M@ and The Dweez'

[Martin Naef]

Hi Tony

On 10.10.2011 6:08, Tony Hardie-Bick wrote:
> On 10/10/11 15:33, Martin Naef wrote:
>> Two TigerSHARC at 600MHz, according to their web page and the
>> manual. That usually translates into 32bit floating point math (the
>> TS201S supports up to 40bit float and 64bit fixed point)
>
> The sane world of high-level languages? Cooool :)

With SIMD and up to 4 instructions in a static superscalar architecture? 
That needs to be a pretty good compiler. At least the math is pretty 
straight-forward.

>> The specs of the AD and DA are good, but not out of this world (D/R
>>  110dB, Crosstalk < -60dB over full bandwidth).
>
> Yeah - I'm usually happy with 110dB D+N as a listener (as is the case
>  here, albeit via some far less impressive D-A converters), although

As a listener, I'm perfectly happy with a CD at 96dB. I don't think my 
speakers and amp are good enough to really reveal the extra detail. I 
assume that the noise floor in our rather quiet home is already reducing 
the effective range below some 80dB or so...

> I'm still exploring how these very good figures might still
> perceptibly influence a musician who practices long hours outside the
> highly compressed perceptive world of rock music.

Personally, I think they simply don't matter. Yes, if you're recording 
with a lot of headroom or extremely dynamic material, it's great to have 
such fine resolution. For the rest, I think it's just peace of mind that 
the conversion is not the weak link. But that's my gut feeling, based on 
no scientific evidence whatsoever...

>> I think somebody just sat down and implemented the algorithms
>> without too much focus on cost-cutting, but probably a bit more
>> focus on making them sound good.
>
> Certainly I'm hearing some clarity, and I'd guess it comes from
> extended precision of some kind, within the unit. Good to hear. I
> thought the crunch on some of the distortions was as good as I've
> heard in the digital domain; extremely usable within the context.

Quite a few algorithms sound good if you either oversample (then you can 
get away with simpler algorithms) or use high-order interpolation (at 
lower sample-rates). The result is less aliasing and other artefacts - 
probably exactly the clarity you're experiencing.

Bye
Martin