Topic: how to tell if you have a true 24bit signal??

[jk labs]

there is a bit meter in wavelab (under the tools menu?). If the signal is padded you will only see the top 16 bits of the meter light up.

bingo, thats what im looking for! cause i can record a 24bit wav with a 16bit dat, so going by whatever wavelab or SF says the file is, doesnt really tell you if the source was true 24bit.  Anyway.. im getting there! I found out, i had to disable the SBM pin as well, because thats the thing that takes the signal converts to 20bits then dithers to 16. so even tho i changed the output pin to 24bit, i was still getting a 16bit signal. when i disabled the sbm pin, i get a 24bit signal now   the signal is way way too hot tho, so i guess there is more to the equation

Todd,

(I make this follow-up public as it might have general interest and/or could catch the sight of people sitting with relevant information).  
  
One thing to have in mind is that the ADC within the SBM-1 is the CXD8493M (per service manual printed in 1995). I assume from the partnumbering that it is made by Sony. The chip itself must then be of '94 vintage at best. (But interestingly enough the analog stage within this ADC is running on +- 4.8 Volts, twice what is standard today).  

I just made a few online attempts at locating a datasheet for this IC but could not. I'd be *very* interested in a copy if anyone has access to this document.    

What does this information indicate? Well, I'd be surprised if this ADC has a S/N ratio better than that equivalent to a bit depth of  17 or 18 (108 dB). What I do not know is how many bits this IC sends to the DSP. Is it really more than 18? Say for the argument it is 20. If you disable the superbitmapping and word reduction within the DSP then the DSP must pad with 0 four times to get 24 bits. So I am in two bits(!) of doubt of how to interpret you now having "24" live bits out from the sbm-1.  

What I expected was for you to possibly/hopefully see A) the noisefloor (with the record volume at 0 [shorted inputs on the ADC itself could be needed] move below 90 dB (and not just because of negative DC offset) and B) possibly as much as 20 active bits.
24 just seems plain unlikely with no dsp processing!

If you amplify the noisefloor signal by 6 dB and do this 15 times (basically you chop off bits from the top and rescale the remaining signal), while zooming in on the timescale so as to see individual samples in time, then an ideal 16 bit noise signal will after 15 iterations sit at -1, 0 and +1 [or -0.5 and +0.5] (overlooking dc offset). I.e. you have shaved off bits and arrived at the lowest lying bit and it's two fluctuations: up one and down one. Any other levels present in the remaining data must be due to lower lying (17th etc) bits. (Having said that one must be open for the fact that the noisefloor could be higher).

Now. The problems you have with both the levels and the left/right channel being messing up, might indicate that there is a problem remaining here that invalidates current assumptions of more than 16 bits present in the output. I suspect for now that this is either an issue of the dsp not being made for "non-sbm" operation (even if it has the pin to disable it), and/or a problem in the transfer of data from the dsp and to the spdif transmitter.

Time will show :-)

Regards

Jon
    
 

[MattD]

It comes up here, but $$ if you want to see it:

http://www.icxinyi.com/EN/zlsearch.php

Edit: 10 bucks ... if you think this will be a successful mod, it might be a good investment.

[Brian]

wow this is great stuff guys! +t's to you both.

[dr.ph0b]

I never knew the SBM had a 24-bit chip in there.

As for the current hot signal problem, do you mean in the digital realm, or do you mean that you're clipping everything you send in (everything in is too hot)?

clipping in the digital realm.  altho when i reocrd to a 32bit float file in wavelab, the levels arent as hot and i get the dial up to 1 before i clip. (over 0)   there are probbaly some other tweaks that need to be done. jon from jklabs has been very cool in giving me info and the pinouts of the IC's and schematics. cause im like a blind man in a maze    

But clarify one thing. Assuming the sweetest music is input to the sbm-1 on it's line inputs.
What comes out through the spdif port? 

Real valid data of improper amplitude or are we talking garbled random data?

Jon

its real valid im assuming by looking at the waveforms. there is an inherent static kinda like whitenoise? in the background tho. After looking at the schematics, im wondering if thats because i have disabled the sbm pin (lifted from the board).  according to the schematic that pin runs to the IC309 chip (labeled control micro computer)  maybe that side need to run to ground....?

[baustin]

^^^^^^

Has anybody gotten to the bottom of this? Can you actually get a clean 24 bit signal from an SBM1?

I'd still be interested.

-b


EDIT: Oops! Think I found it: http://taperssection.com/index.php?topic=48138.0

[udovdh]

^^^^^^

Has anybody gotten to the bottom of this? Can you actually get a clean 24 bit signal from an SBM1?

What can provide a 24-bit signal that is really worthwhile?
(i.e.: has enough signaldata, is portable, small, eats not too many mA's, etc)

[Sid Viscous]

SBM-1 24bits?  hmm interesting.  I always thought that Sony was using a CS5335 20-bit stereo A/D chip set for 16 bit in all bit mapping devices.  I would like more information about how you went about doing this, sense it would require a SMC to be replaced; without the correct soldering equipment it would be futile.

There is no such thing as 20 bit. 20 bit is 24 bit with zeros in the remaining 4 bits.

[Roving Sign]

SBM-1 24bits?  hmm interesting.  I always thought that Sony was using a CS5335 20-bit stereo A/D chip set for 16 bit in all bit mapping devices.  I would like more information about how you went about doing this, sense it would require a SMC to be replaced; without the correct soldering equipment it would be futile.

There is no such thing as 20 bit. 20 bit is 24 bit with zeros in the remaining 4 bits.

[MattD]

Word lengths are all multples of bytes, where 1 byte = 8 bits. 20 is not a multiple of 8, so a "20-bit" word really occupies three bytes of memory, where the least significant byte only has four nonzero bits.

[mdarnton]

Someone correct me if I'm wrong, but doesn't a lot of it have to do with noise level? That is, if you have 95db above noise, you have an effective 16 bits, because the lower 8 (24-16) are filled with useless noise. What does it matter if you have 24 bits if 8 of it is unusable? On the other hand, if the noise level is,say -125db, then you have something beyond 16 bits, which is good, but not a usable full 24. On the other (the third) hand, if you have noise hanging around -145, then you have the full 24 bits above it to fill with music, and everything you paid for.They can sell you a 5 gallon bucket, but if the bottom gallon is already filled, it's really only 4 gallons, as far as you're concerned.

Or is that not the way it works?

[ghellquist]

Measuring the noise level is the only way to know if you have 24 bits. It might be that noise in the analog parts before the AD will mask any information you get in the last bits. Sure, the AD might be working hard and creating all this 24 bit information but a lot of if will be noise -- the question is how much of it is noise. Even the very best AD-s you can buy has difficulty in having true signal in those last bits.

I believe you could use AudoRightmark to create a test file, feed it through a good DA to your unit and then analyze the output file afterwards. (Caveat, I have not tried this myself). One requirement is that the DA has to be better than what you are measuring.

http://audio.rightmark.org/download.shtml

Gunnar

[mhibbs]

there is a bit meter in wavelab (under the tools menu?). If the signal is padded you will only see the top 16 bits of the meter light up.

bingo, thats what im looking for! cause i can record a 24bit wav with a 16bit dat, so going by whatever wavelab or SF says the file is, doesnt really tell you if the source was true 24bit.  Anyway.. im getting there! I found out, i had to disable the SBM pin as well, because thats the thing that takes the signal converts to 20bits then dithers to 16. so even tho i changed the output pin to 24bit, i was still getting a 16bit signal. when i disabled the sbm pin, i get a 24bit signal now   the signal is way way too hot tho, so i guess there is more to the equation

have you run any of this by Doug Oade?  That cat knows the SBM-1 inside out.