John Curl's Blowtorch preamplifier part II

Status
Not open for further replies.
Hi,

Very early trials are promising, but I really lack time to work on this more...


Ciao T

You really should listen to Martin's panphonic mike, there's a patent out there too (at least I read the application). DIY on an unlimited budget or at least the best "junk" box around.

In contrast the mikes for the next article have a $75 BOM but I'm working on a $12 version.
 
Last edited:
There are no real rooms, with real musicians in them (people are noisy) that can't be fit into 16 bits. Modern recording practice of course requires a lot of headroom around this. And if 0VU is defined as the *very* loud 85dB SPL at listening position that is used in mastering studios, and that peaks reach maybe +20dBVU, there are very very few listening rooms that don't fit into 16 bits.

As a separate issue, our hearing covers a wide dynamic range, but not all at the same time. We hear a smaller range that slides up and down dynamically.

Thanks,
Chris

Gee,

Many of the churches I work in have noise levels below 30 db A weighted. The pipe organs have no trouble doing 130 db A weighted by the chest. Then allow for the 30 db signal to noise recovery of the ear and all you need is 22 bits.

If your studio is NC 30 and you close mic an instrument you get way past 16 bits!

Been there done that. 16 bits is not enough for even an artificial reverb in a church.

Then there was the NFL football stadium where the "22" bit DSP really needed to be level tweaked to go from awful to not good!

Now just from the ears ability to perceive sounds below the noise floor and the peak to average ratio ignored by a sound level meter you need 9 bits!

BTY when the noise level is equal to the level of speech you get more than acceptable speech recognition! Speech recognition even works decently when the noise level is above the speech level.
 
Many of the churches I work in have noise levels below 30 db A weighted. The pipe organs have no trouble doing 130 db A weighted by the chest. Then allow for the 30 db signal to noise recovery of the ear and all you need is 22 bits.

If your studio is NC 30 and you close mic an instrument you get way past 16 bits!

Now just from the ears ability to perceive sounds below the noise floor and the peak to average ratio ignored by a sound level meter you need 9 bits!

You run 130dB SPL in a church? Don't tell OSHA.

Studio recording practice requires more bits as headroom as I mentioned.

Special pleading involving the ears' perception should be applied to all noises, so wash out in this discussion.

Thanks,
Chris
 
OSHA uses A weighted slow, clipping is a wee bit faster roughly 40,000 times.

Again the critical band hearing model allows recovery of information below the wide band noise floor. Discrete levels as in digital do not.

A dolby cassette yields the same information content (SNR) as does a 16 bit limited signal!
 
Yet I just mentioned at least one 16/44.1 CD that is shockingly good on several levels.

The issue is not if you can squeeze everything in by good practice, it is "What is a standard that exceeds our discrimination capability?"

16 Bits is more than enough for much music, but not all. 44.1 is fine for some instruments and most vocal, but you are aware of the research that shows there is information way higher in frequency.

As you may be aware RCA did research that showed people disliked music when the bandwidth went above 3000 hertz? Harry Olsen set up an experiment with a band playing in a room with an acoustical low pass filter and showed that people liked the full bandwidth better. The first set of experiments used an electro-acoustic system and the distortion that was at that time considered unimportant was the actual reason for preferring narrow bandwidth reproduction.
 
Again the critical band hearing model allows recovery of information below the wide band noise floor. Discrete levels as in digital do not.

This is a very commonly stated fallacy. Properly dithered digital conversions do not have any "discrete levels". None. Nada. Zip.

The mathematical (and practical!) result of the properly dithered A/D/A conversion of a bandlimited signal is the original signal, plus a small noise (the dither) plus a delay. The resultant signal includes *all* of the original signal, including components smaller than the dither. All the way down to the weeds.

Thanks,
Chris
 
Discrete levels as in digital do not.

Dither? As I mentioned before I find the RR Kodo drum CD unlistenable because there is too much dynamic range even at 16bits. And yes, if you turn up the quiet parts they sound fine, heaven forbid you don't reach for the volume control in time. This is in the context of polite listening not creating an audiophile experience. For that you can go to Dick Burwen's house and listen to his F14 takeoffs recorded at Hanscom field on his 30' concrete horns.

EDIT - No Thorsten he did not record them on those horns but rather played them.
 
Last edited:
Gee,

The dither math I've looked at shows that you also have to account for the sampling bandwidth. So at upper midrange frequencies (where the ear is most sensitive) a perfect dither will only allow the equivalent of 19.5 bits of resolution.

If you have a different math treatment I'd be interested.

Then we could also discuss dither noise shaping or mapping, maybe even why I prefer ratiometric converters.

Scott,

Are you trying an anecdotal argument?

Again acoustic measurements are unless noted IEC "A" weighted. Theoretical are 15.9K bandwidth unweighted.
 
Last edited:
Gee,

The dither math I've looked at shows that you also have to account for the sampling bandwidth. So at upper midrange frequencies (where the ear is most sensitive) a perfect dither will only allow the equivalent of 19.5 bits of resolution.

If you have a different math treatment I'd be interested.

Sorry, I have no idea what you're talking about.
Chris
 
The dither math I've looked at shows that you also have to account for the sampling bandwidth. So at upper midrange frequencies (where the ear is most sensitive) a perfect dither will only allow the equivalent of 19.5 bits of resolution.

You've illustrated by earlier point perfectly. As long as we consider signal-to-noise, noise floor, and resolution as being the same thing, we can make all sorts of incorrect conclusions.
 
You used the term "resolution," then used an equation for dynamic range. You'll have to define "ultimate S/N" to get a sensible answer, and it is not the same as "dynamic range." "Number of bits for a given bandwidth" is another fuzzy term. I don't mean to be pedantic, but if you're going to be quantitative, you have to be clearer about what you're calculating.
 
weighting has already been mentioned but some (Dolby?) use ITU-R 468 instead of "A" so more care is going to be needed in the cassette vs CD comparison - where are you puttting cassette tape peak = digital 0 dbfs, @ 30% 3rd harmonic??

If you do an ANSI standard articulation test of a cassette recording vs a 16 bit digital stream at peak modulation of 100% they measure the same. If you reduce the peak modulation to 1% they still give the same results.

I am not comparing CD to cassette just pointing out that with our hearing mechanism a high performance dolby cassette will yield the same results as a 16 bit linear dsp system.

SY,

I use 20 log (2exp (#bits)) to give a digital equivalent to the Ultimate S/N ratio of an analog system. Ultimate S/N is from the noise floor at 15.9k bandwidth to clipping. Clipping is when the peak output voltage of the DUT is reached and is roughly equal to 5% distortion of a sine wave. This is pretty much the way the engineers I work with measure things.

What seems to be missing is an understanding of monoticity. An "X" bit linear D/A will have the most significant bit and all lesser bits accurate to the least significant bit. But the least significant bit will not be any better. So on an 8 bit D/A you will have 256 levels but if it is putting out 2.55 volts full scale the step size could be .01 volts but when the LSB is toggled it is more likely the change is .049 or .0149 volts. That is why dither does allow more information to pass but at a rising rate of distortion.

Resolution of an "X" bit D/A is as used by my cronies is 1/(2 exp "X").

S/N is the ratio from a standard operating point for Analog this is defined as 0 db(x?) and that is often based on the voltage equal to 1mw into 600 ohms, but not always. The peak operating level is then specified with a given distortion. As the limit is clipping the difference between .1%, .3%, 1% and the traditional 5% is very small, but different folks use different spec.s (.1% to 5% often has a difference of .4 db!)

S/N for a digital system has become from 1/2exp(Bits) to Full scale.

Ultimate S/N as mentioned is from the wide band noise floor (15.9K) to clipping. I use 5% when I make a measurement. To me this is the same as dynamic range.

So far there have been issues with i vs I and Sabin vs Sabine, and other of what I always thought was standard nomenclature.

ES
 
Last edited:
For historical interest this was one of the first test disks from CBS/Sony in 1983. They did not always believe in dither.

Of further interest the "pop" music sample is Johnny Rotten in a Tokyo PIL concert. PIL = Public Image LTD. not exactly bluegrass.
 

Attachments

  • avavav.JPG
    avavav.JPG
    133.9 KB · Views: 200
Last edited:
Status
Not open for further replies.