Topic: AD-20: what is the bit-depth of the SDPIF output?

[jk labs]

Hi!

there seems to be conflicting info on the net.  The AD-20 has 20 bit AD converters and I'd like to know if the digital output ports clock out all 20 bits or if the stream is truncated to 16.  (I am assumning there isn't a third option).

Thanks
Jon

[keepongoin]

16 bits is what i heard prior to making the purchase.

[C.Clark]

the ad-20 is a 20bit a/d but is dithered to 16 bit/ 44.1 khz before hitting the spdf and toslink digi outs, i ran this box for about 8 months and loved it

[JasonSobel]

the ad-20 is a 20bit a/d but is dithered to 16 bit/ 44.1 khz before hitting the spdf and toslink digi outs, i ran this box for about 8 months and loved it

the output is 16-bit, but the 20-bit data stream is truncated, not dithered.  

This little excerpt is taken from Core Sound's website, verbatim:

Some knowledgable folks have asked why the AD-20 doesn't dither its least significant bit, considering that it truncates its 20-bit data down to 16-bits. (In some situations, truncation results in audibly unpleasant artifacts.)
Although the AD-20 uses a true 20-bit A-to-D converter chip (AKM5351, a slight improvement from the AKM5350 found on the ADA1000 converter), its actual dynamic range is just around 98dB. This means that although there may be more signal in the noise, the S/N ratio is really only giving you about 16.5 bits of real information above the noise floor. Oddly enough this works to our advantage when used with 16-bit gear (like CDs and DAT decks) because there is already a natural 'dither' in the 16th bit as it is being recorded from the AD-20. There is no need to add additional dither noise to the signal because the signal is already hanging right off the edge of 16-bits.

If the AD-20 truly delivered 20-bits of real information (that would probably require a so-called 24-bit converter!) then a 20-bit to 16-bit truncation would result in the quantization noise that the human ear would find offensive. In this case, an A-to-D would require additional dither noise to be added just beyond the 16th bit to randomize the signal such that the 16th (truncated) bit was not perfectly quantized.

Summarizing, the AD-20 will sound great with your 16-bit DAT or MD recorder due to the natural dither in the A-to-D converter. And if you go through the effort of capturing 20 or 24-bits (e.g., with a Zefiro ZA2 in a digital audio workstation) then you'll still gain a few extra dB and perhaps a bit more signal below the noise floor.

[jpschust]

wonder what would happen if one was able to bypass the truncation and run into a laptop- curious how it would sound at a true 20bits

[nic]

wonder what would happen if one was able to bypass the truncation and run into a laptop- curious how it would sound at a true 20bits

what software would you use to record at 20 bits?
all I've used were 8, 16,  24, or 32. no 20  

[jpschust]

you could technically record in 24 bit even if it was only 16 bit or 20 bit.  

[Simp-Dawg]

i could test that out...i've got an ad20 and a u24, i'll get started on that right now actually...

[Simp-Dawg]

well...i didn't find anything out.
i ran the ad-20 into the u24 and recorded at 24-bit, and during playback from wavelab, i turned on the bit-depth meters.  all 24 bits lit up, whereas when playing back a 16-bit file it only goes to 16.
however, i tested it again transferring a dat at 24 bit, and got the same results...which means that's not a good test.
any other suggestions?

[Zaphod]

Just curious if JK labs might be thinking of a mod for the AD20?

[nic]

^^ or incorporating a A/D into their Preamps ?

[keepongoin]

Just curious if JK labs might be thinking of a mod for the AD20?

for what?  I have been in contact with them and sent the specs for most of the capacitors and various other parts inside.  I think maybe modding it to run AKG ck6X caps to run with active cables, though I am not too sure.

[Simp-Dawg]

jon,
according to the information above pulled from core-sound's website (and that may be taken with a grain of salt, from what i understand ), hte dynamic range of the adc only equates to approximately 16.5 bits, although it is in fact a "20-bit" adc chip.  
i'm not sure if that means that effectively only 16.5 bits are being transmitted and that there's a "natural" dither occurring during the output to the digital device, or if there is actually a truncation occuring removing the least significant bits, which are essentially empty due to the adc only truly representing 16.5 bits of data.

just food for thought...

i've heard glowing praise about your work and am excited to be able to lend even a small hand in your efforts.  the ad-20 truly is a great piece of equipment, very small and portable, low cost, and high quality.
your proposed modifications would make this unit that much better.  just an fyi, there is not much available room inside, unless perhaps you were to rip out the gain stage guts and only keep the adc, though they do appear to be on the same circuit board...not that i know too much about electronics, but that's an amateur observation.

good luck!

[jk labs]

My suggestion is to think of this issue as a 20 bits ADC with a total (combined digital & analog noisefloor) present in the AD converter stage equivalent to an analog level just about halfway between the 16. and the 17. bit. I.e. bits 17, 18, 19 and 20 are noise.  

In light of this there is nothing to be gained from keeping the last four bits.  Another important issue is that the 16 most significant bits are properly dithered (albeit with only modest control over the noiseshaping spectrum which is naturally created in ALL the preceding analog stages including the thermal motion of air in the mic).  

A useful test would be this. Short the inputs on the AD-20. Turn the gain to a minimum. Record the digital output stream making sure there is no device or software downstream introducing additional dither.  

Ideally this should yield datawords consisting of nothing noise in the 16th bit:  
0000000000000001(xxxx...)
and
0000000000000000(xxxx...)

Now if we amplify this by 6dB a total of 15 times (90 dB i all).  If the original data was 16 bits deep, and the recording device padded the rest (to say 24 bits total) with zero's, then there should be nothing but zeros in the deeper laying bits (17..24). I.e the data after amplification,  should be 10000000000000 or 000000000000000.
If bits 17..20 is noise then clearly a lot more different values will be present when viewed in say audacity.  

My issue was that both the AD converter and the SPDIF transmitter chips used within the unit in question are known to pass 20 bit words.  So where is the truncation to 16 bit words taking place?

Regards
Jon

[Kelso]

The sp/dif, aes interface transmits 32 bits anyway, which is composed of 8 bits of flags and 24 bits of audio. If the machine uses only  16 or 20 bit the LSB are  set to 0. So if the ad 20 transmits 20 bits, we can think a 16 bit machine that receive the signal while just ignore the 4 LSB: that's where the truncation would take place.