[<<][staapl][>>][..]

Sun Dec 2 18:50:57 CET 2007

## signal rates revisited

4 independent frequencies:
* PWM TX rate: determines high-frequency + aliasing noise
* carrier: only important for path (i.e. speaker reso)
* baud rate: bandwidth -> noise sensitivity
* RX rate: selectivity (related to FIR length)
(EDIT: carrier and baud rate are not independent wrt data filter
qualoty. see below)
important for the receiver are :
- baud rate, which limits the maximal integration time (dependent on
symbol length).
- RX rate: enables longer filter lengths, which gives more
selectivity and noise immunity.
it doesn't make sense, for constant baud rate, to up the RX frequency,
but keep the FIR length constant, so:
FIR = k . (RX / BAUD)
RX = Fosc / OPS
where k is the number of symbols the FIR spreads over, probably 1 or
2. and OPS is the amortized number of operations per sample
(processing and aquisition). the filter is 32 in the current
implementation.
this gives about 300kHz at 10MIPS. looks like we have some
headroom..
anything more than 8kHz is probably not going to make much sense.
( I was thinking about noise and dithering, and that at this high
frequency because of absence of noise there will be no 'extra'
sensitivity due to the dithering at levels close to the quantisation
step, but there probably will be extra due to pwm effects. So it
looks like all small bits help.. )
EDIT: another variable i forgot to mention is symbol rate
vs. carrier. using a mixer, it is desirable to have large separation
between the two so a simple data filter can be used.

[Reply][About]

[<<][staapl][>>][..]