phetamine

BOSS GT-8 Guiter Effects Processor :: Specifications

According to the boss website, the gt-8 has 24-bit converters (i'm assuming 24-bit throughout) and a 44.1 kHz sampling rate. The digital output is EIAJ CP1201, S/P DIF.

I'm well aware that a 10kx DSP is locked at 16-bit/48kHz. I can still connect the gt-8 to the audigy drive spdif and it works. And it sounds ok. What i'm wondering is; (excluding direct spdif recording) to what degree is the re-sampled signal suffering from this conversion? Is it significant? I guess it's not really, considering the other tracks in my multitrack setup utilize audigy and live pre's and adc's anyway, so it's not exactly overall very professional.


The main thing i've been wondering, looking at the peak plugin connected to the spdif/src, is this: Should i program the gt-8 to never output any higher than a certain "-xx dB" ?

The reason i ask this is my (probably way off) idea that the bit rate conversion might simply drop the least significant bits. And if i keep the signal quiet, the LSB's are totally insignificant anyway. Is that how it works?

Russ

No, the lower the signal level, the more relevant the LSB bits would become.

i.e.
Assuming that the 8 LSB bits are dropped for a 24 bit signal, and your signal level is low enough to only use the lower 16 bits, if the lower 8 bits are dropped, then you have only 8 bits left of the original signal, not much dynamic range, and in order to hear the signal you probably have to add a lot of gain, which not only applifies your signal, but also amplifies any noise that might be present.

What you want is a signal that is as hot as possible (but without overdriving your inputs/clipping, etc), so that you can pack as much useful info as possible into the MSB bits.

Maddogg6

Yeah, I figure the algo for the SRC is a bit more sophisticated than a simple dropping of any bits. And theres some hint to interpolated SRC in the Driver Compatibility settings, is it the simple interpolation like DANE can use? Perhaps thats only valid in the 44>48 conversion ?? I dunno, but it would be interesting to learn what algo is implemented, and if its software or hardware.

Russ

Maybe, but (when talking about bit depth only) I am not sure how much more it could do, other than maybe dithering.

BTW: I am pretty sure that re-sampling is done by the hardware, and I would imagine that it is a bit more sophisticated than the interp instruction (i.e. (when changing sample rates) probably something like up-sampling by some common multiplier (i.e. zero stuffing), filtering, and decimation, etc).

But of course, all of this is just speculation...

Maddogg6

I was thinking re-scaling so not to cause all loss at one end of the head room.

I dunno - just dropping bits just doesnt sound right in my head...(no pun intended) - I was thinking if all lower 8 bits are dropped - low level sounds are grossly distorted, dropping the higher 8 bits would grossly distort the higher level sound - dropping 4 on each end (4 lowest and 4 highest) with a compander (compress input, expand output), I would think, reduce distortions (edit: I mean, spread the loss out evenly)... ?? but you know how little I know about this digital stuff .

phetamine

Thanks Maddogg6 and Russ for sharing your thoughts. I had a feeling it would be somewhat unknown, the actual methods used to resample (by resample i'm also implying the change in bit depth, assuming it's all part of the same process of starting with one sample rate/bit depth, and finishing with another sample rate/bit depth). I'll try to keep the signal hot so as to retain the highest possible resolution. That makes sense, conversion or no conversion.

I'll tell ya what, all this kx technical stuff has been the ultimate distraction from making music the last few years. It's time to stop worrying about sample conversion, and start making more music. I just can't get over (or accept any alternative to) the kx way of DSP control. More specifically, the KX ProFX way.

If the new support for E-Mu interfaces becomes as feature rich as in 10kx models... maybe a 1616 should be my preference over some firewire interface? I don't know a whole lot about E-Mu's at this stage but it's looking quite promosing. The ADAT could equal 8 high quality pre's of my choice couldn't it? That would be awsome, all routable in a kx dsp window. Is that something to look forward to? (that was way off topic, sorry)

Russ

It is not all at one end, those 8 bits are spread among the full range.

Although the 16 bit range and 24 bit range is different as far as the number of possible values, remember that all the values for both still represent a total signal range of -1..1.

The extra 8 bits adds extra (8 bits = 255) values in between each 16 bit value. These extra values are not representable in 16 bits, so all you can do is round to a nearby 16 bit value. Discarding the 8 LSB bits just has the effect of rounding (Toward Negative Infinity) to the nearest 16 bit value.

i.e.
2 sequential 16 bit values:
[COLOR=Lime]0x1fff[/COLOR] = [COLOR=Lime]0.249969482421875[/COLOR]
[COLOR=Lime]0x2000[/COLOR] = [COLOR=Lime]0.25[/COLOR]

The same 24 bit values (in [COLOR=Lime]green[/COLOR]) (with the 8 LSB bits in [COLOR=DeepSkyBlue]blue[/COLOR]):
[COLOR=Lime]0x1fff[COLOR=DeepSkyBlue]00[/COLOR][/COLOR] = [COLOR=Lime]0.249969482421875[/COLOR]
[COLOR=White]0x1fff[/COLOR][COLOR=DeepSkyBlue]01[/COLOR] = 0.24996960163116455078125
[COLOR=White]0x1fff[/COLOR][COLOR=DeepSkyBlue]02[/COLOR] = 0.2499697208404541015625
...
0x1fff[COLOR=DeepSkyBlue]80[/COLOR] = 0.2499847412109375
...
[COLOR=Lime][COLOR=White]0x1fff[/COLOR][COLOR=DeepSkyBlue]f[/COLOR][/COLOR][COLOR=DeepSkyBlue]e[/COLOR] = 0.2499997615814208984375
[COLOR=White]0x1fff[/COLOR][COLOR=DeepSkyBlue]ff[/COLOR] = 0.24999988079071044921875 [COLOR=Lime]
[COLOR=Lime]0x2000[/COLOR][COLOR=DeepSkyBlue]00[/COLOR] [/COLOR]=[COLOR=Lime] 0.25

[/COLOR](I used hex values above only because it is easier to relate hex values to bits, then it is with decimal integers).

Had the original signal been sampled at both 24 bits and 16 bits to begin with, if you compared the two, you would see similar results you get by just discarding the 8 LSB bits.

Sounds good to me . Yes, kX has spoiled us, and I would hate to not have the flexibility that kX gives us.

Maddogg6

Ahhh - ok, that makes sense.

phetamine

All that binary / hex stuff above made me think a little harder and try to recall the binary stuff that i learned a few years back (which without putting to much use, was somewhat forgotten). Your elaboration there Russ, has put things much more into perspective. I've also had a look into debates regarding general 16-bit vs 24-bit, and techniques like dithering and such.

And so here is MY current perspective. Although there is a huge difference in the number of possible values for a 16-bit sample vs a 24-bit sample (65,536 vs 16,776,960), 16-bit audio isn't bad, and I'm ok with working with it for now. What Russ made me realize with all that binary bullsh is that if my 10kx card were to use a simple bit-chop (truncation) to convert a 24 bit sample to a 16-bit sample, the end result will be pretty close to what it would be if it were sampled at 16 bits in the first place. So thanks Russ for clearing that bit up. The problem with this though, after doing a little research, is that "pretty close to" can be a bad thing when the method for calculating the new sample is a consistant rounding off. If the error (from rounding) is repeating and correlated to the signal, the error that results is repeating and cyclical, and cyclical errors yield undesirable artifacts.

kx cards, being a consumer product (from the yesteryear), could well use some 'crap' method for resampling 24-bit spdif for all i know. At least now though i'm gathering some understanding of how this all works and some of the methods that are out there. If anybody is interested... the wikipedia article on "dither" is quite informative. Dither - Wikipedia, the free encyclopedia

Now it is time to find out what method is used by kx cards. I'm going to start by looking at the spdif standards... maybe dithering (or whatever) is already part of the protocol?

phetamine

Ok, i found a pdf... (1995) Engineering Guidlines: The EBU/AES Digital Audio Interface
http://www.ebu.ch/CMSimages/en/tec_A...tcm6-11890.pdf

"24 bit distribution will be needed in only a few cases" lol, 1995

So... the question is this... Did Creative Labs implement the appropriate subsystem to meet these (clearly well documented) needs? I'm tipping yes... but who knows

Russ

BTW: Truncating would be rounding towards negative infinity, rather than towards zero [COLOR=Gray](I corrected my previous post to avoid confusion)[/COLOR].

I do not have a 10k2 card to test with, so I can only guess at what the hardware might be doing [COLOR=Gray](10k1 models re-sample everything)[/COLOR].

phetamine

thanks for the correction Russ