Digital audio and stress

Status
Not open for further replies.
I still use Cool Edit 2000 regularly. I wish it was a little friendlier with multichannel files, but I may also not know enough about how to work with them in it.

I'm really disappointed to hear about Audacity truncating 24-bit data! I hope they can get that sussed.

- Jim
 
I still use Cool Edit 2000 regularly. I wish it was a little friendlier with multichannel files, but I may also not know enough about how to work with them in it.

I'm really disappointed to hear about Audacity truncating 24-bit data! I hope they can get that sussed.

- Jim

I you google it you will find the developers dissing the need for 24bits. They don't get how nice it is to be able to turn down the mix 10dB and still get well over 16 net bits. Also the latency stuff with ASIO vs something else is a complete don't care for me. I recognize that musicians and recording engineers have their own huge world of requirements.
 
0.055mg Technics EPC-(P)100CMK4 (MM, boron pipe cant.)
Real shame they won't bring that back
I have one left that hasn't gone into cantilever collapse.
55 micro grams? That's absolutely incredible!

As per SY's post, I'm guessing considerable physical fragility came with that amazing lightness. Perhaps that's why Technics won't bring them back? The warranty claims might make it not worth while.

-Gnobuddy
 
I(f) you google it you will find the developers dissing the need for 24bits.
Last time I looked at spec sheet numbers, there were no real 24-bit A/D or D/A converters on the market, in the sense that typically 4 to 6 of those bits were actually below the noise floor, where they were doing no good to anybody.

I'm going off the relationship that the ratio of largest to smallest n-bit binary number is (2^n - 1). Using that, 16 bits translates to 96 dB dynamic range.

That's quite achievable, and there are plenty of D/A converters with noise floors 100 dB or more below clipping. So the information from all 16 bits are preserved.

Using the same equation, 24 bits is equivalent to a dynamic range of 144 dB. I don't know of any room-temperature D/A converters with a noise level 144 dB below clipping - meaning that the information from several of those 24 bits are gone, buried under the white-noise floor.

Meantime, 18 bits is equivalent to 108 dB, and this seems to be roughly where good "24 bit" consumer devices seem to land. That means these "24 bit" devices actually have only 18 usable bits, with the remaining 6 bits contributing nothing.

You'd need a dynamic range of 114 from clipping to noise floor to manage an actual usable 19 bits, and that is a rare noise spec indeed.

20 bit actual resolution? You'd need 120 dB between clipping threshold and noise floor. I haven't seen that good a spec yet in a consumer device.

The formula I used applies to DC voltages, but I am not sure how contemporary digital audio techniques affect it; I suspect noise-shaping and other DSP tricks actually widen the dynamic range you get from "n" bits, which would mean the noise floor in most devices actually limits you to even less than 18 bits.

Has any of this changed since I last looked at it? I would have thought you would need to cool down the converters with liquid nitrogen to get noise levels low enough for true 24-bit operation.

-Gnobuddy
 
The tooling is long gone. Not enough of a market to amortize new tooling, either- fashion audiophiles sneer at moving magnets.
I wonder what might be possible with contemporary MEMS technology? Tip, stylus, and sensing element all fabricated out of a single piece of silicon?

Fun to imagine, but I'll stick with the large collection of ones and zeros on my computer's SSD, though!

-Gnobuddy
 
We wrote the first ever single processor V32 modem on the 56K. Tx, Rx, near and far echo canceller, trellis and AT commands all on the one DSP... All in assembler!
A really nice chip, but at the time, I had very little idea what I was doing, being new to both DSP and assembly language programming.

I had been hired to find a way to make the company's speakers better using DSP where possible. It took most of a year of effort, but in the end I cobbled together bits of test and measurement equipment, lots of Matlab scripts I wrote, assembly code for the 56000, and a few other bits of software and hardware. Other team members worked on the speakers themselves, and various related ancillary issues, including power amplifiers and power supplies for them.

We got it to where we could measure a speaker's quasi-anechoic frequency response (using a gated swept-sine method), work out the inverse frequency response needed to flatten it, fold in the crossover network transfer functions I'd specified, spit out FIR coefficients, and finally, load them into the 56000. Frequency response correction was done separately for the midrange and tweeter, while the woofer received only an active crossover, if I recall right.

End result, a prototype active 3-way speaker with DSP crossover networks, and a flattened (and extremely consistent) frequency response through the mids and highs. The design also incorporated some good ideas to control and manage diffraction.

That was pretty forward-thinking for a prosumer product in 1998/1999. The intention was to go from prototype to production over the next several months.

But unfortunately for all concerned, management was rapidly steering the company finances into the red, even as the dot-com collapse pulled the bottom out of people's disposable income and accelerated the fall.

It ended with layoffs, an employee suicide, corporate bankruptcy, liquidation of assets, et cetera. That product never made it to market. I changed careers, and moved on.

-Gnobuddy
 
That's a sad story, and a shame about the product.
We didn't have matlab to work with -- we wrote some programs in quickbasic to calc the coefficients for FIR and IIR filters from an input file of frequency response points. Those programs were ported from Fortran orginals!! That's showing my age.. 😀
 
I wonder what might be possible with contemporary MEMS technology? Tip, stylus, and sensing element all fabricated out of a single piece of silicon?

Fun to imagine, but I'll stick with the large collection of ones and zeros on my computer's SSD, though!

-Gnobuddy

IIRC it used a thinwall Be tube. Be is a no no these days, so would be have to be an oversized Al tube to match stiffness. Unless you can fab a nanotube stack. Silicon would be no good for a tip.
 
Last time I looked at spec sheet numbers, there were no real 24-bit A/D or D/A converters on the market, in the sense that typically 4 to 6 of those bits were actually below the noise floor, where they were doing no good to anybody.
Looks like you are confusing playback and record/edit here. Sure, 16 bit for playback is fine. For recording and editing, not as good. As Scott mentions, and I'm sure others will too, having the extra headroom when recording is great, and once you get into production, higher bit rates really pay off. It gives you so much more room to work in. 32 bit float works well.

Back when Photoshop would do only 8 bit per channel color, it was easy to run into problems working on complex files. Colors would "break up" if you pushed them too hard. Just not enough dynamic range. For display or printing, 8 bit was fine.

When I was using the computer media server to do a lot of DSP, all that DSP was run 64 bit float, then converted to 24 bit signed for output to the DAC. That gave me all the headroom and finesse I could ever want. And even using 10 or 12dB volume control into a DAC that that might give me 19 bits above the nose, I was still ahead of the game.

I'm sure you know this, I just want it to be clear that bigger bit depths have significant advantages or 16 bit for recording and processing.
 
JP and I are both huge fans. I have one left that hasn't gone into cantilever collapse. IMO, best cartridge ever made.

I need to organize a flight, or talk you in to organizing some needle drops! I don't have an 100CMK4 that behaves any longer, though I do have a rather nice 205CMK3. Still kicking myself on that one. The seller had two and I bought only one.
 
For years I've used Goldwave as my audio editor. I like it because I know it, and it has some cool tools. Goldwave displays the waveform as stair steps. That's great because it lets me get in and edit sample by sample if need be. The stair step makes a good visual representation of the actual samples. That stepped graphic on the screen makes it easy to grab a single sample and change it.
Audacity, to its credit, shows the samples as dots with a line between them.

It does not mean that the analog waveform coming out of the DAC is stepped. It's just a handy way of representing the samples at a high zoom setting.

Actually what the stepped waveform does convey is that once you have a sample, it will not change in value until the next sampling. First order sample & hold.
As noted, it is a convenient representation. Just as a map is a convenient representation of the world, but is not the world. Interstates are normally not red, and country roads generally not yellow. 😎

Jan
 
Last time I looked at spec sheet numbers, there were no real 24-bit A/D or D/A converters on the market, in the sense that typically 4 to 6 of those bits were actually below the noise floor, where they were doing no good to anybody.
-Gnobuddy

That's irrelevant, 10dB of headroom is a huge advantage that's not even 2 bits. There are plenty of converters with ENOB's of 18-19 bits. Looking at noise floor with respect to dynamic range as only the total rms 20-20k is a very narrow view point.
 
Our spare bedroom is always open to you, except when Jan is here, and then it's the futon. 😀

At least it's not the floor 🙂

I've a pile of literature to scan, but here's a sneak peak. This is from the 100CMK4 and P305MCMK2 product launch document. My rough translation is lever, chip, and magnet.
 

Attachments

  • IMG_1680.jpg
    IMG_1680.jpg
    305.5 KB · Views: 170
Status
Not open for further replies.