| ??? 03/08/04 16:32 Read: times |
#66262 - RE: Confusing... Responding to: ???'s previous message |
Hallo Russell, your reply make people think, that MCS1210 is showing extremely low noise performance. But the information you gave is rather confusing and questionable! Maybe I didn't fully understand one of the requirements, but the MSC1210 is a very good part. I will go through you questions and see if I can resolve the confusion. 1. You defined ENOB (effective number of bits) as: ENOB = 24 - log2(STDlsb) What does this mean? I learnt the following definition: ENOB = (SNR - 1.76)/6.02 where 'SNR' is the actual Signal-to-Noise-Ratio, measured in deziBel (dB) and where 'ENOB' is measured in bits. How do the two definitions fit together? The formula you mention is used when you use a full-scale sine wave as your input. The reason I included the definition for ENOB is because of this very confusion. This is the definition we used for ENOB for a DC signal. Even though IEEE has defined an ENOB like this, it is confusing and we will probably be moving to using RMS voltages in our data sheets instead of ENOB. The ADS1256 (our last Delta-Sigma part) includes all the information so that you can use what you are comfortable with. 2. You wrote: If we look at the same 4 Hz data rate (15 Hz data rate) and a gain of 32 (+/- 78 mV) the decimation ratio will be 1023. That will give an ENOB of about 20 which is 149 nV of RMS noise. The importance piece of information here is the decimation ratio. The noise performance will be the same if I am able to increase the sample rate and keep the same decimation. The decimation ratio is what we need to get the information from the charts in the data sheet. You took the ENOB data from the figure of datasheet, which is valid for fdata = 10Hz. But for f-3dB = 4Hz you must take into account the noise performance referring to fdata = 15.3Hz according to the formula (sinc3-filter): fdata = f-3dB / 0.262 That is how I computed that I need a 15 Hz data rate. If you take the data for fdata = 10Hz you get f-3dB = 2.62Hz and for this lowered bandwidth noise performance is some better, of course. A first order approximation yields, that noise level is proportional to square root of bandwidth. This can be verified by reading datasheet of AD7730 and ADuC824. So, very probably, noise performance will be worse than what you have stated by the factor: SQRT(4Hz / 2.62Hz) = 1.24 If you get 149nVeff in your calculation, real data will be 149nVeff x 1.24 = 185nVeff! The noise performance is only related to the decimation ratio as long as the sample rate doesn't exceed 250 kHz. With the highest PGA setting or 64 or 128 the sample rate is 16 times the modulation clock. So at those higher PGA setting you are limited to 15,625 Hz. Which is what the data sheet uses for the sample rate. But since this application was only using a PGA of 32, that gave some flexibility to increase the sample rate. 3. ENOB = 20 you get, if you take the data for 'Buffer OFF' condition. But having the buffer off has some disadvantages. Consequently, when comparing MCS1210 with others, having the buffer on, only 'Buffer ON' data should be taken. And then MCS1210 yields ENOB = 19 ... I wasn't sure what the circuit required. Yes, if you require the buffer to be on then the ENOB will be 19 bits instead of 20. Then you will have a STD of 298 nV. 4. Datasheet of AD7730 states in 'Table I' that for f-3dB = 4Hz and input voltage range of +-80mV output noise is 155nVeff. In 'Table II' they state, that this is equivalent to an resolution of 17.5 bits. In your reply you state: That will give an ENOB of about 20 which is 149 nV of RMS noise. How the hell, can your 149nVeff, which is nearly the same as 155nVeff, give a resolution of 20 bits, when at the same time people of 'Analog Devices' say, that resolution is 17.5 bits??? This is totally confusing!!! The ENOB in the AD7730 data sheet is peak-to-peak resolution. The difference between RMS and peak-to-peak is a factor of 6.6 which is an ENOB difference of 2.7. You can measure RMS, the peak-to-peak is a statistical assumption of gaussian distribution and includes 99.9% of the RMS noise. I guess, that when removing all these smart tricks, MCS1210 isn't any longer that superior! Kai There is no magic here. I am sorry to confuse the issue by using the non-buffered numbers. I guess the best I can do is to increase the sample rate further and increase your decimation ratio to end up with a noise level of 255nV with the buffer on. Best Regards, Russell Anderson |
| Topic | Author | Date |
| ADuC824/834 | 01/01/70 00:00 | |
| RE: ADuC824/834 | 01/01/70 00:00 | |
| RE: ADuC824/834 | 01/01/70 00:00 | |
| RE: ADuC824/834 | 01/01/70 00:00 | |
| RE: ADuC824/834 | 01/01/70 00:00 | |
| RE: ADuC824/834 | 01/01/70 00:00 | |
| RE: ADuC824/834 | 01/01/70 00:00 | |
| RE: ADuC824/834 | 01/01/70 00:00 | |
| RE: ADuC824/834 | 01/01/70 00:00 | |
| Confusing... | 01/01/70 00:00 | |
| RE: Confusing... | 01/01/70 00:00 | |
| No longer confusing... | 01/01/70 00:00 | |
RE: ADuC824/834 | 01/01/70 00:00 |