First step -

If you're preparing your track for mastering - don't use a limiter. Period.

Let your mastering engineer use his/her limiter.

Your mix should have headroom prior to going to the ME. If it doesn't, there's very little they can do to your mix except fix (to varying degrees) the horrible mess made by the plug-in limiter you're using.

Sorry - this doesn't really address the problem of the clipping limiter. I'm just saying, don't use it.

Cheers-
Jeremy

Look for "mystery bass" - something may have some serious low end stuff that isn't showing up on your speakers or in your tracks. If you can't hear it, you may be able to see it with a spectral analysis viewer, etc. It might be a track that shouldn't normally have much bass anyway, so it could be fooling you.

I'm with Jeremy though - don't do any comp/limiting at all if someone else is going to master your tracks.

This also could be your computer crapping out on you. Make sure you aren't using up all your CPU power. A limiter on the master output can really take up some resources

I'm pretty confident that your problem is intersample peaks. Short answer: Don't hit -0 dB, go to something like -0.2 dB or possibly a little lower. That should clean it up a lot.

Long answer: When your sound card converts the digital samples that make up your waveform into a smooth, analog wave, the analog wave does not end up looking exactly like it did when it was digital. If there are two samples together that are higher than the ones on either side of them, the analog wave after conversion can (and probably will) actually go higher than those samples did when they were digital at the point in between where those two higher samples were. So if the digital samples were hitting -0 dB, you might actually end up with a positive number, which means clipping! I believe that's what you are hearing.

Getting new converters might (depending on the design) get rid of the problem...on YOUR machine. This can still be a problem on other people's machines, since their converters might do what yours are doing now. That's why my suggestion is just to master it at -0.2 dB or so, since that will provide a bit of extra headroom for intersample peaks to fall into and still not clip.

There is software that will show you if you are likely to have intersample peaks. And I'm sure analog equipment exists as well that could tell for certain. I'm not sure what that software is (sorry, I've totally forgotten), but I'd recommend using it if you can find it. You might also want to try a different limiter, if one exists that is designed to prevent intersample peaks from happening (again, I have no idea if that exists).

Hope this helps.

Not quite.

The sampling rate has nothing to do with amplitude peaks. That is a function of the bit rate. There is no way to exceed 0dBFS.

If I understand what you're saying, it's that the sample points occur on either side of a sinusoidal wave and that the peak itself extends higher than those sample points...

This isn't correct.

Again - amplitude is a function of bit rate, not sampling frequency.

A signal that is represented as:
111111111111111111111111

is clipped.

A signal that is represented as:
111111111111111111111110
is not
(assuming 24 bit processing).

Actually - edit of sorts - the first representation (24 '1s') is not technically "clipped" until there are two samples (or 3 if you listen to Sony's definition of clipping) that are successive which contain all 1's.

This too is an oversimplification since not all of the bits are truly used for the audio portion, but it's a good enough representation to be used as an example.

Hi

Thanks all for your replies and in answer...

Cucco: I've been mastering my own tracks now for years. The limiter is not put on my main mixes. I mix down, then start a brand new project where apply the mastering chain.

Crankitup: Im getting the problem with only a limiter in the chain. My CPU can handle that..it has before. Which is why i dont understand it. It happened on my old card but went away on its own lol. Now i have an emu 1212m it has returned, and yes it has been fine on my emu previously.

danbronson: This is exactly what i thought at first too, but in experimenting ive dropped it as far as -7 and yet it still is yeilding the same horrible as soon as the limiter kicks in.

I really do appreciate all your answers, this is driving me crazy! lol. It's happening right in the middle of recording an acoustic album. Doing my head in!!

If you can help further id be very greatful.

Acoustic album, eh?

Perhaps it's the inherent wider dynamic range of the instruments that's causing the problem?

What limiter(s) have you tried so far that are giving you trouble?

Well everything really lol.

I have UAD-1 package with the precise limiter. Also i tried waves L2 and voxengo one from a friend.

I dont think it's the software though, they all seem to give the same results.

I've tried sweeping frequencies in the mastering process before the limiter, also doing a heavy compression before going to the limiter to try and squash any naughty dynamics but doesnt seem to make any difference.

I'd start pointing the finger at my alesis monitors if it wasnt for the fact that i had the problem before and it went away. Speakers don't fix themselves. Shame.

Oh yes they do...
Poke the cable about a bit, hear it burst into life. Low end connections, grr.

Maybe your limiter is being set unbelievably harshly or some setting has snuck under your radar and is screwing up the settings. Maybe something like the input gain being up or auto adjustments...

Hmmm...all sorts of possibilities.

Perhaps the output of the limiter is set to a value greater than 0.0?

Perhaps the routing of limiter is the problem? Maybe it's before another effect in the chain? Something I found interesting about my UAD Precision Limiter is that for some reason put into the bus prior to the master output. If you put a master bus EQ on, it comes after the limiter. This will cause clipping for sure.

My point is - don't just assume that because it's turned on, it's assigned in the right order. I'm not sure what software you use, but with Sequoia, you simply click the routing button and you can move the limiter to the last piece in the chain and problem solved.

Cucco wrote: Not quite.

The sampling rate has nothing to do with amplitude peaks. That is a function of the bit rate. There is no way to exceed 0dBFS.

If I understand what you're saying, it's that the sample points occur on either side of a sinusoidal wave and that the peak itself extends higher than those sample points...

This isn't correct.

Again - amplitude is a function of bit rate, not sampling frequency.

A signal that is represented as:
111111111111111111111111

is clipped.

A signal that is represented as:
111111111111111111111110
is not
(assuming 24 bit processing).

Actually - edit of sorts - the first representation (24 '1s') is not technically "clipped" until there are two samples (or 3 if you listen to Sony's definition of clipping) that are successive which contain all 1's.

This too is an oversimplification since not all of the bits are truly used for the audio portion, but it's a good enough representation to be used as an example.

Perhaps I didn't explain it well. Here is a link that explains with helpful diagrams what I'm referring to.
http://www.hometracked.com/2007/11/08/prevent-intersample-peaks/

But nick1982, if you brought the output to -7 and the distortion was still there, I'm not sure what the problem is. Make sure the software you're using can handle the sample rate and bit depth of your recording.

No - the information on that page is plainly incorrect.

They are assuming that the amplitude and frequency are represented by a single numerator (operating in time as a straight line).

It's not. It is represented by a plot of two factors (frequency on the x domain, and amplitude on the y domain.) Since this is sampled 44,100 times per second (for 44.1kHz sampling), the time line is drawn by the assembly of each of these samples.

Seriously - that page is completely wrong and its author should be beaten with a rubber hose.

What he's referring to is simple clipping - if an analog signal goes beyond the capabilities of the digital representation (greater than all 1's), then the signal is clipped and you have 2 or more concurrent 0dBFS samples. There is nothing that occurs between samples. Period. Yes, in analog, there is. In digital, there is not.

This page propogates bad information that is often misunderstood.

Cucco wrote: No - the information on that page is plainly incorrect.

They are assuming that the amplitude and frequency are represented by a single numerator (operating in time as a straight line).

It's not. It is represented by a plot of two factors (frequency on the x domain, and amplitude on the y domain.) Since this is sampled 44,100 times per second (for 44.1kHz sampling), the time line is drawn by the assembly of each of these samples.

Seriously - that page is completely wrong and its author should be beaten with a rubber hose.

What he's referring to is simple clipping - if an analog signal goes beyond the capabilities of the digital representation (greater than all 1's), then the signal is clipped and you have 2 or more concurrent 0dBFS samples. There is nothing that occurs between samples. Period. Yes, in analog, there is. In digital, there is not.

This page propogates bad information that is often misunderstood.

I don't understand what you mean. How is it wrong? Are you saying intersample peaks don't exist (created in the DAC process)?

Keep in mind, nobody (by that I mean myself or the page I linked to) has claimed that the digital level can go higher than 0 dB. Just that if two samples next to each other hit 0 dB, the converted analog signal at the point in between those samples can exceed 0 dB, depending on the converters used. You're saying that's incorrect?

I'm far from the most educated person out there when it comes to this stuff, so I appreciate any correction to this you can give me!

It's just it doesnt explain why it has worked for a year previously. This is what is baffeling me. It's not new hardware or software. It's all worked fine, then it didnt...then it did again for a long while and now it doesnt lol.

That's why im having trouble finding the problem as its worked flawlessy with my current setup for months on end.

:(

In this case I think the best way to think of the digital data is as a sample (an instantaneous point value) along the continuous curve of the analog signal (rather than as a step function as drawn in the linked site). If the analog signal has been band limited (as it should be) the samples will reproduce the incoming analog wave exactly. (This is known as Nyquist's theorem. Doesn't work exactly this way in practice, but close enough for the point being made in the link). Now if you sample a wave form at any set of points other than exactly on a peak, the samples will reproduce the original peak which will be higher than the samples on either side of the peak. But in order to produce a peak higher than the voltage corresponding to 0dBfs you have to put in a signal with a peak higher than that voltage. He can claim all he wants that this is not digital clipping, but it looks like clipping, smells like clipping, and it certainly sounds like clipping. Simple solution -don't input a voltage that big.

Cucco wrote: There is nothing that occurs between samples. Period. Yes, in analog, there is. In digital, there is not.

Most modern DACs use oversampling. In other words, they generate lots of extra samples in-between the ones you send to the DAC...

I think we're getting off topic here. I'm assuming you've restarted your computer

You're right - we're getting way off.

I'll answer only because direct questions were asked.

Bob has already stated the vast majority of the point.

I will add that yes, DACs can freak out if fed consecutive samples at FS. However, most DACs simply do what they're designed - that is they put out a specific voltage (analog) based on a bit representation (digital). If they receive the same bits consecutively, they output the same voltage.

To IIRS -
Yes, smartie...in oversampling, there are additional samples created between the values in the traditional base sampling rate. However, this is done for what purpose? Ah..to move the dessimation and anti-aliasing filters higher. There is still nothing which occurs between the newly created samples (say there's 4x oversampling - the samples are now at 176,400 per second - there's nothing between those newly created samples.

Additionally, there is nothing within those samples that is representative of a new numerical value that wasn't already contained within the 44.1 kHz sampled data. (If there is, I suspect your DAC is a little broken.)

To put this WAY more simply -

(Assumption - operating at 16 bit, 44.1 kHz)

There are 44,100 pictures taken in a second. Each of those pictures represents a sample. That sample is represented by a binary number 16 digits in length. That binary number determines the amplitude of the signal. The range of possible numbers is between 0000000000000000 and 1111111111111111 with over 16,000 possible numbers.

To create a signal over 1111111111111111 would be impossible, therefore all voltages that come out of a DAC should be a specific equivalent to the digital counterpart - always.

Let's look at sample 1 and sample 2. If these are both -0dBFS (or 1111111111111111), there is no picture in between which represents some number theoretically higher in value. That means at that first pulse (sample 1) the DAC will output a specific voltage burst. At the second sample, it will output an identical voltage burst (these "bursts" are the "Pulse" in "Pulse Code Modulation" or PCM). A DAC cannot and should not make assumptions as to data which may or may not occur between these two snapshots. If it does, it is broken.

Does this help?

Thanks, that clears up a lot. So essentially, you're saying intersample peaks are a myth.

They should be - although, some poor implementations of DACs can in fact cause this, but I haven't encountered one (even in $29 DVD players) in a LONG time.

Cucco wrote:
To IIRS -
Yes, smartie...in oversampling, there are additional samples created between the values in the traditional base sampling rate. However, this is done for what purpose? Ah..to move the dessimation and anti-aliasing filters higher. There is still nothing which occurs between the newly created samples (say there's 4x oversampling - the samples are now at 176,400 per second - there's nothing between those newly created samples.

Additionally, there is nothing within those samples that is representative of a new numerical value that wasn't already contained within the 44.1 kHz sampled data. (If there is, I suspect your DAC is a little broken.)

You are assuming that the DAC uses simple linear interpolation when upsampling. Actually they use reconstruction filters which are perfectly capable of generating sample values higher than those fed into them.

( Drum Track )

Did you never notice that an EQ can increase the peak levels of your audio even when cutting frequencies, not boosting?

This is true.

However, one thing that alll of these pages are omitting is -

The potential for that peak in excess of 0dBFS still does not present any significant potential danger to the signal as if this occurs before or during the reconstruction but not yet at the voltage pulse point, the signal will be rendered as 0dBFS. Yes, this is officially "clipping," however, we've already acheived this by this point if we're driving the DAC to this level.

If it occurs after the reconstruction, then it will simply assign a value (in volts) greater than that of the signal represented by -0dBFS. Either the device can handle this or there will be a small fire (tongue in cheek).

In any case, while there is the "possibility" for this based on different means of upsampling, the problem is not as indicated by many of the websites on the subject and if levels are kept even to -.1 dBFS, this should mitigate all or nearly all of these issues as the peak will simply then be rendered as -0dBFS or the peak voltage for the output stage of the DAC.

In any case, it would not cause the problem of excessive clipping sounds coming from the use of a limiter - no way no how.

Maybe the source signal itself is clipped. Try bypassing the limiter and listening.

Or maybe you're brickwalling it by accident...and it's chopping the peaks.

Thanks codemonkey, i got a bit lost some of those posts to how it was gonna help me. But anyways, yeah ive tried resampling. I did try increasing buffers originally first, but after not getting very far i increased them to 100ms (maximum). The horrible audio distortion went away. Although that was a fix, 100ms is blatently un-usable for recording, mixing and mastering as the peaks and stat analysis software is slow and unaccurate which is no help to me.

This leads me to believe that its a driver problem perhaps. So im approaching the problem from this angle and seeing if i can find some answers. EMU's driver isn't renowned for its friendly-ness i believe anyways...

So back to the drawing board for me..

Thanks all. :)

Are you using the new drivers? If not, give them a try and if so, try a rollback. Even just possibly reset to defualts and reboot might do it if you haven't tried.

Thank you nick1982...
Do you know that up until you pointed that out Ive never made the connection between buffer size and slow analysers/meters.