the delay memory discussion (24 or 16 bits)... again...

[Tim Kleinert]

Just wanted to share an interesting observation:

When leafing through the printed 1.2x user manual as a bedtime read (attention: nerd alert!), I stumbled over something interesting:

The second sentence of last paragraph in the section "Voice Allocation And Polyphony" on page 121 reads as follows:

[3phase]

was you the guy that brought up the 16 bit theory?
what memory is the one that gets eaten by the modules? in the graph it looks like one ram for all of it..but...
What is the DSP internal ram doing?... or is that all we have here...hmmm
2,7 seconds on 96 k? is the memory for all slots? so
4 times 256 k... no...thats not enough..there is external ram, or?

[7om]

It is true that the delays and reverb are both 16-bit. Here is a confirmation I received from Clavia:

Hello, yes this is correct the delays and the reverb is 16 bit and
probably will not change in the G2 in its current form but who knows
what the future holds, thanks

Alexander Frodinger
Clavia technical support
support@clavia.se
www.clavia.se

Later, they indicated that it is only the delay or reverb signal itself that is 16-bit. The original signal is still 24-bit. Therefore, if you use reverb and or delay mixed with your patch, it is mixing this 16-bit signal with your original 24-bit signal. Their reasoning is that since you still have your original signal at 24-bit, it doesn't matter, but as many here have written, it does matter for certain kinds of patches.

I recommend that folks write to Clavia to request that a 24-bit delay module be written for the G2, and that they document the information on 16-bit delays and reverb in the manual.

[3phase]

"G2 in its current form.." is an interesting formulation...

[Tim Kleinert]

Thanks 7om for clearing that one up for me once and for good.

So the only possible solution is software-wise: the support of 24bit depth (at the cost of delay time).

But I think they should overhaul their tech specs and the user manual! They are actually quoting wrong specs! I find this a bit... uncool.

[King Rat]

Does anyone claim to be able to notice the difference between 16 and 24 bit in a live situation?
(This is a bald question - I'm genuinely interested how much of a difference it makes)

Angus

[Tim Kleinert]

When using delay lines for vanilla stuff like delay FX, reverb, I don't care so much if it is 16 bits or 24, live or in the studio.

But when you use delay lines for stuff like physical modelling, 16 or 24 does make a considerable difference. Eg. when playing a sustained chord with lots of notes on a simple KS type patch, you notice a clearly audible patina of HF noise (coming from the digital harmonic distortions) build up as the sound fades away.

Having 24 bit would push those 48dB(!) further into the background. That is a CONSIDERABLE improvement.

Sure, when playing live there's alot of other noise going on too that will most probably mask stuff like this. But when your live band is ready to hit the studio...

[cebec]

Yeah, I had also noticed that the latest documentation, 1.3x, states that the signal path and RAM is 24 bits.

from:

http://www.electro-music.com/forum/topic-6323.html

[3phase]

[Tim Kleinert]

[7om]

[Tim Kleinert]

[3phase]

[jksuperstar]

Out of curiosity, I pulled up the Motorola/Freescale spec on the DSP56367, which states:

-------------------------------------
On-chip Memory Configuration
* 5 K - 7 K x 24 Bit Y-Data RAM
* 8 K - 13 K x 24 Bit X-Data RAM
* 40 K x 24 Bit Program ROM
* 32 K - 10K x 24 Bit Program RAM
* Various memory switches available

Off-chip memory expansion
* External Memory Expansion Port
* Off-chip expansion up to two 16 M x 24-bit word of Data memory
* Off-chip expansion up to 16 M x 24-bit word of Program memory
* Simultaneous glueless interface to SRAM and DRAM
-------------------------------------

The external memory interface, after looking at the pinout out, *is* indeed 24-bit. Clavia however used (1) 256K x 16bit memory instead. 24bit RAM is extremely rare, but using a 32-bit and wasting a few bits is not unheard of (My Akai MPC1000 does this, it turns a 256MB card into 128MB).

Of course, Clavia could pack 24bit values in the 16bit RAM no problem, and hopefully, with one delay module at least, they will (Now that I've begun playing with physical modeling and sample loops, I see what you guys are moaning about). I'd bet the internal RAM is used to pass values between modules (the cables, if you will), and for immediate temporary calculation space.

[3phase]

[cebec]

I have also contacted Clavia support numerous times, without response, regarding clarification of the '16 bit issue' and a list of modules affected. I also added that, if, in fact, the delays were 16 bits, then many users have expressed a desire for 24 bit delay(s) even at the expense of delay time.

In addition, I brought the wishlist/enhancement thread to their attention.

Only that latter comment resulted in a response.

[jksuperstar]

Clavia's firmware has the DSP setup for using an external 16bit memory, even though the chip supports up to 24bit. So Clavia is well aware of this happening, and they hopefully do something in their code (such as dithering or similar) to convert 24->16 bits before storing to the external RAM. So it's not that the RAM chip is causing the truncation per, it's the fact that Clavia chose to use a 16bit RAM (32bit SRAM is 3x the $$$) and decided not to store 24bit values in it, but instead store 16bit values to optimize DSP cycles, memory accesses, and delay lengths.

[3phase]

[jksuperstar]

They could (should not?) be truncating... at the least rounding, accumulating the truncated bits and/or adjusting every so many samples, etc. Also, dithering might be a "waste" of DSP cycles, but it's very possible on this DSP.

[3phase]

[Rob]