Dynamic range

[argentino_sf]

I suppose that during the recording, an orchestra volume goes from 50dB to 70dB (macro-transients).

 

How dac and amplifier know that is going from 50dB to 70dB?

 

I know you are going to say something like dealing with dynamic range in the digital media (CD, max. 96dB?; SACD, 120dB?; vinyl?) and voltage/current transformations in the dac and amplifier, but what I would like to know is how the amplifier faithfully reproduce the correct amplitude if the volume knob is fixed (read user not controlling the volume).

 

Just reading the dynamic range recorded in the media?

 

I know that single ended outputs have an "standard" power output (2V Rms), but that does not mean that every dac has the same peak to peak. And differential outputs are usually higher, lets say 6V Rms, but again, not sure about peak levels.

 

I also know that we are able to set a comfortable (on average) volume level with an attenuator when we hit play and the DAC starts (but which average level is the right one to choose? how do I know the average level during the recording?).

 

I think the dynamic range that is fixed in any media and the dac+amplifier gain do not explain how to faithfully reproduce the amplitude in headroom occurring during the recording (the lows and peaks), because dac's and amplifiers do not have an standard gain (the level achieved during full scale output, dBfs).

 

If an amplifier have "low gain/headroom", that real orchestra 50dB to 70dB will be at speakers/headphones, I don’t know, perhaps 50dB to 55dB. If you have "high gain/headroom", are we going to have 50dB to 90dB in our headphones (I presume your not clipping at -0dbfs)?

 

Is more headroom going to be more faithful to reality?

 

Once I visited an audiophile and he had a huge amplifier. Every time I gave him a CD he asked me if it had some volume variation. I thought they hadn’t. Then he was always slowing down the volume at peaks, because we were going deaf. They were audiophile recordings. Therefore, they might have not compression at all during the mastering (but that's just speculation).

 

I would like to hear more opinions about that.

 

p.s.: one post about the same subject http://www.head-fi.org/forums/f113/stax-thread-new-223263/index772.html#post5895823.

 

[argentino_sf]

Did I say something wrong? Or really stupid?

 

[Oystein]

When a recording is done the the recording engineer sets his recording equipment so that the highest volume peaks will not clip. Think of the old cassette players with the VU meters.

In the modern digital world the VU meter should not peak over the 0 dB mark.

 

When your example peaks at 70 dBA ( be aware that volume loudness dBA is not the same as dB in electrical signals trough the stereo set ) this 70 dBA should match at 0 dB recording signal. During normal playing at 50 dB the recorder could show -20 dB.

The peak of the recording at 0 dB is considered equal to 2V rms at the cd player output.

 

The situation with tha audiophile turning dowm the volume at the high peaks would suggest full dynamic range without compression. But at these peaks his cd-player would probably reach 2 V out but at 50 dbA volume the signal will be much lower .. probably less than 100 mV.

Every 3 dBA increase means double output power of the amplifier.

 

If you have a normal volume showing 100 mV then your headroom will the the limit at 2V which is 20x (electrical).

 

[CharlyD]

There might be some misunderstanding about what a decibel is. From Wikipedia:

The decibel (dB) is a logarithmic unit of measurement that expresses the magnitude of a physical quantity (usually power or intensity) relative to a specified or implied reference level.

 

A dB value always implies or specifies a reference value. The 50dB to 70dB you mention for orchestral "volume range" could be interpreted as sound pressure level (SPL) as "volume" is the perception of SPL. When measuring SPL with units of dB, the reference level of 0dB is a SPL of 20 micro Pascals of pressure. SPL levels may also be weighted to account for perceptual changes. A-weighted measurments (dBA) are the values often given to represent volume.[edit] I was curioous about dynamic range of an orchestra and found this paper, http://www.nanophon.com/audio/dynrange.pdf, which claims a maximum SPL of 122dB for loud percussive passages. I doubt there are many playback systems that could achieve that level.[/edit]

 

In digital recording, there is an absolute maximum sample value which cannot be exceeded. This sample value corresponds to some input voltage to the ADC, and this voltage is the reference of 0dB. Because this is the maximum value, the measured voltage from a DAC when represented in dB is always a negative value. There is also a maximum input voltage for analog recording, but the analog system is generally more forgiving about exceeding that value.

 

In the playback chain, amplifers have gain which most often represented in dB. In this case, the reference is the input voltage. An amplifier should reproduce the input signal at the set gain up to the point where the amplifier clips (maximum voltage level). Headroom is the difference between the maximum recorded level and the maximum output level of the system (in volts) represented in dB where the reference is the max recorded level.

 

[Nyal Mellor]

Yeah very few replay systems can even get to 110db without significant distortion, let alone 122db.

 

[PeterSt]

Let's start with this one :

 

Is more headroom going to be more faithful to reality?

 

The general answer is Yes. But what is headroom ?

This is expressed by Dynamic Range phenomenon itself ! Thus :

 

Assuming an orchestra can play the softest at 1dB and the loudest at 120dB, your system needs 120dB dynamic range to cover for that. This means that 16 bits (96dB) is not sufficient, while 24 bits (144dB) is.

But it is not that simple.

 

When the orchestra plays at 1dB (say a most soft hit on a pauk), you will never hear that, because of the environmental noise which is louder. Even molecule noise is much louder (I don't know, but let's say 20dB).

 

Now, to my experience - from audio playback systems - not much more than 70dB dynamic range (!) can be perceived. Why ?

Well, just try this :

Take a normal CD (96dB dynamic range) and digitally attenuate that with 72dB(fs). Now put your amplifier at maximum and listen whether you can hear something. Well, if your amps (and further system) has enough gain (this is also related to the sensiticity of your speakers) and the noise from the system is low, you can. If you can't, just get yourself a more powerful amp, and you can. However ...

Now go back to the maximum digital volume (- 0dBfs) ...

... and at this time your windows will blow out.

 

So you see, you indeed are able to perceive 70dB of dynamic range, but you can't utilize it.

 

While the above is some general explanation for basic understanding, indeed it is so that compression will squeeze the original dynamic range of the performance into ... well ... the maximum allowed (= possible) dynamic range for CD, which is 96dB. Now, if you understood the above, you can see this is of not much use either way, because we are not able to utilize that 96dB anyway. And thus, the compression is needed to let fit the whole lot within the allowed dynamic range of the CD. But then again, it is not that simple ...

 

When a performance of 120dB is squeezed into 96dB, the sound will become more flat. Yes, what about "less dynamical". So, where the peaks of the performance should not be cut off and compression is used, what about cutting off the LOWEST passages ? Remember, when we set our amp to the max allowed volume which let the windows stay in, the lowest passages can't be heard anyway (see earlier). Thus, when the uncompressed maximum is retained and instead the low side is cut off, the dynamics are not compressed and there is no audible penalty.

In the end you could say that compression is a solution to a problem which doesn't exist. It just shouldn't be done at all, as long as the high (volume) side isn't cut off, and instead the low end is.

 

The almost last part is about the absolute volume, or the 1:1 relationship to the original performance and the files it is recorded in (16 vs 24 bits);

 

First off, let's think about 24 bits and 144dB dynamic range. Totally useless. But beware, for "dynamic range" as such it is (useless), not for it's volume resolution. Thus, whether you have a 16 bits file or a 24 bits file, the absolute maximum (say 2VRMS output from the DAC), doesn't change by that. However, now the 120dB playing orchestra can be contained in the file without any compression. Aha ...

 

Assuming indeed a 24 bit file is not compressed, the net result is actually what I just described. Thus, there is no compression, there is no cutting of the peaks (both are just not needed now), and what happens is that the low end is even more lower than the example of the 96dB with 72dB attenuation. Thus, when the fle is played at maximum level (for your windows), there is no difference between the 96dB 16 bit file and the 144dB 24 bit file. Bot play as loud, and when 72dB attenuated, both play as soft. But there is one major difference ! ...

The 96dB file - already attanuated 72dB - can be attenuated 24dB more, while the 144dB file can be attenuated 72dB more !!

 

Is that useful ? nope. We couldn't hear anything from the 72dB attenuated file anyway (at non-windows breaking max levels) so never mind ...

 

The very last part is about the absolute volume of the recording and how it relates to reality;

 

As you suspected, this can't exist. Or maybe it can, but it isn't applied for practical reasons;

Take the famous whatever overture of the orchestra and the canon shots. A canon shot may be 140dB and ... well, the rest of the story is known. The canon shot fits in the dynamic range of 24 bits, but you won't hear the solo violist playing anymore, even when he plays as loud as possible. Bad luck ... (remember, unless you allowed the canon shot blow out your windows just because of air pressure (SPL).

Here is when compression is useful, because letting this go on 96dB max dynamic range, would produce a distorted canon shot. The shot will be way cut off, so it is better to compress it (with the penalty of everything getting compressed !). The result : a less dynamical sounding performance (and an undistorted canon shot which is very unreal for its volume).

 

While this is one part of it (the volume balance between original sounds is lost), another part is the volume setting of your amplifier. Thus, assumed no compression has been applied, what to set the volume to ?

Well, there is one logical way only : listen to e.g. a piano and try to imagine how loud it will sound in your room for reality. Try to match that and the other instruments and all should be in balance. That is, according to the mixing engineer, and what he made of it all in the first place. And here is where it goes "wrong" again ...

 

"Wrong" between quotes, because you really don't want to listen to a piano at the appropriate level, while the drummer hits the cymbals as he did from that short (listening room) distance. Your windows won't break, but your ears will hurt.

 

The audiophile recordings you referred to, generally have this all in the original balance or at least will be as close as possible without hurting your ears really. For a non "audiophile" album try Amarok from Mike Oldfield.

 

On a last note, the less compression has been applied, the steeper the transients are (generally referred to as "dynamical"). This is because the steep peaks haven't been squeezed down in more flat bumps. Thus, this is not only about peaks getting higher from their base to the peak itself, but merely is about the time domain which remained the same, and at reaching a higher peak in the same time, the slopes *must* be more steep. Transients are faster(/steeper) at less compression ...

 

Peter

 

 

[argentino_sf]

Thank you for all explanations. It was all very prolific.

 

Just want to ask a couple of questions.

 

If I were to design a DAC with a digital volume attenuator (in order to connect it directly to a power amplifier without any active or passive preamplifier), would it be better to deal with 24bits files (allowing more attenuation without dynamic compression)?

 

Then, what about 32bits files?

 

If some pc processors nowadays work at 64bits word depth, would it be any beneficial for audio purposes?

 

 

[CharlyD]

Hi Argentino - Have a look at this paper from our friends at Wadia - http://www.wadia.com/technology/technicalpapers/Digital_Volume_Control_2.pdf. One of the points in this note is that it is possible to attenuate a digital sample value without loss of resolution if the word length of the DAC is greater than the word length of the sample. If you're attenuating a stream of 24-bit data before processing by a 24-bit DAC there will always be loss of resolution. With the new 32-bit DACs, it will be possible to implement a digital volume control that could attenuate 24-bit content without noticeable loss of resolution. Using a 64-bit processor and OS may speed up the processing but I doubt it would have impact on the sound quality.

 

[tmfidelis]

Let's no forget to reduce as much environmental and system noise as possible. One may be very pleasantly shocked listening to the results.

 

Regards,

 

Tim Marutani

Emeryville, CA

 

[PeterSt]

Hi Tim,

 

Let's no forget to reduce as much environmental and system noise as possible. One may be very pleasantly shocked listening to the results.

 

True. I have a completely noiseless system (measured under -130dB) and together with 115dB sensitive horn speakers, I am able to perceive a properly dithered 16 bit file 96dB down (leaving 1 bit !) with my ear in the speaker. So, from that 1 bit still music is audible (though very distorted);

Would there be the slightest audible noise (in the area of -100dB) there is no way the noise won't overrule.

 

But in my opinion noise does more;

 

While we could say noise of 96dB down is inaudible, we can also say that technically the music rides on the noise, and the noise expresses on top of it (all adds up). Maybe this is a bit pretentious and maybe not much scienticfic, but I think it works out like that. Thus, if you have analogue noise with a digital value of, say, 50 and a frequency peak at 12000, this adds up to 12050 and the frequency (music) is distorted. You won't hear the noise as such though.

 

Peter

 

[PeterSt]

If I were to design a DAC with a digital volume attenuator (in order to connect it directly to a power amplifier without any active or passive preamplifier), would it be better to deal with 24bits files (allowing more attenuation without dynamic compression)?

 

I didn't read that Wadia link, but I think it will say/imply similar as what I say :

 

If you have a 16 bit (!) file, a 24 bit DAC allows for attenuation (48dB) without *any* loss.

 

If you have a 24 bit file, a 32 bit DAC allows for attenuation (48dB) without *any* loss.

 

If you have a 16 bit file, a 32 bit DAC allows for attenuation (96dB) without *any* loss.

 

No less, no more. But also think of the "lessons" from before ... e.g. :

 

If you have a 24 bit file, a 24 bit DAC which attenuates 48dB removed 48dB on the low end of the available 144dB leaving you with the top end of 96dB. There is NO way you will be able to perceive 96dB down from the maximum volume you will have at that moment (not window breaking !).

 

Peter