "free for all," I'd say the fact that it is not is the reason I've continued to monitor it. About 98% of my activity is in the 2 Channel Forum, which, in spite of being largely inhabited by tube-loving skeptical ol fahrts like me, is also quite a civilized place (most of the time...)

In PM, I've queried about some of the technical references. My own experience is largely empirical, starting with dissatisfaction in the sound qualities of CD vs. LP, a belief that at least part of this was not due to "digital vs. analog," and a lot of experimentation to separate out those issues.

My first experiment in 1998 was extremely successful and has led to improvements in technique and my own understanding of the recording process. That recording, a piano recital (what a place to start...hardest thing next to applause on the planet to record convincingly) by Stuart Wayne Foster. I based my approach on the best of the analog tradition...direct-to-disc and simple mike technique. The mike technique came large from the legendary Mercury single-mike LP's of the early 50's. I used only two, and have remained with two since for stereo, four for surround. The signal path was mike>preamp(vacuum tube, of course...remember where I come from!)>ADC>storage. I immediately realized that most of the ills of commercial CD's lay in the stuff they had in between. The resulting recording has been praised by "golden ears" as being as convincing and clean a piano recording as they've ever heard.

Shawn and Dr. Who, and others here have really enlightened me as to the "whys" of my results, and I appreciate it. Thanks for making this ol' 2 Channel Forum guy welcome!

Dave

[sfogg]

DrWho,

"Do you get the same result by throwing in a low pass filter to take out the upper frequency (instead of just lowering the sampling rate)?"

I didn't try that but if I get a chance I will try it tonight. I have an eigth order low pass at 1800 hz I could put into the chain.

"One other question that comes to mind is how well are your computer speakers reproducing the 19 and 20 kHz tones? In other words, could the 1kHz tone actually be a distortion in the speaker?"

Pretty unlikely. Harmonic distortion would be at a higher frequency. And if I played either 19 or 20kHz tone alone there was no audible sound coming from the speaker at all. Also if I changed the frequency of either tone the beat frequency also shifted along with it.

"what other reason might there be for the flute to sound less grainy when recording at higher sampling rates?"

Are you sure playback levels are matched to 0.1dB? If not you might just be being influenced by SPL differences, it can trip up comparisons very easily and small level differences won't sound like a volume difference. It will be heard as things like impact, clarity..etc..etc.

Any chance the lower sampling rate has additional analog filtering that might be in the audible range?

Any chance the 44.1kHz recording clipped the A/D? Overloading an A/D doesn't sound good.

What happens if you record it at 96kHz but then resample the signal to 44.1kHz or 48kHz?

Have you looked at the higher sampling frequency file on an FFT or other form of a spectrum analyzer? Any chance it has a lot of ultrasonic noise in it? One of the theories on why DSD may sound 'sweeter' then PCM is because DSD has rising levels of noise as frequency increases. It is theorized that perhaps that ultrasonic noise is heating up the tweeter in a system and causing a little bit of power compression which would lower the tweeters level slightly relative to the rest of the system.

" The difference is night and day so I'm pretty sure it's not psychological"

Doesn't rule it out though. People claim night and day differences about a lot of things (cables come to mind) but when tested blind they can't tell them apart.

Shawn

[Mallette]

Shawn:

>What happens if you record it at 96kHz but then resample the signal to 44.1kHz or 48kHz? (I did not copy your informative response here, but it was what you did NOT mention I'd like your take on.)

I usually do 2 channel location recording at 24/88.2, as my experience suggests no detectable difference in 24/88.2 reduced to 16/44.1, though the differnece when coming from non-even multiples is definitely audible (for the worse).

Applause, one of the most difficult of sounds, sounds like rain or frying bacon when recorded either at 16/44.1 or reduced from 24/88.2. OTOH, it is pretty convincing at 24/88.2 and gets better as you move up. My assumption has been that there is, indeed, a deficency in the 16/44.1 CD standard that prevents it from dealing with the very most difficult signals. I've yet to hear applause that would fool me from any CD.

Your thoughts?

Dave

[sfogg]

Dave,

"though the differnece when coming from non-even multiples is definitely audible (for the worse). "

That will depend upon the algorythm that does the re-sampling. With enough DSP power you can re-sample transparently from non-even multiples. The way it works is you first oversample the signal to a higher rate (that is an even multiple of both rates) the re-sample it to the target rate.

"My assumption has been that there is, indeed, a deficency in the 16/44.1 CD standard that prevents it from dealing with the very most difficult signals."

What is the frequency distribution of applause?

" I've yet to hear applause that would fool me from any CD."

I have, but not when listening in two channel. I think a big part of the problem is simply that the applause on a CD in typical stereo listening isn't coming from the correct direction. It shouldn't be on stage yet in stereo listening it always is with maybe a little out of phase aspect to it to try and make it sound a little less distinct.

Shawn

[Mallette]

>That will depend upon the algorythm that does the re-sampling.

What algorythm do you recommend and where can I get it (them)? I'd prefer not to have to run a second machine at 16/44.1 or a multiple when recording primarily at 24/96 or higher for surround.

>What is the frequency distribution of applause?

Extreme high frequencies and air compression are required to be present for it to be convincing. At my age (56), it is obvious I no longer hear as I did in my 20's (out to 22khz-I could hear "silent) burglar alarms), but it is apparent they produce lower signals that impact the reality of the sound.

"I've yet to hear applause that would fool me from any CD."

>I have, but not when listening in two channel.

I am not aware of any CD format that supports more than two channels. Please comment.

>I think a big part of the problem is simply that the applause on a CD in typical stereo listening isn't coming from the correct direction.

In my experience, negatory here. I agree that it's weird to hear applause coming from the orchestra, but the comments about resolution made in my previous post are from my experience...it appears to be a matter of resolution as opposed to direction.

Dave

[DrWho]

----------------

On 5/11/2005 8:00:55 AM sfogg wrote:

"what other reason might there be for the flute to sound less grainy when recording at higher sampling rates?"

Are you sure playback levels are matched to 0.1dB? If not you might just be being influenced by SPL differences, it can trip up comparisons very easily and small level differences won't sound like a volume difference. It will be heard as things like impact, clarity..etc..etc.

----------------

Yes I'm definetly matching the input and playback levels...gain structure has a huge influence on the sound.

----------------

Any chance the lower sampling rate has additional analog filtering that might be in the audible range?

----------------

not sure

----------------

Any chance the 44.1kHz recording clipped the A/D? Overloading an A/D doesn't sound good.

----------------

I'm pretty sure it wasn't overloading. I tend to give myself 3-6dB of headroom when recording instruments like the flute. For louder more dynamic instruments (drums) I'll give myself at least 10dB...metering has a hard time keeping up with the percusive nature of drums and therefore peaks are higher than what your meters read (depends on the meters too).

----------------

What happens if you record it at 96kHz but then resample the signal to 44.1kHz or 48kHz?

----------------

The grainy sound comes right back at ya. I'm using CuBase but I have also done comparisons with other software as well and they're all the same.

----------------

Have you looked at the higher sampling frequency file on an FFT or other form of a spectrum analyzer? Any chance it has a lot of ultrasonic noise in it? One of the theories on why DSD may sound 'sweeter' then PCM is because DSD has rising levels of noise as frequency increases. It is theorized that perhaps that ultrasonic noise is heating up the tweeter in a system and causing a little bit of power compression which would lower the tweeters level slightly relative to the rest of the system.

----------------

Yes I've heard that one before, but so far it's only been discussed as theory. I don't have an FFT readily available so I couldn't perform this test. It's an interesting concept though and it would also have some effects on the amplifier as well...I suppose it's feasible that the distortion is creating a rounding out of the sound which makes it sound less harsh.

----------------

" The difference is night and day so I'm pretty sure it's not psychological"

Doesn't rule it out though. People claim night and day differences about a lot of things (cables come to mind) but when tested blind they can't tell them apart.

Shawn

----------------

Ya, that's why I'm going to go perform some double blind testing. It's one of those things though that when I listen to the mixes of the people I'm training, that I hear it and ask them why they recorded at 44.1 isntead of 96...and then I find out for sure that it was in 44.1 and they can't believe that I could tell the difference (Then I get a "slap your forehead" moment when they tell me that they'll just convert it to 96, lol). I suppose it could be one of those things that I'm really familiar with my equipment and perhaps there's a slightly different analog signal path for the two sampling rates that I associate as a difference between sampling rates (when it could in fact be a difference somewhere else). For what it's worth, I'm using Motu 828mkII's for my ADA processing.

[formica]

----------------

On 5/11/2005 1:56:42 AM DrWho wrote:

I can't imagine that it would be jitter because jitter is a product of the input signal screwing up the time code that the DAC will be reading.

----------------

Actually that is an interesting subject... and wouldn't mind more info on jitter ... although I realise it would be a little off topic here. There seems to be a lot of miss-information floating around about this and I wouldn't mind having a little more "accurate" knowledge on the subject.

A lot of the arguments seem to centre around where these timing errors occur and the result on the sound. My understanding is that once the data is read off the CD, buffered (some say it isn't...but seems unlikely), and then sent to the clock before being sent to the DAC.

What is the source of jitter? Missing data, inconsistent data flow, poor quality clock, misinterpretation by the DAC, etc... ? Would seem to be a product of the DAC circuit and not the transport?

What does it sound like? I realise in digital, errors usually aren't "nuances"... just abrupt changes like ticks, pops, clicks, etc... but is jitter a typical error?

Rob

[sfogg]

"What algorythm do you recommend and where can I get it (them)? "

Don't know I can't help you there, sorry. It may be totally impractical to do it (or not) but it can be done. From a quick calculation if you oversample 96kHz 147x you end up with 14,112kHz. 44.1kHz goes into that evenly 320x.

"Extreme high frequencies "

How high. If you have a 96kHz recording of just applause run it through a software analyzer and look at its distribution. Also check your 16bit recordings to make sure you aren't clipping.

"I am not aware of any CD format that supports more than two channels. Please comment."

I use surround sound processing to extract additional channels out of a CDs two channels based on amplitude and phase relationships between them. One of the things that typically does is take applause and move (steer) it into the four surround channels. Depending upon the mix on some CDs though applause is mixed such that that doesn't happen, it just ends up in the center speaker.

"but the comments about resolution made in my previous post are from my experience...it appears to be a matter of resolution as opposed to direction."

Keep in mind resolution is signal to noise... bit depth/word length, not sampling frequency. Sampling frequency is frequency response.

Shawn

[sfogg]

Formica,

Jitter is another one of those funny things.... almost anything wrong with digital playback people blame on jitter without really knowing what it is. Jitter is much less of an issue then some of the popular magazines make it out to be.

Jitter in a nutshell is timing errors. More on that later....

" What is the source of jitter? Missing data, inconsistent data flow, poor quality clock, misinterpretation by the DAC, etc... ? Would seem to be a product of the DAC circuit and not the transport?"

Actually that is a very good observation. The place jitter matters at all is when the data is fed to the DAC (or a DSP). Your signal can have loads of jitter earlier on but IF the data makes it into the DAC cleanly that earlier jitter was irrelevant.

You have to understand how digital data is actually fed to the DAC chip. It actually is done using typically four different signal lines. (For PCM audio, DSD does it differently) 3 of them are different types of clocks and one is the actual digital audio. Every one of the signals is nothing more the 5v square waves into the DAC, indicating a bit being high or low or a clock pulse being high or low. Logically that is the same thing as a bit being on or off.... a 1 or a 0...etc...etc. But physically it is done with square waves.

The first clock is a master clock that usually runs at some multiple of the sampling frequency. 192x,256x and so on are typical master clock frequencies.

The second clock is the bit clock. This runs at 64x the sampling frequency and shows where bit transitions would be on the data line. It is 64x the sampling frequency because one channels sample of data is fed to the DAC in a block that has 32 bits in it. (Each channel has 32bits so 64 bits total per sample for stereo) Up to 24 bits holds the actual audio data, the other bits are usually zeros by this point but are used during transmition to hold information about the signal.

The third clock is the L/R data clock. This clock indicates if the current 32 bit block is for the left or the right channel. This run at the sampling frequency.

I have a couple of scope pictures of this at home that I'll post tonight which will make this easier to understand.

Jitter is simply a timing error amongst those 4 signal lines such that the data line fed to the DAC is being mis-read as being low when it was really supposed to be high or vice versa. This could happen from bad square waves on the data line itself or it could also occur from problems on the clock lines. For example the way the DAC tells two high bits in a row (voltage stays high for both bits) is by looking at the bit clock. It will transition from high to low at the place the data line transitions from one bit to the next. So if the data signal stays at 5v but the bit clock has transitioned the DAC knows it has two high bits. Same thing for two lows in a row, the bit clock always transitions at bit break points.

Basically anything that can throw those signals off to the DAC (bad grounding, noise in the line, too much inductance in the lines...etc...etc...) is potentially a source of jitter at the DAC.

" I realise in digital, errors usually aren't "nuances"... just abrupt changes like ticks, pops, clicks, etc... but is jitter a typical error?"

Nothing like a pop or a click. If jitter actually causes bad data for a sample basically what happens is when that sample is recontructed to analog it has some extra noise in it that shouldn't be there. I don't believe the noise is typically broad band, more like a very narrow spike at a frequency. When tests for jitter are done they basically look for those noise spikes (at specific frequencies) on the analog output of the system.

Obviously if the jitter caused a problem for just a single sample the noise would be there for an *extremely* short amount of time.

Shawn

[sfogg]

"...metering has a hard time keeping up with the percusive nature of drums and therefore peaks are higher than what your meters read (depends on the meters too)."

If you check the digital file itself clipping should be very obvious. If a signal hits 0dBFS you most likely clipped it.

"I don't have an FFT readily available so I couldn't perform this test."

If you have a small clip I can try analyzing the wave with software I have. There are quite a few decent FFT/MLS computer programs out there. You might enjoy giving them a whirl. I use WinAudioMLS which is pretty good. One of its options is to analyze a wave file on the computer. The author from time to time offers licenses to it on ebay at *greatly* reduced cost. That is how I bought mine.

Shawn

[Mallette]

I fear I am not communicating very well... Let me attempt to rephrase and clarify.

"What algorythm do you recommend and where can I get it (them)? "

>Don't know I can't help you there, sorry...

I meant software. I've found Sound Forge fine since I only since I've not done any reductions not exact duplicates for serious work. I just wondered if you had recommendations on more advanced algorythms that might provide good results from uneven multiples.

"Extreme high frequencies "

How high. If you have a 96kHz recording of just applause run it through a software analyzer and look at its distribution. Also check your 16bit recordings to make sure you aren't clipping.

No clipping. I don't really know how high, but ultrasonic and well into the 20's. Also, the compression wave of cupped hands is extraordinary. That's where ribbon (there is a good reason for the other name, velocity!) mikes catch this accurately. Without it, you get the sort of muffled raindrop sound.

"I am not aware of any CD format that supports more than two channels. Please comment."

I use surround sound processing to extract additional channels out of a CDs two channels...

I am doing so as well now using DPLII and NEO6 after reading some of Jim Fosgate's work. I had been using a DynaQuad (Hafler) circuit on stereo material for the previous 30 years. These are the first I find acceptable for this application.

"but the comments about resolution made in my previous post are from my experience...it appears to be a matter of resolution as opposed to direction."

>Keep in mind resolution is signal to noise...

In my usage (which may not be semantically or technically correct), resolution refers to any information that may improve the final reproduction.

Dave

[sfogg]

" I just wondered if you had recommendations on more advanced algorythms that might provide good results from uneven multiples."

I don't know, I don't do recordings so I don't have experience with those types of tools. Ask around if in the processing of resampling 96kHz to 44.1kHz it first oversamples by 147x to make the math nice and even multiples.

"I am doing so as well now using DPLII and NEO6 after reading some of Jim Fosgate's work. "

I personally don't like NEO6 (too many steering artifacts) but DPLII can work well. It is definitly an improvement over his earlier work.

" In my usage (which may not be semantically or technically correct), resolution refers to any information that may improve the final reproduction."

OK, I'm using in the smallest difference that can be resolved sense. That is the same thing as signal to noise ratio. Once you get under the noise floor (of whatever) you can no longer resolve differences.

Shawn

[DrWho]

RE: JITTER

Here's a great website with pictures and a great explanation of the problem. The cool thing about jitter is that it has no effect when transfering from one digital domain to another. In other words, it's only an issue when the digital signal is being sent to the DAC.

http://www.digido.com/portal/pmodule_id=11/pmdmode=fullscreen/pageadder_page_id=28/

Here's a pic buried in the text that shows it well:

(notice that the time at the 0 crossing changes ever so slightly which is what throws off the time clock...it's not very obvious in the picture but it's there).

[DrWho]

and some more stuff about dither from the same site, which I found very interesting:

http://www.digido.com/portal/pmodule_id=11/pmdmode=fullscreen/pageadder_page_id=27/

[3dzapper]

Here is that Lp sine wave I promised. Not the best pic as the scope was sitting on a Heresy. The signal was a sweep and was changing on the right side.

http://forums.klipsch.com/idealbb/files/LPsine14.jpg