[sdiy] Digital echo chip

[René Schmitz]

At 23:52 05.01.02 +0100, jh. wrote:
>> The technique works very well with a very high clock frequency.
>
>
>Think of it as similar to Sigma-Delta Converters. But unlike these, not
>the actual signal is stored, but the "change of signal": The ADC
>estimates the direction (larger / smaller) for the next sample and then
>proceeds depending on whether this estimation was right or wrong.
>The trick about ADM (adaptive DM) is that the step size for this
>estimation is changed depending on the recent bits. In other words, there's
>a compander built into the ADC. Sample rate and algorithm that is used for
>the compander function can make a huge difference on the quality of ADM.
>Take a look at Deltalab's patents to see how they have refined the
>algorithm (all implemented in hardware).
>This companding function, and the fact that the d/dt of the signal is coded,
>leads to the most severe drawback of the method: It's severely
>slew limiting ! These delays have an impressive SNR and frequency response,
>but *only* for low frequencies and high frequencies with small amplitude.
>Not unlike a slow opamp (;->). Fast and large transients is something
>they don't like. You get excellent SNR figures, but when you excced the
>rise time limit, you get nasty TIM distortion.

Thats true. I built a copy of the codecs as in that Delta lab patents, and it 
sounds really good when I use data rates of 500+ kbits/s. 
The higher the rate, the less chance of reaching that slewrate limit. 
What amazed me is that these bitrates are only two to four times higher than 
that commonly used for mp3, so that shows how efficient this encoding 
technique really is. (Its also often used for audio tracks of digital 
videos.)

Cheers,
 René


-- 
uzs159 at uni-bonn.de
http://www.uni-bonn.de/~uzs159