John Curl's Blowtorch preamplifier part II - Page 1214 - diyAudio
Go Back   Home > Forums > Member Areas > The Lounge

The Lounge A place to talk about almost anything but politics and religion.

Please consider donating to help us continue to serve you.

Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
Reply
 
Thread Tools Search this Thread
Old 18th May 2011, 07:34 AM   #12131
работник
diyAudio Member
 
Rod Coleman's Avatar
 
Join Date: Jul 2004
Location: Warwickshire UK
Quote:
Originally Posted by simon7000 View Post
Scott,

I assume you are familiar with the bit rate error in solid state memory from gamma rays? Small but it is there.
Soft Errors in 1- and 2- transistor memory can be a lot worse than the Semiconductor firms claim. I was at one time responsible for the design of the largest [2-transistor] SRAM consuming fleet of computers in Europe, and witnessed it first hand.

The short version of the tale: single-bit soft errors are caused by alpha-particles, which are unavoidably included in the package epoxy. The SRAM makers mitigated the problem by enveloping the die in Kapton. This Kapton layer arrests alphas with modest energy, but not all [Thorium, from memory, gets through this]. However, the qualification test performed by all firms used Americium 241, and always found zero errors, but this is because the Kapton layer always stops these! In practical use, the errors are worse, at about one FIT (one failure in 1E9 hours, per device) for the SRAM chips of a few years ago.
This soon becomes one computer crash per day, when you have a large population.

Naturally, all of this took an extraordinary amount of effort, measurement, trips to Japan, corrective software, etc.

When your designs reach a very critical level, you learn to stop trusting data sheets, and what anyone else says, and measure things for yourself.
  Reply With Quote
Old 18th May 2011, 09:35 AM   #12132
KSTR is offline KSTR  Germany
diyAudio Member
 
KSTR's Avatar
 
Join Date: Jul 2007
Location: Central Berlin, Germany
Quote:
Originally Posted by SY View Post
I'd love to see Charles's files.
Certainly, please go ahead. It is very important for the incident to find out if the files are identcial or not. Identical is a bit multifacetted here, because anything from the below list could happen :

- complete files (I assume PCM data --.WAV, .AIFF, .FLAC etc) are identical on the binary, file-system level, byte for byte. The most stringent criterion, effectively more stringent than applicable. Most easily checked with DOS command "fc /b", in a DOS-box.

- the sample streams (returned in 16bit ints) recovered from the audio 'data chunks' (*.WAV lingo) of the files are identical, with the representation of the samples in the file not being of any importance (one could be in 32-bit float format and still would deliver the same values). That's audio stream data identity, the one that applies here. Checked by subtraction, giving true zeroes, of the audio data in a sound/wave editor.

- only one big consecutive block of the streams is identical (the center parts), a few samples at the start/end of the files are missing or different (can happen with rips from CD, or when the DAW-operator bounces versions manually). Still there is full audio data identity for the purpose of our comparison. Again best checked with a good wave editor.

- Streams are not bitidentical but rather have the sort and amount of differences that still leave the waveform intact, eg a gapless stream with NO totally wrong or missing samples from unrecoverable errors from the CD readout (with bounced files on hardisks this is irrelevant anyway). This is the point when it gets interesting, for example different dithering would give different "fine grain" in the sample data. That different data may lead to different perceived playback is for sure a no brainer to anybody, only topic would be if the difference has enough magnitude to become audible or not. I've subtracted a special case of different 24-bit ditherings recently (one having DC-offset), the difference was the LSB toggling after offset nulling and the audio contained in that bit was a fully distorted/compressed copy of the track. Could that be audible? In 16bit for sure, still not easily, IMHO.


As for the special case of ripping CD, it is a very proven fact that the better extraction tools which allow for the exact needed setting for the specific drive's read offset all give audio data identity and in most cases even file data identity for .WAV (when options to store non standard non-audio metadata in the file -- which is possible with .WAV -- are turned OFF). No need to worry about that. Otherwise the audio data streams might be offset a little but still are valid and correct. The chance that defects on a CD happen to be exactly such that bit errors would leak through unnoticed through the multiple error correction levels of CD-audio is not zero, but very close to 1/infinity, and even then this would be simple very occasional spot errors, no overall change in sound character.

Last edited by KSTR; 18th May 2011 at 09:40 AM.
  Reply With Quote
Old 18th May 2011, 10:55 AM   #12133
DF96 is offline DF96  England
diyAudio Member
 
Join Date: May 2007
1 in 8 odds don't prove anything. At most they could be an indicator that there may be something worth investigating. But let's assume for a moment that the effect is real: the files really are bit-identical, and people really can tell them apart. We must assume that the data reaching the DAC is the same (that is what 'bit-identical' means) so it must be the timing which is different. This means that the timing depends on either meta-data (e.g. timestamps in the file headers) or context (e.g. block placement on the disc). This is an equipment problem, which could eventually be found and fixed.

Now let's assume that such equipment problems have been eliminated, but the effect persists. Then we would be talking about a little fairy which lives in each file. This fairy gets injured by data-preserving conversions or global NFB, and feels insulted by double-blind tests so pushes off. When happy, their normal state, these fairies enhance our listening experience, which is why we notice when they are missing. Of course, this only happens to those who believe in fairies! The rest of us have to cope with mere reproduction of a waveform to a reasonable level of fidelity.
  Reply With Quote
Old 18th May 2011, 11:01 AM   #12134
SY is offline SY  United States
diyAudio Moderator
 
SY's Avatar
 
Join Date: Oct 2002
Location: Chicagoland
Blog Entries: 1
Quote:
Originally Posted by DF96 View Post
1 in 8 odds don't prove anything. At most they could be an indicator that there may be something worth investigating.
Exactly, which is why I'd like to see those files.
__________________
You might be screaming "No, no, no" and all they hear is "Who wants cake?" Let me tell you something: They all do. They all want cake.- Wilford Brimley
  Reply With Quote
Old 18th May 2011, 12:02 PM   #12135
gk7 is offline gk7
diyAudio Member
 
Join Date: Oct 2007
Location: Vienna, Austria
I would check the files with md5sum. If their md5 hashes are identical they are exactly the same and there is no way they can sound different on the same hardware. If their md5 hashes differ they _might_ sound different.
  Reply With Quote
Old 18th May 2011, 01:35 PM   #12136
Jakob2 is offline Jakob2  Germany
diyAudio Member
 
Join Date: Jun 2004
Location: germany
Quote:
Originally Posted by Charles Hansen View Post
<snip>

So I told him to send me three test pairs and that I would identify them. I got them all right. Of course everyone will tell me that it was just a series of lucky guesses. The chances are 1 in 8. But of course that's never good enough. Nothing is ever good enough.

<snip>
While most certainly some people would in every case refuse to accept surprising audibility results, it would help to test on an accepted level of significance.

In your case, if you would identify 5 pairs (one tailed; actual p=0.031 for this result due to guessing) or 6 pairs (two tailed; actual p=0.031 for this result due to guessing) it would be much harder to call it just lucky guesses.

Otoh, by only comparing 3 pairs the chance for guessing is doubling if both possible identifications were accepted (two tailed; actual p=0.25).
  Reply With Quote
Old 18th May 2011, 02:04 PM   #12137
diyAudio Member
 
john curl's Avatar
 
Join Date: Jul 2003
Location: berkeley ca
Thanks PMA, VERY GOOD WORK!

It is a pretty good guess to get an 8:1 out of the blue. It isn't just once or twice, but Charles Hanson was giving an EXAMPLE of what we do for a living. We can make those sorts of comparisons with audio sound quality. We have trained ourselves to do it. We use these 'decisions' to effect the direction of the next group of products. Charles does it this way, more than I do. However, 15 years ago, I started a design partnership with Bob Crump. This led to the CTC Blowtorch, but also the mods to the HCA3500 to make it a really high quality design, the Parasound JC-1 and JC-2, and even parts of the JC-3.
Bob Crump is deceased now, died about 5 years ago, but in our 10 years of working together, we made progress. He could more easily hear the audio differences than I could, and he was the main contributor to what made the Blowtorch into a world class product, as well as the JC-1 and JC-2 into 'A' rated products.
It was his approach to audio design that I have the most difficulty conveying here, without catcalls and diversion.
  Reply With Quote
Old 18th May 2011, 02:35 PM   #12138
diyAudio Member
 
scott wurcer's Avatar
 
Join Date: Jan 2004
Location: cambridge ma
Quote:
Originally Posted by simon7000 View Post
Ever leave a computer running with a fully functional UPS? How long till it crashes? What is the cause? (Depends on what it is doing... Duh.)

!
My Solaris workstation was up for 9 years 24/7. I could run "diff " on randomly selected files and their copies for months at a time, why bother, as Charles says it's never enough.
__________________
Silence is so accurate.

Last edited by scott wurcer; 18th May 2011 at 02:46 PM.
  Reply With Quote
Old 18th May 2011, 02:47 PM   #12139
DF96 is offline DF96  England
diyAudio Member
 
Join Date: May 2007
Many years ago when I was in IT we would routinely have PDP-11s running for a year doing distributed SCADA at power stations, before they were switched off for routine summer maintenance on the plant. You can imagine my amusement when I read that Microsoft had come up with a new operating system which was so stable that it could run for a week without rebooting!
  Reply With Quote
Old 18th May 2011, 02:55 PM   #12140
diyAudio Member
 
scott wurcer's Avatar
 
Join Date: Jan 2004
Location: cambridge ma
Quote:
Originally Posted by john curl View Post
The audio public finds us 'right'. That is why we are both well known, worldwide.
You can get away with something cheap, or home made, but some people just don't have the time and they have plenty of cash, and they are accustomed to everything working and looking 'just right' This is where the big money comes in. Fit and finish, not the circuitry as much. Just accept an aluminum or a plastic box and prices drop significantly.
Now your talking!

How about this Bristol, new ones go for $400,000 or more. Just think only $100,000 or so in repairs, real quality.

1989 Bristol Beaufighter - 39,500
Silver with black leather. Last owner 11 years, genuine 88,000 miles from new.
Rare lightweight lift-out front roof panel, superb all round condition with some
67,000 spent over the last 15 years on maintenance and upgrades.
__________________
Silence is so accurate.
  Reply With Quote

Reply


Hide this!Advertise here!
Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off



New To Site? Need Help?

All times are GMT. The time now is 04:15 PM.


vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2014 DragonByte Technologies Ltd.
Copyright 1999-2014 diyAudio

Content Relevant URLs by vBSEO 3.3.2