I received a response on another Vegas forum which seems plausible. It suggested that if "Full Scale" (0 dBFS) is defined as maximum level of a 15-bit codeword then the 16th (most significant bit) bit can be used to represent values above the "Full Scale" reference point. This would effectively allow you to represent values above 0 dBFS by using the full 16-bit word. When going to the D/A conversion process only the lower 15 bits would be used to determine signal magnitude.

I don't know if this conjecture is true, but if it is, then the digital level meter would have 6 dB of measurement capability (the difference between the 16-bit and a 15-bit codeword) above 0 dBFS.

--Rob

I'm not sure you understood my question.

My question was related to how values greater than 0 dBFS are digitally represented if 0 dBFS (the "full scale" reading) is defined as a digital word with all ONES.

Sorry for being unclear.

--Rob

Anything above 0dB would be assigned the maximum of the 16bit word, thus if a significant amount of samples went over the 0dBFS 16bit word, then there would be a continuous flat line, thus this would be distortion or a clipped peak. The 15 bit explaination really doesn't hold any water. You can easily verify this for yourself by extracting audio from a CD which has a 0dB peak sinewave and is 16bit. If you play that back in Vegas with all the faders set at 0dB, it will read 0dB peak on the meters. From the 15bit explanation it would display it at -6dB peak, which would be incorrect. Vegas can do floating point math, which means it can do calculations of amplitude well over the 16bit 0dBFS allowable maximum. It can therefore take that information and use it to tell you how far you are over the 0dBFS point and display it on the peak meter.

Again, if you want to test this out for yourself, then take a 0dB peak 1Khz sinewave, and place it on a track. Raise the track fader +6dB, with the master fader set at 0dB. Then render it to a new mix. When you open that mix you will see a sinewave with the peaks clipped off, where the volume exceeded the 0dBFS digital limit.

I think Reds explaination is right-on. It might help clear things up by letting you know that Vegas does all of it's mixing in the floating point domain and then only converts to a fixed binary length (8, 16, 24 bits) at the final stage. Therefore the internal mixing and computation can exceed what the chosen binary output bit depth can represent and that's why it "knows" that it's over 0dB.
Hope this helps more than confuses the topic.
B.

No, this isn't correct.

In fixed-point form the samples are represented in so-called 2's-complement form. This means that positive samples are represented normally from 0000 0000 0000 0000 up to 0111 1111 1111 1111 and negative samples are represented by numbers from 1111 1111 1111 1111 down to 1000 0000 0000 0000 - in effect the most significant bit is a sign bit.

Vegas does all its internal processing in floating-point form so it is able to correctly represent values which would be greater than the 0dBFS level. When the values are converted back to fixed-point form (e.g. when they are saved to a WAV file) any values greater than 0dBFS would of course be clipped to 0dBFS.

I expect someone from Sony will be along soon to explain it better but that is the jist of it as far as I am aware.

Mark

I think we're all saying the same thing, but, in my case, not very clearly.
B.

"No, this isn't correct"
Mark,
Why is this not correct? Yes, I agree with you on the 2's compliment form information. Without going into that detail and complicating things any further, thus making more confusion this information would be correct. It's all relative and irralevent what the value of 0dBFS is in binary, it could be anything and there's actually 3 or 4 different binary representations used in different products.

When I said "This isn't correct" I was specifically refering to the assertion made in the message I was replying to, the assertion that the most significant bit represents some sort of headroom and that this is what allows for the representation of values greater than 0dBFS.

This assertion was presumably made because whoever originally made the assertion had noticed that the maximum positive value of a sample is represented by 0111 1111 1111 1111 (in 16-bit fixed-point, at least). I imagine that they reasoned "Aha, look, it's only using the last fifteen places - I expect the sixteenth place is there for a bit of headroom. Yes, that would allow 6dB of headroom." What they had failed to note was that a fixed-point representaion scheme has to take account of negative values as well, and these are represented with all the values where the top bit is 1 - it is effectively a sign bit.

There is no "headroom" implied by the fact that the top bit in the representation of the maximum positive value and I was simply attempting to correct this misunderstanding in my previous message.

I then went on to explain the real reason why Vegas can represent values greater than 0dBFS - because of its use of floating-point format when manipulating values internally.

I hope that clarifies things a bit further.

Mark

Yes, that is helpful.

Thanks!

--Rob

Thanks to everyone for their thoughtful responses.

One point is still is a little unclear. How is it then on some occassions (most notably when my mic levels were too hot for the available dynamic range of my digitial 8 camcorder - see audio chain below) that upon immedidately capturing and then playing back the stereo digital audio track is Vegas able to show audio levels above 0 dBFS? I understand why I have clipping (mic levels were too high), its just how is the magnitude of this clipping being inferred by Vegas? If the audio recorded by the digital camcorder exceeds the dynamic range of the A/D, then wouldn't an all ONEs digital word be produced by the camcorder?

I buy the fact that Vegas does its number crunching with more precision than 16-bits, but does this also explain this simple capture and playback example? (No mixing involved, or is there?)

The bottom line is keep all your peak levels below 0 dBFS, which I do understand. I'm just trying to "look under the hood" a bit.

Thanks

--Rob

Audio Signal Chain
==================

Rode NT5 mics => Rolls PB224 Dual Phantom Power => Studio1 XLR-PRO BP (XLR to 1/8" stereo plug transformer) => Sony TRV-840 (consumer Digital 8 camera with (unfortunately) full-time AGC).

I think the answer is that Vegas is displaying both peak and average values. The peak value cannot go above O dBFS but the average can on the meters. For example one single peak at the max value will not push the average very high, 100 of them in a row will.

Depending on the metering system there's a peak to average ratio set, maybe 20 dB (yes I know it can be LOTS of values). For tone that means that when the peaks are at 0dBFS the average reading will be -20dBFS. However that makes for a meter that's hard to read, so the readings are normalised so that both the peak and average displays are at 0. That does mean however that your average can go above 0.

Hope I got that right, bit late in the day fro this old brain.

Farass,
Where the heck did you get that information from? If this was true in anyway then it's the exact opposite of what it should be. Peak meter readings are always higher than an average meter reading. I'm not quite sure exactly what the original question was asking, since the signal chain left out how audio is getting transferred from the camcorder to Vegas, which is the biggest part of the puzzle. I'm guessing and assuming the question is regarding why the over or clip indicator turns on. If consecutive samples are set at the max value, then there will be a flat line on the waveform drawing, thus indicating that there is distortion and a peak has been clipped. So if consecutive samples are played back at max value then Vegas does the right thing and tells you, you went over 0dBFS.

Here's a question for you Rtbond. Why don't you try turning your sh*t down and aim for levels at -6dBFS or less like you're suppose too and you won't have to worry about all the clipping and over indicators lighting up all the time?

.> I'm not quite sure exactly what the original question was asking, since
> the signal chain left out how audio is getting transferred from the camcorder to Vegas,

Sorry, its a Firewire data transfer (IEEE 1394A) from the camcorder into Vegas.

>Here's a question for you Rtbond. Why don't you try turning your sh*t down
>and aim for levels at -6dBFS

As I stated in my last message, I understand why I have seen audio clipping, and understand how to prevent this is the future (as you have not so nicely stated, reduce the mic levels).

I was simply curious as if there is clipping at the A/D of the camcorder, what is the Vegas level meter doing to determine that peaks in the captured digital audio are ABOVE 0 dBFS (for example at +1.2 dBFS, which is something I have seen Vegas measure when the camcorder input levels were too high). If Vegas sees a bunch of audio samples at the maximum, and this is defined as 0 dBFS, why does the meter in Vegas measure peaks ABOVE 0 dBFS?

This issue is not holding me up from doing anything, I'm just trying to understand what Vegas is doing in this instance. Not a big deal, however.

Sorry if you are getting tired of the discussion.

--Rob

Not getting tired of the discussion, but you sure do tend to leave a lot of important details out of your questions. You finally mentioned where you're seeing the levels going over the 0dBFS point. You mention an A/D converter in the camera. What type of tape are you recording too? Analog tape? or a digital tape? Let's just pretend everyone here doesn't own exactly the same type of equipment as you and may not be familiar with a "Sony TRV-840 (consumer Digital 8 camera)" Seeing that it says it's a Digital 8 camera, I have to assume it's recording to a digital 8 tape, and therefore recording digitally to the tape. So when you playback, it should directly take the digital data from the tape and put it out the firewire digitally. So where is there an A/D conversion in this process? A little detail of exactly what you are doing when you are experiencing the problem will go a long way. You can use Vegas for a thousand different types of audio or video tasks, how do we know what you're doing unless you give us the details?

I defer to whomever as this may or may not be relevant, but ion many many many devices there is a setting as to what exactly does constitute an over. It can be an average of many consequetive samples. Say on some tascam gear it needs to have more than a few overs before the over light lights up.

Or possibly is it this phenomenon http://www.tllabs.com/files/Digital%20distortion%20white%20paper.pdf ?

>A little detail of exactly what you are doing when you
>are experiencing the problem will go a long way.

Sorry, I wasn't my intention to omit detail, it's just when you are not fully informed on a topic, you may not know what details are important, and yes sometimes make assumptions.

The camera is a Digital8 camera recording to digital tape.

>So where is there an A/D conversion in this process

The A/D is in the audio path of the Digital8 camera.

Let me know if you need for more info.

--Rob

Ok, thanks now this makes sense to me. What I think you are experiencing is similar to what I've seen using Tascam DA-88 recorders, which also used a Hi-8 digital tape to record digital audio. On the tascams if you went over the 0dBFS limit, upon playback you would get something a little worse than distortion. It actually ommitted a kind of zapper/static noise that would scare the hell out of you, when you listened to the playback. I don't think the max binary value for distortion was assigned for that overload, but something random because the recorder didn't know what to assign to it because it ran out of data points. I have a feeling that your Sony Camera probably is doing a similar thing and spitting out some random garbage whenever you overloaded the input to the camera.

Now that my brain is not so foggy and at the risk of getting my head bitten off (again), I'll have another go.

Firstly just how the audio got there is irrelevant, the question relates to the metering in Vegas of a file of digital audio. What A/Ds, mixers, mics or whatever created it is irrelevant.

Most audio metering systems include some form of integration, i.e. they're displaying an average value over a defined time period. A program level meter will have a fairly long integration period, it's showing you the integral of the values over a sampling period. The old VU meters did this mechanically.

Even peak level indicators have some integration over time, if they didn't the thing would be jumping up and down so much it's be useless. So one FS value during the integration period is going to give a different reading to 10 FS values during the integration period. That means the metering system will typically have some headroom. It needs to be able to meaningfully display the differences between one sample at FS and 10 samples at FS during its integration period. The only way to see what is trully happening is to look at the waveform. That's why engineers don't rely solely on meters, they use oscilliscopes.

The way to test out what I'm talking about is with tone burst. Start with a 1KHz tone burst at 10% duty cycles, a pretty typical test signal. Meter it. Now change the duty cycles to 50% or 100%, shouldn't see much change. Now start reducing the duty cycle. You'll reach a point where the peak level reading starts to drop but the peak level hasn't changed has it.

When I've got some free time I'll have a go at this myself using SF. I'll bet also as I change the type of metering system I'll get different results.

Farss,
Sorry you took it, that I was bitting your head off. That wasn't my intent. Just didn't understand what you where trying to say and what you did say sounded backwards to me.

Rednroll,
you're right, my brain wasn't functioning! I knew what I was trying to say but I don't think what I said made much sense either.

Problem I find is trying to explain many things using just words, I write the odd user manual, when I go back and read what I've written some time later damned if even I can make much sense of it without the diagrams. My saving grace is no one bothers to read the damn things anyway.

Maybe that explains the overall decline in the quality of technical doumentation, no reads the stuff cause it's too hard to understand and the writers don't put in much effort because they figure no one will read it anyway and the whole thing spirals downwards.

I believe I follow your explaination of the use of some form of averaging in determining the peak audio levels displayed by Vegas, including the impact on how frequently peaks occur (duty cycle) will have on what the meter displays.

The concept I'm struggling with in your explaination is how any form of averaging would result in average peak values **greater than** the peak value of any individual sample in the set of data points being averaged. If the peak value of any audio sample is limited to 0 dBFS (by definition), then how can any form of averaging of peaks values be greater than 0 dBFS? Sorry if I'm using something.

As I previously stated, I've seen Vegas indicate peaks values greater then 0 dBFS for a digital audio signal acquired from a Digital 8 camcorder which (admittedly) was probably experiencing clipping. (I know how to prevent the clipping in the future, I'm just trying to understand the meter behavior in this situation)

Best Regards,

--Rob

One thing to consider is that when the peaks clip the waveform approaches a square wave more than a sine wave. At any given peak level a square wave contains more energy than a sine wave, up to 41.4213562% more in extreme cases or 3dB. So, if your waveform is clipped great enough that it resembles a square wave it has the same loudness as a sine wave that is up to 3dB higher. Vegas' meters are probably taking this into account when generating their display.