ADS1271 Internal FIR Low-pass, Sample Rate and Nyquist?

Hi Mark,

Welcome to the E2E Forums and thanks for your question!

The FIR filter in the ADS1271 works exactly as you described. The filter gives you a flat response from DC up until 0.453*f_DATA, where it rolls off and hits -3dB at 0.49*f_DATA. The transition band exists from 0.453*f_DATA to 0.547*f_DATA, where the stopband begins. In High-Resolution Mode, you should expect at least 100dB of stopband attenuation until you reach 127.453*f_DATA.

A driver amplifier and anti-aliasing filter is generally used in front of an unbuffered delta-sigma ADC, like the ADS1271. The purpose of the driver is to condition the input signal and provide the charge necessary to the internal sampling capacitor of the ADC. The anti-aliasing filter at the input is intended to cut down out-of-band signals and noise, especially those which occur at multiples of the modulator frequency. Delta-sigma converters have the benefit of shaping most of the quantization noise to these higher frequencies, so the requirements for the anti-aliasing filter are more relaxed.  

Have you given thought to using an anti-aliasing filter at the ADC input? Table 8 shows the expected image rejection that can be achieved with first, second, and third-order anti-aliasing filters.

Best Regards,

Hi Ryan:

Thank you for your response.  However, that still didn't answer my question regarding satisfying Nyquist for the output sampled data at the fDATA output data "sample" rate.  Indeed, as you mention, I do have a 4th order Butterworth low noise analog filter "in front" of the ADS1271, and I condition that Butterworth filter's output using a precision single ended to differential driver to remove common mode voltage while providing the correct differential voltage levels required to maximize the dynamic range of the ADS1271, centered at a Vref of 2.50VDC.  This differential driver I am using successfully drives an input capacitor network consisting of a 10,000pF across the two analog inputs of the ADS1271, along with a 1000pF to ground on each of these two inputs as suggested in the data sheet.  So, input analog signal conditioning is not the issue I am trying to resolve with my request.

My issue is with the frequency content of the resulting sampled data of the fDATA sample rate and to know what the expected effects are of my depending on the internal sharp FIR filter within the ADS1271, to band limit the output data such that the Nyquist criteria is satisfied.  For example, assuming I have band limited my analog input signal with an analog anti-aliasing 4th order Butterworth filter having a -3dB cutoff at  4.1KHz and I am sampling at an fCLK of 65.536KHz in High-Resolution mode, my output data rate would therefore be an fDATA of 1.280KHz.  One would expect that the internal FIR filter within the ADS1271 has further limited my bandwidth to correspond with the fDATA output rate.  However, since the internal FIR filter shows significant signal remaining after fDATA/2 (a Nyquist frequency of 640Hz in this example), should I expect to have some aliasing in the resulting data since the analog input signal I supply was not externally limited to less than a bandwidth of 640Hz?  In other words, if I were to have to create a very sharp analog anti-aliasing filter to go in front of this A/D, then the part's internal FIR filter would therefore not providing me with any functional benefit?  However, if the data being output at the fDATA rate were indeed properly FIR filtered, then the resulting samples would not contain any aliasing beyond Nyquist.

The data sheet is appearing to show the FIR Filter's skirt to extend beyond Nyquist before landing at the "stop band" attenuation level.  With all that said, are my assumptions correct?  Can I depend on using the interal FIR filter as the "sharp filter" in my anti-aliasing scheme?

Thanks

Hi Mark,

I think I see the confusion....

You are assuming the data rate and the sampling rate are the same. However for a delta-sigma ADC, the sampling rate is much faster than the output data rate because the input is being oversampled.

Refer to figure 61 (for HR mode, refer to figure 57 for HS or LP modes), the sampling rate is 128 * fDATA (HR mode only). Therefore, Nyquist frequency is 64 * fDATA. The ADC's digital filter has excellent attenuation up to about 60 * fDATA. Therefore, you need an analog input filter to prevent anti-aliasing of frequencies around 64 * fDATA.

Usually, a first or second order LPF will have sufficient attenuation out at the modulator frequency (modulator frequency = sampling frequency = fCLK / 4), where signals would alias back near DC.

Lets look at the data rate you mentioned:

MARK JOHNSON5 said:

I am sampling at an fCLK of 65.536KHz in High-Resolution mode, my output data rate would therefore be an fDATA of 1.280KHz

The ADS1271, requires a minimum fCLK of 100 kHz (refer to the electrical characteristics requirements on page 4 of the data sheet).

Using fCLK = 100 kHz, then fMOD = fCLK/4 = 25 kHz (this is the actual sampling rate; refer to table 2 as this ratio is different for the various modes). The data rate is then fCLK/256 = 97.65 Hz (refer to tables 3 & 4 for the different modes).

To achieve a 1.28 kHz data rate, you'll need to apply an fCLK of  327.68 kHz (or perhaps 655.36 kHz in frame-sync mode). The sampling rate is then 81.92 kHz (or 163.84 kHz). Your 4th order filter will provide plenty of attenuation near the sampling frequency!

Best Regards,
Chris

Hi Chris:

Thanks for your response to my question.  I do understand the fCLK requirements, and in my application, I am indeed feeding in (as you suggest in your response) an fCLK frequency of exactly 327.68Khz to ultimately achieve the desired 1.280KHz output fDATA rate in the High Resolution mode.  327.68KHz is obviously greater than the minimum 100KHz specified for fCLK for the ADS1271, so I should be OK in that respect. Indeed my analog anti-aliasing low pass filter choice having a -3db cut-off frequency is way, way, way more than sufficient to handle the "digitizing" A/D activity of the ADS1271.  In other words, there is practically no aliasing whatsoever from the A/D input sampling side perspective.  But this is not my question.

It appears that I need to be more specific in my semantics since my question regarding satisfying Nyquist is not an issue with the actual analog sampling by the ADS1271, but instead is with relationship to the output data rate of the ADS1271, at the chip's fDATA frequency which would is1.280KHz in my example.  Although fDATA is not technically an input "sampling frequency" for the ADS1271 chip itself, it can indeed be considered to be an effective output "sampling clock" for the downstream circuitry in that the digitized analog data coming from the part is being retrieved at this lower rate (as if the data were otherwise being "clocked out" of the ADS1271 at this effective data rate).  In other words, I cannot retrieve the data samples from the ADS1271 any faster than fDATA so, fDATA becomes my new effective output "sample clock rate" for the data.

If the FIR inside the ADS1271 does pass frequency content which is more than the output fDATA rate's "Nyquist" frequency (Nyquist = 640Hz for my example) in the output data, then it would seem apparent that mathematically, downstream digital processing could then experience classical aliasing.  This situation is what the diagrams and data for the ADS1271 seem to be suggesting since the descending FIR filter "skirt" shown in Figure 58 (Page 23) slopes downward but exposes significant signal level beyond the output data rate's Nyquist frequency. This is what Figure 58 of the data sheet  appears to be suggesting.

Now, does my question regarding the utility of depending on the use of the ADS1271's internal sharp FIR filter for the anti-aliasing downstream data and its "effective clock out" rate make sense with regards to meeting the output data rate Nyquist criteria?  It would seem to me that TI would have perhaps considered setting the internal FIR filter response cutoff so that the stop band would be at full attenuation (< -100dB) at and beyond the fDATA output rate's effective "Nyquist" frequency.  Is my assumption correct regarding this chip's apparent deficiency here with respect to relying on the internal FIR to do it's job for anti-aliasing for downstream data?  I hope I am wrong and that I can use this chip in this manner.  I thank you in advance for your assistance here.

Thanks!