[sdiy] Harmonic bandwidth

[Richard Wentk]

On 9 Jan 2008, at 19:22, Nicholas Gregorich wrote:

> When I did my test in MATLAB the data was generated and summed in  
> 64-bit floating point and re-quantized to 16 bits for stuffing into  
> the soundcard.

Was this with dither or truncation?

> I don't think bit depth has much to do with the differences heard  
> between the two samples [I also don't think its a sample rate  
> "problem"]. I studied the effects of quantizing a sinusoid with no  
> headroom [best case scenario] and I was shocked at how low the bit  
> depth could be without ANY audible effect. I was studying  
> quantizing as an audio effect and to get the desired results the  
> bit depth had to be below 8 bits, or maybe even 6 bits. I think the  
> reasoning is quantization error is output as white noise which is  
> not objectionable to the human ear. It wasn't until the output  
> waveform looked nearly square that the effect became very pronounced.

Q-noise will be semi-correlated with any sine signal. Also, people  
tend not to find sine distortion very audible - up to 10% THD is  
unobjectionable, depending on the kind of THD. Most (affordable)  
speakers produce a couple of percent of THD, so for a very accurate  
test you need top of the line monitors.

Music tends to be more revealing. Features like dither and jitter  
make a *huge* difference to sound quality, even though they measure  
around +/- 1LSB in a digital system.

As an aside, I've been ripping my CD collection to disk using a tool  
called EAC. It's amazing how much cleaner and more detailed music  
sounds played off a hard disk with the most obvious jitter and error  
correction effects eliminated than when played off a plastic disk in  
a good CD player.

Richard