No. The resolution of the ADC is determined by the hardware. LDmicro always reports the full hardware resolution, which for all supported devices is 10 bits.

In fact, the bottom few bits are just noise. It would be necessary to average multiple samples in order to get even 10 good bits. There's also integral non-linearity (INL), which introduces a few counts of static error no matter how much you average.

It depends on many different factors--for example, the clock speed of the micro, the clock speed chosen for the conversion, the amount of noise on Vdd (or some other reference).

If Vdd isn't horrible, then it's likely that the top seven bits are okay, even after just a single conversion--so probably not worse than 8 counts of noise.

Averaging n times reduces the noise by a factor of sqrt(n), assuming that the noise is uncorrelated. So averaging 32 or 64 samples almost always gets you something like the full 10 bits. A simple average works, though a first-order IIR filter is also easy, and is nicer when considered in the frequency domain.

There's also some static errors (i.e., biases to the mean; so they don't go away if you average). But those are small; for example INL on the 'F887 is +/- 1 count max. By that point the tolerance on your voltage reference is probably more important. Though depending what you're actually trying to measure, it's typical to make a measurement ratiometric, so that it ends up independent of Vref.

Thank you
Exist another microcontroler with 16 Biti or 32 biti can be progaming to LDmicro.

The PIC18s are all 8 bit processors, same as the PIC16 or AVR. And there's no particular relationship between the ADC resolution and the processor word size. Though all the PIC18 parts that I've seen do have a 10 bit A/D as well.

It's possible to buy standalone A/D converters, with a resolution of 16 bits or more. Though higher-resolution A/D converters tend to be used for better dynamic range or noise, not absolute accuracy. The INL on those converters is disproportionately higher, and the rest of your system is probably worse than even a part in 1000--an 0.1% voltage reference or resistor is already somewhat exotic.

The number of bits that an A/D reports is also not necessarily related to the number of useful bits. There's a standard specification based on the signal to noise ratio as measured in the frequency domain, called ENOB (effective number of bits). This is always less than the resolution; and an A/D that reports more bits may actually have a lower ENOB than one that reports fewer.

That might be simple or hard, depending on the bandwidth required. As the bandwidth goes to zero, you can average ever more and drive the (excess, beyond quantization) noise to zero.

Look at

http://www.analog.com/en/analo...converters/products/index.html .

The AD7194, for example, is a 24 bit A/D. Table 8 of its datasheet gives the ENOB versus bandwidth (which is programmable, by changing parameters of an on-chip filter).

Just to take you back to averaging. The IIR filter is digital, but aren't we talking about an analog signal, before the AD conversion happens?

Ok. Thank you very much.

The TMS... parts are Texas Instruments DSPs. Microchip has the dsPICs. DSPs with integrated A/D converters do tend to have slightly better specs than ordinary micros with A/Ds, though not always.

But there's significant compromises involved in building a low-noise A/D on the same piece of silicon as a big fast digital thing. The best A/D converters tend to be separate ICs, and still require some care to avoid coupled noise from other digital circuits.

The only way to know is to read the datasheets, or make the measurement yourself.