Do inter-sample peaks only matter during D/A (Mr Frindle?)?

[ben_allison]

Just a point I need to clarify: I totally get inter-sample peaks. What I wanted to know is if they only occur during the D/A process.

Suppose I have a track with material that has adjacent sample points at 0dbfs, and would (for the sake of argument) create an inter-sample peak of +2. But, this track is going into a mix buss whose fader is down -2.

Problem solved?

Or further, what if I have an EQ (pre-fader) across the buss? Assume I haven't trimmed the input, so it's seeing the two 0dbfs samples.

Is it even conceivable that this EQ could clip at +2 since the EQ, being ITB, is only working within the context of a sample rate?

Are inter-sample peaks even "visible" to most ITB processes since ITB, nothing is technically "inter-sample"?

Are inter-sample peaks really only important at the mastering stage, when preparing audio to be converted by a playback system?

[ben_allison]

Bamp.

[gwailoh]

I'm not PF. Very definitely. But I'm brave and foolhardy, and I've been putting a lot of work in over the last couple of months to implement PF's specific suggestions, going back and re-mixing completed mixes for instance following his guidelines. So I'll take a stab.

Caveat: I'm not a digital audio expert. My education re these issues is from reading and re-reading PF's posts. My answers here attempt to paraphrase some of PF's own answers on PSW and here on GS.

Suppose I have a track with material that has adjacent sample points at 0dbfs, and would (for the sake of argument) create an inter-sample peak of +2. But, this track is going into a mix buss whose fader is down -2.

Problem solved?

If you're mixing to a digital format (internal bounce, Masterlink, etc.) without going through your DAC, the problem is probably not solved. The mastering engineer will bring the level back up to near 0, and with it the hot signals which will clip the listener's DAC downstream. (Unless you tell him or her not to, I suppose!)

If you're mixing through your DAC to analog tape or a DSD recorder or etc., the problem may or may not be solved. You can't really know unless you monitor through meters which will show you the reconstructed signal level, rather than the DAW's sample levels. You're probably safe if you pull the master fader down far enough. But, maybe not.

Or further, what if I have an EQ (pre-fader) across the buss? Assume I haven't trimmed the input, so it's seeing the two 0dbfs samples.

Is it even conceivable that this EQ could clip at +2 since the EQ, being ITB, is only working within the context of a sample rate?

It might not clip the bus, if I understand the question. But the mix bus isn't where the problem lies. The problem is downstream at the DAC, where signals which are legal within the DAW's arithmetic can still turn out to be too hot for unclipped reconstruction during D/A.

Are inter-sample peaks even "visible" to most ITB processes since ITB, nothing is technically "inter-sample"?

I dunno that one. But "inter-sample peaks" aren't really the issue which PF has tried to explain to us all, especially in the wonderful PSW thread where he discusses a number of interrelated issues at length. The crux of the problem is the disparity between DAW headroom (big) and DAC headroom (small), exacerbated by the fact that default DAW meters display sample values, rather than reconstructed signal values.

Are inter-sample peaks really only important at the mastering stage, when preparing audio to be converted by a playback system?

No, because overly hot digital levels will impact monitoring of your track through your own DAC in your studio. In PF's view, the "too hot for the DAC" syndrome is what's responsible for many people's dissatisfaction with the sound of their DAWs, including complaints about narrowing of the stereo image, harshness, lack of depth, etc., compared to analog summing. All of those problems will exist in your own environment if the levels are too hot.

Paul: if you're monitoring GS these days, please correct me if I've misrepresented you!

[ben_allison]

Ok, so what you are saying is that it's not a problem that will manifest itself ITB (in terms of overloading plugs or internal busses).

While it should be addresses ITB, the problem only really rears it's head (clipping is only made manifest) when going through D/A.

[gwailoh]

Originally Posted by ben_allison

Ok, so what you are saying is that it's not a problem that will manifest itself ITB (in terms of overloading plugs or internal busses).

While it should be addresses ITB, the problem only really rears it's head (clipping is only made manifest) when going through D/A.

Exactly. If you're not seeing red LEDs on your DAW meters, or your plugin meters etc., you're not overloading the DAW. This is true on both the channels and the busses.

But, you might very well be creating sample values which are hot enough to cause clipping ("reconstruction errors") in D/A. Your default meters won't show you these D/A reconstruction errors. (Some plugins, such as the Oxford Limiter, will.)

Thus PF's -6dbr rule of thumb. Keep all levels everywhere inside the DAW, including specifically the inputs and outputs to and from all plugins, at -6dbr or lower, and things'll probably be fine. (Or, monitor everywhere with a reconstructing meter, which is more accurate but more work.)

[theblue1]

I should think a mastering engineer worth his salt would go out of his way to reintroducing the possibility of intersample peaks.

One alternative to mixing 'in the dark' with regard to intersample peaks is to use an intersample peak meter like this free one from SSL (accompanied by a nice, but brief explainer on the entire phenomena):

Solid State Logic | Music

[gwailoh]

The manual for the Oxford limiter also has a very nice explanation of the phenomenon: here.

Intuitively you'd think that the mastering process would remove these kinds of overs before transferring them to the final master. PF reports otherwise. Here's a snip from page 7 of the PSW thread:

For example, I have a few discs that I take to work for tests cos they are variously clean, punchy or just manically loud and all of them exhibit the problem. One I have of Shania Twain (a particularly quiet track) is possibly the worst of them all exhibiting bursts of up to 3dB reconstruction overs, even though the peak value meters never reach max.

[ben_allison]

Originally Posted by gwailoh

Thus PF's -6dbr rule of thumb. Keep all levels everywhere inside the DAW, including specifically the inputs and outputs to and from all plugins, at -6dbr or lower, and things'll probably be fine. (Or, monitor everywhere with a reconstructing meter, which is more accurate but more work.)

But the problem reappears when you raise the level 6db for final release... is the idea that you output the signal at -6db, and do your mastering processing outside the box?

[gwailoh]

Originally Posted by ben_allison

But the problem reappears when you raise the level 6db for final release... is the idea that you output the signal at -6db, and do your mastering processing outside the box?

Yes I believe so, if you're following the "rule of thumb" (-6dbr everywhere, including the master fader). If you're using a reconstructing meter I expect that you should be able to be more precise.

The Oxford Limiter BTW has the ability to automatically correct spikes that will cause reconstruction errors. I'm just beginning to experiment with that plug, so I don't know what impact that feature will have sonically. Interesting, though.

[Paul Frindle]

Originally Posted by ben_allison

Just a point I need to clarify: I totally get inter-sample peaks. What I wanted to know is if they only occur during the D/A process.

No they can occur anywhere the sample stream gets reconstructed into a notional signal. This can happen in filters, some dynamics processors and in EQ's where they are oversampled internally - and such things. The excess signal can theoretically rise to +6dBFS, I have seen real signals rise to +5dBFS, but most errors max out at around +3dBFS

Suppose I have a track with material that has adjacent sample points at 0dbfs, and would (for the sake of argument) create an inter-sample peak of +2. But, this track is going into a mix buss whose fader is down -2.

Problem solved?

Yes it's solved for any gear down line from the gain reduction point (provided that the gain is never raised again) - although I would leave an extra dB or so for security :-)

Or further, what if I have an EQ (pre-fader) across the buss? Assume I haven't trimmed the input, so it's seeing the two 0dbfs samples.

Is it even conceivable that this EQ could clip at +2 since the EQ, being ITB, is only working within the context of a sample rate?

That will depend on the kind of EQ, it's internal processes and what platform it's running on. The EQ may well 'see' the overloads internally, whether it clips or not will depend on the kind of internal headroom it has and whether it is running fixed or floating point. TDM processors (and some other exterrnal processing boxes and cards) are fixed point so they 'may' overload if there is no built in headroom. Host and RTAS processes are normally more tolerant of this and are unlikely to clip because the headroom is provided by the floating value representation.

Are inter-sample peaks even "visible" to most ITB processes since ITB, nothing is technically "inter-sample"?

Yes they are visible where any signal reconstruction is going on. For instance I have made the example of something like a LP high cut filter on an EQ where peaks larger than 3dB above the input level may occur at the output - even though you are 'removing' stuff and not adding anything anywhere and the gain is flat.

Are inter-sample peaks really only important at the mastering stage, when preparing audio to be converted by a playback system?

They are important to an extent everywhere, but the mastering stages are often when they are provoked. This can be bad as it can crack up the consumer's DACs in way you won't hear in the studio. Even simple compression that brings otherwise quieter things up in level (like high hats and percussion - even a vocal voiced quietly) can be major causes of inter sample peaks because they figure more prominently than in the original captured sounds.

The problem of inter sample peaks only occurs because your DAW meters are showing only sample values. Since this a pulse code modulation data stream, these sample values give rise to signal only when reconstructed and decoded. Therefore your meters are wrong and can read less than the signal they will represent after being decoded. There is nothing complex or magic about this at all :-)

[ben_allison]

Thank you SO MUCH for popping in here!

So say I don't let anything peak above, -6, anywhere. Great. I'm safe.

How do I then safely bring the level up 6db, and THEN go on to compress/limit to get the level to where it needs to be for release, all the while avoiding inter-sample peaks in the final master?

[staudio]

I prevent intersample peaks in final masters by:

Mastering/doing what I got to do, then checking for IS peaks with a meter like the SSL, if there are none I don't worry about it, if there are peaks I reduce the output ceiling a bit to see if that solves the problem, if I have to reduce the ceiling to below -.5dBFS and it still needs to be a hot master (sigh!) then I will consider using a IS peak limiter like the one in Ozone. I am not against further reducing the ceiling to avoid an extra process if it is acceptable by the client, but most people just want it super loud with a high RMS.

Many IS peaks can be avoided by not over-compressing or limiting, so keep that in mind when doing your processing. Also clipping AD converters in mastering can also create a condition that is conducive to IS peaks because it is often creating a stream of samples at -0dBFS (just like heavy limiting) which generally like to overshoot during an oversampling reconstruction.

Also I leave a max ceiling of -.3dBFS or lower by default which seems to take care of most IS peaks and any further changes in peak level due to sample rate conversion, dithering, and MP3/other format conversion.

[ben_allison]

Originally Posted by staudio

Many IS peaks can be avoided by not over-compressing or limiting, so keep that in mind when doing your processing. Also clipping AD converters in mastering can also create a condition that is conducive to IS peaks because it is often creating a stream of samples at -0dBFS (just like heavy limiting) which generally like to overshoot during an oversampling reconstruction.

Yeah. I mean, I think a dynamic range of 12-14db is a pretty fair compromise of "loud" to "dynamics" for rock (what I'm focused on). Peaks at -.3dbfs

This is possible without many (or any) IS peaks?

[Paul Frindle]

Originally Posted by ben_allison

Thank you SO MUCH for popping in here!

So say I don't let anything peak above, -6, anywhere. Great. I'm safe.

Yes indeed - this effectively solves the issue :-)

How do I then safely bring the level up 6db, and THEN go on to compress/limit to get the level to where it needs to be for release, all the while avoiding inter-sample peaks in the final master?

Ok this is where it gets tricky. Soon as you send your stuff to the mastering stage current fashion dictates that they will once again push the levels to the absolute maximum. That's apparently all anyone wants these days - even though in many cases it's creates illegal program that will sound awful on the end user's equipment - and all bets are off as to what will happen. But listening to almost any popular music CD these days reveals the results - you can hear them on any cheap ghetto blaster :-( One could be forgiven for concluding that all this 'mastering' level war is killing our whole industry; but that's another discussion :-(

You can use something like the Oxford limiter that will detect and effectively fix up the peaks by reducing levels momentarily only during the exact places they happen. This is better than simply turning everything down as the loudness in not affected when the IS peaks don't happen.

But even this will fail if at a later stage the mastering process does anything at all with the program - quite simply the IS peaks can (and will) come back again with EQ, further compression, limiting or almost anything else. This is because the current trend in mastering leaves almost no headroom or dynamic range - a reduction of -0.5dB taken grudgingly and daringly by the mastering engineer in the hope he won't lose his job is just not enough :-(

Ok - so then this gets us back to the subject of the PSW thread. Given that mastering (and mixing) these days is all about actually creating and then handling what are effectively self-generated errors in the signal itself, how these respond and sound can vary considerably depending on how the gear handles these errors.

So for instance one way of dealing with it is to chose a suitable DAC (probably expensive) that does not not crack up too badly with the errors - feed the result to a load of analogue gear (that has headroom - because they can't be made any other way) - then re-coding the whole thing back to digital using an ADC (probably expensive) which will skim off the IS peaks (as it cannot do otherwise) and sound appropriately 'not too bad' as it does so.

However all of the above messing with signal paths to handle self-generated errors of course completely denies us all (musicians, engineers and consumers) the amazing potential advantages of an accurate and repeatable digital signal chain, which is of course that it can (or at least could if implemented correctly) transmit signals without any audible errors or degradation - right into the end users reproduction equipment.. That was supposed to be it's big advantage :-)

[Mike Brown]

Originally Posted by Paul Frindle

Yes indeed - this effectively solves the issue :-)



Ok this is where it gets tricky. Soon as you send your stuff to the mastering stage current fashion dictates that they will once again push the levels to the absolute maximum. That's apparently all anyone wants these days - even though in many cases it's creates illegal program that will sound awful on the end user's equipment - and all bets are off as to what will happen. But listening to almost any popular music CD these days reveals the results - you can hear them on any cheap ghetto blaster :-( One could be forgiven for concluding that all this 'mastering' level war is killing our whole industry; but that's another discussion :-(

You can use something like the Oxford limiter that will detect and effectively fix up the peaks by reducing levels momentarily only during the exact places they happen. This is better than simply turning everything down as the loudness in not affected when the IS peaks don't happen.

But even this will fail if at a later stage the mastering process does anything at all with the program - quite simply the IS peaks can (and will) come back again with EQ, further compression, limiting or almost anything else. This is because the current trend in mastering leaves almost no headroom or dynamic range - a reduction of -0.5dB taken grudgingly and daringly by the mastering engineer in the hope he won't lose his job is just not enough :-(

Ok - so then this gets us back to the subject of the PSW thread. Given that mastering (and mixing) these days is all about actually creating and then handling what are effectively self-generated errors in the signal itself, how these respond and sound can vary considerably depending on how the gear handles these errors.

So for instance one way of dealing with it is to chose a suitable DAC (probably expensive) that does not not crack up too badly with the errors - feed the result to a load of analogue gear (that has headroom - because they can't be made any other way) - then re-coding the whole thing back to digital using an ADC (probably expensive) which will skim off the IS peaks (as it cannot do otherwise) and sound appropriately 'not too bad' as it does so.

However all of the above messing with signal paths to handle self-generated errors of course completely denies us all (musicians, engineers and consumers) the amazing potential advantages of an accurate and repeatable digital signal chain, which is of course that it can (or at least could if implemented correctly) transmit signals without any audible errors or degradation - right into the end users reproduction equipment.. That was supposed to be it's big advantage :-)

Question:

Would a LP filter set to near or below (20k) the nyquist frequency possibly reveal any clipping which may occur from the reconstruction filter?

[Paul Frindle]

Originally Posted by Mayor999

Question:

Would a LP filter set to near or below (20k) the nyquist frequency possibly reveal any clipping which may occur from the reconstruction filter?

Yes - to some extent it will :-)

However it will not be an accurate result because of phase shifting within the filter that actual reconstruction does not (or should not) have.

Also you must bear in mind that what comes out of the filter to be metered is also at the base sample rate - so it too can have IS overs that are not revealed - and so on :-(

What's needed is an oversampled (or multiphase) filter with similar general characteristics of a real DAC...

[zboy2854]

Paul, you mentioned that the Oxford Limiter can deal with IS peaks. What about the limiter section of your DSM?

[Paul Frindle]

Originally Posted by zboy2854

Paul, you mentioned that the Oxford Limiter can deal with IS peaks. What about the limiter section of your DSM?

No, the limiter section of the DSM does not process inter sample peaks.

The DSM was not really primarily conceived to replace a fully featured mastering limiter - but it can be used in front of an IS limiter like the Oxford without problems.

In this case it's best to let the DSM do the program limiting and maximisation - and use the Oxford to deal only with the inter sample peaks.

I can provide a set-up for the Oxford that will do this if it helps.. I don't have any other IS limiters to try though..

[Monobasser]

Originally Posted by Paul Frindle

I can provide a set-up for the Oxford that will do this if it helps.. I don't have any other IS limiters to try though..

Hi Paul, just bumping this thread! Yes I would be interested in this.

[vx2000]

bump